Meaning, Mind, and Knowledge

In this collection of essays, most of which are of recent vintage, and seven of which
appear here for the first time, Christopher S. Hill addresses a large assortment of
philosophical issues. Part I presents a deflationary theory of truth, argues that
semantic properties like reference and correspondence with fact can also be charac-
terized in deflationary terms, and offers an account of the value of these ‘thin’
properties, tracing it to their ability to track more substantial properties that are
informational or epistemic in character. Part II defends the view that conscious
experiences are type-identical with brain states. It addresses a large array of objec-
tions to this identity thesis, including objections based on the alleged multiple
realizability of experiences, and objections based on Cartesian intuitions about the
modal separability of mind and matter. In the end, however, it maintains that
theories of experience based on type-identity should give way to representationalist
accounts. Part III presents a representationalist solution to the mind-body problem.
It argues that all awareness, including awareness of qualia, is governed by a Kantian
appearance/reality distinction—a distinction between the ways objects and proper-
ties are represented as being, and the ways they are in themselves. It also presents
theories of pain and visual qualia that kick them out of the mind and assign them to
locations in body and the external world. Part IV defends reliabilist theories of
epistemic justification, deploys such theories in answering Cartesian skepticism,
responds critically to Hawthorne’s lottery problem and related proposals about the
role of knowledge in conversation and practical reasoning, presents a new account of
the sources of modeal knowledge, and proposes an account of logical and mathem-
atical beliefs that represents them as immunune to empirical revision.
Meaning, Mind,
and Knowledge

Christopher S. Hill

1
OUP CORRECTED PROOF – FINAL, 5/2/2014, SPi

3
Great Clarendon Street, Oxford, OX2 6DP,
United Kingdom
Oxford University Press is a department of the University of Oxford.
It furthers the University’s objective of excellence in research, scholarship,
and education by publishing worldwide. Oxford is a registered trade mark of
Oxford University Press in the UK and in certain other countries
# in this volume Christopher S. Hill 2014
The moral rights of the author have been asserted
First Edition published in 2014
Impression: 1
All rights reserved. No part of this publication may be reproduced, stored in
a retrieval system, or transmitted, in any form or by any means, without the
prior permission in writing of Oxford University Press, or as expressly permitted
by law, by licence, or under terms agreed with the appropriate reprographics
rights organization. Enquiries concerning reproduction outside the scope of the
above should be sent to the Rights Department, Oxford University Press, at the
address above
You must not circulate this work in any other form
and you must impose this same condition on any acquirer
Published in the United States of America by Oxford University Press
198 Madison Avenue, New York, NY 10016, United States of America
British Library Cataloguing in Publication Data
Data available
Library of Congress Control Number: 2013943744
ISBN 978–0–19–966582–2
As printed and bound by
CPI Group (UK) Ltd, Croydon, CR0 4YY
Links to third party websites are provided by Oxford in good faith and
for information only. Oxford disclaims any responsibility for the materials
contained in any third party website referenced in this work.
For Jay, Kate, and Elena
Funny, Fun, and Formidable
Table of Contents

Acknowledgments viii

1. Introduction 1

Part I. Meaning
2. “gavagai” (1972) 23
Postscript to “gavagai” (2013) 30
3. Rudiments of a Theory of Reference (1987) 32
Postscript to “Rudiments” (2013) 49
4. A Substitutional Theory of Truth, Reference, and Semantic
Correspondence (2006) 51
5. How Concepts Hook onto the World (2013) 66

Part II. A Type-materialist Theory of Experience

6. In Defense of Type-materialism (1984) 97
7. Imaginability, Conceivability, Possibility, and the Mind-Body
Problem (1997) 117
8. The Identity Theory (2013) 136

Part III. A Representationalist Theory of Experience

9. OW! The Paradox of Pain (2005) 155
Postscript to “OW!” (2013) 176
10. Locating Qualia: Do They Reside in the Brain or in the Body and
the World? (2012) 177
11. Visual Awareness and Visual Qualia (2013) 197
12. The Content of Visual Experience (2013) 218

Part IV. Knowledge

13. Process Reliabilism and Cartesian Skepticism (1996) 239
14. Hawthorne’s Lottery Puzzle and the Nature of Belief,
(written with Joshua Schechter, 2007) 253
15. Conceivability and Possibility (2013) 273
16. Concepts, Teleology, and Rational Revision (2013) 297

Index 323
Acknowledgments

My most extensive debts are to Anil Gupta. I am deeply grateful for his encour-
agement, inspiration, guidance, and criticism. I have also been helped substantially
by David Bennett, Ned Block, Justin Broackes, Anthony Brueckner, Alex Byrne,
David Chalmers, David Christensen, James Dreier, Katherine Dunlop, Ivan Fox,
Simone Gozzano, David Hefer, Katrina Hill, Jaegwon Kim, Alexandra King, Joseph
Levine, Jack Lyons, Vann McGee, Brian McLaughlin, Peter Momtchiloff, Adam
Pautz, Joshua Schechter, Susanna Siegel, Michael Tarr, Fritz Warfield, Lee Warren,
William Warren, Mark Wilson, and Steven Yamamoto.
Most of the chapters of this book originally appeared in other venues. Here is a
list:
1. “ ‘gavagai,’ ” Analysis XXXII (1972), 68–75.
2. “In Defense of Type Materialism,” Synthese LIX (1984): 295–320.
3. “Rudiments of a Theory of Reference,” The Notre Dame Journal of Formal
Logic 28 (1987): 200–19.
4. “Process Reliabilism and Cartesian Scepticism,” Philosophy and Phenomeno-
logical Research, LVI (1996): 567–81.
5. “Imaginability, Conceivability, Possibility, and the Mind-Body Problem,”
Philosophical Studies 87 (1997): 61–85.
6. “Ow! The Paradox of Pain,” Murat Aydede (ed.), Pain: New Essays on its
Nature and the Methodology of it Study. Cambridge, MA: MIT Press, 2005,
75–98.
7. “Précis of Thought and World,” Philosophy and Phenomenological Research
LXXII (2006): 174–81.
8. “Replies to Marian David, Anil Gupta, and Keith Simmons,” Philosophy and
Phenomenological Research LXXII (2006): 2005–222.
9. “Hawthorne on the Lottery Problem and the Nature of Belief ” (co-authored
with Joshua Schechter), Philosophical Issues, 17 (2007): 103–22.
10. “The Identity Theory.” In Timothy Bayne (ed.), The Oxford Encyclopedia of
Consciousness. Oxford: Oxford University Press, 2009, 359–63.
11. “Introduction” (co-authored with Simone Gozzano) in Simone Gozzano and
Christopher S. Hill (eds.), New Perspectives on Type Identity. Cambridge:
Cambridge University Press, 2012, 1–15.
12. “Locating Qualia: Do They Reside in the Brain or in the Body and the World?”
in Gozzano and Hill, 127–47.
 ACKNOWLEDGMENTS ix

13. “Conceivability and Possibility,” in Herman Cappelen, Tamar Gendler, and

John Hawthorne (eds.), Oxford Handbook of Philosophical Methodology.
Oxford: Oxford University Press, forthcoming.
14. “Concepts, Teleology, and Rational Revision,” in Albert Casullo and Joshua
Thurow (eds.), The A Priori in Philosophy. Oxford University Press, forthcoming.
I thank Cambridge University Press for permission to reprint “Locating Qualia” and
portions of the Introduction; Duke University Press for permission to reprint
“Rudiments of a Theory of Reference”; MIT Press for permission to reprint
“OW!”; Oxford University Press for permission to reprint “ ‘gavagai’,” portions of
“The Identity Theory,” “Conceivability and Possibility,” and “Concepts, Teleology,
and Rational Revision”; Springer Publishing Company for permission to reprint “In
Defense of Type Materialism” and “Imaginability, Conceivability, Possibility, and the
Mind-Body Problem”; and Wiley for permission to reprint “Précis of Thought and
World,” “Replies to Marian David, Anil Gupta, and Keith Simmons,” and
“Hawthorne on the Lottery Problem and the Nature of Belief.”
 1
Introduction

This book draws together essays that have been written over a period of 40 years,
though most of them appeared in the comparatively recent past, and seven are
published here for the first time. In this Introduction I will provide brief descriptions
of the chapters, discuss their interconnections, and say something about their
relationships to my present views. Surprisingly, perhaps, I still have a fairly high
degree of commitment to the claims and arguments in all of the chapters, including
the claims and arguments in the papers written in the distant past. This is true even of
the papers in Part II, which envision a reduction of experiences to states of the brain.
I no longer accept this theory of experience, but I still feel that those papers make a
strong case for an important view.

I. Meaning
The first part of the book is concerned with meaning. More specifically, it addresses
questions about truth-related semantic properties—properties of sentences and
thoughts like truth and correspondence, and also properties of words and concepts
like reference and denotation. I say that reference and denotation are truth-related
because the question of whether a sentence or a thought is true depends on the
reference and denotation of its constituent words or concepts. For example, whether
the sentence “Obama is a politician” is true depends on whether the person to whom
the singular term “Obama” refers is denoted by the general term “politician.”
Although two of the essays in Part I are concerned with the meanings of words and
sentences, I will focus here, as I do in the two other chapters in Part I, on the
meanings of concepts and thoughts. There is a close correspondence between these
two sets of semantic properties. Thus, a sentence is true if it is used to express a true
thought, and a word refers to an object if it is used to express a concept that refers to
that object. In view of these relationships, a theory of meaning for concepts and
thoughts can provide the foundation for a theory of meaning for words and
sentences.
It will be useful to have three terminological conventions on board. First, I will
always use “thought” as a term for mental representations that are composed of
concepts, have logical structures, and can be assessed for truth and falsity. Second,
I will sometimes use “proposition” as an equivalent of “thought,” because that will
2 INTRODUCTION

allow me to deploy the adjective “propositional.” Unlike “proposition,” “thought” has

no corresponding adjective. Third, in addition to using “reference” as a term for the
semantic relation that singular concepts bear to objects, I will sometimes use it as a
term for all of the semantic relations that concepts bear to portions of extra-
conceptual reality. Thus, where no confusion will result, I will use it as a blanket
term for the semantic relation that singular concepts bear to particular objects, the
relation that general terms bear to the items they denote, the relation that general
concepts bear to the properties that they express, and so on. These terminological
conventions are observed in the essays in Part I as well as in this Introduction.
To be more specific about the contents of Part I, it is principally concerned with
these five questions:
1. What is it for a thought to be true?
2. What is it for a thought to correspond to a state of affairs?
3. What is it for a concept to refer to an object?
4. What exactly are the relationships between truth-related semantic relations and
the various empirical relations with which semantic relations seem to be
aligned, such as the relation that a concept bears to a property P if it is used
to encode information about items that possess P?
5. Is it true, as it seems to be, that reference links concepts determinately to extra-
conceptual items, or is it rather the case, as Quine and others have maintained,
that there is often no fact of the matter as to whether a concept refers to an
item? (See Quine 1960.)
Part I can be described by saying that it proposes deflationary answers to the first
three questions, explores several models of the relationships between truth-related
semantic relations and other, more clearly empirical relations between mind and
world, and offers various grounds for thinking that reference is determinate.
My proposals are deflationary in the sense that they imply that semantic concepts
can be exhaustively explained without incurring any substantive empirical or meta-
physical commitments. Two different forms of deflationism are proposed in the
essays—one in the second chapter and another in the third and fourth chapters. The
first proposal preceded the second by a number of years. I still think that it has merit.
Moreover, it can be seen on reflection to provide important motivation for various
aspects of the second proposal. At present, however, I think the latter faces the issues
more directly. Among other things, it maintains that our truth-conditional semantic
concepts are very close to logical concepts. More specifically, it maintains that they
are definable in terms of a vocabulary that consists principally of substitutional
quantifiers—logical operators that resemble the familiar quantifiers “some” and
“all.” If the proposal is correct, semantic concepts can be grasped a priori.
My proposals are presented in some detail in the forementioned chapters. For
present purposes it will suffice to describe a comparatively simple but closely related
theory that is due to Paul Horwich. (See Horwich 1998. For arguments for the
 INTRODUCTION 3

superiority of my approach, see Hill 2002 and Hill 2011.) According to Horwich, the
concept of propositional truth is defined by the class of all propositions that have the
following form:
(T) The thought that p is true just in case p.

Here, for example, is a component of Horwich’s definition:

The thought that wolves are territorial is true just in case wolves are territorial.

Horwich claims that propositions like this are constitutive of the concept of propos-
itional truth, and this claim seems correct. No one who doubted such propositions
could be said to possess the concept. It follows that such propositions are epistem-
ically available independently of one’s empirical or metaphysical commitments. One
is entitled to accept them simply in virtue of possessing the concept. Hence, given
that Horwich’s theory maintains that truth is defined by these propositions, the
theory is fully deflationary in character. Its definition of propositional truth is
empirically and metaphysically neutral.
A companion idea is that the concept of singular reference is defined by the class of
propositions that look like this:
(R) For any object x, the concept of a refers to x just in case a is identical with x.

Thus, for example, the following proposition is a component of the definition of

reference:
(O) For any object x, the concept of Barack Obama refers to x just in case x is identical with
Barack Obama.

In combination with the fact that Barack Obama is identical with Barack Obama, and
the fact that Barack Obama is identical with the person who is the President of the
United States in 2013, this proposition entails that the concept of Barack Obama
refers to Barack Obama, and also that the concept refers to the person who is
President of the United States in 2013. These consequences are both true. Moreover,
reflection shows that (O) doesn’t entail any false claims about reference. Where b is a
particular that isn’t identical with Obama, it would be false to say that the concept of
Obama refers to b. There is no risk of deriving any such claim from (O); for (O)
entails that for any object x whatsoever, if the concept of Obama refers to x, then x
will be identical with Obama. In general, the consequences of the present account of
reference match our intuitions about the reference of particular concepts.
Note that instances of (R) are no less platitudinous than instances of (T). They can
be seen to be true independently of empirical and metaphysical commitments. It
follows that the present account of the concept of reference is deflationary. It implies
that the concept is insubstantial. Note also that the account implies that the concept
can be grasped a priori. To grasp the concept, it suffices to appreciate that all of
the propositions that have form (R) are to be accepted. One can appreciate this
4 INTRODUCTION

independently of one’s experience of the world. (A qualification: what is genuinely a

priori is not (O) itself, but rather the conditional proposition that if Barack Obama
exists, then for any object x, the concept of Barack Obama refers to x just in case x is
identical with Barack Obama.)
Deflationary views like Horwich’s and mine face a number of rather difficult
challenges. I address a range of these challenges in Part I, but I am principally
concerned in those chapters to defend the idea that truth-related semantic concepts
are insubstantial and a priori. In the view of many philosophers, the concept of
reference cannot be insubstantial or a priori because its main job is to report facts
involving information. Thus, for example, one might register the fact that a concept is
used to encode information about rabbits by saying that the concept refers to rabbits.
If it is the main job of the concept of reference to register manifestly empirical facts of
this sort, how can the concept be insubstantial or a priori? Another reason for
doubting that it is insubstantial or a priori is that it plays an essential role in empirical
laws. Here is an example:
If C is a concept that refers to a perceptually accessible object, o, and x is an agent who
possesses C, then x is disposed to form beliefs involving C as a result of acquiring perceptual
information about o.

A third reason, closely related to this second one, is that the concept of reference
figures prominently in empirical explanations. If you are able to encode information
about Barack Obama by forming beliefs about him, part of the explanation will be
that you possess a concept that refers to Obama. These and other considerations
seem to show that semantic concepts are robust concepts, with a full share of
empirical meaning. I have come to think that it is this view that is primarily
responsible for the very strong resistance to deflationism that I have encountered
in many friends and acquaintances. It is so protean that it cannot be easily refuted,
but I have tried to lay it to rest in Part I.
I turn now to the questions about the determinacy of meaning. After asking us to
imagine that a certain foreign term “gavagai” is used to signal the presence of rabbits,
Quine went on to argue that there is no fact of the matter as to whether “gavagai”
refers to rabbits, to undetached parts of rabbits, or to temporal stages of the life
histories of rabbits. The key idea was that, necessarily, a whole rabbit is present just in
case an undetached rabbit part is present, and an undetached part of a rabbit is
present just in case a stage in the history of the rabbit is present. It follows that when
“gavagai” is used to signal the presence of a rabbit, it also necessarily signals the
presence of an undetached rabbit part and a temporal stage of a rabbit. (See Quine
1960.) Various other authors followed Quine in expressing skepticism about the
determinacy of reference. It can easily be shown, however, on the basis of principles
like (R), that the reference of concepts is fully determinate. Thus, for example, as we
have seen, it follows immediately from (R) that the concept of Obama refers to
Obama and nothing else. Further, if we agree that it is determinate whether an agent
 INTRODUCTION 5

possesses a given concept, and that it is also determinate whether the agent uses a
given word to express that concept, we will be able to infer that, contrary to Quine,
linguistic reference is determinate. It remains to be seen, of course, whether questions
about the possession of concepts really do have determinate answers; but even the
present very quick line of thought suffices to show that the classic arguments for
indeterminacy of reference were misguided. At the deepest level, inquiries into
determinacy should focus on the possession conditions for concepts, not on the
reference of words.
In addition to the question of whether semantic relations link words and concepts
determinately to unique portions of reality, there is also a question of whether there
are empirical relations that have this feature. Among other things, the fourth essay in
Part I addresses this further question, arguing that the answer should be affirmative,
at least in the case of concepts that refer to perceptible items. Concepts of this sort are
linked uniquely to their referents by informational relations, which are understood to
be broadly empirical. The first essay in Part I provides some background for this
discussion, and also introduces an idea on which the discussion builds. (It is a
consequence of these two essays that there can be determinate possession conditions
for perceptual concepts.)
Meaning has more than one dimension. In addition to its truth-related dimension,
there is also a dimension that has to do with the cognitive roles of concepts—that is,
with the roles that concepts play in inference and propositional attitudes. I say very
little about this other dimension of meaning in Part I, and make no attempt to discuss
it systematically anywhere in the volume. I do, however, consider one aspect of it
in “Concepts, Teleology, and Rational Revision,” the last chapter in Part IV. The
cognitive role of a concept depends in part on rules or procedures for updating beliefs
involving the concept in the light of experience. The essay in question argues that
updating procedures are shaped by the cognitive interests that concepts serve, and
points out that in some cases, the cognitive interests that govern a concept are at least
partially independent of such empirical concerns as prediction and explanation.
When this is true, the updating procedures that govern a concept may imply that
fundamental principles involving the concept are immune to revision.

II. A type-materialist theory of experience

I first encountered the mind-body problem in college, as a result of reading Freud’s
Interpretation of Dreams. (Freud 2010) Originally I thought of the problem in
comparatively simple terms, as an intuitive conflict between the data of introspection
and the facts about the brain that science reveals. In effect, I reasoned as follows:
“When I experience a pain, what I am aware of seems strikingly different from the
neural activity that is described in textbooks and registered by various types of
apparatus. Introspective awareness provides no inkling of axons, dendrites, and cell
bodies, nor, a fortiori, of the electrochemical activity in which these entities are
6 INTRODUCTION

involved. Moreover, it seems that there is no hope of explaining away the differences
between experiences and brain processes by saying that the former are merely
misleading appearances of the latter. We are directly acquainted with pains, and as
a result, awareness of pain provides no foothold for talk of a contrast between
appearances and reality. But surely, if pains are nothing but brain states, and we
are immediately aware of them, then we should have some clue as to their physical
nature.” At the time, this introspection argument seemed to me to preclude reduc-
tionism absolutely, and I still think of it as one of the two most fundamental sources
of anti-physicalist sentiment. (The other is the “unbridgeable gulf ” intuition that it
would be possible to have a full understanding of the brain without having a clue as to
the existence of pain and other sensory states. This second source is pretty clearly the
underlying motivation for the knowledge argument and the explanatory gap argu-
ment. There are of course a number of other anti-reductionist arguments, such as the
Cartesian modal argument; but in my view, they all depend in one way or another on
the facts emphasized in the introspection argument, or on the unbridgeable gulf
intuition, or on both.)
Since I was not attracted to dualism, I thought that the arguments against
physicalistic accounts of experience could only be accommodated by reducing the
physical world to congeries of possible and actual experiences. My hopes lay with
what I thought of as constructive idealism—that is, with projects like the ones in
Russell’s Our Knowledge of the External World, Carnap’s Aufbau, and Goodman’s
Structure of Experience. This sort of solution continued to appeal to me through
graduate school; but eventually it gave way to a half-hearted physicalism, largely
because, after a period of denial and self-deception, I came to accept Chisholm’s
argument that constructive idealism is in principle bound to fail. Chisholm’s argu-
ment allows room for non-constructive idealism, which asserts that physical objects
are composed of experiences but makes no effort to associate specific classes of
experiences with specific objects. However, no one can long admire a theory that
paints a picture of the world in bold strokes but denies that there is a need to provide
details. (Very roughly, Chisholm argued that it is impossible to specify the experi-
ences that compose a given physical object without making assumptions about other
physical objects. He went on to conclude that any systematic attempt to reduce
objects to experiences would be regressive or circular. (See Chisholm 1948.))
Eventually, in the early 1980s, I became persuaded that it is possible to account for
the apparent contrasts between the data of immediate experience and brain states by
appealing to differences between the concepts we use to classify sensory phenomena
and the concepts we use to classify brain states. The former are concepts that we
apply in virtue of being in brain states; the latter are concepts we apply in virtue of
perceiving brain states, or perceiving changes in the apparatus we use to register brain
states, and in virtue of deploying theories that interpret such observations. Pains and
other experiential states do not seem to us to have physical structure because the
concepts we use to classify them do not articulate the details that are revealed by
 INTRODUCTION 7

perception and theory-construction. In other words, they have no logical relations to

our perceptually grounded beliefs about the brain. I also began at that time to have
doubts about Putnam’s multiple realization argument, which maintains that crea-
tures often have brains with quite different structures despite having similar mental
states, and that this prevents us from identifying mental states with brain states.
Among other things, it seemed that the multiple realization argument had less
intuitive backing than Putnam claimed for it: I was impressed by a tendency
I noticed among informants to withdraw or qualify attributions of mental states to
organisms like snakes and grasshoppers when they came to appreciate the vast
difference between the brains of such creatures and human brains. Newly optimistic
about the prospects of strongly reductionist theories of experience, and type-identity
theories in particular, I published “In Defense of Type Materialism,” the first essay in
Part II, in 1984. Seven years later I attempted a full dress presentation of type-
materialism in a book called Sensations (Hill 1991).
“Possibility, Imaginability, Conceivability, and the Mind-Body Problem,” the
second chapter in Part II, is a further attempt to buttress type-materialism against
dualist arguments. It focuses on Kripke’s version of the Cartesian modal argument,
but the intended scope is actually wider than that. For example, the scope includes
the argument from introspection that I described a few paragraphs back. The third
essay is my take on the status of type-materialism in 2013. In a way, though I am no
longer fully committed to the theory, I find its credentials more impressive than I did
at the time of writing “In Defense of Type Materialism.” This is largely due to
increasing doubts about the multiple realization argument, and a correspondingly
stronger commitment to the idea that there are extremely robust correlations
between experiences and brain states. Unlike other psychophysical correlations (for
example, the correlation between pain in one’s left hand and physical damage in that
hand), these correlations are exceptionless and ideally specific. Perfect correlations
demand identifications, or so it can seem. The third essay spells out the grounds for
rejecting the multiple realization argument in some detail. It also joins the two earlier
essays in recommending conceptual dualism as a defense against arguments for
property dualism. Conceptual dualism no longer seems to me to refute those
arguments decisively, but it does temporarily return the burden of proof to the
shoulders of the dualist.
Although type-materialism has a lot going for it, there are also very substantial
considerations that call it into question. One consideration derives from the phe-
nomenon that has come to be known as transparency. The properties that are
revealed in experience are not properties of brain states, but rather properties of
such extramental phenomena as physical objects in the visual environment and
disturbances in the periphery of the body. Thus, for example, vision is concerned
with properties like being small, being red, and being elliptical, and also with prop-
erties like looking small from here, looking red in this light, and looking elliptical from
this angle. These are all properties of external objects. As Harman famously said,
8 INTRODUCTION

[W]hen you see a tree, you do not experience any features as intrinsic features of your
experience. Look at a tree and try to turn your attention to the intrinsic features of your visual
experience. I predict you will find that the only features there to turn your attention to will be
features of the presented tree . . . . (Harman 1990, p. 39)

Equally, when one turns one’s attention to an experience of itching, what one seems
to encounter are features of the itch, which means that they are features of a bodily
disturbance, perhaps located in the left calf. If we are to honor transparency, we won’t
be able to identify experiential properties with properties of brain states. Another
problem for type-materialism is that conceptual dualism is ultimately unsuccessful. It
is arguably adequate as a first-line defense, but it fails in later stages of the dialectic.
This is because it is simply false that awareness of sensory and other qualitative
phenomena requires conceptualization. It is the job of concepts to effect classifica-
tions. To do this job, they must prescind away from the particularities of experience,
tracking commonalities rather than specific details. But awareness of sensory and
other qualitative phenomena is generally exquisitely detailed. Consider, for example,
what it is like to be visually aware of a complex pattern in a carpet, or of the jagged
outline of a mountain range in the distance. When you are visually aware of such
phenomena, you are acquainted with levels of detail that could only be captured
conceptually by long descriptions, and then only to a rough approximation. More-
over, in addition to being concerned with greater complexity than conceptualized
awareness, experiential awareness generally presents us with higher levels of deter-
minacy. Conceptualized awareness puts us in touch with determinable features of
objects, but what we see is highly determinate—for example, highly determinate
shades of red, and highly determinate degrees of curvature. In sum, there are
differences between conceptualized awareness and experiential awareness in point
of both complexity and specificity, and these differences are essential, given that
concepts have classificatory functions that are foreign to experiential representations.
(For a discussion of attempts to rescue conceptual dualism by appealing to demon-
strative concepts and concepts that have experiences as constituents, see Hill 2009,
pp. 53–6.)
In my view, despite these and other difficulties, type-materialism is still one of the
two main options for physicalists. Further, as I say at the end of “The Identity
Theory,” I believe that type-materialism will be around for some time to come. It is
desirable to honor transparency, but it is also desirable to honor the extremely tight
correlations linking experiential states to brain processes. Alas, we cannot do both.
Because each of the desiderata is important, it is reasonable to expect that the
sensibilities of theorists will continue to resonate to both of them, yielding by turns
to the influence of the first and the influence of the second. As the appeal of each
desideratum waxes and wanes, the pendulum will swing back and forth between
type-materialism and its physicalist rivals, the chief of which I take to be
representationalism.
 INTRODUCTION 9

III. A representationalist theory of experience

The chapters in Part III maintain that all awareness, including awareness of sensa-
tions and other qualitative phenomena, depends on representations, and that this fact
is the key to solving the mind-body problem. There are two main reasons for
accepting a representationalist account of awareness. One derives from contempor-
ary cognitive science, which explains psychological processes in terms of series of
transformations that take representations as inputs and yield representations as
outputs. This is the story that it tells of information processing in general, and of
the processing that underlies awareness in particular. The other main reason for
accepting representationalism differs from this first one in that it is concerned
specifically with perceptual and other forms of experiential awareness, and in that
it is more or less independent of current science. It is a platitude that states of
perceptual awareness have such properties as scope, perspective, and resolution.
When one is perceptually aware of an object, one is aware of some of the neighboring
items but not others. This is scope. Further, one is aware of some of the object’s sides
or facets but not all of them. This is perspective. And one is aware of some of the
object’s parts while other, smaller parts fall outside one’s ken. This is resolution. Now
a representationalist can explain the scope, perspective, and resolution of a state of
awareness by appealing to the contents of the representations that constitute the
state. When I look at the chair across the room, my state of awareness involves a
representation that has a particular content. The state represents the chair as next to a
table, as having a front side with a certain shape and color, and as having such parts
as arms, legs, and a seat. But it does not represent the chair as being in the same room
as the radiator behind me, or as having a backside that has a different shape and color
than its front, or as having atoms as constituents. These facts about the content of the
representation in question provide proximal explanations of the relevant features of
my state of awareness. Non-representational theories have a great deal of trouble
explaining those features. This is true particularly true of Russellian theories that take
awareness to consist in a metaphysically simple relation of acquaintance. Since such
theories maintain that facts of awareness are simple and fundamental, there are no
structural features of awareness to which they can appeal in explaining high level
properties like scope.
Here, then, are two powerful grounds for thinking that awareness is representa-
tional. It is sometimes maintained that there are also introspective grounds for this
view. Thus, for example, advocates of transparency sometimes express their position
by saying that when they introspectively consider a perceptual experience, they are
aware of the representational content of the experience but not of its intrinsic
properties. This view is misguided. To be sure, introspection reveals that an experi-
ence is an experience of a certain item, and an advocate of transparency would be
right in claiming that this is all that an experience reveals. But to say that an
experience has of-ness is not the same as saying that it has representational content,
10 INTRODUCTION

for there is more than one way of explaining of-ness. For instance, a Russellian can
say that an experience is an experience of blue because it consists in acquaintance
with a blue item. To move from introspectively accessible facts involving of-ness to a
representationalist conclusion requires arguments. Representationalism can’t simply
be read off from the data of introspection. It has to be shown that it provides the best
explanation of of-ness.
A representationalist theory of awareness will help with the mind-body problem in
two ways. First, it will permit us to draw an appearance/reality distinction with
respect to awareness of qualitative phenomena like pains and perceptual appear-
ances. If awareness of qualia is representational, there will be a Kantian distinction
between the ways in which qualitative phenomena are represented and the ways in
which they are in themselves. There will, for example, be a difference between a
pain qua represented and a pain qua item in the world. Among other things, this
distinction will afford a basis for explaining why awareness of pain does not disclose
the structural features that the physical correlates of pain possess (and in doing so, it
will undercut antireductionist arguments like the introspective argument at the
beginning of the last section.) Second, a representational theory will permit us to
explain away the unbridgeable gulf intuition—the intuition that it is possible to have
a full understanding of the physical world without having any clue as to the existence
of qualitative phenomena. That intuition arises because the representations that are
deployed when we are aware of qualitative phenomena from the first person per-
spective are quite different than the representations that subserve third-personal
awareness of physical structure. Consider first-personal, experiential awareness of
pain, and compare it with the states of perceptual awareness that provide evidence for
theories of the various physical correlates of pain, such as processes in peripheral
nociceptive nerve fibers and the activity in the somatosensory cortex that registers
such processes. The representations that are involved in first-personal awareness of
pain are altogether disjoint from the representations that are involved in perceptual
awareness of the physical correlates, so it is no surprise that it can seem that
awareness of pain is concerned with an entirely different domain than awareness
of the correlates. Moreover, it is no surprise that philosophers should resist the idea
that this impression of distinctness can be explained by saying that our first-personal
grasp of pain is provided by a distinct set of representations. As we noticed a moment
ago, introspection fails to attest to the representational nature of experiential aware-
ness, and it therefore fails to reveal that experiential awareness of qualia is governed
by a Kantian appearance/reality distinction. Further, folk psychology is an obstacle to
recognizing the representational nature of awareness of pain. Indeed, it counts
directly against drawing an appearance/reality distinction here, because it implies
that one really is in pain whenever it seems to one that one is in pain (that is,
whenever one has an experience as of pain). If one consults a physician about a pain
in one’s foot, the physician may say that there is nothing wrong with your foot, but
she won’t say that your pain is an illusion. Equally, even if your foot is seriously
 INTRODUCTION 11

damaged, no one will insist that you’re in pain unless you actually have an impression
of pain. (For a more detailed exposition of this line of thought, and for the lines of
thought in the two preceding paragraphs, see Hill 2009, chapters 3 and 4.)
As the reader may have observed, this representationalist strategy for dealing with
the mind-body problem is structurally similar to the strategy that we considered
earlier under the name “conceptual dualism.” In each case, the strategy attempts to
explain away the impression that qualitative phenomena are distinct from physical
phenomena by invoking the different ways in which the two kinds of phenomena are
represented. The chief difference between the strategies is that conceptual dualism
seeks to explain the impression in terms of a special lexicon of concepts, while
representationalism seeks to explain it by appeal to a special category of experiential
representations.
The representationalist strategy is superior to that of conceptual dualism because it
doesn’t involve a commitment to the false proposition that all awareness is concep-
tualized. But also, its prospects of explaining specific aspects of qualitative awareness
are superior to those of conceptual dualism. To give just one example, it is usually
maintained that our lexicon of atomic conceptual representations is finite, and that
complex representations can only be produced from it by concatenation, but there
are no such restrictions on experiential representations. Indeed, it is widely allowed
that the systems of representation that subserve perception may be analog in
character. Now if this hypothesis is true, there is an extremely natural explanation
of our ability to experience an immense variety of highly determinate perceptual
qualities, such as the different shades of gray. We can experience a virtually limitless
variety of shades of gray because the system we use to represent colors is at least
virtually analog, affording us a virtually limitless variety of representations of con-
crete shades. (For additional arguments against conceptual dualism, see the penul-
timate paragraph of “The Identity Theory.”)
The version of representationalism that I wish to defend is quite different than the
versions that have been endorsed by other authors. One difference has to do with its
account of the metaphysical nature of qualia. Generally speaking, representationalists
tend to identify perceptual qualia with intrinsic, physical properties of external
objects. I agree that they are properties of external objects, but I think that they are
relational, viewpoint-dependent properties, such as looking small from this distance
and looking elliptical from this angle. A major goal of my work on qualia has been to
find a detained theory of such properties. The view I have wound up with is presented
in “The Content of Visual Experience.” My account of pain is also different than
other representationalist accounts, in part because it maintains that the common-
sense concept of pain is badly flawed, in that the concept conflates three quite
different phenomena, and in part because of its complex treatment of pain intensity.
My views about pain are presented in “OW! The Paradox of Pain” and “Locating
Qualia: Do They Reside in the Brain or in the Body and the World?”
12 INTRODUCTION

My view also differs from those of other representationalists in a second respect.

Representationalists like Byrne, Dretske, and Tye are committed to the view that the
qualitative character of experience supervenes on its representational content. (See
Byrne 2001, Dretske 1995, and Tye 2000.) I accept the idea that qualia are always
objects of awareness, and are therefore included in representational content, but
I think it’s necessary to appeal to features of representations other than positive facts
about their contents in order to explain fully how qualia appear to us—that is, in
order to explain all of the judgments about qualia that experience inclines us to make.
Consider the ways that colors look to us in broad daylight. It is plausible that these
color qualia are complex, relational properties of objects—say, relational properties
that objects have in virtue of reflecting light that affects cone cells in particular ways.
Now it is clear our experience of color qualia does not apprise us of their complexity
or relationality. Instead, it inclines us to think of them as simple and as intrinsic
properties of the surfaces of objects. As I see it, in order to explain this fact, it is
necessary to invoke the fact that our visual representations of color qualia do not fully
articulate their physical structure. More specifically, it is necessary to invoke the fact
that the representations in question are atomic, or syntactically simple: they do not
have representational constituents corresponding to the different aspects of the
physical structure of color qualia, and because of this, they are unable to represent
those aspects explicitly. Similar remarks apply to the representations of the other
visual qualia that strike us as simple, and also to representations of a broad range of
auditory, olfactory, and gustatory qualia. This picture can be summarized by saying
that there are two types of phenomenal character that an experience can have. In the
first place, there is the class of qualia that the experience represents. Generally, these
are properties of bodily sensations or relational properties of external objects. Second,
there are the properties that experiential representations lead us to attribute to qualia
because of their intrinsic features. These features may or may not be possessed by the
qualia to which they are attributed.
To summarize, the version of representationalism that I defend in Part III makes
four main claims: first, all awareness, including awareness of qualitative phenomena,
is representational in character; second, because of this, it is possible to draw an
appearance/reality distinction with respect to qualitative awareness; third, qualia are
not properties of experiences, but rather properties of extra-cranial occurrences in
the body and relational, viewpoint-dependent properties of external objects; and
fourth, it is necessary to invoke intrinsic features of representations to explain some
aspects of our overall take on qualia. (To be strictly accurate, the third claim
maintains only that pain and other somatic qualia are mainly properties of extra-
cranial events. The theory is compatible with the existence of headaches!)
It is worth noting that the first, second, and fourth doctrines are independent of
the third. It would in principle be possible to develop a version of representationalism
according to which qualia are physical properties of occurrences in the brain, and
awareness of qualia involves representations of such properties. This alternate
 INTRODUCTION 13

version of representationalism has some virtues. For one thing, reflection shows it
supports a response to the mind-body problem that is much the same as the one that
is supported by the original version. Moreover, it does a better job than the original
version of accommodating the tight correlations linking qualitative states to brain
states. I discuss this version at some length in “Locating Qualia,” the third chapter in
Part III, and the version also makes an appearance in the fourth chapter. In the end
I reject it, in part because it has trouble accommodating transparency, and in part
also because the neural systems in the brain that support qualitative awareness seem
to be dedicated to tracking extra-cranial entities. (Among other things, they all
involve maps of extra-cranial terrains.)
One could argue that it is possible to get around the transparency problem by
maintaining that the mechanisms that bind representations of qualia to representa-
tions of objects are systematically flawed, connecting representations of properties of
brain states with representations of extra-cranial entities. That is, one could maintain
that systematically inappropriate binding leads us to “project” properties of brain
states onto external objects. I consider this proposal in Chapter 12, but do not pursue
it at length. It seems to lead to intractable problems.
At all events, as I now see the situation, the version of representationalism that is
defined by the foregoing four claims is the theory of experience that has the best
prospects of success.
There is considerable thematic overlap between the essays in Part III and my 2009
book Consciousness, but I hope that this won’t deter readers of the earlier work from
looking at them. The issues are sliced and diced differently in the essays than in the
book; presentations of key ideas are different and in many cases more intuitive; and
there are additional arguments. I am confident that readers who made it through
Consciousness without being fully persuaded will be unable to resist these alternative
formulations and new lines of thought!

IV. Knowledge
The chapters in Part IV are all concerned in one way or another with knowledge, but
on the whole, they are less cohesive than the chapters in earlier Parts. Thus, while the
first two chapters in Part IV have overlapping concerns, the third and fourth chapters
branch out in different directions, focusing on two issues that are quite different than
the ones considered in the first two chapters, and also quite different from each other.
The first chapter, “Process Reliabilism and Cartesian Skepticism,” considers the
nature of epistemic justification—the type of justification that a belief must possess in
order to count as knowledge. It has two goals. One is to defend process reliabilism,
according to which, roughly speaking, a belief is epistemically justified just in case it is
produced by a cognitive process that is highly reliable, in the sense that the beliefs to
which it gives rise are very likely to be true. The other goal is to propose an answer to
Cartesian skepticism. The second paper, which I wrote with my colleague Joshua
14 INTRODUCTION

Schechter, is related to this first one because it defends the idea that there is a close
relationship between knowledge and probability of truth. As with the first paper, its
main goal is to make the world safe for reliabilist accounts of epistemic justification.
In pursuit of this goal, the chapter addresses the following three questions: (i) Is
it true that one can easily come to know any proposition that one knows to be
implied by other propositions that one knows? (ii) Is it true that one must know a
proposition in order to be entitled to assert it? And (iii) is it true that one must know
a proposition in order to be entitled to use it as a premise in practical reasoning? John
Hawthorne has argued for answers to these questions that tend to count against
reliabilist accounts of epistemic justification (Hawthorne 2004). Schechter and
I criticize his arguments and defend opposing answers. In addition, we offer an
account of practical reasoning that meshes with currently popular “dual system”
theories of cognition. (In brief, dual system theories maintain that the mind can be
understood as having two types of cognitive faculty, one that makes use of quick and
dirty heuristics, and another that relies on methods that are computationally
demanding and/or make use of multiple sources of information. See Kahneman
2011 for an extended account of the dual systems approach.) The chapter maintains
that human beings employ two methods of practical reasoning, one that is quick and
relatively undemanding but that can only deliver approximate answers in many
cases, and another that delivers more accurate answers but requires more elaborate
computations. Which method we use depends upon how high the stakes are in a
given decision problem, and also on how appropriate it is to assimilate the relevant
probabilities to 0 or 1.
The third chapter is concerned with the epistemology of modal knowledge.
Philosophers have generally maintained that we can come to know that a proposition
is possible by conceiving of a situation in which the proposition is true. They have
also held that we can come to know that a proposition is necessary by trying to
conceive of a situation in which its denial is true, and finding that we are unable to do
so. The third chapter criticizes these traditional doctrines, and seeks to replace them
with more adequate epistemological principles. Thus, for example, building on the
criticisms of the Cartesian modal argument in Part II, the chapter maintains that
conceivability must be severely constrained if it is to provide knowledge of possibility.
It also argues that an adequate set of constraints would significantly restrict the role
that conceivability can play in producing knowledge of necessity. Roughly, the reason
is this: if conceiving is to be a reliable guide to possibility, then its deliverances must be
compatible with the class of necessary truths. It follows that the proper use of
conceivability presupposes that the agent has an a priori understanding of which
propositions are necessary, and exercises that understanding in moving from facts
about conceivability to conclusions about possibility. But this means that knowledge of
necessity must have some source other than conceivability. After considering various
proposals, the chapter draws the conclusion that we can only come to know that a
proposition is necessary if it is stipulated to be necessary by an implicit definition of
 INTRODUCTION 15

necessity. In the final section the chapter goes on to criticize a suggestion about
knowledge of metaphysical necessity that I put forward some years ago, according to
which such knowledge can be reduced to knowledge of counterfactual conditionals
(Hill 2006).
As I mentioned much earlier, the fourth chapter in Part IV is concerned with the
relationship between the cognitive interests that a concept serves and the rules or
procedures that define its epistemic role. To illustrate, consider Quine’s claim that the
concept of truth is used as a device for endorsing propositions without explicitly
mentioning them. (See Quine 1970.) This is a plausible view. Thus, for example, it
seems to be the point of (K):
(K) All of Kant’s main doctrines are true

that it enables us to endorse or accept all of Kant’s main doctrines without going to
the trouble of explicitly saying what those doctrines are. Now in addition to this
Quinean purpose of the concept of truth, there are other purposes that are closely
related but nonetheless distinct. Thus, the concept also serves the purpose of enabling
summaries or compilations of sets of propositions—compilations that can be easily
stored in both long term memory and working memory, and can be easily accessed in
either venue. It is clear, I think, that (K) is able to serve as a compilation of this sort.
In general, propositions like (K) make it possible for us to compress many proposi-
tions into single propositions that are fairly compact. But this would not make such
propositions useful unless they could also serve the further purpose of generating the
specific propositions that they summarize. Thus, for example, if (K) is to be genuinely
useful as a compilation, it must be possible to use it as a pump for producing
instances of (M):
(M) If it is one of Kant’s main doctrines that p, then p.

Now if (K) is to serve this generative purpose, it must be the case that any agent who
embraces (K) is entitled to combine it with instances of (T*):
(T*) The thought that p is true only if p.

For example, the agent must be entitled to combine (K) with (S):
(S) The thought that space is transcendentally ideal is true only if space is transcendentally
ideal.

In other words, if (K) is to play a compressive and generative role in an agent’s

cognitive economy, then all instances of (T*) must be epistemically available to the
agent as additional premises. This illustrates one of the main claims of the chapter—
the thesis that the purpose of a concept shapes the epistemic status of various rules
and propositions in which it figures as a constituent.
The second main claim is that some concepts serve purposes that are non-
empirical. And the third claim is that when this is true, the epistemic rules or
16 INTRODUCTION

procedures governing the use of a concept may protect certain propositions contain-
ing it from being called into question by empirical evidence. Like the first claim, these
additional claims can be illustrated by the concept of truth. It seems that we would be
interested in deploying the concept of truth no matter what empirical evidence we
had accumulated. Thus, for example, we can imagine a context in which an agent
encounters evidence that leads him to abandon his faith in science, and induces him
to place his trust in a certain guru instead. In this context he will still need the concept
of truth, for it will be useful for him to formulate (G) and store it in memory:
(G) Everything the guru tells me is true.

Again, it seems that an agent would still need the concept of truth if he came to
believe that the senses are completely untrustworthy, for he would want to embrace
(N) in order to register his skepticism:
(N) No perceptually grounded proposition is true.

Moreover, the agent would want to be able to derive the denials of particular
perceptually grounded propositions from (N). For example, if he were to have visual
experiences as of a snowstorm, it would be important to be able to derive the denial of
the proposition that he was seeing snow. In general, it seems that no matter what our
epistemic situation might be, we would have a use for a concept with the compressive
and generative powers of truth. Now reflection shows that the epistemic availability
of propositions of form (T) (see p. 3) is a necessary condition of the concept’s having
the full range of these powers. Accordingly, we can conclude that the concept of truth
is, or at least ought to be, governed by rules or procedures that guarantee the
epistemic availability of such propositions in every empirical context. It follows
from this that propositions of form (T) are, or at least should be, immune to
empirical revision.
Earlier I said that “Process Reliabilism and Cartesian Skepticism” proposes an
answer to Cartesian skepticism. That is true, but the answer I give in that chapter
carries some presuppositions that not everyone will endorse. I should have empha-
sized at some point that it is possible to extend the argument in such a way as to make
it independent of those presuppositions, but I neglected to do so. I will try to correct
this omission here, and will then say something about the relationship between the
present argument and the one in the chapter. As in the chapter, I will understand
Cartesian skepticism to be the claim that we are not epistemically justified either in
holding our perceptual beliefs or in holding any of the other empirical views that are
based on our perceptual beliefs.
In evaluating this claim, we need to keep in mind the familiar distinction between
internalist and externalist conceptions of what it is for a belief to be justified. An
internalist conception claims that the factors that provide justification for our beliefs
are independent of the facts of the external world. Roughly, the idea is that all such
factors are introspectively available. An example of such a view is the doctrine,
usually known as explanationism, that we are justified in believing whatever theory
 INTRODUCTION 17

of external reality provides the best explanation of our perceptual experiences.

On the other hand, an externalist conception claims that justification depends on
the external facts that are the truth-makers for our beliefs. Process reliabilism is a
paradigmatic externalist view. As we saw, it maintains that whether a belief is justified
depends on whether it is produced by a cognitive process whose outputs are in
general very likely to be true. It is common to see internalist and externalist
conceptions as offering competing accounts of epistemic justification, but there is
actually no need to regard them as opposing positions. Even if epistemic justification
turns out to be mainly externalist (or internalist) in nature, it could still be true that
there is another important form of justification that admits of an internalist (exter-
nalist) characterization.
Now as I argue at some length in the chapter, the Cartesian skeptic would be on
extremely weak ground if she were to maintain that we are not externalistically
justified in holding our perceptual beliefs. For this is a very strong thesis! To defend
it, the skeptic would have to show that the cognitive processes that produce our
perceptual beliefs are very unreliable, and she could only show this if she were to
make a very strong claim about the nature of extramental reality. More specifically,
she would have to claim that the external world is quite different than our perceptual
beliefs represent it to be. But the skeptic’s whole point is that we do not have an
epistemic right to make claims about the external world. Hence, if she were to
attempt an argument of the given sort, she would be caught in a self-referential
inconsistency. Moreover, even waiving this first point, it seems that an effort to
discredit our perceptual beliefs would run into problems. To succeed, the skeptic
would have to amass empirical evidence showing that our perceptual beliefs are quite
often false. It seems, however, that such evidence would be hard to come by, for it
seems that our beliefs about, say, the shapes and sizes of objects in the immediate
environment are quite trustworthy. If you were to judge a nearby object to be
rectangular, or to be the same size as your hand, there is an excellent chance that
your judgment would be true—or at least approximately true, which is all that a
defense of commonsense perceptual judgments really needs to maintain.
In view of these considerations, it is clear that if the skeptic is to make a case for her
view, she must focus on internalist justification. She must show that we are not
internalistically justified in holding our actual empirical beliefs. But it would not be
easy to reach this goal. Thus, an explanationist can argue plausibly that the hypoth-
esis that our empirical beliefs are roughly correct provides a better explanation of our
perceptual experiences than any skeptical hypothesis that has ever been entertained.
One such argument is that skeptical hypotheses have little or nothing to say about the
specific casual mechanisms by which evil demons or super-computers generate our
perceptual experiences. Another argument is that the explanations of our experiences
that are afforded by skeptical scenarios are parasitic on the explanations that are
provided by the empirical beliefs we actually hold. (If we are to see what predictions
about our future experiences follow from the evil demon hypothesis, we must first
18 INTRODUCTION

work out what predictions follow from our actual empirical beliefs. For all we know
about the evil demon is that it wants to create a massive delusion that is systemat-
ically unified, so we won’t be able to penetrate it.) In view of these considerations, it is
clear that the “explanations” provided by skeptical scenarios are actually quite poor,
if indeed they are explanations at all. Accordingly, an explanationist is in a position to
claim that we are internalistically justified in holding our empirical beliefs. Moreover,
it is pretty clear that internalists of a number of stripes can arrive at similar
conclusions. (For a more systematic exposition of the dialectic between explanation-
ists and skeptics, see Vogel 1990. And for a rather different internalist reply to the
skeptic, see Huemer 2001.)
There is a case, then, that whether epistemic justification is externalist or intern-
alist in character, the Cartesian skeptic will have trouble defending her central claim.
At this point, however, it might be said that the plausibility of the skeptic’s
argument (see Section II of “Process Reliabilism and Cartesian Skepticism” for an
account of it) shows that there must be some version of internalism that is both dear
to our hearts and vulnerable to the skeptic’s attack. The plausibility of the skeptic’s
argument is a non-constructive proof of the existence of such a version. But this view
is highly questionable. As the foregoing discussion indicates, there are several
appealing and reasonably well worked out versions of internalism on which the
skeptic’s reasoning fails. This diminishes the importance of any version on which it
might succeed. But also, it increases the plausibility of diagnoses of skepticism which
trace it to a confusion of some sort. If the most attractive known versions of
internalism are immune to skepticism, and attempts to identify attractive versions
that are vulnerable to skepticism have come to naught, we should expect that the
appeal of the skeptic’s reasoning is illusory. (In my view, skeptics tend to confuse
having a good reason for holding a belief with having sensory evidence for the belief.
It isn’t always easy to distinguish between these properties, but reflection shows that
they are different: If a belief has explanatory power, that might count as a good
reason for holding it, but it certainly wouldn’t count as sensory evidence for the
belief. Skeptics are also confused about the status of reasons. Thus, a skeptic might
dismiss explanatory power as irrelevant to her concerns because it isn’t truth
conducive. But if, as I maintain, the skeptic must make an internalist case for her
position, reasons for holding beliefs can be relevant even if they aren’t truth
conducive.)
This is not to say that it is possible to prove that we are epistemically justified in
holding our perceptually grounded beliefs. If, as I believe, externalism provides the
correct account of epistemic justification, then any attempt to show that we are
epistemically justified in holding our perceptual beliefs would have to make use of
premises about extramental reality, and would therefore presuppose that such
premises could be known to be true. This would beg the question. Fortunately, as
I argue in the chapter, this observation has no tendency to show that we lack
 INTRODUCTION 19

externalist justifications for our perceptual beliefs. It is possible to have an externalist

justification for a belief without being able to prove that one has one.
If the skeptic had been content to express doubts about proofs of justification
instead of issuing challenges to justification itself, she would have been on much
stronger ground. To draw a general moral, if epistemic justification is ultimately
externalist in character, those who inquire into justification should have modest
ambitions. It is impossible to settle the deepest questions by proof.
This set of views I have just been reviewing is closely related to the ones expressed
in “Process Reliabilism and Cartesian Skepticism,” but there is a significant differ-
ence. The present argument has the form of a dilemma: it purports to show that the
skeptic will encounter problems whether epistemic justification is externalist or
internalist in character. On the other hand, in criticizing skepticism the paper, I in
effect assume that externalism gives a correct account of epistemic justification, and
then proceed to show, on the basis of this assumption, that the skeptic cannot hope to
establish her position. I don’t offer a parallel argument pointing out that the skeptic
also runs into trouble on internalist scenarios.

References
Byrne, A. (2001). “Intentionalism Defended,” The Philosophical Review 110: 199–240.
Carnap, R. (2003). The Logical Structure of the World and Pseudoproblems in Philosophy.
Chicago: Open Court.
Chisholm, R. M. (1948). “The Problem of Empiricism,” The Journal of Philosophy 45: 512–17.
Dretske, F. (1995). Naturalizing the Mind. Cambridge, MA: MIT Press.
Freud, S. (2010). The Interpretation of Dreams. New York: Basic Books.
Goodman, N. (1977). The Structure of Appearance. New York: Springer.
Harman, G. (1990). “The Intrinsic Quality of Experience.” In J. Tomberlin (ed.) Philosophical
Perspectives, Vol. 4. Atascadero, CA: Ridgeview Publishing Company, 31–52.
Hawthorne, J. (2004). Knowledge and Lotteries. Oxford: Oxford University Press.
Hill, C. S. (1991). Sensations: A Defense of Type Materialism. Cambridge: Cambridge Univer-
sity Press.
Hill, C. S. (2002). Thought and World: An Austere Portrayal of Truth, Reference, and Semantic
Correspondence. Cambridge: Cambridge University Press.
Hill, C. S. (2006). “Modality, Modal Epistemology, and the Metaphysics of Consciousness.”
In Shaun Nichols (ed.), The Architecture of the Imagination. Oxford: Oxford University
Press, 205–35.
Hill, C. S. (2009). Consciousness. Cambridge: Cambridge University Press.
Hill, C. S. (2011). “Review of Paul Horwich’s Truth—Meaning—Reality,” Mind 120: 1262–70.
Horwich, P. (1998). Truth, 2nd edition. Oxford: Oxford University Press.
Huemer, M. (2001). Skepticism and the Veil of Perception. Lanham, MD: Rowman & Littlefield.
Kahneman, D. (2011). Thinking Fast and Slow. New York: Farrar, Straus, and Giroux.
Quine, W. V. O. (1960). Word and Object. Cambridge, MA: MIT Press.
20 INTRODUCTION

Quine, W. V. O. (1970). Philosophy of Logic. Englewood Cliffs, NJ: Prentice Hall.

Russell, B. (1914). Our Knowledge of the External World as a Field for Scientific Method in
Philosophy. Chicago: Open Court.
Tye, M. (2000). Consciousness, Color, and Content. Cambridge, MA: MIT Press.
Vogel, J. (1990). “Cartesian Skepticism and Inference to the Best Explanation,” The Journal of
Philosophy 87: 658–66.
PART I
Meaning
 2
“gavagai”*

When Quine tries to bring leading themes of his philosophy down to earth, he
usually relies on his “gavagai” example. I have no quarrel with the leading themes,
but I am suspicious of his observations about “gavagai.” So as to fix ideas, I open with
a version of the example, and follow with an account of Quine’s views as to how logic
fares under translation. Then, working from these summaries, I argue that the
example falls short of illustrating the leading themes.

I
Suppose that a linguist has made a certain amount of progress in translating an exotic
tongue. We can imagine that he has found the exotic sentence “Gavagai” to coincide,
in stimulus meaning, with the “Rabbit” of his own people, and that he has been
moved thereby to write five English sentences next to “Gavagai” in his tentative
lexicon. The sentences are: “Rabbit,” “Undetached rabbit part,” “Rabbit stage,”
“Rabbithood,” and “Rabbit fusion.” Not that our linguist is an ordinary one. Rather,
in pursuing the philosophy of translation, we are concerned with the behavior of
every imaginable linguist, and therefore, with linguists who are influenced by con-
temporary domestic philosophers. Such philosophers may demur, or venture only a
“Well, perhaps,” when prompted by a rabbity-stimulation and one of the four less
familiar sentences; but their behavior suggests that each of the special sentences
might someday become responsive to stimulation, and thereby, to the same stimu-
lations as “Rabbit” itself.
Having entered the five translations of “Gavagai” in his lexicon, our linguist comes
to the question of whether “gavagai,” as a term now, is to be rendered by our term
“rabbit,” by “rabbit part,” by “rabbit stage,” by “rabbithood,” or by “rabbit fusion.”
Since their respective occasion sentences are imagined to coincide in stimulus
meaning, the question cannot be answered by checking among patterns of stimula-
tion and assent. Can further behavioral data, say correlation between pointing
gestures and speech, help to resolve the problem?

* I wish to thank W. V. Quine and Israel Scheffler for commenting on an earlier version of this note.
24 PART I : MEANING

Consider, then, how. Point to a rabbit and you have pointed to a stage of a rabbit, to an integral
part of a rabbit, to the rabbit fusion and to where rabbithood is manifested. Point to an integral
part of a rabbit and you have pointed again to the remaining four sorts of things; and so on
around. (Quine 1960, pp. 52–3)

To get at the extension of “gavagai,” therefore, our friend will need to supplement his
pointing with questions about sameness, difference, and number. He will need to ask
the equivalent of “Is this the same thing as that?” or of “Do we have one rabbity-thing
here, or are there more than one?” But, Quine says, such questioning requires of the
linguist a command of the native language far beyond anything that stimulus
meaning can teach. Stimulus meaning leaves it open, for example, whether a certain
exotic term is to be translated by our identity predicate, or by our “belongs with”
instead. For this reason,
when in the native language we try to ask “Is this gavagai the same as that?” we could as well be
asking “Does this gavagai belong with that?”. Insofar, the native’s assent is no objective
evidence for translating “gavagai” as “rabbit” rather than “undetached rabbit part” or “rabbit
stage” (Quine (1969a, p. 33)).
Thus, neither pointing nor questioning gives a unique answer to the linguist’s problem. He will
of course make a choice, and in making it he will doubtless be guided by some consideration or
other. But any such consideration will pertain less to linguistic behavior of natives than to his
own preference for one of the five domestic terms. Quine concludes that translation of
“gavagai” is indeterminate: that there are five different ways, equally compatible with linguistic
evidence, of correlating “gavagai” with a domestic term. And since the five domestic terms in
question differ from one another in reference or extension, he adds that the reference of
“gavagai” is inscrutable (Quine (1969a, p. 35)).

II
Turning now to consider Quine’s views about translation of logical words, we find
that he is optimistic about signs for the truth functions: he urges that there are
“substantial” behavioral criteria for recognizing negation, conjunction, and the rest
(Quine (1969a, p. 104)). On the other hand, he despairs of translating quantifica-
tional sentences uniquely. His reason is that they “depend for their truth on the
objects . . . of which the component terms are true; and what those objects are is not
uniquely determined by stimulus meanings” Quine (1960, p. 610)). Since general
terms of different languages cannot be uniquely correlated in point of extension,
translation of their quantificational contexts cannot be unique.
Quine allows, however, that there is a point to the question of whether a given
exotic expression counts as a quantifier. This attitude emerges as he reflects on
substitutional quantification:
Behavioral conditions for interpreting a native construction as existential substitutional quan-
tification, then, are readily formulated. We fix on parts of the construction as candidates for the
 “ GAVAGAI ” 25

roles of quantifier and variable; then a condition of their fitness is that the natives be disposed
to dissent from a whole quantified sentence when and only when disposed to dissent from each
of the sentences obtainable by dropping the quantifier and substituting for the variable.
A second condition is that natives be disposed to assent to the whole whenever disposed to
assent to one of the sentences obtainable by dropping the quantifier and substituting for the
variable. (Quine (1969a, pp. 105–6))

Now, there are many different kinds of substitutional quantification: quantification

which interacts with singular terms, quantification which interacts with verbs, and
even quantification which interacts with punctuation marks. Quine does not say, in
the passage cited, whether it is possible to distinguish between such kinds of
quantification at the level of assent and dissent. Let us look into the matter, focusing
on quantification which interacts with singular terms.
A linguist will have recognized certain large classes of expressions, known as
grammatical categories, in pursuing an inductive definition of the class of native
sentences. Will he have any reason to correlate one of these categories with our
English category of singular terms? One expects that the answer is “Yes.” Thus, as
Strawson has urged, it is to be expected that truth value gaps, sentences which are
neither true nor false, will be seen to depend systematically on members of one
grammatical category. (Strawson (1969, p. 100)). There is such a category in English
grammar: the category of singular terms.
Whether a sentence is true or false depends on the success or failure of the general term; but the
failure of the singular term appears to deprive the general term of the chance of either success
or failure. (Strawson (1969, p. 100))

If, therefore, a study of truth value gaps leads our linguist to category C, he will have
one reason to correlate C with our domestic singular terms. (Truth value gaps can be
made into grist for the linguist’s mill, as Quine points out, by taking refusal to assent or
dissent, exotic indifference, as a sign that a gap is at hand (Quine (1969b, pp. 320–1)).)
After looking into truth value gaps, our linguist will perhaps begin to study
changes in truth value. If he finds that members of a proper subset of C turn up
frequently in sentences with fluctuating truth value, and that they never turn up in
sentences with constant truth value, he will have found new justification for his initial
correlation. For in English, sentences with fluctuating truth value normally have
parts which occur in sentences with constant truth value, and parts which do not; and
parts of the second kind are by and large singular terms. (Assent and dissent will be
clues as to truth value.) Moreover, further justification might be found by looking for
another category (perhaps it will turn out to be D) which interacts with C as follows:
sentences can be built from a single member of D by adding two members of C, but
sentences never result from adding two members of D to (one occurrence of) a single
member of C. If he can find a category which exemplifies this pattern, he will have a
reason to correlate it with our transitive verbs, and at the same time, he will have
found another point of resemblance between our singular terms and members of C.
26 PART I : MEANING

Not that such cues must coalesce immediately: for example, the study of truth
value gaps might focus attention equally on C and on H. But one expects that these
cues and others will in the long run lead our friend to a native category which is closer
to our category of singular terms than to any other. And when he finds the right
category, he will be one step away from substitutional quantifiers which interact with
singular terms.1
In the quoted passage, Quine describes an experiment for detecting quantifiers. Such
experiments will sometimes indicate that informants are using objectual, or normal,
quantifiers. For if there is at least one existentially quantified sentence which com-
mands assent, while each of its substitution instances tends to provoke dissent, the
exotic people are thereby stamped as objectualists, and at the same time, as believers in
nameless objects. On the other hand, when informants are found assenting and
dissenting as Quine describes, their quantifiers can be viewed either way. Certainly
they could be substitutional quantifiers. It is possible, though, that the natives have an
ontology which is bigger than their stock of names, but that no predicate of their theory
is satisfied by just the nameless objects. Since both hypotheses accord equally well with
patterns of assent and dissent, Quine says that they come to the same thing: in this
situation, he is unable “to distinguish objectively between referential quantification and
a substitutional counterfeit” (Quine (1969a, p. 67)).
Be that as it may, we should at least credit the linguist with having learned from his
quantificational studies. He can now map exotic sentences uniquely onto logical
congruence classes of English sentences. (Sentences are congruent when they have
the same logical form up to permutation of substitutional and objectual quantifiers.)
This means that he can follow proofs in the exotic language, for his congruence-
mapping of sentences determines a congruence-mapping of proofs. In a phrase, he will
be able to recognize logical similarities between sentences and proofs of the two
languages. Linguists can sometimes do better, but they can always do at least this well.

III
Consider now the primary laws of identity:
(1) x = x
(2) [x = y & F(x)] ! F(y);
and let us agree to view (2) as a schematic picture of the infinite set of sentences
obtainable by replacing “F ” with an English predicate. As is well known, these laws
provide for all the general truths about identity, in the sense that each such truth can
be proved, in elementary logic, from (1) and one or more instances of (2). But more:

1
It should be mentioned that the line of thought ending here departs from the letter of Quine’s writing
on comparative grammar; see, e.g., Quine (1960), middle of p. 53. It is clear, however, that this procedure is
fully Quinean in spirit.
 “ GAVAGAI ” 27

they provide a criterion for determining whether or not an exotic predicate should be
translated by our identity sign. For it is evident that an exotic relation sign will count
as expressing identity if and only if it satisfies these two conditions: First, informants
must be disposed to assent to every sentence formed by inserting two occurrences of
a singular term in its argument places. And second, when an informant is found
assenting to a sentence which contains the sign and two other terms, say “a” and “b,”
and also to a sentence which contains “a” (or “b”) at one or more places, he must be
disposed to assent to a third sentence which is like the second except for having
occurrences of “b” in place of one or more occurrences of “a” (or of “a” in place of
occurrences of “b”). Except that it is couched in the language of “term” and “relation
sign,” this two-part criterion might be found anywhere in Quine. But as we saw
earlier, this kind of language can be given a behavioral interpretation. It seems, then,
that essentially Quinean techniques can be used to determine whether an exotic
people share our concept of identity.
When, in Section I, our linguist tried to ask an informant whether this gavagai is
the same as that one, it seemed that he could just as well be asking whether this
gavagai part (or this gavagai stage) belongs with that part (or with that stage). In view
of the criterion, however, we see now that this problem will yield to research. In
particular, if the linguist is worried about the first possibility, that he might have
asked whether this gavagai part belongs with that one, he will presumably proceed as
follows. First, he will create a variegated animal, by applying a little dye to the original
rabbit. Second, he will consult his lexicon of observation sentences, looking for the
correct translation of “Red.” We can suppose that it turns out to be “Fleep.” Next,
running through his list of exotic predicates, he will pick out the one which is closest
in form to “Fleep.” At this point, he will have three predicates: “fleep,” “gavagai,” and
the exotic relation sign, say “varx,” which is troubling him. Blending them with the
exotic existential quantifier and a few other logical items (which, we can suppose,
turn out to have the same form as corresponding devices of his own people), he will
obtain a sentence which looks like (3):
(3) (Ǝx) (Ǝy) (gavagai (x) & gavagai (y) & fleep (x) &—fleep (y) & varx (x, y)).
(3), it will be noted, is similar to (4) and (5) in logical form:

(4) (Ǝx) (Ǝy) (rabbit (x) & rabbit (y) & red (x) &—red (y) & x = y);
(5) (Ǝx) (Ǝy)(rabbit part (x) & rabbit part (y) & red (x) &—red (y) & belongs
with (x, y)).
If gavagais are in fact parts of rabbits, (3) will command the assent of anyone
stimulated by the variegated animal. (4) is demonstrably false, given an appropriate
instance of (2). Hence, if an informant is found assenting to (3), and there is no way
of changing his mind by an argument from (3) and an exotic instance of (2), (4)
cannot be the correct translation of (3). If gavagais are parts of rabbits, therefore, the
linguist can with a little effort come to see that his original question, in Section I, had
no bearing on rabbits or identity.
28 PART I : MEANING

One can feel, initially, that assent to (3) does not rule out all hope of seeing
informants as partial to rabbits and of pairing the mysterious “varx” with identity.
For, one thinks, the two occurrences of “fleep” might have different meanings:
“fleep” might be an indicator word, roughly equivalent to “red here.” Or again,
“fleep” might mean “red at one place” at its first occurrence, and “red at another
place” at its second occurrence. Notice, however, that these hypotheses can be tested.
If the informant makes no sign that “fleep” is an indicator word, if no pointing
behavior can be detected, then “red here” runs afoul of the first principle of transla-
tion: translate so as to preserve behavioral patterns. Or take the second suggestion. “x
is red at one place” is substantially a relational expression, and as such, it supports two
existential inferences. The suggestion can be tested, then, by checking whether “fleep
(x)” has two existential consequences of its own. If not, then assent to (3) counts as the
final word on the matter.
There are analogous procedures to help linguists who wonder whether they have
been asking about rabbits or about stages thereof. Still other procedures can resolve
indecision as to whether a linguist has been asking about rabbits or about rabbit
parts, where the parts in question are spatio-temporal fragments of a big fusion
embracing all rabbits. And again, indecision between “rabbit” and “place where
rabbithood is manifested” can be resolved along similar lines: the rabbit moves; a
predicate is found which applies to its first position but not to the second; and an
appropriate variant of (3) is presented to the informant.
Further kinds of indecision, as between “rabbit” and some other term, are imagin-
able; and they call for slightly different techniques. Thus it is imaginable that
“gavagai” is a singular term meaning “the rabbit fusion,” and that the linguist has
been asking, in effect, “Is the rabbit fusion here the same as the fusion over there?”
The simplest approach here is of course to check whether “gavagai” belongs to the
category of singular terms. But if additional, independent evidence is wanted, it can
be obtained by the following procedure: Point to a red rabbit and a white one, asking
“Fleep?” each time. Any friend of the rabbit fusion can be expected to assent twice;
for, holding that both acts of ostension pertain to one and the same thing, a friend of
the fusion is bound to hold also that both acts pertain to a fleep-ish thing. To spare
such people the imputation of craziness, therefore, one would have to render their
“fleep” by “is partly red,” and their “gavagai” by “the rabbit fusion.”
It is also imaginable that “gavagai” is a singular term meaning “rabbithood,” and
that the linguist has been asking whether the rabbithood over here is the same as the
rabbithood over there (i.e., whether there are two rabbit universals or only one).
Insofar as this conceit leads only to indecision between “rabbit” and “rabbithood,”
the issue can be cleared up in about the same way as in the preceding case: essentially
the same pattern of assent can be elicited. The only difference is that in translating
“fleep,” it will be necessary to resort to a barbarism like “is partially coinstantiated
with redness.”
 “ GAVAGAI ” 29

Counterfeit identity predicates can be distinguished from the real thing, and
identity judgments which pertain to rabbits can be distinguished from judgments
which pertain rather to the rabbit fusion or to rabbithood. Thus, given that the exotic
people are partial to rabbits, their partiality can be recognized. Moreover, deviant
partialities can be distinguished from one another. Four deviant general terms have
been imagined: one applies to spatial parts of rabbits, another applies to temporal
stages, a third applies to spatio-temporal sections of the rabbit fusion, and a fourth
applies to places where rabbithood is manifested. Such terms can be distinguished by
experiments involving identity predicates. Someone partial to rabbit parts would
acquiesce in identity judgments which friends of stages, sections, and places would
reject; for he would hold that any twice-ostended part was one and the same object,
whereas the others would see a different object each time. (Strictly speaking, a friend
of places would agree with a friend of parts until the rabbit moved.) Again, someone
partial to stages would acquiesce in identity judgments (involving simultaneous
ostensions) that a friend of sections would reject, and would reject judgments that
friends of parts and places would accept. A friend of sections would be a pretty
negative fellow: he would reject all identity judgments except ones involving simul-
taneous ostensions directed at the same point. And so on.
We have also imagined exotic singular terms standing for rabbithood and for the
rabbit fusion. By any standard account of these entities, there are parts of the rabbit
fusion (e.g., the part consisting of your pet rabbit and the tail of mine), which do not
instantiate rabbithood. Once one such part is specified by a series of pointing gestures
and identity statements, a couple of questions about it will relieve indecision between
“rabbithood” and “the rabbit fusion.” And by the same token, the answers will
indicate whether informants are working with a relation sign for pairs of instances
and universals, or with one for pairs of parts and fusions. (Should there be non-
standard views about rabbithood or the rabbit fusion in the background, the problem
would call for more subtle techniques.)
When certain natural assumptions are satisfied, therefore, patient scrutiny can put
our linguist in touch with the extension or reference of “gavagai.” The major
assumptions are: that there is an exotic grammatical category similar to our category
of singular terms; that the exotic vocabulary is sufficiently rich to support a full-
fledged identity predicate; that exotic pointing gestures can be identified and correl-
ated uniquely with our own. Insofar as exceptions and borderline cases are imagin-
able, there is still room for a Quinean claim: not, perhaps, for the claim that reference
to rabbity entities is inscrutable, but certainly for the claim that such reference is local
to a proper subset of the range of conceptual schemes (cf. Quine (1960, p. 53)). On
the other hand, one would have to travel in thought to Mars or beneath the sea to find
exceptions; by and large, reference to rabbity entities is common coin.
Quine’s leading themes pertain less to rabbits than to the nature of scientific
theories; and as is well known, they do not imply his observations about “gavagai.”
Yet the example tends in its salience to divert attention from the themes. I hope that
this note will help to clear the air.
30 PART I : MEANING

References
Quine, W. V. (1960). Word and Object. Cambridge, MA: MIT Press.
Quine, W. V. (1969a). Ontological Relativity and Other Essays. New York: Columbia University
Press.
Quine, W. V. (1969b). “Reply to Strawson.” In D. Davidson and J. Hintikka (eds.), Words and
Objections. Dordrecht: D. Reidel Publishing Company.
Strawson, P. F. (1969). “Singular Terms and Predication.” In D. Davidson and J. Hintikka
(eds.), Words and Objections. Dordrecht: D. Reidel Publishing Company.

Postscript to “‘gavagai’”
In Section II I claim that it is necessary and sufficient for a dyadic predicate to count
as an identity predicate that it satisfy the following conditions: “First, informants
must be disposed to assent to every sentence formed by inserting two occurrences of
a singular term in its argument places. And second, when an informant is found
assenting to a sentence which contains the sign and two other terms, say ‘a’ and ‘b’,
and also to a sentence which contains ‘a’ (or ‘b’) at one or more places, he must be
disposed to assent to a third sentence which is like the second except for having
occurrences of ‘b’ in place of one or more occurrences of ‘a’ (or of ‘a’ in place of
occurrences of ‘b’).” This claim is close to the truth, but it is not quite true as it stands.
If an agent’s vocabulary has an impoverished stock of predicates, then a dyadic
predicate might satisfy the two conditions, as used by the agent, even if it is not used
by the agent to express identity. To see this, suppose that the universe of discourse
with which a certain agent is concerned is limited to persons, and that different
members of the same income group are indiscriminable with respect to the predi-
cates in the agent’s vocabulary. (That is, suppose that if a predicate applies to one
member of the universe of discourse, then it applies to every other member who has
the same income as the given member.) Suppose now that the agent introduces a new
predicate, “=,” and that the agent uses the predicate in such a way that it conforms to
the foregoing two conditions. Does the predicate mean is identical with? Not
necessarily. It might mean belongs to the same income group as instead. Thus,
suppose that it in fact has the latter meaning. Then, in the first place, the agent will
be disposed to assent to every sentence of the form “a = a,” for each such sentence
will just affirm the triviality that a given member of the universe of discourse belongs
to the same income group as himself. And second, the agent will inevitably assent to
“Fb” whenever he assents to “Fa” and to “a = b,” for the agent will see “a = b” as
meaning that a and b belong to the same income group, and will also see “Fa” as
attributing a property to a that is also possessed by every other member of the income
group to which a belongs. (See W. V. Quine. (1969). Set Theory and Its Logic, 2nd
edition. Cambridge, MA: Harvard University Press, p. 15.)
 “ GAVAGAI ” 31

To avoid this difficulty, it is necessary to strengthen the second condition in such a

way that it is concerned, not just with the predicates that an agent’s lexicon actually
contains, but also with possible extensions of the agent’s vocabulary as well. That is to
say, it must be strengthened so that it says, of a putative identity predicate “=,” that
the agent’s use of “=” conforms to Leibniz’s Law with respect to all of the predicates
that belong to his actual lexicon, and also that the agent would continue to honor
Leibniz’s Law, with respect to his use of “=,” if his lexicon were enriched in any way
by additional predicates. This condition is quite strong. It can be counted on to draw
a line between predicates that genuinely express identity and counterfeits that serve
simply to connect objects that happen to be lexically indiscriminable.
For an argument that a predicate counts as an identity predicate if it would
conform to Leibniz’s Law under all possible extensions of a lexicon, see Vann
McGee and Brian McLaughlin, Terrestrial Logic (Oxford: Oxford University Press,
forthcoming). (The argument at the end of a chapter entitled “Everything: Its Scope
and Limits.”) I am grateful to McGee for a helpful correspondence about this topic.
 3
Rudiments of a Theory
of Reference*

The last 15 years have witnessed a vigorous and extremely interesting debate
concerning two competing views about the nature of semantic reference. One of
these views is perhaps best described as a counterpart of the Redundancy Theory of
Truth. Its most basic component is the thesis that pairs of sentences like (1) and (2)
are more or less equivalent in point of assertional content.
(1) “Snow” refers to this stuff.
(2) This stuff is snow.
The other view claims that our term “refers” stands for an empirically manifest
relation that has causal and explanatory significance, and by implication it asserts
that there is a large gulf between the content of sentences like (1) and the content of
sentences like (2). Both views have been presented and compared in a number of
recent writings, including influential papers and books by Devitt (1981, 1984), Field
(1972), Friedman (1979), Leeds (1978), Putnam (1978), and Soames (1984).
The present chapter sketches a theory of reference which I believe to have the
merits of each of these views and the flaws of neither. In Section I I try to fix ideas by
describing the two views in greater detail and by presenting their main shortcomings.
Sections II and III set the stage for the theory I wish to recommend by describing two
semantic concepts that are employed by an imaginary linguistic community. Both
concepts are simpler in several respects than any of the concepts we actually employ,
but, as it turns out, there are some striking similarities between our thought and talk
about reference and thought and talk about reference in the imaginary community.
Building on these similarities, Section IV states the central hypothesis of my theory of
reference and presents some supporting arguments. Finally, in Section V I cite several
features of our concept of reference that lie beyond the scope of my central hypoth-
esis, and I argue briefly that it may be possible to extend the hypothesis in such a way
as to accommodate them.

* Over the years I have learned a great deal about reference in conversations with Ivan Fox, Anil Gupta,
and Hilary Putnam. I am also indebted to Gupta for advice and encouragement concerning an earlier
version of this paper. Finally, I have been helped considerably by the comments of an anonymous referee
for the Notre Dame Journal of Formal Logic.
 RUDIMENTS OF A THEORY OF REFERENCE 33

I
According to the first view, which will hereafter be called the Redundancy Theory of
Reference, it is possible to account for the most basic features of our use of “refers” by
supposing that we are guided in its use by schemata like (3) and (4).
(3) If t exists, then “t” refers to an object just in case the object is identical with t.
(4) “P” refers to an object just in case the object is (a) P.
As can be seen from the grammatical forms of (3) and (4), “t” marks a position that is
accessible only to singular terms, and “P” marks a position that is accessible only to
monadic general terms. There are similar schemata that represent sentences involv-
ing terms from a number of other grammatical categories. All such schemata have
come to be called Disquotation Schemata.
Advocates of the Redundancy Theory claim that instances of Disquotation Sche-
mata are partially constitutive of the concept of reference, and that coming to
appreciate this fact is the most fundamental step in acquiring the concept.
These initial insights can be developed in a number of different ways. One version
of the Redundancy Theory claims that there is no single purpose or semantical role
that underlies all uses of “refers.” This version explains simple contexts like (1) by
appealing to Disquotation Schemata, but it takes a quite different line in explaining
quantificational contexts like (5).
(5) Every word in this list refers to a large North American city.
It asserts that sentences like (5) are equivalent to sentences that have substitutional
quantifiers in place of “refers.” Thus, for example, it asserts that (5) is equivalent to
something like (6):
(6) ( t)(if “t” is a word in this list, then t is a large North American city)
where “(t)” is a substitutional quantifier. (It is, of course, part of this view that
substitutional quantifiers are capable of binding variables that stand within quotation
marks. Quantifying into quotation marks is thought to be problematic only when the
quantifier is objectual.)
There is another version of the Redundancy Theory which assigns a single role to
“refers.” It claims that all English discourse concerning the reference of English terms
can be explained by supposing that speakers of English are guided by definitions like
the following ones:
(7) x refers1 to y if and only if (t) (x = “t” and y = t)1
(8) x refers2 to y if and only if (P) (x = “P” and y is (a) P)

1
When I deploy objectual quantifiers and variables, I am assuming that their use is governed by a free
logic.
34 PART I : MEANING

where “(t)” and “(P)” are substitutional quantifiers. This claim meshes nicely with
the initial insights of the Redundancy Theory: it is natural to maintain that the concept
of reference is based on definitions like (7) and (8) while also maintaining that
instances of Disquotation Schemata are the core of the concept, for it is natural to
see definitions like (7) and (8) as summaries of instances of schemata like (3) and (4).
This brings us to the second of the two theories that I cited at the outset. According to
this view, which will hereafter be called the Empirical Relation Theory, the concept of
reference is more or less on a par with such concepts as the concept of valence. The concept
of reference stands for a component of the natural order: generalizations about reference
can have the status of laws of nature, and reference can play a role in causal explanations.
Champions of the Empirical Relation Theory usually acknowledge that they are
not yet in a position to capture all of the main features of reference in a set of general
principles. Rather it is their practice to cite several sample principles, and to claim
that others will follow after the scientific study of language has gained further
momentum. Here are several generalizations that either have been or might be
cited as examples by champions of the view:
(9) The terms in a mature science typically refer.
(10) If n is a proper name that can be used to refer to o at a given time t, then
normally o was dubbed with n in a name-giving ceremony that took place at some
time prior to t.
(11) If an individual knows that n refers to o, then normally others can use
sentences containing n to secure the individual’s help in projects that involve o
(provided that the individual has the desire and the opportunity to help).
(12) If the members of a group know that n refers to a place p, then normally they
can use sentences containing n to arrange to meet one another at p.
Champions of the Empirical Relation Theory claim explanatory significance for
generalizations of this sort, and it is clear that they are right to do so. Thus, suppose
that “Ben” is the name of a dog owned by my friend Michael. Suppose further that
Michael telephones me to say that Ben has run away, and to ask me to look for him in
my neighborhood. Suppose finally that at the time of the call I am ready, willing, and
able to come to Michael’s assistance. When these facts are combined with (11), we
have a (partial) explanation of my setting forth to find Michael’s dog, and of my
grabbing the dog by the collar when I find him trotting down my street.
Unfortunately, both of these views about reference have serious shortcomings.
Advocates of the Redundancy Theory are confronted with three main difficulties:
(i) The Theory fails to provide adequately for the fact that speakers of English can
make fully meaningful claims about the reference of terms in other languages. In the
hope of accommodating this fact, advocates of the Redundancy Theory have fre-
quently tried to account for the concept of reference that can be applied to expres-
sions in other languages by explaining it in terms of the concept of translation and
 RUDIMENTS OF A THEORY OF REFERENCE 35

the concept of reference that we use in talking about English terms. Thus, it has been
suggested that a sentence of the form “‘t’ refers to o in L,” where “L” stands for a
foreign language, is equivalent to something like “ ‘t’ as a term of L can be adequately
translated by an English term s such that s refers to o in English.” However, this
suggestion seems wrong. For one thing, the suggestion is probably circular, for it is
probably necessary to employ the concept of reference in analyzing the concept of an
adequate translation. Moreover, it fails to accommodate contexts of the following
form: “‘t’ refers to something in L; but I can’t specify the referent since there is no
equivalent term in English.” It is plain that such contexts exist. For example, where L
is a language spoken by theoretical physicists who are much more advanced than
their English-speaking counterparts, and x is a term that belongs to the upper-most
reaches of their theory, it would be true to say that x refers to something in L
that cannot be specified in English. But the suggestion implies that all such claims
are false.
(ii) The theory also fails to provide for the fact that speakers of English in the actual
world can make true claims about the reference of English terms in various coun-
terfactual situations. For example, the following sentence is true: “If we used ‘Kripke’
in the way that we actually use ‘Hesperus,’ then ‘Kripke’ would refer to a planet.”
Definitions like (7) and (8) wrongly imply that sentences of this sort are false.2
(iii) Although the Redundancy Theory accommodates the fact that we can make
true generalizations about reference, it does not do full justice to the intuition that
some of these generalizations are laws of nature. Presumably, if a generalization
about reference is a law of nature, then it should apply to all expressions that belong
to a given category. However, as is shown by objection (i), concepts based on
definitions like (7) and (8) are too narrow in scope to apply to all members of a
category. It follows that generalizations involving such concepts must inevitably be
limited in scope.
As for the Empirical Relation Theory, its chief failing is that it is unable to account
for the role that substitution instances of Disquotation Schemata (hereafter called
Disquotation Sentences) play in guiding our use of “refers.” Disquotation Sentences
are epistemically more basic than other propositions about reference: we take them as
fixed points in checking to see whether other propositions about reference are
acceptable. (Here, of course, I mean to except such other sentences as instances of
the following schema: “If t exists, then ‘t’ refers to t.” This schema is very closely
related to (3).) In particular, we are guided by Disquotation Sentences in assessing
generalizations like (9)–(12). Unfortunately, the Empirical Relation Theory cannot

2
To see that (7) implies that the given counterfactual is false, notice first that the counterfactual cannot
be (nonvacuously) true unless the sentence “ ‘Kripke’ refers to Hesperus” is true at some possible world w.
Second, observe that (7) implies that “ ‘Kripke’ refers to Hesperus” is true at w if and only if some sentence
of the form “ ‘Kripke’ is identical with ‘t’ and t is identical with Hesperus” is true at w. And finally, observe
that the first conjunct of a sentence of this form is true at w if and only if the second conjunct is false at w.
36 PART I : MEANING

explain this fact. It is unable to explain how we can know that an indefinitely large
number of particular propositions about reference are true before we come to accept a
body of general principles about reference.
The situation involving Disquotation Sentences appears to be much the same as
the situation involving instances of (13).
(13) “S” is true if and only if S.
Since Tarski’s classic papers (cf. Tarski (1949, 1956)) instances of (13) have played
the role of data in philosophical investigations of truth. Philosophers have recognized
that an analysis of truth is unacceptable unless it implies all (non-paradoxical)
instances of (13), and that a general principle about truth is wrong unless it is
compatible with them. In a phrase, philosophers have come to recognize the epi-
stemic primacy of instances of (13). A theory of truth must explain this primacy.
The point is not simply that there are facts which the Empirical Relation Theory
does not explain. If the problem were only this, then there would be room for hope
that it could be solved by supplementing the theory in some appropriate way. The
problem is rather that the Empirical Relation Theory appears to impede the explan-
ation of certain facts. If the concept of reference stands for an empirically manifest
relation, then it should be possible to account for acceptance of truths involving the
concept in more or less the same way as we can account for acceptance of truths
involving other concepts of empirically manifest relations. But this seems not to be
the case. Thus, consider the following propositions:
If 17,391,020,651 exists, then “17,391,020,651” refers to an object just in case the object is
identical with 17,391,020,651.
“Quark” refers to an object just in case the object is a quark.

I accept both of these propositions. It cannot be the case, however, that this
acceptance is based directly on perception; for I came to accept them without benefit
of perceptual contact with any of the extralinguistic entities with which they are
concerned. Nor is my acceptance based on a prior knowledge of general principles of
reference, for, as we noticed earlier, one’s acceptance of propositions of the given type
is more fundamental than one’s acceptance of any general principles of reference. (As
we noticed earlier, we assess general principles of reference by checking them against
the class of Disquotation Sentences.)
It appears that each of our two views about reference is able to do justice to a
dimension of the concept of reference that is ignored or even distorted by the other.
The Redundancy Theory recognizes the epistemic primacy of Disquotation Sen-
tences, but it fails to do justice to the interlinguistic scope and modal import of the
concept of reference, and it fails to account for the causal and explanatory force of the
concept. The Empirical Relation Theory acknowledges and explains the three fea-
tures of reference that call the Redundancy Theory into question, but it is unable to
 RUDIMENTS OF A THEORY OF REFERENCE 37

account for the epistemic primacy of Disquotation Sentences. In view of this config-
uration of offsetting strengths and weaknesses, it is natural to ask whether it is
possible to obtain a theory of reference that enjoys the advantages of the Redundancy
Theory and the Empirical Relation Theory by taking the average of their main
features. I think that the answer is “yes,” and I want to turn now to the task of
arguing that this answer is correct.

II
In order to formulate the theory I wish to recommend it is helpful to appeal to an
imaginary community whose members use a version of English that is much simpler
than ours. I will therefore begin by discussing the users of an imaginary language
called Senglish. Senglish is a language that is very much like ours in grammatical
structure and vocabulary; however, it differs from ours in that it lacks hopelessly
vague terms, indexical terms, and ambiguous terms.
Speakers of Senglish have a concept of reference that is introduced by the following
definitions:

(7) x refers1 to y if and only if (t) (x = “t” and y = t)

(8) x refers2 to y if and only if (P) (x = “P ” and y is (a) P)
where of course “(t)” and “(P)” count as substitutional quantifiers. Thus, this
Senglish concept literally has the properties that are attributed to the English concept
of reference by the Redundancy Theory.
Since (7) and (8) are components of a definition, it is appropriate to see them as
necessary truths, and it is also appropriate to hold that speakers of Senglish know
them to be true a priori. Moreover, the same is true of all Senglish instances of (3) and
(4). This is because all such instances can be derived from (7) and (8) when (7) and
(8) are supplemented with a few other truths that are both necessary and a priori.3
Now as soon as the concept defined by (7) and (8) is in hand, it is possible for
speakers of Senglish to state generalizations which connect the concept to concepts
that stand for various projectible properties. Many of these generalizations turn out
to be true, and they are called laws of reference. Examples include (9)–(12), for it
turns out that the laws of reference for Senglish are quite similar to the laws of
reference for English. Here are several additional examples:

3
The additional truths are identity axioms of the standard sort and the following two principles:
For all t1 and for all t2, if “t1” = “t2,” then for every object y, y = t1 if and only if y = t2.
For all P1 and for all P2, if “P1” = “P2,” then for every object y, y is (a) P1 if and only if y is (a) P2.
Here all four of the initially placed quantifiers are substitutional, and the two internal quantifiers are
objectual.
38 PART I : MEANING

(14) Normally, if n is a proper name that refers to o, then speakers who have
mastered the use of n can use sentences containing n to call one another’s
attention to their propositional attitudes about o.
(15) Normally, if n is a proper name that refers to o, then there is a set of
sentences a such that: (i) the members of a contain n, and (ii) facts involving o
are among the causally necessary conditions of speaker’s acceptance of some of the
members of a.4
(16) Normally, if n is a proper name that refers to o, then speakers who have an
interest in o make an effort to learn and remember n.5
(17) Normally, if n is a proper name that refers unambiguously to o, S is a synthetic
sentence that contains n, i is an individual who has mastered the use of n, and i
has had considerable perceptual contact with o, then, provided that i has no other
information that is germane to S (such as well-confirmed beliefs that imply it), i will
consult memories of o in trying to decide whether or not to accept S.
(18) Normally, if P is a monadic general term that unambiguously refers to the
members of a, and a is a natural kind of perceptible objects, then there is a group
of speakers who have the ability to recognize members of a, and other speakers
tend to defer to members of this group in determining whether to accept contin-
gent sentences that contain P.6
(19) Normally, if P is a monadic general term that unambiguously refers to
members of a, a is a class of theoretical entities, i is an individual who has mastered
the use of P, and S is a synthetic sentence containing P, then, provided that i has no
other information that is germane to S (such as well-confirmed beliefs that imply
it), i will consult observable events that i believes to be caused by members of a in
trying to decide whether or not to accept S.
(20) Normally, if P is a monadic general term that refers unambiguously to the
members of a class a, there are sentences containing P such that speakers’
responses to those sentences vary with the distribution of certain features within
the boundaries of members of a (in the way in which, e.g., responses to tokens
of “That rabbit is white” vary with the distribution of whiteness within the
boundaries of ostended rabbits).7
(21) Normally, if P is a monadic general term that unambiguously refers to the
members of a, a is a class of perceptible objects, i is a cooperative individual who
has mastered the use of P, and S is Quinean occasion sentence that is obtained
from P, then i will assent to a token of S accompanied by an ostensive gesture if the
gesture ostends a member of a.
(22) Normally, if P is a monadic general term, a is a class of perceptible objects, S
is a Quinean occasion sentence obtained from P, and individuals are disposed to

4 5
See Evans (1973). See Evans (1982, p. 379).
6 7
See Putnam (1978). See Evans (1975).
 RUDIMENTS OF A THEORY OF REFERENCE 39

assent to a token of S when a member of a is ostended and to dissent from a token

when a member of some other class is ostended, then either: (a) P refers to the
members of a, (b) P refers to the members of some other class that individuals are
unable to distinguish from members of a on the basis of unsupplemented obser-
vation, or (c) individuals tend to construe pointing gestures directed at members
of a as cases of deferred ostension.
There are two points about laws like (9)–(12) and (14)–(22) that should be empha-
sized. First, they have a variety of epistemological characteristics. While some link the
concept defined by (7) and (8) to concepts that pick out forms of behavior and other
observable phenomena, others link it only to concepts that are largely or even entirely
theoretical. Second, the examples given thus far are only a tiny fragment of the set of
all laws of reference, and there are laws quite different than any of (9)–(12) and (14)–
(22). (For example, there are laws of reference that express correlations between
semantic facts involving numerals and certain formal facts involving numerals.)
The laws of reference provide a foundation for a second concept of reference. After
the first concept has been acquired, and the existence of laws of reference has been
noted, speakers of Senglish begin to appreciate that it would be useful to have a
concept of reference that would enable them to say the same kinds of things about
foreign terms as the first concept enables them to say about domestic terms. Accord-
ingly they introduce a second concept by a definition that comes to this:

(23) A set of ordered pairs R is a reference relation for a language L in a possible

world w iff the left field of R consists of terms that belong to L in w, and it is true in
w that R has the projectible properties that the Senglish laws of reference attribute
to the relation (i.e., to the set of ordered pairs) that is defined by (7) and (8).
Since this second concept owes its existence to laws that can only be seen to hold on the
basis of experience, it can appropriately be called the a posteriori Senglish concept of
reference. The first concept can appropriately be called the a priori Senglish concept.

III
These concepts have several features that should be considered before we go on to
look at other matters.
First, (23) contains two terms that seem prima facie to carry a lot of semantic
weight, namely, “define” and “attribute.” It might be held that their presence
trivializes the project of defining reference, or that reference is somehow invoked
by one or the other of them, and that (23) therefore introduces a circularity into the
Senglish conceptual scheme. However, these worries are groundless. When the terms
are used with the senses that they have in (23), they can be eliminated in favor of
expressions that are ultimately reducible to primitives with no semantic significance.
To see this, notice that it is possible to obtain an equivalent definition by replacing “R
has the properties in w that the Senglish laws of reference attribute to the relation
40 PART I : MEANING

defined by (7) and (8)” with “R satisfies the Senglish laws of reference in w under the
interpretation that correlates ‘refers’ with R and all other terms with their normal
referents.” Here “normal referent” is to be seen as an expression that is explained by (7)
and (8), and “satisfies” is to be seen as a term that is reducible via a Tarskian definition
to a set of expressions that contains no semantic terms beyond “normal referent.”
Assuming that the laws of reference are finitely axiomatizible, there is a second way
of expressing the sense of (23) without using “define” and “attribute.” First, speakers of
Senglish would need to change the laws of reference by replacing every open sentence
of the form “v1 refers to v2” with an open sentence of the form “v1 bears R to v2 as used
by speakers of language v3 in possible world v4,” where “v3” and “v4” are individual
variables that do not occur anywhere in the original laws of reference, and “R” is a
variable that ranges over quaternary relations (considered as relations in intension). To
complete this step, it would be necessary to prefix universal quantifiers that bind “v3”
and “v4” to each of the laws. Second, they would have to form a single open sentence
that captured the content of the sentences that resulted from the first transformation.
This would be done by simply taking the conjunction of the axioms that summarize the
laws of reference. When they had completed these steps, speakers of Senglish would be
in a position to replace (23) with a new definition. Thus, if “A (R)” was the conjunction
of the axioms, they could define the a priori concept of reference as follows:
x refers to y as used by speakers of L in a possible world w iff x is a term of L in w and there exists
a quaternary relation R (i.e., a relation in intension) such that A(R) and x bears R to y and L in w.

In short, speakers of Senglish could obtain a set of open sentences that implicitly
defined the a posteriori concept by simply replacing a term in the laws of reference
with a variable that ranged over relations, and they could then convert this implicit
definition into an explicit one by expressing the laws as a conjunction of axioms and
binding the new variable with an existential quantifier. This would enable them to do
without “define” and “attribute.”8
This brings us to the second fact that should be considered. Although it is extremely
convenient to use substitutional quantifiers in defining a priori reference, it seems
that, strictly speaking, it is not necessary to do so. If Senglish had lacked substitutional
quantifiers, it would have been possible to gain the effect of (7) and (8) by first defining
two concepts of primitive a priori reference in terms of disjunctive lists, and by then
defining two concepts of derived a priori reference by recursion from the primitive
concepts. Thus, where “n1,” “n2,” . . . , and “nm” represent all of the proper names of
Senglish, and “P1,” “P2,” . . . , and “Pj” represent all of the logically simple monadic
general terms, primitive a priori reference could have been defined as follows.
x primitively refers1 to y iff either x is identical with “n1” and y is identical with n1 or . . . or x is
identical with “nm” and y is identical with nm.9

8
Here I am indebted to Lewis (1983).
9
This construction derives ultimately from Tarski, but I am also indebted to a paper by Field. See Tarski
(1956) and Field (1972).
 RUDIMENTS OF A THEORY OF REFERENCE 41

x primitively refers2 to y iff either x is identical with “P1” and y is (a) P1 or . . . or x is identical
with “Pj” and y is (a) Pj.

The first of these biconditionals is in effect what we get when we restrict (7) to
singular terms that are logically simple, and the second is what we get when we
restrict (8) to simple monadic general terms. As for derived a priori reference,
limitations of space and knowledge make it impossible for me to provide a detailed
account of it here. However, perhaps it is possible to convey the idea I have in mind
by stating several of the simplest components of a definition. Here are four such:
If x is a logically simple singular term, then x derivatively refers1 to y iff x primitively refers1 to y.
If x is a logically simple monadic general term, then x derivatively refers2 to y iff x primitively
refers2 to y.
If x is a monadic general term that is obtained by inserting “or” between two other monadic
terms z and w, then x derivatively refers2 to y iff either z derivatively refers2 to y or w
derivatively refers2 to y.
If x is a monadic general term that is obtained by prefixing “non-” to a monadic term z, then x
derivatively refers2 to y iff z does not derivatively refer2 to y.

In addition to these clauses, an adequate definition would contain clauses for all other
ways of forming logically complex referring expressions from logically simple ones.
The definition would be extremely long and extremely complex, and our current
ignorance of the syntactic structures of natural languages makes it impossible to
envision all of its details. On the other hand, since there seem to be good reasons for
holding that the class of referring expressions of a natural language is recursively
specifiable, we apparently have the right to hold that it is in principle possible to
construct a recursive definition of the right sort.10
Third, despite the fact that Senglish contains a number of dyadic general terms and
also a number of general terms of higher adicities, neither the a priori concept of
reference nor the a posteriori concept applies to such terms. However, it would be
easy to fill these lacunae in Senglish semantics. The scope of the a priori concept
could easily be extended by supplementing (7) and (8) with new definitions of the
same type. Each new definition would bring new laws of reference in its wake.
Fourth, the a posteriori concept introduced by (23) is best seen as a concept of
primitive reference, that is, as a concept that applies only to terms that are compo-
nents of the lexicon, terms that are logically simple. Unlike logically simple terms,
which cannot be said to belong to a language unless they are put to use in commu-
nication from time to time, logically complex terms are not always put to use in
communication. This follows from the fact that every language contains an infinite
number of logically complex terms. Further, the laws of reference fail to pin down the

10
The main ideas in the second half of the paragraph derive from Putnam’s first John Locke Lecture.
See Putnam (1978, pp. 9–17).
42 PART I : MEANING

reference of terms that are never used. They state necessary (or nearly necessary)
conditions of reference and sufficient (or nearly sufficient) conditions for many
terms, but such conditions always involve features of the use of terms or features
of psychological states that underlie use. Hence, they cannot be said to state condi-
tions of reference for terms that never get used. And by the same token, since the
foregoing definition is based on the laws, it cannot provide information about what it
means to say that an unused term refers to something.
If they wanted an a posteriori concept that applied to complex terms as well as to
simple ones, speakers of Senglish could erect a recursive definition of a new a posteriori
concept on the foundation provided by the primitive a posteriori concept. That is to
say, they could construct a recursive definition with the following properties: the basis
clause would be something like (23), and each of the recursion clauses would corres-
pond to one of the standard ways of forming more complex terms from simpler ones.
Fifth, the definition of the a posteriori concept implies that the words of a language
will refer to different things in different possible worlds. Thus, according to the
definition, reference depends upon the way in which the words of a language are
used. Since the terms of the language are used in different ways in different worlds,
reference can vary from world to world.
Sixth, (23) implies that the laws of reference hold no less for the a posteriori
concept than for the a priori concept. In other words, it implies that there are two sets
of laws of reference. The sets are just alike except that one contains a predicate that
expresses the former concept and the other contains a predicate that expresses the
latter concept.

IV
This brings us to the theory of reference that I wish to recommend. According to the
central hypothesis of this theory (which is hereafter called the Similarity Hypothesis),
speakers of English (and other actual natural languages) have two concepts of
reference, and the concepts are similar in content and in logical properties to the
two Senglish concepts that are discussed in Section II.
Like many other hypotheses about portions of our conceptual scheme, the Simi-
larity Hypothesis does not claim that the structures it postulates are explicitly
represented at the level of conscious awareness. Thus, the Hypothesis is compatible
with the fact that we are at best tacitly or marginally aware that we have two concepts
of reference, and also with the fact that we do not explicitly formulate definitions like
(7), (8), and (23) in everyday life. The Hypothesis is not to be tested by trying to gain
introspective access to our semantic concepts, but rather by determining whether it
provides the best explanation of the main features of the thought and talk in which
our semantic concepts are manifested.
We have already taken note of four features of our thought and talk that have
caused problems for other theories of reference. They are as follows: (i) we are able to
make true statements about the reference of terms in other languages; (ii) we are able
 RUDIMENTS OF A THEORY OF REFERENCE 43

to make true statements about the reference of our own terms in various counter-
factual situations; (iii) we are able to use generalizations about reference in giving
causal explanations of psychological states and behavior; and (iv) Disquotation
Sentences are epistemically more fundamental than any general principles of refer-
ence. We can test the Similarity Hypothesis by checking to see whether it provides an
adequate basis for explaining these four features.
It is clear that the Hypothesis can handle (i)–(iii) quite nicely: the Hypothesis
asserts that we have an a posteriori concept of reference with roughly the same
content as the Senglish concept that is based on (23), and it is clear that the latter
concept can be put to use in making statements about terms in languages other than
Senglish, in making statements about the properties of Senglish terms in counter-
factual situations, and in making statements that figure in causal explanations of
Senglish psychological states and Senglish behavior.
As for (iv), recall from Section II that Senglish contains two sets of Disquotation
Sentences that are instances of (3) and (4), and that the members of these sets follow
immediately from the definitions that introduce the Senglish a priori concept of
reference. Second, notice that if the Similarity Hypothesis is true, then speakers of
English have an a priori concept of reference with the same properties as the Senglish
a priori concept. In view of these two facts, we can see that the Similarity Hypothesis
predicts that speakers of English will be able to recognize the truth of Disquotation
Sentences before they recognize the truth of general principles of reference.11
It appears, then, that the Similarity Hypothesis is able to explain (i)–(iv), and that it
is to be preferred both to the Redundancy Theory and the Empirical Relation Theory.

V
Although there are a number of striking similarities between the portion of our domestic
conceptual scheme that is concerned with reference and the corresponding portion of
the Senglish conceptual scheme, there are five important features of the former that are
not possessed by the latter. In the present section I will attempt to develop and refine the
Similarity Hypothesis in such a way as to accommodate one of these five features.

11
There is a curious feature of the Similarity Hypothesis that may be worth pointing out: it provides two
explanations of the epistemic primacy of Disquotation Sentences.
If the Similarity Hypothesis is correct, then Disquotation Sentences are ambiguous. This is because
Disquotation Sentences contain the term “refers,” and because the Similarity Hypothesis implies that we
have two different concepts that can be expressed by this term.
According to the argument just given, the Hypothesis predicts that Disquotation Sentences will enjoy
epistemic primacy when “refers” is used to express the a priori concept of reference. But it also predicts that
Disquotation Sentences will enjoy epistemic primacy when “refers” is used to express a posteriori
reference. This follows from two facts: first, Disquotation Sentences are implied by the English counterpart
of (23) no less than by the English counterparts of (7) and (8); and second, in order to adopt a definition
like (23), it is not necessary to know any particular laws of reference (one need only have reason to believe
that it will eventually be possible to find some laws).
44 PART I : MEANING

Moreover, while a full discussion of the four remaining features lies beyond the scope of
the present chapter, I will urge briefly that an advocate of the Similarity Hypothesis may
well be able to deal with the other features in a satisfactory way.
We have a concept of reference with a modal dimension that is foreign to both of
the Senglish concepts. There are two senses in which a term that is used by a
linguistic community in a possible world w1 can be said to refer to an object in
another possible world w2. (i) A term that is used in w1 can be said to refer to an
object in w2 in virtue of semantic properties that it has in w1. Thus, for example, it is
one of the semantic properties of the English term “red” that it expresses the property
being red in the actual world. In virtue of this fact, “red” can be said to refer to objects
that have the property in other possible worlds. (ii) A term that is used in w1 can be
said to refer to an object in w2 in virtue of semantic properties that it has in w2. To
continue the example, since “red” is used in some other possible worlds to express
the property being blue, it can be said to refer to objects that are blue in other worlds.
As we noticed in Section III, definition (23) authorizes speakers of Senglish to say
that their words refer to objects in other possible worlds. However, it only authorizes
them to make claims of this form when “refers to objects in other worlds” is used in
the second of the foregoing senses. Speakers of Senglish are unable to make such
claims when “refers to objects in other worlds” is used in the first sense. In short, the
Senglish definitions make no provision for intermundane reference.
In attempting to refine the Similarity Hypothesis so as to accommodate inter-
mundane reference, I will first introduce a concept that can be used in discussing the
intermundane reference of proper names. Let “A” stand for the set of ordered pairs
〈x, y〉 such that: (i) x is English proper name, and (ii) x refers1 to y (where “refers1” is
understood to have the sense that it is given in definition (7) earlier). Further, let “B”
stand for the set of all laws of reference that are satisfied by A. Using these two terms,
it is possible to define a concept of intermundane reference as follows:

(26) A set of ordered pairs R is a name-reference relation of a language L in a

possible world w iff the left field of L consists of the words that are the proper
names of L in w and it is true in w that R has the projectible properties that the
members of B attribute to the relation A.
(27) x name-refers to y at world w2 as used by speakers of L in world w1 iff the
pair 〈x, y〉 is in the name-reference relation of L in w1 and y exists in w2.
Of course (27) presupposes the view that a proper name refers to the same object in
all possible worlds. If this view should turn out to be wrong, it would be necessary to
account for the intermundane reference of names in some other way.
The task of accounting for the intermundane reference of monadic general terms is
a bit more complex. It has several parts. (i) It is necessary to begin by explaining what
it means to say that an English monadic general term connotes a property. Let us
suppose that the operator “being an object such that” can be used to convert monadic
 RUDIMENTS OF A THEORY OF REFERENCE 45

open sentences into names of properties. This assumption authorizes us to construct

the following definition:
An English monadic open sentence x connotes the property y if and only if (S) (x is identical
with “S” and y is identical with being an object such that S)

where “(S)” is a substitutional quantifier. Once it has been explained what it is for
an English monadic open sentence to connote a property, it is possible to explain
what it is for an English monadic general term to connote a property. The definition
runs as follows:

(28) An English monadic general term x connotes the property y if and only
if there is a monadic open sentence S such that: (a) x is a general term in the
predicate of S, (b) there is no general term in the predicate of S of which x is a
proper part, and (c) S connotes y.
(ii) The next step is to bring a certain fact into focus, namely, the fact that there is a
set of laws that link the concept defined by (28) with other concepts. The members of
this set (hereafter called “C”) are similar in many respects to the laws of reference.
Here are two examples:
Normally, if it is true (a) that x is a monadic general term, (b) that y is a property, (c) that
individuals who have mastered the use of x are disposed to use sentences containing x when
they wish to induce others to take a certain attitude or to behave in a certain way toward things
that they believe to have y, and (d) that they would also be so disposed in possible situations in
which y is not believed to be coextensive with the properties with which it is believed to be
coextensive in the actual world, then x connotes y.
Normally, if x is a monadic general term that unambiguously connotes y, y is a theoretical
property, i is an individual who has mastered the use of x, and S is a synthetic sentence
containing x, then, provided that i has no other information that is germane to S (such as well-
confirmed beliefs that imply it), i will consult observable events that i believes to be caused by
instances of y in trying to determine whether or not to accept S. Moreover, i would also consult
such events in possible situations in which y is not believed to be coextensive with the
properties with which it is believed to be coextensive in the actual world.

(iii) The third step is to explain what it means to say that a property is connoted by a
monadic general term in some language other than English or by an English monadic
general term in some possible world other than the actual world. It is possible to
explain statements of these two kinds by the following definition:

(29) A set of ordered pairs R is the connotation relation of a language L in a

possible world w iff the left field of R consists of the words that are the monadic
general terms of L in w and R has the projectible properties in w that the laws
belonging to C attribute to the relation D.
(Here “D” is a term for the relation picked out by (28).) (iv) Finally, it is necessary to
state a definition like (30).
46 PART I : MEANING

(30) x refers2 to y at world w2 as used by speakers of L in world w1 iff there exists a

property z such that: (a) the pair 〈x, z〉 is in the connotation relation of L at w1,
and (b) y has z in w2.
(30) gives us the desired concept of intermundane reference for monadic general
terms.
In order to obtain a concept of intermundane reference that applies to descrip-
tions, one must provide a series of recursive definitions that are based on concepts of
intermundane reference for more basic linguistic categories. Thus, for example,
before one can obtain a concept that applies to descriptions, one must supplement
the foregoing definitions with a definition that applies to dyadic general terms.
We are now in a position to consider a version of the Similarity Hypothesis that is
more precise and has greater explanatory power than the version with which we
started. According to this version, we have two concepts of reference. One, which
may be called the a priori concept, is based on (7) and also on the following definition.

(31) x refers2 to y iff x is a monadic general term and there exists a property z
such that x connotes z and z is instantiated by y.
(The sense of “connotes” that is relevant to (31) is the sense that is introduced by
(28).) Second, there is a concept (hereafter called the a posteriori concept) that is
based largely on laws that involve the a priori concept of reference and the a priori
concept of connotation. In other words, a large part of its content is given by
definitions (27) and (30).
At the beginning of this section I mentioned that the portion of our domestic
conceptual scheme that is concerned with reference has five important features that
are not shared by the corresponding portion of the Senglish conceptual scheme. We
have found that it is possible to account for one of these features by changing and
supplementing our first version of the Similarity Hypothesis. It is time now to
consider the four remaining features.
In the first place, unlike speakers of Senglish, we make use of indexical expressions
and are able to ascribe reference to them. What are the features of our concept of
reference that support such ascriptions?
It is not my intention to offer an answer to this question that is in any sense final or
complete. I wish only to point out that it is possible for an advocate of the Similarity
Hypothesis to supplement the Hypothesis with principles like these:

(32) A token of “I” refers to i iff i is the individual who produced the token.
(33) A token of “you” refers to i iff i is the intended audience of the individual
who produced the token.
(34) A token of “here” refers to p iff p is the place at which the token was produced.
(35) A token of an expression consisting of “this” followed by a monadic
general term refers to o iff o is an object such that: (a) the monadic term refers
to o, and (b) o is made especially salient by the previous discourse, or by an
 RUDIMENTS OF A THEORY OF REFERENCE 47

indexical gesture, or by o’s relationships to other entities within the nonlinguistic

context.
These principles are at best a fragment of a theory of indexicals. A full treatment
would require additional principles concerning English indexicals, principles con-
cerning foreign indexicals, and also a detailed account of the various forms of salience
that are presupposed by principles like (35). It is fairly clear, however, that it would be
possible to extend the list of principles without encountering any substantial diffi-
culties. Further, there is no reason to think that the project of analyzing salience
would require concepts or principles that are unavailable to an advocate of the
Similarity Hypothesis.
Second, we differ from speakers of Senglish in that we are able to ascribe reference
to singular terms that are in some sense vacuous. Even though there is no such thing
as Achilles, we can say, truly, that “Achilles” refers to Achilles. How is this possible?
Further, how can it be possible to say that “Achilles” refers to the slayer of Hector, or
that “`åغºı” refers in Greek to Achilles?
Here I wish to claim only that an advocate of the Similarity Hypothesis is not
ex officio excluded from developing a position that has some plausibility. An
advocate can, for example, avail himself or herself of the following view: if S is a
true sentence containing a singular term for an imaginary object, then, normally, it is
appropriate to see S as an abbreviation of a sentence like “It is true in fiction that S” or
“It is true in the world of make believe that S.” This view, which has of course enjoyed
considerable popularity among philosophers, encourages us to hold, among other
things, that “ ‘Achilles’ refers to Achilles” is an abbreviation of “It is true in the world
of Homer that ‘Achilles’ refers to Achilles.”
It is possible to apply the view to sentences that ascribe reference to vacuous terms
without insisting on a uniform interpretation of the occurrences of “refers” that are
found in such sentences. If the term to which reference is ascribed belongs to a foreign
vocabulary, then “refers” must be seen as expressing our a posteriori concept of
reference. On the other hand, if it belongs to our domestic vocabulary, then “refers”
may be seen as expressing either our a posteriori concept or our a priori concept.
I do not wish to suggest that the “invisible operator” approach is the only way or
even the best way of bringing the Similarity Hypothesis into line with the fact that we
can ascribe reference to terms that are in some sense vacuous. There are several
interesting alternatives, and one may turn out to be preferable to the invisible
operator approach.
Third, we are able to ascribe reference to terms with semantic content, that is, to
terms that contain “refers” and other semantic expressions. For example, in English
it is fully meaningful to say “ ‘The referent of “George Washington”’ refers to George
Washington.” However, our talk of this kind is known to lead to paradox (e.g., the
Berry Paradox). At some point, a theory of reference must face this problem. Perhaps
a theory should also suggest ways of solving the problems that are posed by the other
semantic paradoxes.
48 PART I : MEANING

The options that are available at this point to an advocate of the Similarity
Hypothesis are quite different than the options that are available to advocates of
other pictures of reference. However, I see no reason to think that they are less
promising. We are not in a position today to conclude that a theory of reference is
wrong because it conflicts with a theory of the paradoxes that we know to be correct.
Rather our position is just the opposite. It seems that our best hope of understanding
the paradoxes is to examine theories that purport to capture the point or the content
of the concept of reference.12
Fourth, we have thus far failed to come to grips with referential ambiguity.
A proper name can refer to two or more individuals, and a monadic general term
can have several disjoint extensions. The foregoing definitions fail to take these facts
into account.
Fortunately, while referential ambiguity shows that we do not yet have an adequate
version of the Similarity Hypothesis, it does not necessarily constitute an insur-
mountable obstacle. Thus, for example, it may well be true that if a term is charac-
terized by an n-fold ambiguity at the level of reference, then it is also characterized by
an n-fold ambiguity at the level of assertibility conditions (in the sense that some
contingent sentences containing the term are governed by n distinct sets of assert-
ibility conditions). If this proposition is true, then it is possible to accommodate
referential ambiguity by relativizing reference to sets of assertibility conditions.
Equally, we can accommodate it by maintaining that it is not a term taken alone
that has reference, but rather a term qua indexed by a set of assertibility conditions.
I have already mentioned several reasons for thinking that the theory of reference
presented in these pages is worthy of further attention. I would like to conclude by
mentioning one additional reason, namely, that the theory seems to be philosophic-
ally fruitful. It has implications concerning a number of important problems in the
philosophy of language, and these implications seem on the whole to agree with fairly
vivid intuitions. (For example, it provides a basis for responding to certain counter-
intuitive skeptical views about the uniqueness of schemes of reference.13)

12
Although the range of options associated with the Similarity Hypothesis is in some respects narrower
than the ranges associated with other pictures of reference, it is wider in certain other respects. This can be
illustrated by a passage in a familiar paper on substitutional quantification by Marcus (see (Marcus 1972)).
Marcus cited a definition of a truth predicate that can be expressed as follows:
x is true iff for some S, x is identical with “S” and S.
In discussing this definition, Marcus pointed out that it is necessary to impose certain restrictions on its
quantifier (which is of course a substitutional quantifier) in order to satisfy the requirements of adequacy
for definitions. Further, she in effect suggested that if the truth predicate we use in English was based on a
definition of this sort, then the paradoxes of truth could be seen as a side effect of unwitting violations of
these restrictions. In view of the evident similarities between Marcus’s definition and several of the key
components of the Similarity Hypothesis, there is reason to think that this second observation could be
combined with the Similarity Hypothesis to obtain a partial solution to the paradoxes of reference.
13
One can obtain a fairly strong objection to such views by combining the ideas in the present chapter
with the ideas in Hill (1985).
 RUDIMENTS OF A THEORY OF REFERENCE 49

References
Devitt, M. (1981). Designation. New York: Columbia University Press.
Devitt, M. (1984). Realism and Truth. Princeton, NJ: Princeton University Press.
Evans, G. (1973). “The Causal Theory of Names,” Aristotelian Society, Supplementary Volume,
47: 187–208.
Evans, G. (1975). “Identity and Predication,” The Journal of Philosophy 72: 343–63.
Evans, G. (1982). The Varieties of Reference. New York: Oxford University Press.
Field, H. (1972). “Tarski’s Theory of Truth,” The Journal of Philosophy 69: 347–75.
Friedman, M. (1979). “Truth and Confirmation,” The Journal of Philosophy 76: 361–82.
Hill, C. (1985). “Animadversions on the Inscrutability Thesis,” Pacific Philosophical Quarterly
66: 304–13.
Leeds, S. (1978). “Theories of Reference and Truth,” Erkenntnis 13: 111–29.
Lewis, D. (1983). “How to Define Theoretical Terms.” In his Philosophical Papers, vol. 1.
New York: Oxford University Press, 78–95.
Marcus, R. (1972). “Quantification and Ontology,” Nous 6: 240–50.
Putnam, H. (1978). “The Meaning of Meaning.” In his Language and Reality (Collected Papers,
vol. 2). Cambridge: Cambridge University Press.
Putnam, H. (1978). Meaning and the Moral Sciences. Boston, MA: Routledge Kegan Paul.
Soames, S. (1984). “What is a Theory of Truth?” The Journal of Philosophy 81: 411–29.
Tarski, A. (1949). “The Semantic Conception of Truth.” In H. Feigl and W. Sellars (eds.),
Readings in Philosophical Analysis. New York: Appleton Century Crofts, 52–84.
Tarski, A. (1956). “The Concept of Truth in Formalized Languages.” In his Logic, Semantics,
and Metamathematics. New York: Oxford University Press, 152–278.

Postscript to “Rudiments”
Some of the laws of reference cited in “Rudiments” presuppose that beliefs and other
propositional attitudes are specifically concerned with, or about, objects and prop-
erties in the world. This is true, for example, of law (14):
(14) Normally, if n is a proper name that refers to o, then speakers who have
mastered the use of n can use sentences containing n to call one another’s attention
to their propositional attitudes about o.
Other such laws include (16) and the two laws of connotation in Section IV.
It might be thought that laws of this sort presuppose a notion of reference that
applies to concepts. Moreover, someone who has this thought might worry that there
is something regressive about presupposing conceptual reference in explaining the
reference of words. This is a valid worry, but there are two considerations that should
assuage it. First, there are many, many laws that do not presuppose semantic notions.
Accordingly, it would be possible to jettison the ones that do without affecting any of
the essential features of the proposal. Second, as the next two chapters will make
50 PART I : MEANING

clear, it is possible to define the semantic notions that apply to propositional

attitudes and concepts explicitly in terms of substitutional quantification, and to
characterize the relevant substitutional quantifiers exhaustively in terms of their
conceptual roles. It follows that there is no regress of semantic concepts, even if
laws like (14) are retained.
 4
A Substitutional Theory of Truth,
Reference, and Semantic
Correspondence*

I will begin by sketching a deflationary theory of truth that I present in greater detail
in an earlier work, Thought and World (Hill 2002). The principal target of the theory
is the concept of propositional truth—that is, the concept that figures in claims like
the proposition that snow is white is true. The theory that I will sketch is deflationary
in that its centerpiece is a logical concept, the notion of substitutional quantification.
I hope the present sketch will make the theory more accessible.
After presenting this substititional account of truth, I will show how to extend it so
as to encompass semantic relations. My main targets will be reference and semantic
correspondence, construed as a relation between propositions and facts or states of
affairs; but as I will show, it is possible to generalize the account I will give of these
two relations to include others, such as the relation that a general concept bears to the
property it expresses. Since my analysis of semantic relations is based largely on
substitutional quantification, it can be combined easily with the earlier account of
truth to provide a fully general theory of truth conditional semantic properties. I call
the resulting theory semantic substitutionalism, or substutionalism for short.
Now if substitutionalism is sound, our truth conditional semantic concepts are
thin concepts. Most of their content derives from substitutional quantification, which
(as I will argue) is closely related to ordinary objectual quantification, and therefore
counts as a logical device. Many people, however, have thought that concepts like
truth, reference, and semantic correspondence are thick concepts, possessed of a
robust share of empirical meaning, and this has led them to reject deflationary
theories like substitutionalism. After presenting substitutionalism, I will consider this
opposing view at some length. In some cases, it derives from a desire to “naturalize”

* This chapter is largely a composite of the two contributions I made to a symposium on my book
Thought and World ((Hill 2006a) and (Hill 2006b)). The first two sections of the present chapter
correspond to a précis of Thought and World that opened the symposium, and the remaining sections
correspond to parts of my replies to my fellow symposiasts, Marian David, Anil Gupta, and Keith
Simmons. I am grateful to these philosophers for their illuminating comments, and also to Gupta for
reviewing drafts of the précis and my replies.
52 PART I : MEANING

concepts and propositions by reducing them to more basic phenomena, such as words
and sentences in a language of thought, and/or a desire to naturalize semantic
properties and relations by reducing them to properties and relations that are mani-
festly empirical. In other cases, the objection derives from the perception that semantic
concepts play essential and substantial roles in laws of nature, and therefore in the
explanations of empirical phenomena that the laws support, together with the com-
panion view that this would be impossible unless semantic concepts were substantive
concepts with a full share of empirical meaning. I will argue that these sources of the
opposing view fail to provide it with an adequate rationale. To be sure, substitution-
alism makes no effort to “naturalize” semantic relations: it makes no attempt to reduce
them to empirical relations like causation and information. As we will see, however, it
undercuts the perception that there is a need for reduction of semantic relations by
showing that they can be fully explained in terms of notions that are transparently
legitimate. It removes the main grounds for metaphysical or epistemological concern
about the relations. Moreover, substitutionalism is fully capable of accounting for the
role that semantic concepts play in laws of nature, and therefore in the explanations
that invoke those laws.
There are other sources of the idea that semantic concepts are empirically robust
that I will not be able to discuss in the present essay. One of these additional sources
is the perception that we can best explain the intuitions that motivate semantic
externalism, such as the intuition that the ability to refer to water presupposes that an
agent stands in a causal or informational relation to actual samples of water, by
supposing that reference is reducible to an empirical relation like causation or
information. I criticize this perception in Thought and World. Another important
source is the view that it is reference and other relations that provide us with our
cognitive purchase on specific portions of extramental reality, together with the view
that this would be impossible unless semantic relations were empirically robust.
I discuss this view in the immediately following chapter, “How Concepts Hook
onto the World.”
Although I will be much concerned with propositions, I will generally use the term
“thought” in place of the term “proposition,” hoping thereby to avoid some ambigu-
ities. In my view, thoughts are the things that serve as the objects of propositional
attitudes, and that are expressed by sentences when we use them to make assertions.
Further, following Frege, I take thoughts to have a mereological structure—for
example, I take the thought that snow is white to consist of the concept snow, the
concept white, and a predicate-forming device corresponding to the word “is.”
A thought is individuated in part by its logical organization and in part by the concepts
that serve as its constituents. I will not offer a detailed theory of concepts, but will
presuppose that it is somehow possible to explain how concepts are individuated in
terms of long-armed conceptual roles. More specifically, I will presuppose that con-
cepts are individuated in part by core aspects of their internal cognitive roles and in
part by the roles they play in encoding information about the external world.
 A SUBSTITUTIONAL THEORY OF TRUTH 53

I. A substitutional theory of truth

Paul Horwich explains truth in terms of thoughts that have the following form:

(T) The thought that p is true just in case p.

To be more precise, he offers an implicit definition of truth that has infinitely many
clauses, one for each of the thoughts of form (T) (Horwich 1998). This theory can
account for many fundamental facts about the concept of truth. Consider, for
example, the fact that we have a priori knowledge of (1):

(1) The thought that love is blind is true just in case love is blind.
Horwich’s theory can easily explain this fact: (1) is a component of a definition, so
of course it can be known a priori. But more: unlike Tarski’s theory, Horwich’s account
can explain why (1) is a triviality—that is, why we can come to know it without drawing
on a complex theory of syntax, and without engaging in laborious argumentation.
Unfortunately, despite having these virtues, Horwich’s theory is inadequate. As is
inevitable, given that it offers only an implicit definition, it fails to fix the extension of
the concept of truth. What is much worse, it is incapable of explaining our a priori
appreciation of certain generalizations about truth—for example, the generalization
that a disjunction is true just in case at least one of its disjuncts is true. If one is to
explain our knowledge of this generalization, one must show that it can be derived
from other generalizations that we know. It follows that if a definition of truth is to
play a role in explaining our knowledge of the generalization, the definition must
itself be general in form. But Horwich does not use a generalization to define truth.
Instead he uses a host of particular propositions, each of which explains what it is for
a specific thought to be true. (This problem was first pointed out by Anil Gupta in a
searching examination of Horwich’s theory (Gupta 1993). For further discussion of
Horwich’s theory, see Hill 2011.)
Seeking to avoid these difficulties, substitutionalism uses substitutional quantifi-
cation to enfold all of the clauses of Horwich’s implicit definition into a single explicit
definition. To be specific, it defines propositional truth as follows:

(S) For any thought x, x is true just in case (p)(x =the thought that p and p).
Like Horwich’s theory, (S) can easily explain the a priori character of (1), and also its
triviality, for it is an easy matter to derive (P) from (S), and (1) follows immediately
from (P) by universal instantiation:

(P) (p)(the thought that p is true just in case p).

(S) also has other advantages. It completely fixes the extension of the concept of
truth; and since it is a generalization, it provides a basis for explaining our knowledge
of other general propositions in which the concept of truth plays a role. Also, it
dovetails beautifully with a very plausible account of the cognitive function of the
54 PART I : MEANING

concept of truth. Quine and other writers have argued convincingly that we need
the concept of truth because it provides us with the ability to endorse a large set of
thoughts without having to endorse each member of the set individually, and with the
ability to endorse single thoughts without having to formulate the thoughts explicitly
(Quine 1970). (For additional discussion of the motivation for the concept of truth, see
the Introduction to the present volume.) In short, according to Quine, the concept of
truth is a device that enables general and indefinite endorsements. Assuming that this
view is correct, the concept of truth is fundamentally akin to the substitutional
quantifiers, for it is clear that the universal substitutional quantifier is a device that
enables general endorsements (consider (Pp)(if Joe asserts that p, then p)), and that the
existential substitutional quantifier is a device that enables indefinite endorsements
(consider (Sp)(Fermat’s Last Theorem = the thought that p and p)). In effect, then,
since (S) implies that there is a deep relationship between truth and substitutional
quantification, (S) ratifies Quine’s view and provides additional support for it.
The idea that truth can be analyzed in terms of substitutional quantification is far
from novel; but the idea has never been received with much enthusiasm, because it
has seemed that definitions like (S) are implicitly circular. The reason for this
perception is the belief that it is necessary to explain substitutional quantifiers by
describing the truth conditions of thoughts that contain them. Thus, for example, it
has been held that it is necessary to explain the existential substitutional quantifier by
a proposition like (2):

(2) A thought of the form (Sp)( . . . p . . . ) is true just in case at least one substitu-
tion instance of the matrix ( . . . p . . . ) is true.
(2) makes essential use of the concept of propositional truth. Accordingly, if (2) was
the basis of our grasp of the existential substitutional quantifier, it would be entirely
inappropriate to view (S) as a definition of truth.
Substitutionalism responds to this concern in two more or less independent ways.
First, it calls attention to the fact that substitutional quantification appears to be present
in our commonsense conceptual scheme. Consider the following propositions:

(3) It holds without exception that if Kurt believes that things stand thus and so,
then things do stand thus and so.
(4) Generally speaking, when Kurt maintains that matters are arranged in such-
and-such a way, matters really are arranged in such-and-such a way.
(5) If Kurt predicts that so-and-so, then, whatever so-and-so may be, it turns out
that so-and-so.
(6) It never fails: if Kurt maintains that things stand thus and so, and Peter
contradicts him, maintaining instead that matters are arranged in such-and-such
a way, then whatever the particular nature of their respective contentions, it turns
out to be the case that things stand thus and so, and that matters are not arranged
in such-and-such a way.
 A SUBSTITUTIONAL THEORY OF TRUTH 55

Grover, Camp, and Belnap speak of prosentences, meaning thereby to indicate

components of sentences that are counterparts of pronouns (Grover, Camp, and
Belnap 1975). Equally one might say that certain components of thoughts are
prothoughts. A prothought stands in a position that could be occupied by a whole
thought. Examples are things stand thus and so, matters are arranged in such and
such a way, and so-and-so. Now it is extremely plausible that these prothoughts have
the role of substitutional variables in (3)–(6), and extremely plausible also that the
quantifiers that bind them are substitutional quantifiers. Thus, it is clear, for example,
that (3) summarizes an infinite sheaf of thoughts that have the same form as (7):

(7) If Kurt believes that the universe is expanding, then the universe is expanding.
To see this, observe that it is clearly appropriate to infer (7) and all other thoughts of
the same form from (3). Observe also that the quantifier it holds without exception
that provides no ground for inferring thoughts from (3) that do not have the same
form as (7). One may of course infer other thoughts from (3), such as the disjunction
of (3) and the thought that snow is white, but one cannot justify those additional
inferences by appealing to the logical powers of the quantifier.
One might reply that these facts do not establish conclusively that it holds without
exception that is a substitutional quantifier. They can all be explained, it might be
said, by supposing that it is an objectual quantifier that ranges over a domain
consisting of propositions. This reply is initially plausible, but reflection shows that
it will not work. The hypothesis that it holds without exception that is an objectual
quantifier is ruled out by the fact that, as (3) shows, it can bind variables that occur in
hyperintensional contexts. This is a distinctive mark of substitutional quantifiers.
Unlike their objectual brethren, substitutional quantifiers can reach inside ascrip-
tions of propositional attitudes, operators like it has been proved that, and even
quotation marks. Since it holds without exception that binds a variable that is
governed by believes that in (3), it must be a substitutional quantifier.
It appears, then, that substitutional quantification plays a role in commonsense
logic. But this means that it is possible to explain the formal substitutional quantifiers
S and P without resorting to (2). One can explain them by simply equating them
with commonsense counterparts.
In addition to this first truth-independent way of assigning meaning to the formal
substitutional quantifiers, there is also a second such way. Since the substitutional
quantifiers are logical concepts, it is possible to confer meaning on them by formu-
lating rules that are to govern their use in inference. More specifically, it is possible to
confer meaning on each formal quantifier by stating two rules of inference for it—a
rule that specifies the conditions under which it is permissible to introduce the
quantifier (that is, the conditions under which it is permissible to infer a thought
containing the quantifier from another thought), and the conditions under which it
is appropriate to eliminate the quantifier (that is, the conditions under which it is
appropriate to infer another thought from a thought containing the quantifier).
56 PART I : MEANING

Substitutionalism provides introduction and elimination rules for both the existential
substitutional quantifier and the universal substitutional quantifier. Here, for
example, are its introduction and elimination rules for the propositional existential
quantifier:
Existential Introduction

(...T...) (...q...)
————————— —————————
(p)( . . . p . . . ) (p)( . . . p . . . )

Here T is a particular, determinate thought, and ( . . . T . . . ) is the thought that comes from
replacing all free occurrences of the propositional variable p in the open thought ( . . . p . . . )
with T. Further, q is a propositional variable, and ( . . . q . . . ) is the open thought that comes
from replacing all free occurrences of the propositional variable p in the open thought
( . . . p . . . ) with free occurrences of q.
Existential Elimination
(p)( . . . p . . . )
If ( . . . q . . . ), then T
———————
T
Here T is a thought, q is a propositional variable, and ( . . . q . . . ) is the open thought that comes
from replacing all free occurrences of the propositional variable p in the open thought
( . . . p . . . ) with free occurrences of q. Further, for Existential Elimination to be properly
performed, q must satisfy three additional conditions: (i) it cannot have a free occurrence in
T; (ii) it cannot have a free occurrence in (p)( . . . p . . . ); and (iii) it cannot have a free
occurrence in any premise on which the thought T depends.

In proposing an explanation in terms of introduction and elimination rules, I am

simply following a long tradition, which stems from Gerhard Gentzen, in the
philosophy of logic. Moreover, it can reasonably be claimed that the rules I am
proposing are entirely appropriate, given that the quantifiers are to be counterparts of
the commonsense concepts some and all. Rules like the present ones are structurally
akin to rules that can be seen on reflection to govern some and all. There are only two
differences. One is that the rules proposed by substitutionalism are intended to bind
variables that stand in positions that can be occupied by thoughts. In their most usual
deployments, some and all bind variables in positions that can be occupied by
singular terms. The other difference has to do with the fact that the substitutional
rules are meant to govern substitutional quantifiers. As noted a bit earlier, substitu-
tional quantifiers differ from their objectual colleagues in that they can bind variables
that stand in hyperintensional contexts. It is of course necessary to allow for this
difference in formulating rules of inference. One can do so by formulating the rules
for substitutional quantifiers in such a way that they apply to all contexts—to florid
hyperintensional contexts as well as to prosaic extensional contexts. The
 A SUBSTITUTIONAL THEORY OF TRUTH 57

substitutional rules have precisely this character. Thus, for example, the rule of
Existential Introduction permits passage from the hyperintensional thought Kate
believes that Jon is talking to the thought (Sp)(Kate believes that p). On the other
hand, in formulating rules for objectual quantifiers it is necessary to restrict their
deployment to extensional contexts.
To summarize: Substitutionalism characterizes the formal substitutional quantifiers S
and P by formulating rules that are to govern their use in inference. The rules in question
are structurally akin to the rules that govern the commonsense objectual quantifiers some
and all. These rules can be seen on reflection to be constitutive of the concepts of
existential and universal quantification. That is why they have received the approval of
generations of logicians. To be sure, the substitutional rules have some features that are
not present in normal codifications of the logic of some and all, but these differences were
to be expected, given that substitutional quantifiers are meant to bind propositional
variables, and given also that they are meant to be substitutional quantifiers.

II. A substitutional theory of semantic relations

I will now offer substitutional accounts of our relational semantic concepts, starting
with the following analyses of reference, denotation, and expression:

(R) For every object x and every object y, x refers to y if and only if (a)(the
concept of a is a singular concept and x = the concept of a and y = a).
(D) For every object x and every object y, x denotes y if and only if (F)(the
concept of a thing that is F is a monadic general concept and x = the concept of a
thing that is F and y is an F).
(E) For every object x and every object y, x expresses y if and only if (F)(the
concept of a thing that is F is a monadic general concept and x = the concept of a
thing that is F and y = the property F-ness).
In addition to these analyses of relational concepts that apply to concepts, substitu-
tionalism also offers an analysis of semantic correspondence, a relational concept
that applies to whole thoughts. It runs as follows:

(SC) For any thought x and any state of affairs y, x semantically corresponds to y if
and only if (p)(x = the thought that p and y = the state of affairs that p).
Arguably (SC) delivers all of the truths about semantic correspondence that guide
and motivate our use of the concept. Thus, it can be used to derive every proposition
of the form The thought that p semantically corresponds to the state of affairs that p.
Once (SC) is in hand, it is possible to introduce a correspondence-based concept of
truth via the following definition:

(CT) For any thought x, x is true if and only if there is a state of affairs y such that
(a) x semantically corresponds to y, and (b) y actually obtains.
58 PART I : MEANING

I leave it to the reader to decide whether (CT) is a better account of our ordinary
concept of truth than the account provided by (S). I myself favor (S) on grounds of
simplicity, but there are many philosophers who prefer an account like (CT). My
present point is that substitutionalism can do full justice to correspondence intu-
itions. (CT) is virtually identical to standard versions of the correspondence theory.

III. Substitutionalism and reduction

Turning now to the explanation and defense of this theory of semantic relations,
I begin by emphasizing that theory is not intended to fulfill the reductionist dreams of
naturalistically-minded philosophers. Beginning with Bertrand Russell, twentieth-
century philosophers sought accounts of the aboutness of mental states and the
reference of words that would naturalize them. Some hoped to reduce the relations
to dispositions to apply concepts and words to objects; others to reduce them to
relations definable in terms of the concept of information; and still others to reduce
them to teleosemantic relations. The motivation for these reductionist programs was
complex, but generally speaking, it included a desire to lay to rest worries originating
with Brentano as to whether reference and other semantic relations might pose a
threat to physicalism and the unity of science. It is after all quite impressive, and also
quite puzzling, that reference can connect the mind to such disparate phenomena as
wars fought in the distant past, bridges that will be built next year, particles that are
imperceptible, and abstract entities like colors and shapes. How can a single relation
be so marvelously promiscuous? It is also puzzling that while reference generally
links concepts to single items, or to single sets of items, empirical relations tend to
link concepts to multiple entities. We observe this difference, for example, when we
compare reference to the empirical relation encodes information about: while the
concept of Barack Obama refers to Obama and nothing else, the concept encodes
information about a number of things, including Obama, retinal imagers of Obama,
and thalamic projections of such images. It is not surprising that philosophers have
been concerned that reference might somehow transcend the natural world, thereby
posing substantial epistemological and metaphysical challenges. There were many
eloquent expressions of this concern in the twentieth century. Here is Fodor’s
formulation: “It’s hard to see . . . how one can be a Realist about intentionality
without also being, to some extent or other, a Reductionist. If the semantic and the
intentional are real properties of things, it must be in virtue of their identity (or
maybe their supervenience on?) properties that are themselves neither intentional
nor semantic. If aboutness is real, it must really be something else” (Fodor 1987).
My treatment of semantic relations has a completely different motivation, and a
completely different content, than the forementioned reductionist proposals. Instead
of explaining reference and semantic correspondence in terms of empirical relations
that are of interest to science, the theory maintains that they are induced by certain
formal, mereological relations among concepts. Consider the theory’s definition of
reference:
 A SUBSTITUTIONAL THEORY OF TRUTH 59

(R) For every object x and every object y, x refers to y if and only if (a)(the
concept of a is a singular concept and x = the concept of a and y = a).
(R) defines a relation between concepts and objects, but it does so by exploiting a fact
about our practice of forming canonical names for concepts. The concept that serves
as a name for the concept C is formed by attaching the concept-forming operator the
concept of to C. Thus, for example, the concept that serves as the canonical name for
the concept Abraham Lincoln is formed by prefixing the operator the concept of to
Abraham Lincoln. Because of this fact about our practice, the canonical name for the
concept C can be said to contain C as a constituent. (R) exploits this formal
relationship between the canonical name of C and C itself. The same is true, mutatis
mutandis, of the definition that substitutionalism provides for the concept of semantic
correspondence:

(SC) For any thought x and any state of affairs y, x semantically corresponds to y if
and only if (p)(x = the thought that p and y = the state of affairs that p).
(SC) defines a relation between thoughts and states of affairs, but it does so by
exploiting the fact that the concepts that serve as our canonical names for thoughts
share constituents with the concepts that serve as our canonical names for states of
affairs. The canonical name the proposition that p shares the thought p with and the
canonical name the state of affairs that p.1
It is of course perfectly appropriate to invoke formal relations between names in
specifying a relation between objects. If someone writes the names of certain people
on index cards, he can then specify a relation between the people by pairing the cards.
To be sure, not every pairing of names can be counted on correspond to a pairing of
objects: the procedure can be trusted to work only when one has grounds for thinking
that the named objects actually exist. But we have the appropriate existential assur-
ances in the case of (R) and (SC). We know both that the concept of Abraham
Lincoln exists and that Abraham Lincoln exists, and we have similar knowledge
about every pair that we regard as standing in the relation of reference. Equally, we
know both that the thought that Abraham Lincoln was wise exists and that the state
of affairs that Abraham Lincoln was wise exists, and we have similar knowledge about
every pair that we regard as standing in the correspondence relation.2

1
In Thought and World this proposal concerning semantic correspondence is accompanied by explan-
ations of the logical properties of the concepts the thought that and the state of affairs that. One might
expect that my explanation of the latter concept would treat it as expressing the function that maps the
thought that p onto the state of affairs that p, for every p. But this is not the case, for a very good reason: if
I had taken that line, my account of semantic correspondence would have been circular. The function in
question is tantamount to the relation of semantic correspondence, so it would be circular to invoke it in a
definition of the relation. Instead, I take concepts of the form the state of affairs that p as abbreviations for
descriptions of the form the state of affairs x such that, necessarily, x is actual just in case p. See Thought and
World, pp. 46–8.
2
There is an argument in Thought and World (pp. 41–6) which is designed to show that we do in fact
know that states of affairs exist.
60 PART I : MEANING

It is clear that neither (R) nor (SC) counts as a naturalization. A naturalization of a

semantic relation would explain the relation in terms of other relations that are
empirically or mathematically robust. It is clear that neither (R) nor (SC) attempts to
do this. Fortunately, it is equally clear that there is nothing metaphysically question-
able about the relations defined by (R) and (SC), and that there are no challenges
associated with grasping them epistemically. Like any other relations, the relations
defined by (R) and (SC) can be seen as collections of ordered pairs—or, to be precise,
as functions that map possible worlds onto collections of ordered pairs. The only
thing that is unusual about them is that they are made salient by, and specified by
appeal to, formal relations between names of the objects that figure in the pairs. The
relations themselves are no more mysterious, and no more in need of a legitimizing
naturalization, than the function that maps each possible world onto the set consist-
ing of the pairs <1, 2>, <2, 3>, and <3, 1>. Moreover, to gain epistemic access to them,
one simply has to take note of the formal relations between names that are invoked in
(R) and (SC).
It appears, then, that there is no problem with using formal properties of repre-
sentations to define relations. Still, it might be objected that there is a problem with
supposing that concepts and thoughts have formal properties. I have presupposed
that the concept a is a constituent of the concept the concept of a, and that the
thought p is a constituent of both the concept the thought that p and the concept the
state of affairs that p. More generally, I have presupposed that concepts and thoughts
have roughly the same combinatorial capacities as words and sentences in a spoken
language. It might be doubted that these presuppositions are appropriate. After all,
there are prominent theories of concepts that make no mention of combinatorial
powers. This is true, for example, of the theories that seek to explain the role of
concepts in categorization by representing them as prototypes or sets of exemplars.
Moreover, psychologists who favor connectionist architectures have often main-
tained that we can live without the assumption that the mind uses representations
with syntactic structures. Is it risky to view concepts as having word-like syntactic
properties in trying to explain their semantic properties?
It would be difficult to work out a theory of concepts that reconciles their
combinatorial capacities with their other powers, such as their role in categorization;
but challenging though it may be to devise such a theory, I think we have no choice
but to seek one. For it is essential to assume that concepts have word-like syntactic
properties if we are to account for the roles that thoughts play in inference. The
thought that 2 +2 = √16 has different inferential properties than the thought that
2 +2 = 4, even though the two thoughts are a priori equivalent, and the same is true of
the thought that Bill and Edith live here and the thought that it’s not the case that
either Bill doesn’t live here or Edith doesn’t live here. It is easy to establish experi-
mentally that this is so—for example, by showing that it takes longer to process
inferences involving the thought that 2 +2 = √16 than to process inferences involving
the thought that 2 +2 = 4. Now differences of these sorts can be explained by
 A SUBSTITUTIONAL THEORY OF TRUTH 61

supposing, first, that the thoughts I have cited are composed of different concepts,
and second, that their constituent concepts have different inferential roles. Thus, we
can explain why inferences involving the thought that 2+ 2 = √16 take longer to
process than inferences involving the thought that 2 + 2 = 4 by supposing that the first
thought contains the concept of √16, and that this concept has different inferential
powers than the concept of 4. Further, we can explain this difference between the two
concepts by supposing that the concept of √16 has a different constituent structure
than the concept of 4, and by pointing to the inferential powers of its constituents.
These explanations are satisfactory—so much so that it is hard to imagine a future
science of mind that would not avail itself of them. Hence, as Frege appreciated,
whatever difficulties may attend the doctrine, it appears that we are compelled to
acknowledge that concepts and thoughts resemble linguistic expressions in having a
syntactic organization.
To summarize: A number of philosophers have believed it necessary to “legitim-
ize” semantic relations by reducing them to empirically robust relations that are
better understood. They have hoped to reduce the number of problems confronting
philosophy by showing that semantic relations do not give rise to any new or
proprietary epistemological or metaphysical concerns. I have tried to offer an alter-
native perspective, according to which semantic relations are epistemologically and
metaphysically legitimate because it is possible to define them explicitly using only
logical devices that are reasonably well understood. These definitions exploit formal
properties of our canonical names of concepts, thoughts, and states of affairs, but this
is perfectly acceptable. The practice of exploiting formal properties of representations
in specifying relations is familiar and straightforward.
This is perhaps the place to acknowledge that substitutionalism does not protect
the concept of truth and other semantic concepts from paradoxes. On the contrary,
all of the familiar semantic paradoxes can be derived easily from the definitions of
semantic concepts that substitutionalism provides. Moreover, it is possible to derive
paradoxes directly from the rules of inference that govern the substitutional quan-
tifiers. (Hill 2002, pp. 121–6). Substitutionalism has a paradoxical core.
Now in view of these difficulties, it cannot be claimed that substitutionalism
disposes of all of the problems associated with truth and the semantic relations. It
protects them from most of the worries that motivated semantic reductionism, but it
doesn’t show that they are altogether innocuous.
In considering this fact, it is useful to distinguish between the enterprise of
analyzing our commonsense semantic concepts, and the enterprise of revising our
commonsense concepts in ways that prevent the derivation of paradoxes. One reason
for seeing the two enterprises as distinct is the intuitive appeal of the principles that
lead to the paradoxes. Until we appreciate that they have unacceptable consequences,
these principles seem obviously correct. Indeed, they seem to be forced upon us. This
strongly suggests that they are constitutive of our everyday semantic concepts, which
of course means that the paradoxes are constitutive of those concepts as well.
62 PART I : MEANING

Another reason for seeing the enterprises as distinct is that theories of truth that
succeed in blocking the paradoxes, such as Kripke’s theory of grounding and Gupta
and Belnap’s revision theory, are extremely complex, presupposing considerable
amounts of logic and mathematics. (See Kripke 1975 and Gupta and Belnap 1993.)
It simply isn’t believable that commonsense concepts could harbor such complexity.
Now in the version I am proposing here, substitutionalism is concerned exclusively
with analyzing our commonsense semantic concepts. Accordingly, it is a virtue
rather than a fault that it provides a basis for deriving the standard semantic
paradoxes. Indeed, since our commonsense concepts seem to be inherently paradox-
ical, substitutionalism would fail to reach its objective if its definitions did not lead to
contradictions.
To emphasize the main point, I mean to be proposing substitutional analyses, not
rational reconstructions. But it should be mentioned that it is also possible to use
substitutional quantifiers in reconstructive efforts. To do so, it would be necessary to
revise the rules of inference for the quantifiers in such a way as to block the derivation
of contradictions. Fortunately, there are ways of doing this. To be more specific, the
standard solutions to the semantic paradoxes can be adapted so as to block the
derivation of paradoxes involving the substitutional quantifiers (McGee 2000). To be
sure, it is not yet clear which ideas about the semantic paradoxes are best, and it is
therefore not yet clear how best to revise substitutional quantification. Those ques-
tions are still under active investigation. But it is reasonable to suppose that they have
answers, and that current research is gradually leading us to them. If this is so, then it
will someday be possible to use substitutional quantification to construct counter-
parts of our current concepts that are altogether free from difficulties.

IV. Substitutionalism and the laws of folk psychology

Reflection shows that the laws of folk psychology include many principles like the
following:
Law of informational uptake: If C is a concept that refers to a perceptually accessible object, o,
and x is an agent who possesses C, then x is disposed to form beliefs involving C as a result of
acquiring perceptual information about o.
First law of reliable indication: If an agent x comes to believe that p as a result of immediate
inference from a perceptual experience, then, where z is the state of affairs that corresponds
semantically to the thought that p, it is quite likely that z is actual, and quite likely also that z is
a distal cause of the fact that x comes to believe that p.
Second law of reliable indication: If an agent x comes to believe that p as a result of immediate
inference from an episodic memory, then, where z is the state of affairs that corresponds
semantically to the thought that p, it is quite likely that z is actual, and quite likely also that z is
a distal cause of the fact that x comes to believe that p.
First law of reliable inference: If it is the case (a) that x arrives at the thought w from the
thoughts y1, . . . ,yn by deductive inference, and (b) that w and y1, . . . ,yn correspond semantically
 A SUBSTITUTIONAL THEORY OF TRUTH 63

to the states of affairs u and z1, . . . ,zn, respectively, then it is quite likely that u is logically
implied by z1, . . . ,zn.
Second law of reliable inference: If it is the case (a) that x arrives at time T at the thought w by
inductive inference from the thoughts y1, . . . ,yn, (b) that w corresponds semantically to the state
of affairs u, where u involves a future time T’, and (c) that y1, . . . ,yn correspond semantically to
z1, . . . ,zn, then the conditional probability P(u/ z1, . . . ,zn) is quite likely to be high.
First law of desire: If N is a need such that in x’s experience instances of the property P have
consistently fulfilled N, then, generally speaking, on those occasions when N is particularly
pressing and x’s perceptual state carries information to the effect that there is an instance of P
in the current environment, x forms a desire that semantically corresponds to a state of affairs
involving x’s obtaining that instance of P.
Second law of desire: If x desires that p, and does so because the prospect of its being the case
that p has an intrinsic appeal, then, where z is the state of affairs that corresponds semantically
to the thought that p, it is quite likely that z is similar to states of affairs that have fulfilled one of
x’s needs in the past or to states of affairs that have caused x to experience some form of
pleasure in the past.
Law of action: If x desires that p, then, where u is the state of affairs that corresponds
semantically to the thought that p, and z1, . . . ,zn are the states of affairs that correspond
semantically to x’s beliefs, x is disposed to act in ways that would tend to bring about u in
possible worlds in which z1, . . . ,zn were actual.

Now I wish to claim that these laws, together with similar laws involving our other
semantic concepts, are partly responsible for the perception that semantic relations
are robustly empirical. Inspection shows that two semantic relations, reference and
semantic correspondence, figure prominently in the laws, and reflection shows that
the contributions they make to the laws are essential. Moreover, the laws imply that
there are stable and systematic links between these semantic relations and thick
empirical relations like indication and information. To be more specific, they imply
that the former relations enjoy nomological tracking relationships to the latter. But if
this is so, then the former relations must themselves be empirical. How could it be
otherwise?
I will respond to this line of thought by arguing that it misconstrues the role that
semantic concepts play in the foregoing generalizations. They don’t add to the
empirical content of the generalizations by bringing additional empirical relations
to the fore, but rather enable a level of generality that could otherwise be achieved
only by using substitutional quantification. In arguing for this interpretation of the
laws, I will focus on the following simplified version of the Law of Informational
Uptake:

($) If C is a nominal concept that refers to a perceptually accessible object o, then C

is used to encode perceptually acquired information about o.
I will maintain that it is actually quite reasonable to suppose that ($) depends
obliquely on a mereological relationship between names of concepts and names of
64 PART I : MEANING

perceptually accessible objects. As the reader will appreciate, it is possible to adapt

everything I will say about ($) to other laws containing semantic concepts, and in
particular, to all of the laws that are considered earlier.
($) is concerned with a semantic relation, but it is also concerned with a range of
facts involving an empirical relation—the relation that a nominal concept bears to an
object when the former is used to encode perceptually acquired information about
the latter. Suppose that we wish to consider a particular fact in which this empirical
relation plays a role. Suppose, for example, that we wish to consider the fact that the
nominal concept that refers to my brother in law, Mike, bears the relation to Mike.
To do this we must entertain the following thought:

(#) The concept of Mike is used to encode perceptually acquired information

about Mike.
(#) involves two singular concepts—the concept of Mike and Mike. It is clear that these
singular concepts are linked by a formal, mereological relationship: the former
concept is obtained from the latter by combining it with the concept-forming operator
the concept of. Thus, we have used a pair of concepts that stand in a formal,
mereological relationship to describe a fact involving a thick empirical relation.
Moreover, it is not an accident that we have done this. We were concerned to describe
a relationship between an object and a concept of that object. Clearly, if we are to reach
a goal of this sort, we must make use of a concept that refers to the object and a concept
that refers to the concept. But where C is any concept, if we are to form a concept that
refers to C, the simplest and most straightforward way to proceed is to combine C with
the operator the concept of. Thus, given that we were concerned to describe a
relationship between Mike and a concept that refers to Mike, it was natural to make
use of two concepts that were linked by a formal, mereological relation.
Now when we reflect on (#), we see that it illustrates a general pattern. There are
countless informational relationships between concepts and objects like the one that
is described by (#), and our canonical descriptions of these facts will all have the same
form as (#). In short, there are countless true propositions like (#). Clearly, it is
desirable to have some way of summarizing these propositions. Substitutional quan-
tification provides us with a way of doing this:

(%) (a)(if the concept of a is a nominal concept and a is a perceptually accessible

object, the concept of a is used to encode perceptually acquired information
about a).
We can use substitutional quantification to summarize the class of true propositions
like (#) because all such propositions involve names of objects and names of
concepts, and the former names are constituents of the latter.
Now it can be shown that if reference is defined substitutionally, in the way
indicated in Section II by (R), then the substitutional generalization (%) is equivalent
 A SUBSTITUTIONAL THEORY OF TRUTH 65

to the objectual generalization ($). In effect, (%) uses substitutional quantification to

accomplish what ($) accomplishes by deploying the concept of reference and objec-
tual quantification. In view of this fact, we can claim to have an explanation of why
reference enjoys the stable and systematic relationship to perceptually acquired
information that is captured by ($). As we have seen, it is a consequence of our
convention for forming canonical names for concepts that we are able to use
substitutional quantification to describe the general pattern that is illustrated by (#).
Given that the concept of reference simply provides us with an alternative means of
describing this same pattern, it is inevitable that there be a stable and systematic
relationship between reference and information.
This completes the argument. The key idea is that ($) does not use the concept of
reference to bring a new empirical relation into the picture, but rather to achieve a
desired level of generality in representing a set of facts about the use of concepts to
encode information—a level of generality that could otherwise be achieved only by
using substitutional quantification.
As noted earlier, this argument can be generalized so as to apply to all of the laws
that are cited earlier. In its most general form, it fully and decisively undercuts the
impression that the laws provide the foundation of a case for viewing semantic
relations as contingent and empirical, and by the same token, it shows that it is
impossible to argue from the representational powers of the mind to a conclusion
that is at odds with substitutionalism.

References
Fodor, J. A. (1987). Psychosemantics. Cambridge, MA: MIT Press, 97.
Grover, D., Camp, J. L., Jr., and Belnap, N. (1975). “A Prosentential Theory of Truth,”
Philosophical Studies 27: 73–125.
Gupta, A. (1993). “Minimalism,” Philosophical Perspectives 7: 359–69.
Gupta, A. and Belnap, N. (1993). The Revision Theory of Truth. Cambridge, MA: MIT Press.
Hill, C. S. (2002). Thought and World: An Austere Portrayal of Truth, Reference, and Semantic
Correspondence. Cambridge: Cambridge University Press.
Hill, C. S. (2006a). “Precis of Thought and World,” Philosophy and Phenomenological Research
LXXII: 174–81.
Hill, C. S. (2006b). “Replies to Marian David, Anil Gupta, and Keith Simmons,” Philosophy and
Phenomenological Research LXXII: 2005–222.
Hill, C. S. (2011). “Review of Paul Horwich, Truth—Meaning—Reality,” Mind 120: 1262–70.
Horwich, P. (1998). Truth, 2nd edition. Oxford: Oxford University Press.
Kripke, S. A. (1975). “Outline of a Theory of Truth,” Journal of Philosophy 72: 690–716.
McGee, V. (2000). “The Analysis of ‘x is True’ as ‘For Every p, if x = ‘p,’ then p’.” In A. Chapuis
and A. Gupta (eds.), Circularity, Definition, and Truth. New Delhi: Indian Council for
Philosophical Research, 255–72.
Quine, W. V. (1970). Philosophy of Logic. Englewood Cliffs, NJ: Prentice Hall.
 5
How Concepts Hook
onto the World*

I will be concerned in this chapter with relations that link concepts to extraconcep-
tual reality. More specifically, I will be concerned with three such relations: the
semantic relation of reference, the representational relation that a concept bears to
an item when it is used to encode information about the item, and the epistemic
relation that a concept bears to an item when it enables agents to discriminate
between that item and a range of other things. I will attempt to bring these con-
cept/world relations into sharper focus, to explain their mutual relevance, and to
describe their importance for cognition and action.

I. Deflationism
Deflationism is a theory of semantic properties like truth, reference, and denotation.
Roughly, it denies that these properties have a robust nature that can be elucidated by
science, metaphysics, or normative inquiry. Instead, truth, reference, and denotation
can be explained in terms of a very meager conceptual base. Thus, for example,
according to a prominent version of deflationism that is due to Paul Horwich, truth,
reference, and denotation can be implicitly defined by enumerating all of the
propositions that have the following forms:

(T) The proposition that p is true just in case p.

(R) For any y, the concept of a refers to y just in case y is identical with a.
(D) For any y, the concept of an F denotes y just in case y is an F.
Inspection shows that the propositions that are used to define truth contain no
concepts over and above the ones that figure in the proposition whose truth is
being explained. Similar remarks apply to the definitions of reference and denotation.1

* I have benefited considerably from a number of conversations with Anil Gupta and Mark Wilson. The
objection to deflationism that is the subject of Sections III and IV is an attempt to capture the common core
of worries that they have expressed on different occasions. I am also grateful to audiences at the University
of Pittsburgh, the University of Texas, and the NC State Conference on concepts, for insightful comments.
1
See Horwich (1998). For discussion of his theory, see Hill (2011a).
 HOW CONCEPTS HOOK ONTO THE WORLD 67

According to another version of deflationism, instead of thinking of the implicit

definitions of semantic concepts as having infinitely many separate components, it is
best to think of them as single propositions that make use of substitutional quanti-
fiers.2 On this second version of deflationism, truth, reference, and denotation can be
explained as follows:

(T*) (p) the proposition that p is true just in case p.

(R*) For any y, (a) the concept of a refers to y just in case y = a.
(D*) For any y, (F) the concept of an F denotes y just in case y is an F.
Here (Pp) is a substitutional universal quantifier whose associated substitution class
consists of propositions; (Pa) is a substitutional quantifier whose associated substi-
tution class consists of singular concepts; and (PF) is a substitutional quantifier
whose associated substitution class consists of general concepts. I have elsewhere
tried to show that it is acceptable to combine substitutional quantifiers with variables
for propositions and concepts, and that there is in fact reason to think that we deploy
constructions of these sorts in everyday thought and talk.3
Horwich uses the label “minimalism” to stand for the first version of deflationism,
and I use “substitutionalism” to stand for the second version. I have presented both
versions here because they have complementary virtues. On the one hand, it is
somewhat easier to understand minimalism, and to appreciate the motivation for
it. On the other hand, substitutionalism has more explanatory power than minim-
alism, as a result of the technical apparatus that it deploys. In particular, unlike
minimalism, substitutionalism can explain why we are disposed, just on the basis of
possessing the concept of truth and the standard logical concepts, to accept various
generalizations about truth. For example, because it makes use of an operator that
expresses generality in its definition of truth, substitutionalism can explain why

2
Hill (2002). See also the preceding chapter, “A Substitutional Theory of Truth, Reference, and
Semantic Correspondence.”
3
In chapter 2 of Hill (2002), I explain the substitutional quantifiers in terms of rules of formal
inference—more specifically, in terms of introduction and elimination rules that resemble rules for the
objectual quantifiers. I also argue that counterparts of the formal substitutional quantifiers play a role in
our everyday thought and discourse. The argument consists in citing propositions like these:
(a) When Joe claims that matters are arranged in such and such a way, then it always turns out that
matters are in fact arranged in that way—no matter what being arranged in that way may involve.
(b) If the content of a thought is that matters stand thus and so, then the thought is true just in case
matters really do stand thus and so. This holds for any thought whatsoever.
To be sure, there are various ways of articulating the logical structures of claims of this sort, and not all of
them represent such claims as making use of substitutional quantification. While acknowledging this point,
however, we should also acknowledge that substitutional interpretations are very much in the running.
After all, it is arguable that the introduction and elimination rules that govern the relevant quantifiers are
substitutional in character. (Thus, for example, starting with (a) as a premise, it is clearly possible to infer If
Bill claims that Biden likes trains, then Biden does in fact like trains.)
For an explanation of the substitutional quantifiers in terms of rules of inference, see Section I of the
preceding chapter.
68 PART I : MEANING

anyone who possesses the concept of truth is disposed to accept the generalization
that a conjunctive proposition is true just in case both of its conjuncts are true.4
In addition to the foregoing characterizations of truth, reference, and denotation,
the two versions of deflationism offer accounts of the relation of semantic
correspondence—that is, of the thought-world relation we see at work in proposi-
tions like this one:
The proposition that snow is white corresponds to the state of affairs that snow is white.

Minimalism can explain correspondence by saying that it is defined by the totality of

all propositions of the following form:

(SC) The proposition that p corresponds to the state of affairs that p.

And substitutionalism can explain correspondence by saying that it is defined in this way:

(SC*) For any y, (p) (the proposition that p corresponds to y just in case y = the
state of affairs that p).
Thus, given a background ontology that includes states of affairs, both minimalism
and substitutionalism have the resources to explain intuitions to the effect that truth
consists in correspondence with actual states of affairs—or in other words, with facts.
In addition to the version of substitutionalism that I have just described, which
offers implicit definitions of truth, reference, denotation, and correspondence, there
is also a second version that uses substitutional quantifiers to provide explicit
definitions of our semantic concepts. Here, for example, is its definition of truth:

(T**) For any object x, x is true if and only if (p)((x = the thought that p) and p).
(Here (Sp) is a substitutional existential quantifier whose associated substitution
class consists of propositions.) Although I will focus on the previous version of
substitutionalism here, because it is simpler and therefore a bit easier to discuss, it
is important that truth and other semantic concepts can be explicitly defined in terms

4
To explain our acceptance of the generalization in terms of a commitment to (T*), it is necessary to
assume that we are also committed to the following three substitutional generalizations:
(i) For any object x, if x is true, then (p) x = the proposition that p
(ii) (p)(q)(if the proposition that p = the proposition that q, then p if and only if q)
(iii) For any x and any y, (p)(q)(if x = the thought that p and y = the thought that q, then the
conjunction of x and y = the thought that p and q)
Given (T*) and these three additional propositions, it is possible to deduce the proposition that for any x, y,
and z, if z is the conjunction of x and y, then z is true just in case x and y are both true. Incidentally, when
the implicit definition (T*) is conjoined with (i) and (ii), the result is equivalent to the following explicit
definition of truth:
(T**) For any object x, x is true if and only if (p)((x = the thought that p) and p).
In view of the triviality of (i) and (ii), it follows that (T*) is quite close to an explicit definition.
For further discussion, see chapter 2 of Hill (2002), especially note 22.
 HOW CONCEPTS HOOK ONTO THE WORLD 69

of substitutional quantification. Among other things, this shows that they are very
closely related to logical concepts. The contributions that semantic concepts make to
our conceptual scheme are essentially the same as the contributions of two logical
devices, the universal and existential substitutional quantifiers.

II. Arguments for deflationism

There are four reasons for thinking that deflationism is true.
First, it can easily explain why propositions of the following forms can be seen to
be true a priori:
(a) The proposition that snow is white is true just in case snow is white.
(b) The concept of Barack Obama refers to Barack Obama.
(c) The concept of a rabbit denotes rabbits.
Minimalism explains the a priori character of these propositions by pointing out that
they are clauses of definitions. Substitutionalism explains it by pointing out that they
follow immediately from definitions.
Second, both theories can easily explain why propositions like (a), (b), and (c) are
utterly trivial—that is, why they can be seen to be true without invoking theories or
engaging in argumentation. The theories imply that they are trivial for the same
reasons that they are a priori—they are either components of definitions or imme-
diate consequences of definitions.
It follows from this that both theories can easily explain why semantic skepticism
is absurd. Someone influenced by Quine might conceivably urge that there is no basis
for deciding whether the concept of a rabbit denotes rabbits or instead denotes
rabbitty entities of some other kind, such as undetached rabbit parts or temporal
stages of rabbits.5 Equally, someone influenced by Kripke might conceivably main-
tain that there is no basis for deciding whether the concept of addition refers to
addition or to a mathematical operation that diverges from addition on arguments
that are inaccessible to human calculators.6 Clearly, both of these claims are absurd,
and deflationism can explain why. According to deflationism, the proposition that
the concept of a rabbit denotes rabbits is either a component of a definition or a
trivial consequence of a definition. Moreover, deflationism makes a similar claim
about the proposition that the concept of addition refers to addition.
These observations do not imply that Quine and Kripke are themselves guilty of
manifest absurdities. Quinean skepticism is concerned with the denotation of words
rather than the denotation of concepts, and the same is true of Kripkean skepticism
about reference. Even if one grants that the concept of a rabbit determinately denotes
rabbits, and that the concept of addition determinately refers to addition, one might

5 6
Quine (1960), chapter 2. Kripke (1982), chapters 1 and 2.
70 PART I : MEANING

still doubt that the denotation of the word “rabbit” is determinate, and one might
have similar doubts about the reference of the word “addition.” Thus, for example,
one might doubt that it is determinate whether an agent uses the word “rabbit” to
express the concept of a rabbit, as opposed to the concept of an undetached rabbit
part or the concept of a temporal stage of a rabbit. This doubt would naturally lead to
skepticism as to whether the word “rabbit” determinately denotes rabbits. Accord-
ingly, to silence Quinean skepticism, it would be necessary to show that it is generally
determinate whether a given word expresses a given concept. I think it would not be
hard to show this, but that is a different question. The present points are just that
deflationism gives an immensely persuasive account of why skepticism concerning
conceptual reference would be absurd, and that this explanation does not immedi-
ately apply to Quinean and Kripkean skepticism about linguistic reference.
The third reason for accepting deflationism is that it provides an especially simple
and natural explanation of the value of semantic concepts. As a number of defla-
tionists have pointed out, the concept of truth is useful because it makes it possible
for us to assert propositions without explicitly formulating them, as in the following
claims:
All of Obama’s remarks at the press conference were true.
All of the consequences of the Peano Axioms are true.

If someone wanted to endorse all of Obama’s remarks at the press conference, she
would probably assert the first proposition, because it would be too cumbersome or
too time consuming to recall all of Obama’s remarks and assert them individually.
And of course, if someone wanted to endorse all of the consequences of the Peano
Axioms, she would have no choice but to assert the second proposition, given that the
consequences of the Peano Axioms are infinite in number. To generalize from these
examples, we need the concept of truth because we need to assert propositions that
we are either unwilling or unable to formulate explicitly. The concept of truth gives
us the means to assert propositions without formulating them—that is, without
making their contents explicit. Now as is easily seen, both versions of deflationism
actually predict that the concept of truth will be used for this purpose. For they both
imply that where S is any set of propositions, the propositions in S are all equivalent
to propositions of the following form:
The proposition that p is true.

Because of these equivalences, it is possible to assert the members of S by asserting the

corresponding propositions that have the given form. Further, instead of asserting
the latter propositions individually, we can assert them collectively by asserting the
following generalization:
For any object x, if x is a member of S, then x is true.
 HOW CONCEPTS HOOK ONTO THE WORLD 71

Thus, by implying all instances of the equivalence schema (T), the two versions of
deflationism explain how truth serves our interest in asserting propositions without
formulating them. Of course, any theory of truth that implies the instances of (T) will
also be able to explain this fact, but it will not be able to explain it as simply and
straightforwardly as the two theories we are considering. As we have already seen,
they deliver the instances of (T) in an ideally efficient way.
Fourth, deflationism is supported by a line of thought I call the diversity argument.
Roughly, the argument maintains that the items that have semantic properties and
stand in semantic relations tend to be quite diverse—so diverse that it is implausible
that they fall within the extensions of natural properties and relations, where a
natural property is a robust property that would require substantive metaphysical
or scientific concepts for its elucidation, and a natural relation is a robust relation that
meets this same condition. Generally speaking, if two items fall in the extension of a
natural property, the items must have something in common—they must be intrin-
sically similar in one or more respects. Equally, if two ordered pairs fall in the
extension of a natural relation, they must have something in common. But members
of the extensions of semantic properties and relations tend to be quite different. Thus,
for example, the pairs of concepts and objects that stand in the reference relation tend
to be no more similar to each other than they are to the pairs of concepts and objects
that do not stand in the relation.
To see this, recall that there are concepts that refer to physical objects and also
concepts that refer to a range of mathematical entities. Could it be true that concepts
of both these kinds bear the same natural relation to their referents? This seems very
unlikely. The relations that physical concepts bear to their referents are principally
causal, indicational, and informational relations. As most writers have agreed, any
naturalistic explanation of the reference of physical concepts would have to invoke
one or more of these relations at some point. But it is impossible for concepts to bear
relations of these kinds to mathematical entities. It follows that it is impossible to
reduce reference to a single natural relation. The best we could hope for is a reduction
to a disjunction of very different natural relations. But what could tie the different
disjuncts together?
We should also observe that physical concepts come in more than one flavor.
There are concepts that refer to past objects, concepts that refer to present objects,
and concepts that refer to future objects. (In the 1870s, engineers and workers used
the concept Brooklyn Bridge to refer to a structure that had not yet been built.) There
are concepts that refer to events that last only an instant, and concepts that refer to
properties that we encounter again and again. There are concepts that refer to first
order properties and concepts that refer to properties of second and higher orders.
There are concepts that refer to objects we are innately equipped to see, hear, and feel,
and concepts that refer to properties that can only be known via abstract theories and
complex inferences. There are concepts that refer to objects we can delimit percep-
tually, and concepts that refer to objects that are much too small or much too large to
72 PART I : MEANING

be discriminated from other things. It seems very unlikely that there is a single
natural relation that links concepts of all of these kinds to their referents. A reductive
naturalistic account of their reference would have to be highly disjunctive in order to
be materially adequate. So even if we restrict our attention to the special case of
empirical concepts, we are unable to answer the question, What do all the pairs that
stand in the reference relation have in common?
Unlike naturalistic accounts of reference, deflationism can answer this question,
and this is a very strong point in its favor. According to deflationism, there is a non-
natural property that all the relevant pairs have in common. They are alike in that the
canonical name for the first member of a relevant pair contains the canonical name
for the second member as a constituent. Consider, for example, the pair consisting of
the concept of red and the color red. The canonical name for the first member of this
pair is “the concept of red” and the canonical name for the second member is “red.”
Clearly, the second name is a constituent of the first name. All of the pairs that stand
in the reference relation are related in this way. Deflationary definitions of reference
exploit this formal relationship between names, and deflationary definitions of truth
and denotation exploit similar formal relationships. This is why they can imply all of
the instances of schemas (T), (R), and (D), despite the diversity of the entities that
mentioned in those instances.7
In my view, and, I’m sure, in your view as well, these arguments provide very
strong support for deflationsim. It seems that we are obliged to embrace it.

III. Problems with deflationism

Alas! Reflection shows that there are also very strong arguments that weigh against
deflationism.
As we have just seen, in combination with the diversity argument, deflationism
entails that there is no substantial, natural relation that concepts bear to their
referents. Rather, reference is just a relation-in-extension, a set of ordered pairs
whose members have no more in common than that they can be generated by a
certain formal rule. More specifically, where P is any pair that belongs to the
reference relation, P has nothing in common with all of the other members of the
relation beyond the fact that our canonical name for the first member of P bears a
certain formal relation to the name of the second member. (In the interests of
economy, I will henceforth use “refers” and “reference” exclusively in talking about
semantic relationships between concepts and their extensions. That is, I will use these
terms to cover both the semantic relation that singular concepts bear to their
referents and the semantic relation that general concepts bear to the items they
denote.)

7
I give a somewhat more detailed version of the diversity argument in Hill (2010).
 HOW CONCEPTS HOOK ONTO THE WORLD 73

Now if all of this is so, then concepts cannot be said to have causal powers in virtue
of their reference, for the fact that one item belongs to the same ordered pair as a
second item has no tendency to show that the first item bears a causal relation to the
second item. Indeed, it has no tendency to show that the first item bears a causal
relation to any other item. Being a member of a certain pair is a trivial fact about an
object, as is being a member of a pair that belongs to a certain set of pairs. Every
object is a member of countless pairs, and every pair containing a given object is a
member of countless sets of pairs. On the other hand, possessing a causal power is
not a trivial fact. Many objects (e.g., numbers) lack causal powers altogether; and if an
object does possess one or more powers, it will possess only a comparatively small set
of them. But of course, if membership in pairs is trivial, and possessing causal powers
is substantial, the latter cannot derive from, or be dependent on, the former. Indeed,
there is no interesting relationship of any kind between membership in pairs and
causation.
These observations have two easy corollaries.
First, a concept cannot be said to contribute to an agent’s epistemic capacities in
virtue of its reference. Epistemic capacities include the ability to discriminate objects
from other things, the ability to recognize objects, the ability to identify them, the
ability to encode information about them, the ability to predict their future behavior,
and the ability to explain the behavior that they have already exhibited. Making a
contribution to one of these abilities would inevitably involve the exercise of causal
powers, and as we have just seen, a concept cannot be said to have causal powers in
virtue of its reference.
Second, for similar reasons, a concept cannot contribute to an agent’s practical
capacities in virtue of its reference. Thus, for example, a concept cannot contribute to
an agent’s ability to perform actions in virtue of its reference, nor to an agent’s ability
to pursue or achieve his or her goals.
It appears, then, that reference is too frail a reed to provide support for either our
epistemic or our practical abilities. But more: if the diversity argument is sound,
reference cannot be said even to contribute to our abilities vicariously. To make a
vicarious contribution, it would have to be coextensive with a relation that supports
causal powers. But only a natural relation can support casual powers, and the
diversity argument implies that reference is not coextensive with a natural relation.
Unfortunately, these results are highly counterintuitive. When we reflect, it seems
obvious that conceptual reference provides the foundation for various epistemic and
practical abilities. Evidently, we must possess concepts that refer to items in the world
in order to be epistemically en rapport with specific items, and in order to select and
pursue specific goals. For example, if I didn’t have a concept that referred to Porsches,
I wouldn’t be able to think about Porsches, and a fortiori, I wouldn’t be able to form
beliefs that encode information about their properties. Equally, if I didn’t have a
concept that referred to Brussels sprouts, I wouldn’t be able to adopt the goal of
finding some Brussels sprouts for my dinner. To have a cognitive purchase on
74 PART I : MEANING

specific items in the world, and to aim at goals that involve such items, I must possess
concepts, and more particularly, I must possess concepts that are connected to those
items in determinate ways. “Reference” is our name for the determinate connections
between concepts and worldly items that make cognition and action possible. Or so
it seems.
Could it be that it is not reference but some other relation that enables thoughts
and deeds that are concerned with specific entities? No. To support exactly the same
cognitive and practical abilities as reference appears to support, a relation would have
to be coextensive with reference. For example, it would have to link the concept of a
Porsche determinately and exclusively with Porsches, and the concept of a Brussels
sprout determinately and exclusively with Brussels sprouts. Otherwise it could not
explain how the concept of a Porsche enables me to think about Porsches, rather than
entities of an altogether different kind, or a range of entities that includes both
Porsches and other items. And it could not explain how the concept of a Brussels
sprout makes it possible for me to choose actions that are focused determinately and
exclusively on Brussels sprouts.
For reasons like these, philosophers have thought that reference must be a
substantial relation, a relation that is capable of doing a great deal of causal and
explanatory work. Or if they have not thought that, they have at least thought that
reference is coextensive with a relation that can do such work. By the same token,
they have thought that reference is an extremely important relation. Moved by these
beliefs, they made it a central topic of twentieth-century philosophy. For decades, it
was one of the major targets of epistemology and metaphysics, as well as philosophy
of logic, philosophy of language, and philosophy of cognitive science. Today it
continues to play prominent roles in a variety of literatures.
In view of these reflections, we must conclude that deflationism is involved in a
paradox. There are considerations that mandate its acceptance, but it seems incap-
able of explaining the contribution that reference makes to our being epistemically
and practically en rapport with the world, and it therefore seems incapable of
explaining the importance of reference. Deflationism is true, but incapable of sus-
taining our intuitions about the explanatory power of reference. Alternatively, we can
say that deflationism is true, but that this is extremely unfortunate, because it implies
that we do not have the cognitive and practical command of the world that we take
ourselves to have. No one could be happy with this result!

IV. A partial fix
In combination with the diversity argument, deflationism implies that reference is
not itself a natural relation, and therefore that it is devoid of causal powers; but
it could still be true that it is linked to natural relations in lawful ways. In other
words, it could still be true that there are true generalizations that connect reference
with natural relations, and that these natural relations explain at least some of the
 HOW CONCEPTS HOOK ONTO THE WORLD 75

epistemic and practical abilities that we have been considering. If that was true, then
reference would have a kind of vicarious explanatory power.
This turns out to be the case. Here are a few generalizations that are true and that
also seem to be capable of figuring in useful explanations:
(a) If x is a concept that refers to a perceptually accessible object, y, and z is an
agent who possesses x, then z is disposed to form beliefs involving x as a result
of acquiring new perceptual information about y, provided only that z has no
defeaters for those beliefs.
(b) If x is a concept that refers to a perceptually accessible object, y, z is an agent
who possesses x and has had considerable perceptual contact with y, and w is a
synthetic proposition containing x, then z is disposed to consult perceptual
memories that encode information about y in trying to decide whether or not
to accept w.
(c) Let x be a natural kind concept that refers to a natural kind y, and let it be the
case that y is observationally accessible. Given these assumptions, if z is an
agent who possesses x at time T, then it must be the case that z satisfies the
following condition: it has been true at times prior to T, generally speaking,
that when z has been led by normal perceptual processes to make a recogni-
tional judgment involving x, one of the causes of z’s judgment has been a
member of the kind y.
(d) If x refers to y, where y is a type of basic action, then desires containing x serve
as proximal causes of actions of type y.
In addition to these and other generalizations about reference, there are also gener-
alizations linking semantic correspondence to natural properties. Here is an example:
(e) If activity in the visual system produces a perceptual belief that semantically
corresponds to a state of affairs y, then y is a cause of the activity.
As with (a)–(d), (e) claims that a certain natural relation is a necessary condition of a
semantic relation in a certain limited range. In effect, it tells us that in certain limited
range of contexts, semantic correspondence tracks a causal relation. It is possible to
determine that the causal relation obtains by determining that the semantic relation
obtains.
Because of the tracking relations that are described by (a)–(e), we can to some
extent understand the importance that is frequently attributed to semantic relations.
In many cases, at any rate, if we want to explain why an agent has a certain epistemic
or practical ability, an ability that involves a relationship between the agent and one
or more items in the world, it will suffice to cite the reference of one or more
concepts, because our citation will imply the existence and explanatory relevance
of one or more informational or causal relations. Just as reference tracks various
natural relations, so also explanations involving reference track more substantial
explanations involving natural relations. The same holds, mutatis mutandis, for
76 PART I : MEANING

semantic correspondence. Accordingly, (a)–(e) provide us with a certain amount of

relief from the foregoing paradox.
Is deflationism compatible with the truth of (a)–(e)? Yes. This is easiest to see in
the case of substitutionalism. According to substitutionalism, each of the foregoing
generalizations is equivalent to another generalization that lacks the concept of
reference but contains a substitutional quantifier instead. Thus, for example, (a) is
equivalent to the following generalization:
For any z, (a)(if a is a perceptually accessible object, and z is an agent who possesses the
concept of a, then z is disposed to form beliefs involving the concept of a as a result of acquiring
new perceptual information about a, provided only that z has no defeaters for those beliefs).

The truth of the generalizations poses no threat to substitutionalism because there

are variants of those generalizations that deploy exactly the apparatus that substitu-
tionalism claims to be essential to understanding reference.8

V. The persistence of the paradox

We have been noticing that there are natural relations that are necessary conditions
of reference. Natural relations support causal powers. Hence, although we cannot say
that concepts have causal powers in virtue of their reference, we can say that they
have causal powers in virtue of standing in natural relations to the world that are
closely tied to reference. Conceptual reference is not itself a source of causal powers,
but it tracks relations that are such sources. This provides us with a basis for
explaining how concepts can contribute to cognitive and practical abilities. They

8
If reference owes its very being to a purely formal relationship between names of concepts and names
of objects, as deflationism contends (see Section II, penultimate paragraph), isn’t it a miracle that reference
turns out to be aligned with causally robust natural relations? Isn’t it a miracle that laws like (a)–(d) hold?
No. Suppose that a concept stands in a certain natural relation N to an object. Suppose also that we wish to
report on this fact. Clearly, we will have to use a name for the concept and a name for the object. Now as we
have seen, our canonical names for concepts are obtained by applying the operator the concept of to names
of objects. Hence, in reporting the fact in question, we will have to use a name for the object twice—once as
a free standing component of our report, and another time as a constituent of the name of the concept.
That is, our report will have the form the concept of O bears the relation N to O. It follows that if we wish to
generalize the claim, we will be able to do so using a substitutional quantifier. Thus, for example, we might
construct a generalization of the following form:
(a)(if the concept of a is a concept of type T, then the concept of a bears the relation N to a).
Now given that reference is definable in terms of substitutional quantification, it is possible to transform
this generalization into one that contains only objectual quantifiers:
For any x and any y, if x is a concept of type T and x refers to y, then x bears the relation N to y.
Hence, instead of being miraculous, the connection between reference and N is due to three rather
humdrum factors—a naming convention, the fact that substitutional quantification gives us a means of
obtaining generalizations by exploiting this convention, and the fact that generalizations about concepts
that contain substitutional quantifiers are equivalent to generalizations that contain objectual quantifiers
and the concept of reference.
For further discussion, see Section IV of the preceding chapter.
 HOW CONCEPTS HOOK ONTO THE WORLD 77

do not contribute to abilities of these kinds in virtue of their reference, but they do
contribute to them in virtue of relations to the world that reference can be said
to track.
We must now observe, however, that insofar as the natural relations tying concepts
to the world are only necessary conditions of reference, as opposed to being both
necessary and sufficient, they are not capable of tying concepts uniquely to the objects
to which they refer, and, by the same token, they fail to provide a basis for explaining
how concepts provide us with cognitive and practical abilities that are directed
uniquely on their referents.
It seems that we really do possess such abilities. Thus, for example, it seems that if
an agent possesses the concept of red, then the agent is able to distinguish between
things that are red and things that exemplify other colors. That is, the agent has a
discriminative ability that is focused on red. Accordingly, insofar as the natural
relations that link concepts to the world are only necessary conditions of reference,
they fail to explain abilities that we undeniably possess, and that we undeniably
possess in virtue of possessing concepts.
In view of these considerations, it is clear that our effort to identify naturalistic
necessary conditions of reference has not fully solved the problems that we noticed in
Section III, such as the problem of explaining why reference is interesting. Evidently, to
solve those problems fully, we will need to adopt a new line of inquiry. Now there is in
fact a line that seems prima facie to merit our attention. The diversity argument implies
that there are no natural relations that are necessary and sufficient for reference,
considered as a relation linking concepts of all kinds to items in the world, but it
could still be true that there are a number of natural relations N1, . . . , Nk such that each
Ni provides a necessary and sufficient condition for the reference of some limited range
of concepts. In other words, it could still be true that it is possible to factor the class of
concepts into a series of categories K1, . . . , Kk, and to pair these subclasses with natural
relations in such a way that certain principles of the following form were true:

(P) For any concept x, if x belongs to the category Ki, then for any object y, x refers
to y just in case x bears Ni to y.
Or at least, it could be true that principles of this sort hold for some subclasses of
concepts, perhaps including the subclasses consisting of concepts that enjoy espe-
cially intimate relations with perception. Clearly, this possibility should be explored.
Unfortunately, there are powerful reasons for doubting that this strategy can be
successful. More specifically, there are reasons to doubt that any categories of
concepts, including even the various categories of perceptually grounded concepts,
are tied uniquely to their referents by natural relations. These reasons have all figured
prominently in the literature for some years, and they are therefore quite familiar.
They have proved to be quite difficult to answer or dismiss. They include the causal
chain problem, the problem of error, Quine’s gavagai argument, the holism problem,
78 PART I : MEANING

the Kripkenstein paradox, and arguments from such model-theoretic results as the
Löwenheim-Skolem Theorem and the Non Standard Models Theorem for arith-
metic.9 In one way or another, all of these lines of thought cast doubt on the idea that
naturalistic relations can be sufficient for reference, even if questions about reference
are relativized to particular categories of concepts.
In the rest of this chapter I will be concerned to argue that, despite initial
appearances to the contrary, concepts are often tied uniquely to their referents by
natural relations. I will be concerned exclusively with perceptual concepts, but I will
construe “perceptual” broadly, so that perceptual concepts include concepts for
perceptual kinds, such as the concept of a rabbit and the concept of water, and also
concepts that stand for perceptible individuals, such as specific human beings. I will
focus on what I take to be the three main reasons for doubting that perceptual
concepts are linked uniquely to their referents—the causal chain problem, the
problem of error, and Quine’s gavagai argument.

VI. The causal chain problem

It is widely acknowledged that perceptual concepts are linked to their referents by
informational relations, and it is also widely acknowledged that informational rela-
tions can be analyzed, in part at least, in terms of causal relations. Moved by these
perceptions, philosophers were for some time inclined to think it possible to explain
reference in terms of causation. Unfortunately, this explanatory project ran into
difficulties. The causal chain problem was one of them.
This problem arises because the causal relations between concepts and their
referents generally involve entities that are spatio-temporally intermediate between
the two end-terms. For example, when a passing rabbit causes an agent to form a
belief involving the concept of a rabbit, the rabbit will also cause an image of a rabbit
to form on the agent’s retina, and this image will count as another cause of the given
belief. And there will be a number of other intervening entities with these causal
properties, including a leporiform bundle of light at a half-way point between the
rabbit and the retinal image, and a projection of the retinal image onto the thalamus.
In general, if an external entity causes an agent to deploy a concept, many other
entities will cause that deployment as well. Because of this embarrass de richesses, it is
not in general true that that concepts are tied uniquely to their referents by causal
relations. By the same token, it would be false to say that having the property of
causing an agent to deploy a concept is a sufficient condition for being a referent of
the concept. Unfortunately, because of the close links between causation and

9
For the causal chain problem, see Fodor (2008, pp. 205–20). For the problem of error, see Fodor
(1990). The holism problem is formulated in Cummins (1989, pp. 62–6), the gavagai problem in chapter 2
of Quine (1960), the Kripkenstein paradox in chapters 1 and 2 of Kripke (1982), and the argument from
model theory in Putnam (1980).
 HOW CONCEPTS HOOK ONTO THE WORLD 79

information, it seems to follow that informational relations cannot be sufficient

conditions of reference, either. Whether a concept refers to an entity is, to some
degree, anyway, independent of its informational relations to that entity.
Although the causal chain problem has occasioned a great deal of discussion,
reflection shows that it is not especially serious, at least insofar as perceptual concepts
are concerned. Thus, it is generally possible to distinguish between the referent of a
concept and the other links in a causal chain by exploiting the fact that the referent
lies at the point of intersection between a number of different actual and potential
causal chains that link the concept to the world. For example, in addition to the actual
causal chain that links the concept of a rabbit both to a rabbit and to a retinal image
of a rabbit, there are other causal chains that link the concept to the same rabbit but
to quite different retinal images. The images in question differ from one another
along several dimensions, including size, shape, luminance, hue, and angular velocity.
Because of this, deployments of the concept of a rabbit can be said to track the
comings and goings of rabbits much more robustly than they track the comings and
goings of retinal images—and also, much more closely than they track the comings
and goings of thalamic projections of such images and of leporiform packets of light.
This point can be illustrated by any case in which a subject is led by a perceptual
experience to deploy the concept of a rabbit and then goes on to deploy several other
concepts that are anaphorically related to the first deployment. In any such case, the
conditional probability that a single rabbit continues to be present throughout the
series of deployments is vastly greater than the conditional probability that a single
intermediate entity continues to be present throughout the series. Deployments of
the concept prescind away from variations in proximal stimuli and other intermedi-
ate entities.
The point I mean to be making here is related Fred Dretske’s claim that it is
possible to triangulate on the objects that serve as referents of a perceptual concept by
considering a variety of intersecting causal histories.10 But the point I have in mind is
more general than Dretske’s thesis. Dretske observed that if a concept can be
triggered by a signal running through a subject’s visual system, it is often true that
the concept can be triggered by signals running through one or more of the subject’s
other sensory systems as well. For example, in addition to being responsive to causal
chains that involve retinal images, the concept of a rabbit may also be responsive to
chains that involve activity in a subject’s olfactory epithelium, and to chains that
involve activity in the subject’s tympanic membrane. This observation is true enough,
but it isn’t sufficiently general to serve my purpose here. It isn’t generally true that
people who possess the concept of a rabbit are able to discriminate between rabbits
and other animals by hearing or by smell. I would like to be able to explain how

10
Dretske (1986). Jerry Fodor has also advocated a triangulation response to the causal chain problem.
See Fodor (2008, pp. 209–15).
80 PART I : MEANING

people who lack such discriminative abilities are able to achieve objective reference.11
The immediate ancestor of the point about concepts that I have in mind is Tyler
Burge’s claim that the objective reference of perceptual representations is due to
constancy mechanisms of various kinds.12 As we all know, the visual system pre-
scinds away from the vagaries of peripheral stimuli by computing constancies of
position, size, shape, luminance, and so on. My present point can be put by saying
that perceptual concepts track their referents across changes in peripheral stimula-
tion, and also across more objective changes, such as changes in location in three
dimensional space. As a result, they are much more closely keyed to the objective
features of enduring external objects than to transitory causal traces of such objects.
I conclude that the concept of a rabbit bears a naturalistically specifiable relation R
to rabbits that it does not bear to such intervening entities as leporiform packets of
light, retinal images, and thalamic projections of such images.

VII. The problem of error

We are trying to see whether it is possible to formulate a naturalistic condition that is
necessary and sufficient for the reference of perceptual concepts, with a view to
explaining how concepts provide us with cognitive purchase on perceptible phenom-
ena. A guiding thought has been that to have cognitive purchase on an item it is
necessary and sufficient to be able to discriminate between that item and others. The
problem of error arises because we often fail in our efforts to discriminate. We often
make mistakes in our use of concepts, applying them to presented objects to which
they do not refer, and refraining from applying them to objects to which they do
refer. Now clearly, this failure of alignment between our discriminative abilities and
reference calls into question the effort to provide a naturalistic account of perceptual
reference in terms of discriminative abilities. If, as seems to be the case, there is no
alignment in a significant range of cases, no condition based on discrimination can be
necessary and sufficient for reference.
This can be made more vivid by considering (D1) and (D2):

(D1) x would judge that a presented object falls under the concept of beige if the
object was beige.
(D2) x would deny that a presented object falls under the concept of beige if the
object wasn’t beige.

11
Here is Dretske’s formulation of his proposal: “If we think of [a] detection system . . . as having the
function of enabling the organism to detect F, then the multiplicity of ways of detecting F has the
consequence that certain internal states . . . can indicate . . . that F is present without indicating anything
about the intermediate conditions that ‘tell’ it that F is present” (p. 607). In this comparatively abstract
version, Dretske’s claim comes to the same thing as the point I wish to make. But when Dretske developed
the point, he did so in terms of causal chains running through different sensory systems. For the reason
given above, I prefer to develop it is terms of multiple causal chains running through a single system.
12
See Burge (2010, chapters 8 and 9).
 HOW CONCEPTS HOOK ONTO THE WORLD 81

If x satisfied these conditions, then, it seems, the concept of beige would be tied
uniquely and determinately to the property of being beige. Unfortunately, however, x
will almost certainly be prevented from satisfying them by tendencies to make certain
kinds of error. Consider, for example, a situation in which a presented object is beige,
but in which x fails to appreciate that fact because the lighting is dim. As a result of
this perceptual error, x will not appreciate that the presented object falls under the
concept of beige. But this means that the situation will be one in which x will fail to
satisfy (D1).
There are six types of error that are pertinent to (D1) and (D2). (1) As we’ve just
been noticing, x might fail to register that a presented object was beige because the
lighting was weak or otherwise abnormal. (2) x might fail to register that a presented
object was beige because the light reflected from the object was distorted by some
factor intermediate between the object and x. Thus, for example, x might be wearing
tinted lenses. (3) x might fail to register that a presented object was beige because of a
processing glitch in x’s visual system. (4) x might erroneously judge that a presented
object was beige because of an illusion of some kind, due either to the external
circumstances of perception or to faulty internal processing. (5) x might fail to
register the actual color of a presented object because x believed falsely that the
lighting was inadequate, or that there were intermediate factors that perturbed the
reflected light, or that a processing glitch had occurred. And (6) x might fail to
register the color of a presented object because he had a false antecedent belief about
its color, and x allowed this false belief to override his current sensory evidence.
Given that these six possibilities are very likely actualized in a significant range of
cases, it would be a serious mistake to maintain that x satisfies (D1) and (D2). But if x
fails to satisfy (D1) and (D2), and x therefore fails to have the dispositions that (D1)
and (D2) describe, then we cannot say that the concept of beige is linked by those
dispositions uniquely and determinately to the color beige. If the concept is linked
uniquely and determinately to the color by any dispositions, they will have to be
dispositions that are less demanding than the ones described by (D1) and (D2).
Can we avoid this difficulty by weakening (D1) and (D2)? That is, can we make it
easier for x to satisfy them by introducing various qualifications? Perhaps, but what
exactly should those qualifications say?
Let us begin by focusing on errors of types (1), (2), (3), and (4). Since they arise
from inadequate or misleading lighting, from noise in the transmission of reflected
light, and from glitches in the processing of information by various mechanisms in
human perceptual systems, it seems possible to weaken (D1) and (D2) appropriately
by requiring (i) that the lighting must be such as to permit accurate color discrim-
inations by normal retinal processors, (ii) that the information channels linking
relevant objects to human sense organs must be highly reliable, and (iii) that the
relevant processing mechanisms must be functioning properly. If we add these
restrictions to (D1) and (D2), errors of type (1), (2), (3), and (4) will not prevent x
from satisfying them.
82 PART I : MEANING

Using the phrase “optimal perceptual circumstances” to abbreviate conditions (i)–

(iii), the present proposal can be expressed by saying that errors of types (1)–(3) and
(4) can be accommodated by changing (D1) and (D2) to read as follows:

(D1*) x would judge that a presented object falls under the concept of beige if x was
in optimal perceptual circumstances and the object was beige.
(D2*) x would deny that a presented object falls under the concept of beige if x was
in optimal perceptual circumstances and the object wasn’t beige.
It would be much easier for a typical agent to satisfy these conditions than to satisfy
(D1) and (D2). Accordingly, they come much closer to what we have in mind when
we suppose that an agent is able to discriminate between objects that are beige and
objects that are not. After all, we believe that human beings generally possess this
ability. We would not believe this if we had in mind an extremely demanding
conception of the ability like the one captured by (D1) and (D2).
This brings us to errors of type (5) and type (6). Errors of type (5) occur in
situations in which x is enjoying a perceptual experience as of an object of a certain
color, but in which x has one or more false beliefs that call into question the
veridicality of the experience, by alleging that its etiology is sub-optimal. Errors of
this sort can be blocked simply by restricting the scope of (D1*) and (D2*) to
situations in which x is taking the evidence of his senses at face value. Suppose that
x is looking at a beige object and that the circumstances of perception are optimal. In
a situation of this sort, the presented object will actually look beige to x. Accordingly,
if x is taking the evidence of his senses at face value, he will judge that the presented
object is beige. Errors of type (6) occur when x has an erroneous antecedent belief
about the color of a presented object, and allows this antecedent belief to outweigh
current sensory evidence. This sort of error can be blocked by restricting the scope of
(D1*) and (D2*) to situations in which x bases assessments of the color of presented
objects solely on the way the objects appear at the time. Suppose that a presented
object is some color other than beige, and that the conditions of perception are
optimal. In a situation of this sort, the presented object will not look beige to x.
Accordingly, if x bases his assessment of the color of the presented object solely on
the way the object looks at the time, he will judge correctly that the object is not beige.
It is clear that these two restrictions on the scope of (D1*) and (D2*) are perfectly
compatible with naturalism.
We can draw three conclusions. First, because x has the dispositions described by
(D1*) and (D2*), understood as governed by the indicated scope restrictions, we can
conclude that x has the ability to discriminate between objects that are beige and
objects that are not—or, to be more precise, that x has the ability to make such
discriminations in situations that satisfy certain constraints. Second, because this
ability involves the concept of beige, we can conclude that the concept is connected
by the ability to the color beige. And third, because the ability can be specified in
 HOW CONCEPTS HOOK ONTO THE WORLD 83

naturalistic terms, we can conclude that relation between the concept and color that
is established by the ability is fully natural in character.13
In sum, even though x will sometimes erroneously apply the concept of beige to
things that fail to be beige, it is nonetheless true that as used by x, the concept of beige
bears a naturalistically specifiable relation to all beige things that it doesn’t bear to the
things that x erroneously classifies as beige. This disposes of the problem of error. Or
so I believe.

VIII. Quine’s gavagai argument

Quine’s gavagai argument is concerned with a field linguist who is trying to deter-
mine the reference of an imaginary jungle term. Suppose that the linguist has found
that uses of the term “gavagai” are highly correlated with situations containing
rabbits. Can she conclude that the term refers to rabbits? According to Quine, the
answer is “no.” For all the linguist has thus far determined, it could be true that
“gavagai” refers instead to undetached parts of rabbits or to momentary stages of
rabbits. We can put this by saying that questions about the reference of “gavagai”
cannot be settled by considering causal or correlational relations between the use of
“gavagai” and rabbits, because any such relations will be matched by causal or
correlational relations between the use of “gavagai” and undetached rabbit parts,
and by causal or correlational relations between “gavagai” and stages of rabbits. It
could not be otherwise, for it is a necessary truth that if a rabbit stands in a causal or
correlational relation to another entity x, then an undetached part of the rabbit will
also stand in a causal or correlational relation to x, and the same will be true of a stage
of the rabbit.
Even though Quine’s argument is focused narrowly on a question about the
reference of a linguistic expression, it nonetheless seems to have important and fairly
immediate implications for our present concerns. Thus, if we keep Quine’s argument
in view, we will be inclined to doubt that the use of the concept of a rabbit is linked
uniquely to its referents by causal or correlational relations. For the argument turns
on a point about causal and correlational relations that is perfectly general—
specifically, the point that if a representation R is tied to items of any kind K by a

13
Someone might object by invoking the property P that is defined as follows: an object has P just in
case (i) it is beige, (ii) it is in a situation in which the perceptual circumstances are optimal, and (iii) it is
being viewed by an agent who is taking his current perceptual experience at face value and is basing his
judgments solely on his current experience. Reflection shows that if the dispositions described by (D1*)
and (D2*) link the concept of beige to the property beige, then there must also be dispositions that link the
concept to the quite different property P. And it can seem to follow from this that the concept of beige is
not linked uniquely to beige by a natural relation. Fortunately, however, this is an illusion: when we reflect,
it becomes clear that the natural relations described by (D1*) and (D2*) are different than the ones that
connect the concept of beige with P. There is no one natural relation that links the concept of beige both to
the property beige and to P.
84 PART I : MEANING

causal or correlational relation, it will also be tied by a relation of the same sort to
undetached parts of K’s and to stages of K’s.
I will use the expression “the new Quinean argument” to refer to the argument that
consists in applying this general point to the concept of a rabbit. Now strictly speaking,
the conclusion of this new argument is concerned only with causal and correlational
relations. I will view it here, however, as implying that the concept of a rabbit is not
linked uniquely with its referent by any natural relation. After all, it is plausible that
natural relations between concepts and their referents of other types, such as informa-
tional relations, are largely determined by causal and correlational relations.
It would be a mistake to characterize the new Quinean argument by saying that it
poses a challenge to the uniqueness of conceptual reference. The argument has no
tendency to imply that the concept of a rabbit does not refer uniquely to rabbits.
Indeed, as we saw earlier, this idea is absurd. Given that conceptual reference can be
given a deflationary characterization, it is fully determinate that the concept of a
rabbit refers to rabbits and not to rabbit parts or to stages of rabbits. Accordingly, in
characterizing the implications of the new Quinean argument for conceptual refer-
ence, we should be careful to focus on the question of whether the concept is linked
uniquely to its referents by natural relations. In view of the deflationary account of
reference, it wouldn’t be interesting to consider whether the concept is linked
uniquely to its referents by reference itself.
It would also be a mistake to characterize the new Quinean argument by saying
that it involves or rests on a skepticism as to whether it is possible to distinguish
between the conceptual role of the concept of a rabbit and the roles of other rabbitty
concepts, such as the concept of an undetached rabbit part, or by saying that it
involves or rests on a skepticism concerning the prospects of determining whether a
given agent possesses the concept of a rabbit or possesses one of the other concepts
instead. As I will illustrate in a moment, we know quite a bit about the cognitive role
of the concept, and this knowledge is quite robust—it would be foolish for a Quinean
to attempt to challenge it. Moreover, the knowledge makes it possible for us to
distinguish the concept of a rabbit from other concepts, and to determine whether
an agent has it in his or her conceptual repertoire.
The new Quinean argument accepts it as a given that the concept of a rabbit refers
to rabbits, and also that the concept plays a distinctive role in cognition. The
skepticism that it purports to establish is quite limited in scope. This skepticism
claims only that there is no causal or correlational relation that the concept bears
uniquely to rabbits. If it bears any such relation to rabbits, it will bear that relation to
one or more other rabbitty entities as well.
What do we know about the cognitive role of the concept of a rabbit? One
important fact about the concept is that non-deferential possession of it involves a
sensitivity to the spatial and temporal boundaries of rabbits.14 To possess the

14
Gareth Evans was the first to appreciate the relevance of this fact to Quine’s gavagai argument. See
Evans (1975).
 HOW CONCEPTS HOOK ONTO THE WORLD 85

concept, one must know how to trace the boundaries that separate a rabbit from its
spatial context, and must be able to determine, in principle, at least, when a rabbit
comes into existence and when it passes away. Now it is also true that one must be
sensitive to the spatial boundaries of rabbits in order to possess the concept of an
undetached rabbit part, and that one must be sensitive to the temporal boundaries of
rabbits in order to possess the concept of an undetached rabbit stage. This is because
an undetached rabbit part is necessarily a constituent of a whole rabbit, and because
an undetached rabbit stage is necessarily a constituent of a series of stages that
constitutes an enduring animal. But the deployment of one of these other concepts
requires more than a sensitivity to boundaries. It also involves using attention to
select a specific part or a specific stage. Thus, to deploy a concept of the form this
undetached rabbit part, one must determine that a whole rabbit is present, but one
must also use attention to select a specific part of that rabbit, and one must also
appreciate that the part is a constituent of the whole. Equally, to deploy a concept of
the form this undetached rabbit stage to a perceived item, one must determine that an
enduring animal is present, but one must also use attention to select a specific stage of
that animal, and one must appreciate that the stage belongs to the appropriate series
of stages. To put the point in slightly different terms, it takes two types of spatial
sensitivity to guide the use of the concept of an undetached rabbit part, one that has
to do with the typical shape of a rabbit, and another that has to do with the
boundaries of a particular spatial region that has been selected by attention. And
by the same token, it takes two types of temporal sensitivity to guide the use of the
concept of an undetached rabbit stage. I am assuming here that whether the use of a
concept is governed by a particular form of attention is always a factual question, and
that the answer can in principle be determined by cognitive science. Of course, since
Quine wrote from the perspective of behaviorism, he would no doubt have rejected
this assumption. But from our current perspective, it seems entirely reasonable.
There are other ways in which the cognitive role of the concept of a rabbit is
distinctive. Thus, for example, if an agent has mastered the use of the concept and has
also mastered the use of the concept of an undetached rabbit and the concept is a part
of, then the agent will believe that every undetached rabbit part is a part of a whole
rabbit. This belief will be recognizable as such, because the concept is a part of has a
distinctive cognitive role, consisting in the role it plays in propositions that we use to
register perceptual data (e.g., propositions of the form This region is a part of that
region), and also in the role that it plays in the a priori principles of mereology
(which, by the way, have a recognizable formal structure). Further, as Quine himself
pointed out, the concept of a rabbit interacts in distinctive ways with numerical
concepts and the concept of identity: an agent will reject the proposition There are
many rabbits here in contexts in which the agent will accept There are many
undetached rabbit parts here, and also in contexts in which the agent will accept
There are many rabbit stages here. Quine might say that this fact cannot be used to
distinguish the concept of a rabbit from the other rabbitty concepts because it is
86 PART I : MEANING

impossible to discriminate numerical concepts and the concept of identity from

various other concepts that have similar formal properties. But I think we can see
that any such objection would be misguided. Consider, for example, the following
feature of the cognitive role of the concept of identity: If one is committed to a
proposition of the form A is identical with B, then, for every predicative concept C
that belongs to one’s conceptual scheme, one is disposed to accept the proposition
that A is C just in case one is also disposed to accept the proposition that B is C, and
moreover, the same is true for every concept C* that does not actually belong to one’s
conceptual scheme but might be added to it. Reflection shows that this is a feature of
the concept of identity, and also that it is not shared by other concepts.15
In view of these considerations, we can see that possession of the concept of a
rabbit can in principle be distinguished from possession of either of the other rabbitty
concepts. We can also see that it is possible to determine which of the three concepts
is being deployed on a particular occasion. But more: we are in a position to see that
the concept of a rabbit is linked uniquely to rabbits by the acceptance conditions of
certain propositions containing it. To illustrate, we have observed that the concept of
identity is distinguished from similar concepts by facts concerning its conceptual
role. It follows from this that numerical concepts have distinctive roles, for those
concepts enjoy distinctive logical links to the concept of identity. But this means that
thoughts like There are exactly two rabbits here can be recognized as such. What
I wish to claim now is that the acceptance conditions of such thoughts link the
concept of a rabbit uniquely to rabbits. The given thought is acceptable just in case
two whole, enduring rabbits are present. It isn’t enough that the context contain two
undetached rabbit parts or two undetached stages.
I should add that this is only part of the story. In order to arrive at beliefs involving
the concept of a rabbit, we must privilege information that is pertinent to the shapes
and locations of rabbits over information about the shapes and locations of detached
parts and we must also privilege information about properties that can only be

15
The lines of thought of the last two paragraphs receive further elaboration in my Hill (1972) and Hill
(2011b).
I emphasize that the feature I am claiming to distinguish identity from other concepts is concerned with
predicative concepts that might be added to an agent’s conceptual scheme as well as concepts that actually
belong to the scheme. Without the reference to possible extensions of the scheme, my claim would reduce
to the thesis that the concept of identity is uniquely governed by the following principle: if one is
committed to a proposition of the form A is identical with B, then, for every predicative concept C that
belongs to one’s conceptual scheme, one is disposed to accept the proposition that A is C just in case one is
also disposed to accept the proposition that B is C. This thesis is false. If an agent’s conceptual vocabulary
has an impoverished stock of predicative concepts, then a concept might obey this principle, as used by the
agent, even if it was not the concept of identity. (See, for example, Quine (1969, p. 15).) But identity is
uniquely governed by the much stronger principle in the text, which in effect says that if a concept is to
count as the concept of identity, then it must be true that the agent’s use of the concept conforms to
Leibniz’s Law, with respect to the predicative concepts that are actually included in the agent’s conceptual
lexicon, and it must also be true that the agent’s use of it would continue to conform to Leibniz’s law under
any expansion of that lexicon. For discussion see the “Afterward” that follows my chapter “ ‘gavagai’ ” in
the present volume.
 HOW CONCEPTS HOOK ONTO THE WORLD 87

instantiated by temporally extended objects over information about properties that

can be instantiated by stages. Does this talk of privileging of streams of information
of certain kinds make sense in naturalistic terms? Yes. To be sure, it would not have
made much sense to Quine, whose psychological commitments had a markedly
behaviorist character; but it is not hard to see that it is sustained by contemporary
cognitive science.16
The new Quinean argument focuses on concepts rather than linguistic expres-
sions. This opens the door to appealing to facts about the possession conditions of
concepts in evaluating the argument. I have maintained that the possession condi-
tions for perceptually grounded sortal concepts are such as to distinguish them from
other concepts, in that they reflect perceptible spatio-temporal boundaries, and that
their distinguishing features include natural relations that tie them uniquely to their
referents. If my arguments are right, they vindicate the intuition that sortal concepts
give us a robust cognitive purchase on specific portions of reality.

IX. Conclusion
Reference connects concepts with extraconceptual reality, but not in a way that is
intrinsically interesting or important. Reference is derivatively interesting, however,
due to its associations with causal, correlational, and informational relations, and
with the ability to discriminate between different objects and properties. I have been
concerned in this paper to defend the view that in the case of perceptual concepts, at
least, reference tracks these important natural relations quite closely.
This view is independently plausible. If I am able to track my friend Ivan through
time, recognizing him after having been out of touch with him for a while, and using
his past behavior as a basis for predicting his future words and deeds, this is in large
part due to the fact that I have a concept that refers to him. If I am able to adopt the
goal of having Brussels sprouts for supper, and succeed in reaching this goal, this is
largely due to the fact that I have a concept that refers to Brussels sprouts. And if I am
able to arrange to meet Ivan at Irv’s Restaurant, and actually connect with him at
Irv’s, this is largely due to the fact that we both possess a concept that refers to that
establishment. Now intuitively speaking, it would not be possible for these concepts
to support the cognitive and practical activities in question if they were not connected
in determinate and causally robust ways to their referents. But if this is true, then
reference must be in step with one or more natural relations, at least in cases like the
ones described here, in which concepts provide agents with a cognitive purchase on
specific entities.

16
John Campbell also argues that facts about the ways in which information streams are privileged by
cognitive equipment within the subject can be used to undercut Quinean funny business, but there are
significant differences in our overall approaches to the gavagai problem. See Campbell (2002, pp. 216–23).
88 PART I : MEANING

Despite its prima facie plausibility, however, this commonsense view has been
challenged by powerful arguments. My main concern in recent sections has been to
respond to these challenges. To the extent that my efforts have been successful, we are
entitled to renew our commitment to the commonsense view.
According to one version of the commonsense view, it is possible to formulate
necessary and sufficient conditions for the reference of particular categories of
perceptual concepts in naturalistic terms. That is to say, it is possible to establish
that some generalizations of the following form are true:

(P) For any concept x, if x belongs to the category Ki, then for any object y, x refers
to y just in case x bears Ni to y,
where Ki is a category of concepts and Ni is a natural relation. The arguments in recent
sections tend to support this idea. More specifically, for the case where Ki is the
category of perceptual concepts, I have argued that it is possible to specify, in fully
naturalistic terms, three relations that collectively anchor perceptual concepts pretty
firmly to their referents. I haven’t as yet attempted to combine these three relations
into a single, unified relation, but I will now say that they can all be classified as forms
of sensitivity. Thus, in discussing the causal chain problem, we saw that the use of a
concept may be sensitive to the comings and goings of its referent while being insensi-
tive to the comings and goings of, say, retinal projections of the referent. Further, in
considering the problem of error, we observed that an agent may display a sensitivity to
differences between the referent of a concept and other items in ideal perceptual
circumstances, even if he has a tendency to confuse them under less favorable condi-
tions. And, finally, we noticed that the acceptance conditions for propositions involving
the concept of a rabbit are, in effect, sensitive to the differences between the presence of
whole, enduring rabbits and the presence of other rabbitty entities. In view of these
considerations, it appears that sensitivity comprehends all three of the naturalistic
relations we have considered. Accordingly, I think we can say that, insofar as we are
prepared to rely on sensitivity in formulating a naturalistic account of the reference of
perceptual concepts, we have little to fear from the standard objections to naturalism.

Appendix
I have maintained that basic perceptual concepts are connected uniquely with entities in the
world by a natural relation. How far into the higher reaches of our conceptual scheme can the
argument be extended? What about the theoretical concepts of the natural sciences? Is there
any reason to think that they too are anchored securely to the world, and can therefore ground
our cognitive purchase on their referents? What about mathematical concepts?
I cannot do full justice to these questions here. My goal is simply to sketch a few lines of
inquiry that seem worth pursuing. In the interests of brevity I will focus on two comparatively
simple systems of conceptual representation. One, which I will call G, consists of the concepts
that belong to a theory of gravitation, together with the thoughts that can be formulated using
those concepts. I will assume that the syntax used to generate the thoughts is quite simple,
 HOW CONCEPTS HOOK ONTO THE WORLD 89

approximating to the syntax of a first order language. The other system, P, consists of the
concepts that belong to Peano arithmetic, together with the thoughts that can be formulated
using those concepts. Again, I will assume that the relevant grammatical structures are simple.
We noticed earlier that there are many functions mapping concepts onto objects and
properties. As a special case of this general fact, there is a large set GS of interpretation functions
mapping the concepts of G onto objects and properties. There is also a large set PS of functions
that do the same for the concepts of Peano arithmetic. I will begin by describing several
naturalistic constraints on the members of GS—constraints designed to reduce GS to a smaller
totality consisting only of interpretations of G that are admissible. It would be nice if we could
conclude that the constraints narrow the class of interpretations to one. Unfortunately, however,
the vagueness of one of the constraints will preclude our reaching any such precise conclusion. In
effect, the argument will show that there are some fairly powerful naturalistic necessary condi-
tions on the reference of theoretical concepts, but it will not provide us with a sufficient
condition. Although it may well be possible to approximate a sufficient condition more closely
by sharpening the constraints, any such undertaking is beyond the scope of the present chapter.
I will conclude by making some related observations about the class PS of interpretations of P.
The first constraint is that if an interpretation function is to count as admissible, it must map
each perceptual concept onto the entity to which it refers. This constraint can be given a
naturalistic construal because, as we saw in the main body of the chapter, there are naturalistic
relations between perceptual concepts and their referents that uniquely determine the referents.
The second constraint is that an interpretation function must map each theoretical concept C
onto an entity E such that users of C are disposed to use it to encode information about E. Now
this constraint will exclude many interpretations of G, but it will presumably allow a number of
other interpretations to count as admissible. Thus, for example, if a concept C in the lexicon of G
is used to encode information about an entity x, it will also be true that it is used to encode
information about the other entities in the causal chain running from x to C. How many
interpretations will be excluded? To answer this question, we would need a much more exacting
formulation of the present constraint. As it stands, it is too vague to permit a definite answer. The
upside is that it is sometimes possible to tighten informational constraints on reference so as to
make them both more demanding and more precise. We saw this to be true in Sections VI–IX.
Perhaps it is possible to achieve a similar tightening of the present informational constraint.
The third naturalistic constraint is that an interpretation function must confer truth on the
thoughts that count as axioms of G (and therefore on all the thoughts that follow from those
initial thoughts by the rules of logic). In putting this constraint forward, I don’t mean to invoke
the commonsense concept of truth—the concept that substitutionalism claims to analyze.
Rather I have in mind a technical concept that is a counterpart of the model-theoretic notion of
truth under an interpretation. The model-theoretic notion can be explicitly defined in terms of
concepts drawn from syntax and set theory, so it is fully naturalistic in character. The
counterpart notion that I have in mind is susceptible of a similar definition, and is therefore
also fully naturalistic, but it differs from the model-theoretic notion in that it applies to entities
with a logical grammar that is different than the logical grammar of the artificial-language
sentences to which the model-theoretic notion applies.17 A further difference is that we do not
today know how to characterize the grammar of concepts and thoughts. Perhaps it is closely

17
I defend the appropriateness of ascribing syntactic structures to thoughts in Section III of the
preceding chapter, “A Substitutional Theory of Truth, Reference, and Semantic Correspondence.”
90 PART I : MEANING

akin to the grammar of sentences from natural language, but we are not yet in a position to say
for sure whether this is so. All that can be said is that at the end of the day, when the syntax of
concepts and thoughts is well understood, it will be possible to characterize the notion of truth
I am invoking here in terms of syntax and set theory, in the same way that Tarski originally
characterized the corresponding notion that applies to simple artificial languages.18 The notion
is in principle naturalizable, though we cannot claim that it has already been naturalized.
We know from model theory that if a set of sentences comes out true under one interpret-
ation, it will also come out true under countless others. The same must be true of thoughts: if a
set of thoughts, such as the axioms of G, comes out true under one interpretation, there will be
countless other interpretations that also confer truth on the thoughts, most of which make no
contact with the entities to which the conceptual constituents of the thoughts refer. Thus, for
example, there will be truth-conferring interpretations of the axioms of G that assign properties
of numbers to concepts like mass and gravitation. Hence, the third constraint is a weak
constraint. When combined with the others, however, it can have considerable bite. Another
way to put this is to say that in combination, the three naturalistic constraints amount to a
fairly demanding necessary condition on reference. How close they come to constituting a
sufficient condition is a question that I will have to leave open here.
This brings us to the mathematical theory P and the class PS of its associated interpretations.
Just as it is in principle possible, by using an essentially Tarskian procedure, to define a
relational truth predicate that applies to the thoughts in G, so also it is in principle possible
to use a Tarskian procedure to define a relational truth predicate that applies to the thoughts in
P. Now the axioms and theorems of P will be true under a number of these interpretations, and
we can exclude many of the members of PS by focusing on the ones that have this property.
This restricted class will include an interpretation function that mimics the reference relation
for P in its assignments of objects and properties to concepts, but it will also include countless
others that diverge abruptly from the reference relation. Hence, as with G, it is only a weak
constraint to say that admissible interpretations must confer truth on the axioms and theorems
of P. But still, if it is possible to formulate other naturalistic constraints on interpretations of P,
this one might combine with them to good effect.
Is it possible to formulate other naturalistic constraints? Philosophers and mathematicians
have often maintained that mathematical theories are concerned exclusively with ideal
objects—objects that lie outside space and time and are devoid of causal properties. On this
view, it is very hard to imagine additional naturalistic constraints. But the picture brightens
considerably if we assume that the concepts that figure in P are the same as the concepts we use
in counting and in manipulating the numerical results that counting delivers. Suppose, for
example, that the concept of 2 that figures in P is the same as the concept we use in asserting
that there are 2 ducks on the pond, 2 quarters on the table, 2 clouds hanging over the
mountain, and 2 squads of soldiers on the road. On this assumption, it is natural to think
that we use the mathematical concept of 2 to encode information about a certain second order
property—the property of having 2 instances. It seems to make good sense to speak of
information here, and by the same token, good sense to speak of counterfactual dependence
and nomological covariation. You would not have asserted that there are 2 ducks on the pond
if the property of being a duck on the pond had not had the second order property of having 2

18
See Tarski (1983).
 HOW CONCEPTS HOOK ONTO THE WORLD 91

instances. Equally, your use of the concept of 2 in contexts of the form There are two F’s
covaries nomologically with situations in which perceptually accessible first order properties
exemplify the second order property in question. It seems that we can exploit relationships of
this sort in formulating a second naturalistic constraint on the interpretations of P.
All of this seems right, or at least defensible, but there is a problem. It is presumably true that
numerical concepts increase in complexity as their referents increase in magnitude. For
example, it is presumably true that the concept of 2,222 is more complex than the concept
of 2. Moreover, it is no doubt true that there is an upper limit to the complexity of the concepts
that the human mind can entertain. In principle, however, there are infinitely many numerical
concepts, each of them the output of a recursive process that mirrors the behavior of the
successor function. Now all of these concepts belong to the theory P, which is an ideal entity,
not affected by the performance limitations of human cognitive faculties. Hence, it can seem
that there are constituents of P that could not be linked to the world by naturalistic relations of
the sort we have just been considering. Concepts that refer to smaller numbers are linked to the
world in these ways, but it can seem that concepts that refer to very large numbers could not be.
This can seem to be true even if we think of the links as dispositional. As Kripke emphasized in
a similar context, there are problems with supposing that human beings have infinitely many
dispositions.19 It seems that the limited capacities of cognitive faculties like working memory
impose restrictions on the dispositions that human beings can possess. If working memory is
too small to hold a given concept, how can it be true that we are disposed to use that concept to
encode information about a second order property?
I think this problem can be solved by distinguishing between the cognitive abilities that we
actually possess and the abilities we would possess if certain of our cognitive faculties were
extended in relevant directions. In other words, it can be solved by considering idealized
versions of ourselves. Consider a Turing machine with a finite tape—or, rather, consider a
physical calculating device that realizes such a machine. Suppose the tape has j squares, and
suppose also that the machine is programmed to compute the successor function. Given the
size of its tape, the machine can only compute the successor function for a limited number of
arguments. But it is nonetheless true that if the machine’s tape were extended, then it would be
able to compute values of the function for larger arguments.20 Moreover, it’s true that if its tape
were extended indefinitely, then for any numerical argument n, the machine would be able to
compute n + 1. It is more difficult to describe the relevant enhancements in the case of human
beings, but the following counterfactual does a reasonably good job:

If an agent x were able to use the concept of n to encode the information that n F ’s were
present, and another F was added to the total, then, assuming that x was able to entertain the
concept of n + 1 and also able to use it in thought, if x was able to detect the additional F, x
would use the concept of n + 1 to encode the information that n + 1 F’s were present.

19
Kripke (1982, chapter 2).
20
Of course, a physically realized Turing machine will have parts, and those parts will wear out (at
different rates) across time. But as long as those parts are replaced with molecular twins of the originals, the
physical dispositions of the machine will remain constant. Since the computational powers of any machine
are fixed by its physical dispositions, it follows that the computational powers of a Turing machine are
independent of the specific physical parts of which it is composed at origin. (I am indebted to Steve Angle
and Jamie Dreier for discussion of this point.)
92 PART I : MEANING

It seems natural to suppose that human agents satisfy counterfactuals of this sort, and also that,
because they satisfy such counterfactuals, all of the numerical concepts in P are linked by
natural relations to numerical properties—including even the numerical concepts that are too
complex for human beings to actively entertain.
Kripke rejected this strategy because he thought that in imagining enhancements of faculties
like memory, and imagining the attendant extensions of our life span, we might be imagining
situations in which the enhancements and extensions perturbed or disabled other features of
the our cognitive profile that are relevant to the use of numerical concepts. Perhaps the closest
worlds in which human agents live for significantly longer periods of time are worlds in which
we go crazy through boredom, and therefore worlds in which we are incapable of caring about
the cardinalities of very large sets. To be sure, we can try to guard against such disruptions by
building “ceteris paribus” clauses into our descriptions of the relevant counterfactual situ-
ations, but Kripke worried that such clauses might be equivalent to conditions like “and x
retained the abilities relevant to using numerical concepts to encode information about second
order numerical properties.” If we were to replace “all things being equal” with a condition of
this sort, the resulting conditional would be a tautology. It couldn’t be used to articulate a
substantive natural relationship between a numerical concept and its referent.
It is true that enhancements of some human faculties might degrade others, but it seems that
we could stipulatively focus on (possibly remote) situations in which this does not happen by
simply listing all of the relevant faculties and specifying exactly the changes in each that we
wish to consider. We cannot of course do this today, because we don’t have a sufficiently deep
understanding of the psychology of counting and numerical computation, but that purely
epistemic limitation has no tendency to discredit the strategy, which is after all a proposal for
dealing with a metaphysical problem. Today we must rely on a “ceteris paribus” condition in
specifying the relevant counterfactual situations, but that is not an intrinsic feature of the
approach. This claim is supported by the Turing machine analogy. Turing machines are
sufficiently simple that we can specify them exhaustively, and when we do so, it becomes
easy to describe the specific enhancements we are imagining, and to specify the aspects of a
machine that are to remain constant, without relying on nebulous “ceteris paribus” conditions.
If human beings were as well understood as Turing machines, we could handle the “ceteris
paribus” worry about specifications of enhanced human faculties in the same way.21
It appears, then, that there are non-trivial naturalistic constraints on interpretations of
Peano arithmetic. It remains to be seen whether numerical concepts are tied uniquely to their
referents by natural relations. That question will have to await another occasion.

21
Another way to put this is to say that if we knew enough, we could identify those parts of the
functional architecture of the human brain that are responsible for computing function like successor or
plus, and also the hardware elements that support this architecture. This would in effect enable us to pick
out a physical embodiment of a Turing machine. We could then consider the computational behavior of
this machine in an infinite series of possible worlds in which its “memory tape” was successively expanded.
Its behavior in these worlds could ground our saying that a human being whose brain contains the Turing
machine in the actual world is disposed to compute all of the values of the relevant numerical function.
Connectionists might wish to challenge the assumption that a component of a human brain can
naturally be described as a Turing machine, on the grounds that human computational faculties are
inextricably intertwined with the finite memory stores that the brain makes available. I find this incredible
for cases like the present one, in which relevant algorithms can be executed on paper.
 HOW CONCEPTS HOOK ONTO THE WORLD 93

Anyway, however that question is answered, there will be many open questions concerning
the interpretations of other scientific and mathematical theories. It can hardly be claimed that
G is a typical scientific theory, given that I have assumed that it has an axiomatic structure, or
that the features of Peano arithmetic are characteristic of mathematical theories generally.
Thus, for example, theories in neuroscience and psychology lack the formal structure of G, and
as a result, it is much less clear how their theoretical concepts are related to observation. By the
same token, it is much more difficult to assess the constraints that interpretations of their
observational vocabularies impose on interpretations of their theoretical concepts. As for
Peano arithmetic, the fact that its numerical concepts are used in counting observable objects
creates links between those concepts and the physical world that are much more transparent
than the corresponding links involving the concepts of real number theory or complex
analysis. And of course, the domain of Peano arithmetic is much smaller than the domains
of those other theories, and vastly smaller than the domain of set theory. There seems little
chance of showing that we have a cognitive purchase on those domains by appealing to
dispositions, even if the dispositions involve considerable idealizations. In general, there
seems to be little chance of generalizing the arguments of this Appendix so that they apply
to scientific and mathematical theories across the board.
I conclude by claiming (i) that there are a number of different ways in which concepts are
linked to objects by natural relations, (ii) that the linkage is unique in some cases but not in all,
and (iii) that in cases in which the linkage is not unique, there is considerable variation in the
degrees to which uniqueness is approximated.22

References
Burge, T. (2010). Origins of Objectivity. Oxford: Oxford University Press.
Campbell, J. (2002). Reference and Consciousness. Oxford: Oxford University Press.
Cummins, R. (1989). Meaning and Mental Representation. Cambridge, MA: MIT Press.

22
I began Section III by assuming that deflationism “defines” or “specifies” a determinate set of ordered
pairs <x, y> such that x is a concept and y is the entity that counts as the referent of x. Having adopted this
assumption, I went on to ask whether the members of the pairs are linked to one another determinately and
uniquely by natural relations. In the case of perceptual concepts, I argued that the answer to this question is
affirmative. More recently, however, with reference to other cases, I have maintained that the answer is
less clear.
I would like now to call attention to the possibility that my initial assumption is in a sense hostage to the
outcome of the subsequent inquiry into natural relations. If we were to find that a certain category of concepts
C fails to be linked uniquely and determinately to extraconceptual entities by natural relations, then, it seems,
we could question my initial assumption—or at least, the part of it that is concerned with C. That is, it seems
that we could question the hypothesis that deflationism succeeds in “specifying” a set of ordered pairs that
connect members of C with particular entities. After all, if the members of a subset of concepts (say the subset
consisting of concepts of infinite classes) are not firmly anchored to the world by a set of natural relations, one
might wonder how they could participate in the “specification” of a determinate set of entities.
The task of examining this line of thought is complex and delicate, and cannot be undertaken here. But
I hope enough has been said to make it clear that when I speak of deflationism as specifying a determinate
set of pairs, my claim is meant to be provisional. It requires defense. Insofar as concepts turn out to be
linked uniquely to extraconceptual entities by natural relations, the claim will be vindicated. But if it should
turn out that one or more categories of concepts are not anchored to the world in this way, there will be
reason to question it.
94 PART I : MEANING

Dretske, F. (1986). “Misrepresentation.” In Radu Bogdan (ed.), Belief: Form, Content, and
Function. Oxford: Oxford University Press, 597–610.
Evans, G. (1975). “Identity and Predication,” The Journal of Philosophy LXXII: 343–63.
Fodor, J. A. (1990). “A Theory of Content II.” In his A Theory of Content and Other Essays.
Cambridge, MA: MIT Press, 89–136.
Fodor, J. A. (2008). LOT II: The Language of Thought Revisited. Oxford: Oxford University
Press, 205–20.
Hill, C. S. (1972). “ ‘gavagai,’ ” Analysis XXXII: 68–75.
Hill, C. S. (2002). Thought and World: An Austere Portrayal of Truth, Reference, and Semantic
Correspondence. Cambridge: Cambridge University Press.
Hill, C. S. (2010). “Intentionality Downsized.” In Ernest Sosa and Enrique Villanueva (eds.),
Philosophical Issues 20: 144–69.
Hill, C. S. (2011a). “Review of Paul Horwich, Truth—Meaning—Reality,” Mind 120: 1262–70.
Hill, C. S. (2011b). “Can Carey Answer Quine?,” Behavioral and Brain Sciences 34: 132–3.
Horwich, P. (1998). Truth, 2nd edition, Oxford: Oxford University Press.
Kripke, S. A. (1982). Wittgenstein on Rules and Private Languages. Cambridge, MA: Harvard
University Press.
Putnam, H. (1980). “Models and Reality,” Journal of Symbolic Logic 45: 464–82.
Quine, W. V. (1960). Word and Object. Cambridge, MA: MIT Press.
Quine, W. V. (1969). Set Theory and Its Logic, 2nd edition. Cambridge, MA: Harvard
University Press.
Tarski, A. (1983). “The Concept of Truth in Formalized Languages.” In J. Corcoran (ed.), Logic,
Semantics, Metamathematics, 2nd edition. Indianapolis: Hackett, 152–278.
PART II
A Type-materialist Theory
of Experience
 6
In Defense of Type-materialism*

I
Conscious experiences have two kinds of qualities that are of importance to philo-
sophers: some of their qualities are representational and others are phenomenal. As
their name suggests, representational qualities are properties that conscious experi-
ences have in virtue of representing or signifying various objects or states of affairs.
They include being an image of my grandmother, being a thought about Vienna, and
being the thought that my friend Ray will laugh when he hears this joke. On the other
hand, the class of phenomenal qualities can be defined, at least in a rough and
preliminary way, as the class of nonrepresentational qualities of conscious experi-
ences of which one can have knowledge of an especially high grade through intro-
spection. Phenomenal qualities include being a pain, being a yellowish-orange
afterimage, and being a bitter gustatory sensation.
A phenomenal property is a phenomenal kind if it is expressible by a term that can
play a role in lawlike psychological generalizations.
Further, a property is a neurological quality if it is expressible by a term that can be
constructed from the primitive predicates of some adequate formulation of the laws
of neuroscience. A neurological quality is a neurological kind if it is expressible by a
term that can play a role in lawlike neurological generalizations.
In this chapter I discuss a view about phenomenal qualities and phenomenal kinds
that is sometimes known as type-materialism. This view may be formulated by
saying, first, that every phenomenal quality is identical with some neurological
quality and, second, that every phenomenal kind is identical with some neurological
kind. Thus, according to the view I wish to discuss, if X is any phenomenal quality,
then there exists a neurological quality Y such that X and Y satisfy these three
conditions: first, Y is a neurological kind if X is a phenomenal kind; second, it is
logically impossible for an event to be an instance of X without also being an instance
of Y; and, third, it is logically impossible for an event to be an instance of Y without

* I have benefited considerably from the philosophical talents and maieutic skills of Ivan Fox. I have also
received valuable help from William G. Lycan, R. J. Nelson, David Westendorf, and an anonymous referee
for Synthese. My research was supported by a N.E.H. Summer Stipend.
98 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

also being an instance of X. (Recall here that representational qualities do not count
as phenomenal qualities.)
Although it is defended briefly in Putnam’s latest book (Putnam 1981), type-
materialism has not fared well in the recent literature. It is one of the least popular
positions in philosophy of mind, if not all of philosophy. It seems to me, however,
that its bad reputation is largely undeserved. I hold that when one takes a careful and
critical look at the arguments that are widely held to refute type-materialism, one is
led inevitably to the conclusion that they are incapable of playing the dialectical role
that has been claimed for them. Further, although I recognize that we are not
currently in a position to conclude that it is true, I believe that the progress of science
may well someday put us in a position in which it will be necessary to accept type-
materialism. I try to justify these claims in the present chapter.1

II
Type-materialism presupposes a view about the relationship between phenomenal
kinds and neurological kinds that I call the Correlation Thesis. According to the
Correlation Thesis, if X is a phenomenal quality, then there exists a neurological
quality Y such that X and Y satisfy these two conditions: first, Y is a neurological kind
if X is a phenomenal kind; and second, an individual has a conscious experience that
is an instance of X when and only when an instance of Y takes place in the
individual’s brain. Much of the literature on type-materialism is centered around
this thesis, and much of what I will say in the present chapter is connected with it in
one way or another.
With the Correlation Thesis in hand, it is possible to impose a useful tripartite
division on the class of objections to type-materialism. First, there are objections
which purport to show that type-materialism must be false even if the Correlation
Thesis is true. Second, there are objections which purport to call the Correlation
Thesis into question on conceptual or a priori grounds. And third, there are empirical
objections to the Correlation Thesis.

1
I make no attempt to discuss objections to type-materialism that are also objections to token-
materialism. (Token-materialism implies that conscious experiences are identical with brain processes,
but it does not claim that phenomenal qualities are identical with neurological qualities.) Thus, for
example, I say nothing about the Location Objection, the Grain Problem, or the Appeal to Bats. My
main reason for excluding such arguments is that there seems to be something of a consensus at present
that some form of token-materialism (more specifically, some form of Functionalism) is true. If this is
correct, then I can count on most readers to share my view that the standard arguments against token-
materialism do not work.
I do not discuss Davidson’s argument against type-materialism (see, e.g., Davidson 1970) because it
seems to me that it is not directed against the form of type-materialism that I accept. As I understand it,
Davidson’s objection counts primarily against the view that propositional attitudes can be identified with
neurological qualities. I share Davidson’s doubts about this view.
 IN DEFENSE OF TYPE - MATERIALISM 99

After a section that states what I take to be the main justification for believing that
type-materialism is true, I try (in Sections IV and V) to answer two objections that
belong to the first of our three major categories of objections: the objection that type-
materialism is ruled out by certain Cartesian intuitions about possibility, and the
objection that it is incompatible with the “striking qualitative differences” between
conscious experiences and brain processes. I then turn (in Sections VI and VII) to
consider two a priori objections to the Correlation Thesis: the objection that it is
incompatible with our perceptions concerning the mental states of certain beings
(e.g., C3P0) that do not fall within the domain of neurology, and the objection that it
presupposes claims about the conscious experiences of other organisms that no
individual could know to be true. These are the objections from the first and second
categories that seem to me, qua lay historian of analytic philosophy, to have played
the largest role in turning public opinion against type-materialism.
Although my primary objectives in the present chapter are conceptual, I also briefly
address the main empirical objections against the Correlation Thesis. I do not attempt
to construct decisive refutations of such objections, but I do try (in Section VIII) to
show that they leave the type-materialist plenty of logical room in which to maneuver.
As for empirical considerations that count in favor of the Correlation Thesis,
I believe that we are at present in possession of a certain amount of supporting
evidence, and that it is not unreasonable to expect that such evidence will continue to
accumulate.2 However, I also believe that we do not yet have enough facts to be able

2
Some of the evidence for the Correlation Thesis comes from studies of the relationships between
sensory disorders and brain injuries, and some comes from operations during which the brains of
conscious human patients have been subjected to electrical stimulation. Evidence from these sources has
led to the adoption of some non-trivial hypotheses.
Take for example, the realm of visual experience. Before modern stimulation techniques were developed,
investigation of the relationships between visual experience and brain processes consisted primarily of
studies of correlations between lacunae in the visual field and brain injuries. Information from these studies
suggested an hypothesis that postulated an exhaustive mapping from specific positions in the visual field to
specific regions in the striate cortex. (A diagram of this mapping and a review of the evidence that first
suggested it may be found in Holmes (1945).) In more recent years, confirmation of the hypothesis has
been provided by the fascinating experiments of such investigators as Brindley and Dobelle. Thus, Brindley
and Lewin summarize some of their research as follows (1968, p. 479): “For weak stimuli the map of the
visual field on the cortex [suggested by stimulation experiments] agrees with the classical maps . . .” (See
also Brindley (1973).) Dobelle and Mladejovsky reach the same conclusion (1974, p. 570): “The phosphene
produced by any given electrode appears at a position which corresponds roughly to expectations based on
classical maps showing the projection of the visual field onto the cortical surface.” (A cortical phosphene is
defined as a visual sensation that is caused by direct stimulation of the brain. Phosphenes come in a variety
of colors and shapes (they can take the form of a ball, a disc, a star, a cloud, and even a pinwheel), but they
are typically small spots of white light. Multiple phosphenes may be produced simultaneously by using
multiple electrodes.)
The somatosensory field affords another example. Building on the work of Foerster, Penfield was able to
obtain considerable evidence for the hypothesis that the somatosensory field is projected onto the
postcentral gyrus. The relative positions of images of locations in the sometosensory field were found to
be constant over a wide range of subjects. (See Penfield and Rasmassen (1950, pp. 21–45). Penfield was also
able to obtain many other relevant results, including a correlation between stimulation of a region he called
the interpretive cortex and the experience of déjà vu. See Penfield (1959).)
100 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

to determine whether the Correlation Thesis is empirically true or empirically false.

At present it is at best a conjecture.

III
In the present section I sketch a conditional justification of type-materialism. That is
to say, I argue that if it should turn out that the Correlation Thesis is true, then there
would be a strong justification for accepting type-materialism.
The argument I have in mind can be summarized as follows:
First Premise: If a theory T provides a good explanation of a set of phenomena, and it is also
true that the explanation is better than any explanation provided by a competing theory, then it
is permissible and even obligatory to believe that T is true.
Second Premise: Type-materialism provides a good explanation of the psychophysical laws that
are claimed to exist in the Correlation Thesis. Moreover, the explanation that it provides is
superior to the explanations provided by all competing theories.
Conclusion: If the Correlation Thesis should turn out to be true, then it would be permissible
and even obligatory to believe that type-materialism is true.

In short, I wish to claim that it may be possible to justify type-materialism by appealing

to its explanatory power.3
It seems reasonable to suppose that a principle like the first premise (hereafter
called the Best Explanation Principle) would be a component of any adequate theory
of rational belief, for it seems to be necessary to use some such principle in explaining
why it is rational to accept scientific theories that postulate unobservable phenomena.
Moreover, as Harman has often argued (e.g., Harman 1966 and Harman 1968), the
Best Explanation Principle underlies a great deal of the non-deductive reasoning that
takes place in everyday life. For example, when a detective concludes that the butler is
the murderer, he often does so because he has a high opinion of the explanatory
power of that hypothesis.4

3
Several other authors have noted that materialism can be defended by appealing to its explanatory
power. Thus, for example, although certain worries deter him from accepting type-materialism, Lycan
states an argument that is quite similar to the one given above in his interesting paper on psychological
laws (Lycan 1981, p. 10). (Lycan writes as follows: “What better way to explain a lawlike correlation
between A’s and B’s than by supposing, in the absence of significant evidence to the contrary, that in fact
A’s are just B’s.”) Again, it seems quite likely that Smart had a similar argument in mind in writing his
seminal paper (1959, last two paragraphs). (The situation is somewhat obscured, however, by the fact that
Smart twice describes the argument as an appeal to “the principles of parsimony and simplicity.” It may be
that simplicity and explanatory power are related in certain interesting ways, but it is nonetheless necessary
to distinguish sharply between arguments like the one given above and arguments for materialism that are
based on Occam’s Razor. See note 5.)
4
The Best Explanation Principle can reasonably and usefully be seen as a descendant of the Principle of
Sufficient Reason. Roughly, the former is the credible core that remains when the latter is revised to
accommodate certain a priori objections and the a posteriori objections that are based on Quantum
Mechanics.
 IN DEFENSE OF TYPE - MATERIALISM 101

It is clear, I think, that type–materialism provides a good explanation of the

psychophysical laws that are claimed to exist in the Correlation Thesis. Suppose,
for example, that conscious experiences of a certain kind A turn out to be correlated
with brain processes of kind B. Surely, if someone were to ask for an explanation of
this correlation, it would be perfectly appropriate to respond by saying, “The
correlation obtains because being a conscious experience of type A is the very same
thing as being a brain process of type B.” (Compare: “Miss Lane, why does Clark Kent
always show up in the same places as Superman?” “Because, Jimmy, Clark is
Superman.”)
The explanations that type-materialism provides compare favorably with the
“explanations” that are offered by dualistic theories. Consider, for example, the
“explanation” of correlation laws that one finds in Leibniz. According to Leibniz,
the relationship between mind and body can only be explained by appealing to God’s
plan for the universe: it is like the relationship between two clocks that are con-
structed in such a way that they remain in phase throughout their existence despite
being causally isolated from one another. This account is inferior to the account
provided by type-materialism in a number of respects. For example, it is inferior in
point of intelligibility. Although some philosophers hold that the concept of identity
poses difficult problems, few if any deny that it is less problematic than the concept of
God. Again, the explanation differs from the account provided by type-materialism
in that it raises more questions than it answers. If one were to explain psychophysical
laws by appealing to the creative energy of God, then one’s account would inevitably
lead to such questions as “Why did God want there to be events of two different kinds
in the first place?” and “Why did God want events of the two kinds to be connected
by the laws that actually exist?” On the other hand, there would be no point to the
question “Why are phenomenal qualities identical with neurological qualities?” If a
statement is constructed solely from the identity predicate and rigid designators, it is
neither necessary nor possible to explain why the statement is true.
It would be a mistake to look elsewhere among dualistic theories for better
explanations. A dualistic “explanation” of a psychophysical law is usually little more
than a euphemistic way of confessing that the law has the status of an unexplained
primitive.
One encounters a somewhat different situation, however, when one turns to
consider the explanations of psychophysical laws that are offered by token-
materialism. For present purposes, a theory counts as a version of token-materialism
if it implies both of the following claims: first, every event that is a token of some
phenomenal property is identical with some physical event; and second, no phe-
nomenal property is identical with any neurological property. Now the most popular
versions of token-materialism are based on the view that the Correlation Thesis is
false. All such versions deny that there is anything to be explained. However, an
advocate of token-materialism could accept the Correlation Thesis. Let TM be the
version of token-materialism that is based on these claims: first, the Correlation
102 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

Thesis is true; second, every event that is a token of some phenomenal property is
identical with an event that is a token of the neurological correlate of that phenomenal
property; and third, no phenomenal property is identical with any neurological
property. Like type-materialism, TM can explain all psychophysical laws claimed
to exist in the Correlation Thesis without introducing metaphysical concepts that are
highly problematic. Moreover, when considered from a certain point of view, its
explanations seem to be structurally similar to the explanations that are provided by
type-materialism.
On the other hand, despite their similarity, the explanations provided by these two
forms of materialism are different in a key respect. To appreciate the difference,
notice that both theories imply that there are a number of true generalizations of the
following form:

(1) Every brain process of type Y has the property being a conscious experience of
type X.
Type-materialism can explain such generalizations, for substitution instances of (1)
can be explained by substitution instances of “The property being a brain process of
type Y is identical with the property being a conscious experience of type X.” But TM
has no explanations to offer. Thus, TM replaces one set of facts with a new set of facts
that are roughly similar, and it differs from type-materialism in that it fails to explain
the members of the new set.
In view of this difference, it seems correct to say that type-materialism is more
successful than TM in explaining psychophysical laws. But suppose otherwise.
Suppose that the failure of TM to explain substitution instances of (1) does not
detract in any way from its adequacy as an explanation of such laws. (Here, of course,
I am assuming temporarily that substitution instances of (1) do not count as
psychophysical laws.) Even if this supposition is true, it is still possible to use the
Best Explanation Principle to justify a preference for type-materialism. For even if it
is true that type-materialism is no more successful than TM in explaining psycho-
physical laws it is nonetheless quite clear that it is more successful than TM in
explaining substitution instances of (1).5

5
Since materialists typically appeal to Occam’s Razor in attempting to justify their views, the reader
may wonder why I prefer to appeal to the Best Explanation Principle. The reason is that I have not yet
found a version of Occam’s Razor that is as plausible as the Best Explanation Principle. Suppose you are in
my office and you hear a noise coming from the next room that can be explained by each of the following
hypotheses: first, there is one extremely efficient secretary in the next room who is typing at great speed;
and second, there are two less efficient secretaries in the next room who are both typing at moderate speed.
According to most of the versions of Occam’s Razor I have encountered, it is obligatory to prefer the first
hypothesis to the second, since it does not multiply secretaries beyond necessity. I think it would be much
more reasonable to suspend judgment; and I can see no strong reason for holding that the situation is
significantly different in cases that are of greater philosophical interest. (Some versions of Occam’s Razor
do not imply that one should keep the number of entities to a minimum, but imply instead that one should
minimize the number of different kinds of entities. I believe that these versions encounter problems that are
similar to the one posed by the secretaries.)
 IN DEFENSE OF TYPE - MATERIALISM 103

IV
This brings me to the first of the four a priori objections to type-materialism that
I wish to discuss. Although this objection has received considerable attention in the
recent analytic literature (see, e.g., Kripke 1980), it derives ultimately from Descartes.
I will therefore call it the Cartesian Argument.
First premise: Conceivability is an adequate test for possibility. That is to say, if one can clearly
and distinctly conceive of its being the case that p, then it is genuinely possible for it to be the
case that p.
Second premise: One can clearly and distinctly conceive of events which possess phenomenal
qualities but which lack the correlated neurological qualities, and one can clearly and distinctly
conceive of events which possess neurological qualities but which lack the correlated phenom-
enal qualities.
Third premise: If it is possible for there to be a case of the property X which is not identical with
a case of the property Y, then the property X is not identical with the property Y.
Conclusion: Phenomenal qualities are not identical with neurological qualities.

A number of gifted philosophers have already raised objections to this argument; but
the philosophers in question have typically accepted some form of Functionalism,
and Functionalism has typically played an essential role in their objections.6 As a fan
of type-materialism (which is incompatible with all forms of Functionalism), I am
obliged to try to develop an objection of a different kind.
In discussing the Cartesian Argument, I will focus on the special case of it that can
be obtained by restricting the second premise and the conclusion to the phenomenal
quality being a pain and the neurological quality being a case of C-fiber stimulation.
(Here I follow the tradition in pretending that the latter has been found by neuro-
scientists to be correlated with the former.) This restriction will simplify the discus-
sion considerably.
In my opinion, there is very good reason to reject the first premise of the Cartesian
Argument. Thus, there is very good reason to believe that possibility resembles
actuality in being objective: whether something is possible is largely independent of
whether we think it to be so. If this is true, then it is extremely risky to maintain that

6
The chief exception to this is an objection that has been raised by Richard Boyd (Boyd 1980, pp. 83–5).
(Boyd endorses a version of Functionalism, but his objection to the Cartesian Argument is independent of
his Functionalist views.)
Although I feel that Boyd’s argument is interesting and important, I do not think that it is immune to
criticism. His objection stands or falls with the claim that Cartesian intuitions (i.e., intuitions to the effect
that phenomenal and neurological qualities are not necessarily coinstantiated) are due to the fact that we
have a tendency to conceive of brain processes in terms of the contingent qualities that guide us in
identifying them. But someone could maintain that Cartesian intuitions can still occur even when one is
careful to conceive of brain processes in terms of the essential qualities postulated by the scientific theory of
the brain (i.e., that they can still occur when we view brain processes through the axioms of a neurological
theory rather than through their contingent tendencies to produce certain kinds of effects). The objection
I develop is immune to this reply to Boyd’s objection.
104 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

conceiving of its being the case that p (which is, after all, a mere psychological state) is
a guarantee that it is genuinely possible that p. It is much more reasonable to accept
the following view: conceiving of its being the case that p is a prima facie ground
for believing that it is genuinely possible that p ; but if (i) there is a good reason for
thinking that it is not possible that p, and (ii) there is at least one plausible explanation of
why one can conceive of its being the case that p that is compatible with the claim that it
isn’t possible that p, then conceiving of its being the case that p fails to justify the claim
that it is possible that p. (In other words, the first premise of the Cartesian Argument
should give way to this principle: an intuition to the effect that it is possible that p should
normally be taken at face value; but if (i) there is a good reason for distrusting the
intuition and (ii) it is possible to explain the intuition away, it may be rejected.)
In Naming and Necessity (1980, pp. 144–55), Saul Kripke argues that it is impossible
to explain away our Cartesian intuitions (i.e., our intuitions to the effect that phenom-
enal and neurological qualities are not necessarily coinstantiated). If this view was
correct, then our Cartesian intuitions would pose a serious threat to materialism. For
our new principle about modal intuitions would lead to the conclusion that it is
necessary to take Cartesian intuitions at face value. As against Kripke, however,
I hold that it is possible to explain our Cartesian intuitions away, and by the same
token, I hold that it is possible to show that condition (ii) is satisfied in the present case.
If I can justify these views by giving an adequate explanation of the intuitions, then the
intuitions must be consigned to dialectical limbo until we acquire enough empirical
data to be able to assess the Correlation Thesis. For consider: If the Correlation Thesis
should turn out to be empirically true, then, by the line of thought of Section III, there
would be a good reason to believe that type-materialism is true. Equally, there would be
a good reason for distrusting our Cartesian intuitions. Hence, given that my explan-
ation of Cartesian intuitions is adequate, if the Correlation Thesis should turn out to be
true, then conditions (i) and (ii) would both be satisfied, and our principle about modal
intuitions would permit us to conclude that Cartesian intuitions are delusive.
In order to explain our Cartesian intuitions away, it is enough to appeal to the
following generalizations:

(2) If there are no logical connections between the concept of an F and the
concept of a G, then it is possible to conceive of something as an F without
conceiving of it as a G. (Here and elsewhere, I assume that contexts governed by
“concept of ” and “conceive” resist interchange of coreferential designators even
when the designators are rigid.)
(3) If it is possible to conceive of something as an F without conceiving of it as a G,
then it is possible to conceive of there being something which is an F and which
fails to be a G.
(4) If it is conceivable that p, then it seems to be possible that p.
When these generalizations are combined with the fact that there are no logical
connections between the concept of pain and the concept of C-fiber stimulation, we
 IN DEFENSE OF TYPE - MATERIALISM 105

can deduce that it will sometimes seem possible for there to be pains that do not fall
under the concept of C-fiber stimulation, and that it will sometimes seem possible for
there to be events which fall under this concept that are not cases of pain. Thus, since
at least one of the generalizations (viz., (4)) is lawlike, they can be said to provide an
explanation of the remaining intuitions.
Is type-materialism compatible with the fact that the concept of pain has no logical
connections with the concept of C-fiber stimulation? To see that the answer is
affirmative, consider the following fact: the concept of pain is formed in response
to the experiences that one has in virtue of BEING IN a state of mind that partially
instantiates being a pain, and the concept of C-fiber stimulation is formed in
response to the experiences one has in virtue of OBSERVING a brain state that
partially instantiates being a case of C-fiber stimulation. Type-materialism requires
one to conceptualize this fact in a way that is not required by other theories. Thus, if
type-materialism is true, then the fact comes to this: the concept of pain is formed in
response to the experiences one has in virtue of BEING IN a brain state of a certain
sort, and the concept of C-fiber stimulation is formed in response to the experiences
one has in virtue of OBSERVING a brain state which is a state of the very same sort.
However, even when it is stated in the form that is required by type-materialism, the
fact shows that there is an enormous difference between the factors which shape the
concept of pain and the factors which shape the concept of C-fiber stimulation. In
view of this difference, it is clear that it would be silly to expect the two concepts to
have any common content, or to stand in any other logical relationship.

V
The second objection I wish to discuss is found in a number of places in the literature,
and it appears in a number of different forms. Here is Jerome Shaffer’s version
(1968, p. 46):
Let us return to the man reporting the red afterimage. He was aware of the occurrence of
something or other, of some feature or other. Now it seems to me obvious that he was not
necessarily aware of the state of his brain at this time (I doubt that most of us are ever aware of
the state of our brain) nor, in general, necessarily aware of any physical features of his body at
that time. He might, of course, have been incidentally aware of some physical feature, but not
insofar as he was aware of the red afterimage as such. Yet he was definitely aware of something,
or else how could he have made that report? So he must have been aware of some nonphysical
feature. That is the only way of explaining how he was aware of anything at all.

Armstrong considers a closely related version in the course of cataloguing the main
objections to his position. Thus, he imagines (1968, p. 78) an opponent who argues as
follows:
Central-state Materialism holds that when we are aware of our mental states what we are aware
of are mere physical states of our brain. But we are certainly not aware of the mental states as
106 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

states of the brain. What then are we aware of mental states as? Are we not aware of them as
states of a quite peculiar, mental, sort?

One also encounters the objection when one hears it said that there is an “unbridge-
able gulf ” between conscious experiences and brain processes, and when one is told
that there is no way of accounting for the “striking qualitative differences” between
the essential natures of events of these two kinds. Another manifestation is the claim
that, for example, the difference between a red sensation and a sour sensation is self-
evidently a different difference than the difference between the neural processes that
accompany them. Yet another manifestation is the idea that phenomenal qualities
differ from neurological qualities in that the former are self-evidently simple and
unanalyzable while the latter are analyzable in terms of the properties and relations of
micro-events. And, less obviously, it underlies the claim that conscious experiences
enjoy a “luminosity” or a “phosphorescence” that brain processes lack.
When Shaffer’s line of thought is set out as a formal argument, and transposed to
the terminology of the present chapter, it comes to this:
First premise: When one is introspectively aware of a conscious experience, one is aware of it as
having a phenomenal quality.
Second premise: If phenomenal qualities were identical with neurological qualities, then when
one is aware of an event as an event that has a phenomenal quality, one would be aware of it as
an event that has a neurological quality.
Third premise: When one is introspectively aware of a conscious experience, one is not aware of
it as an event that has any neurological qualities.
Conclusion: Phenomenal qualities are not identical with neurological qualities.

This formulation is too sober and clinical to do justice to the power of the intuitions
that give rise to the argument; but in its own way, it captures all of the ideas that play
a logically essential role.
In order to reply to this argument, I need a premise about awareness that seems
uncontroversial, namely, the view that to be aware of a particular as an F is to have an
experience that leads one to subsume the particular under the concept of an F.
Assuming that this view is correct, the second premise of the argument carries the
following presupposition: If phenomenal qualities were identical with neurological
qualities, then the experiences that lead us to subsume events under our concepts of
phenomenal qualities would partially or entirely coincide with the experiences that
lead us to subsume events under our concepts of neurological qualities. As far as I can
tell, there is no reason to believe that this presupposition is true.
As we noticed earlier, the circumstances in which the concept of pain is acquired
are completely different than the circumstances in which the concept of C-fiber
stimulation is acquired. In virtue of the circumstances in which the concept of pain is
acquired, to have an experience which leads one to subsume an event under the
concept of pain is simply to have a pain. That is to say, to have an experience which
 IN DEFENSE OF TYPE - MATERIALISM 107

leads one to subsume an event under the concept of pain is to have an experience
which itself falls under the concept of pain. Further, in virtue of the circumstances in
which the concept of C-fiber stimulation is acquired, to have an experience that leads
one to subsume an event under the concept of C-fiber stimulation is to have an
experience that is produced by using one’s sensory apparatus (supplemented or
unsupplemented) to observe a state of the brain. Now the question is this: should
materialism encourage us to expect that experiences of the first sort (i.e., experiences
that fall under the concept of pain) will either partially or fully coincide with
experiences of the second sort (i.e., experiences produced by using one of the senses
to observe the brain)? It is clear, I think, that the answer is “No.”

VI
The third objection I wish to discuss is sometimes known as the Multiple Realization
Argument (MRA). It runs as follows:
First premise: It is logically possible for there to be individuals (hereafter called Aliens) which
satisfy three conditions: (i) they have the same behavioral capacities and tendencies as human
beings; (ii) they have an internal device for collecting, processing, and transmitting informa-
tion that is functionally isomorphic to the human nervous system; and (iii) they have no
internal states that can be said to possess neurological qualities. (The states of the internal
computing device are composed of microstates which differ both in structure and in type of
matter from the corresponding states of human beings.)
Second premise: Aliens enjoy conscious experiences with the same phenomenal qualities as
human beings, and their experiences are connected with behavior by the same laws as are
known to obtain in the human case.
Conclusion: Phenomenal qualities are not identical with neurological qualities.

This argument can be found in a number of places in the literature, including Block
and Fodor (1972, p. 161), Block (1978, section 3.0 and all other parts dealing with
Chauvinism), and Lewis (1980).
MRA stands or falls with the second premise, and it is therefore necessary to ask
whether the second premise can be justified. I will argue that the answer is negative.
According to an interpretation which appears to be favored by at least one of its
advocates (McGinn 1982, pp. 21–2), MRA is simply a special case of the Cartesian
Argument we considered earlier. Thus, according to one interpretation, the second
premise is to be justified by adverting to the fact that we can conceive of Aliens as
beings who enjoy conscious experiences like our own, and by inferring from this fact
that it is objectively possible for Aliens to have such experiences. Insofar as this
interpretation is correct, we are already in a position to conclude that MRA is
defective.
Is there another way of defending the second premise? The advocates of MRA do
not normally face this question directly; but it seems quite likely that if they were to
108 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

do so, they would respond by giving a Best Explanation Argument. That is to say, it
seems likely that they would reason as follows: the hypothesis that Aliens have
experiences with the same phenomenal qualities as ours provides the best explan-
ation of their behavior; hence, in view of the Best Explanation Principle, we are
justified in concluding that the hypothesis is true. This argument appears prima facie
to be quite strong. In fact, it is hard to imagine an a posteriori argument for the same
conclusion that would enjoy a greater amount of prima facie plausibility.
Upon closer inspection, however, it turns out that we are not in a position to apply
the Best Explanation Principle in the present context. For it is possible to account for
Alien behavior by another hypothesis, and there is no reason to think that this other
hypothesis is inferior to the hypothesis that Aliens have internal states with phe-
nomenal qualities. According to the second hypothesis, the internal states of Aliens
lack phenomenal qualities, but they have qualities which play roles in the mental lives
of Aliens that are similar to the roles played by phenomenal qualities in our mental
lives. Thus, while the first hypothesis can be seen as the claim that Aliens satisfy the
totality of psychological laws that human beings satisfy, including the laws that refer
to phenomenal qualities, the second hypothesis can be seen as the claim that Alien
psychological laws can be obtained from domestic laws by reinterpreting all predi-
cates that denote phenomenal qualities over a set of qualities that (i) are instantiated
by the internal states of Aliens, and (ii) are functionally equivalent to the properties
that the predicates normally express.
It follows trivially from this description of the second hypothesis that the two
hypotheses are equivalent in explanatory power. The first hypothesis provides an
explanation of a pattern of Alien behavior if and only if the second hypothesis
provides an isomorphic explanation. However, it might be thought that the first
hypothesis should be preferred to the second on the grounds of simplicity. After all,
the second hypothesis calls for a duplication of psychological laws, thereby compli-
cating our overall theory, and it might be thought that the duplication is unnecessary.
Further, it is often held that there is a methodological principle which counsels
conservatism. Since the second hypothesis calls for a change in our overall theory
that can seem to be gratuitous, it might be thought that this methodological principle
calls the second hypothesis into question.
These complaints about the second hypothesis would be legitimate if it was true
that the explananda which come to the fore when we focus on Alien behavior are
identical to the explananda that come to the fore when we focus on human behavior.
But in fact the explananda are different, and the foregoing appeals to simplicity and
conservatism are invalidated by their differences. The explananda are different
because human behavior is only identical to Alien behavior up to an insomorphism.
To see this, consider a case in which someone withdraws his or her hand from a hot
stove, and in which we explain the motion of the hand by referring to a pain. Here the
explanandum is the motion of a human hand, where a human hand is conceived as
something that is a member of a certain natural kind. No such explanandum comes
 IN DEFENSE OF TYPE - MATERIALISM 109

to the fore when we consider Aliens. To be sure, it is possible to say that Aliens have
hands; but when we say this, we are using “hand” in a new and extended sense. Alien
hands do not belong to the same natural kind as human hands. Indeed, unlike the
paw of a dog, an Alien hand does not even belong to a neighboring kind. For it is an
essential feature of a human hand and of a dog’s paw that they contain nerves, but
nerves are entirely foreign to Alien hands. Thus, when we use “hand” in talking about
Alien appendages, either it stands for a different kind than it does when we talk about
human hands, or our use of it is temporarily guided by a criterion of what is to count
as a hand that is purely functional and/or structural.
Up to this point it has been assumed that Aliens have no neural tissue whatsoever.
Let us now suppose instead that Aliens have nerves at their peripheries, and that they
differ from us only in having central computing devices that are devoid of neuro-
logical qualities. (We might suppose that Aliens of this sort are created by replacing
the brains of certain human beings with artificial brains, and by equipping the
latter with some means of communicating with peripheral nerves.) Under this new
assumption, Alien behavioral explananda are type-identical with domestic behavioral
explanada. However, the totality of Alien explananda still differs considerably from
the totality of all domestic explananda, for the former totality includes central
explanada that are quite different than all of the explananda in the latter totality.
To see this, notice that it is reasonable to accept this pair of propositions: first, there
are causal relations between phenomenal events and human brain processes; and
second, these causal relations mediate many of the causal relations between phe-
nomenal events and human behavioral events. (I cannot defend these propositions
here. If the reader doubts that there is currently enough evidence to make it
reasonable to accept them, then he or she should think of the present line of thought
as an argument that may become effective at some later point in time.) Assuming that
the two propositions are correct, if an hypothesis claims that the etiology of Alien
behavior is isomorphic to the etiology of human behavior, then it must explain some
of the events that take place in Alien central computing devices in terms either of
phenomenal events or counterparts of phenomenal events. Since all events that take
place in Alien central computing devices belong to different kinds than human brain
processes, it follows that some of the Alien explananda that confront our two
hypotheses are central events that belong to different kinds than the corresponding
human explananda.
Simplicity and conservatism may require us to explain similar phenomena (i.e.,
phenomena that belong to the same kind) in similar ways, but they do not require us
to explain different phenomena in similar ways. Thus, appeals to simplicity and
conservatism are inappropriate in the present context. We cannot use them to
establish the claim that the first hypothesis is superior to the second.
(Of course, a type-materialist will hold that the counterparts of phenomenal
qualities that are postulated by the second hypothesis are ultimately identical with
certain of the Alien counterparts of neurological qualities.)
110 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

It appears, then, that it is impossible to justify the second premise of MRA by an a

posteriori argument. Is it possible to justify the second premise by an a priori
argument? Is it possible to show a priori that beings must have internal states
with phenomenal qualities if they are like us in point of behavior and in point of
functional qualities of internal states? Well, if it was possible to show this a priori,
then either there would be logical connections between our concepts of phenomenal
qualities and some of our functional or behavioral concepts, or it would be the case
that we rely on our functional or behavioral concepts in fixing the reference of our
phenomenal concepts. However, both of these possibilities can be ruled out by a few
commonplace observations about pain. There can be no logical connections between
the concept of pain and any of our functional or behavioral concepts; for it is easy to
conceive of pains that are not accompanied by events with any functional or
behavioral qualities, and it is equally easy to conceive of events with any given
functional or behavioral qualities that are not accompanied by pains. Moreover, it
would be wrong to say that we rely on functional or behavioral concepts in fixing the
reference of the concept of pain. Since we are directly acquainted with pain itself, it is
possible to fix the reference of the concept of pain without making use of any
concepts that apply to the contingent effects of pain.

VII
Type-materialism presupposes that we can know that the Correlation Thesis is true. It
is time now to turn to consider an extremely important objection to this presuppos-
ition. According to the objection I have in mind, since it is impossible to be directly
acquainted with the conscious experiences of others, no one is in a position to know
that beings with similar nervous systems (i.e., other people and members of other
highly developed biological species) are capable of having conscious experiences. But,
the objection continues, if it is impossible to know that other beings with similar
nervous systems have conscious experiences, then it is impossible to know that the
Correlation Thesis is true. In short, the objection challenges one of the main presup-
positions of type-materialism by appealing to the traditional problem of Other Minds.
Although I cannot do full justice to this objection in the present chapter, I must at
least create some doubt as to whether it is ultimately successful. I will try to create the
doubt by refurbishing and defending the Argument from Analogy (AA).
I wish to claim that the following version of AA is inductively strong:
When I consider my own conscious experiences, I find that they play essential roles in all of my
sensory explorations of the external world, and also that they are causally linked to all or
almost all of the patterns of behavior in my behavioral repertoire. Further, I find that many of
the sensory explorations and behavioral ventures that depend on my conscious experiences are
essential to my physical well being and survival. And, finally, when I compare my basic sensory
and behavioral capacities to those of other people, I find that they are biologically normal, in
 IN DEFENSE OF TYPE - MATERIALISM 111

the sense that they are shared by almost all other members of my species. Hence, probably all
other biologically normal members of my species enjoy conscious experiences like my own.
Moreover, it is likely that the experiences of such beings are linked to other factors in
perceptual processes and to patterns of behavior by laws that are quite similar to the laws
I have found to hold in my own case.

The conclusion of this argument refers only to the class of human beings, but it is
possible to construct related arguments that lead to analogous conclusions about the
members of other species.
Like all other versions of AA, the foregoing argument differs in two respects from
the analogical arguments that are cited as paradigms in logic texts. In the first place,
the argument is based on a sample that has only one member. And second, it is
logically impossible to check up on the correctness of the argument’s conclusion by a
confirmation procedure that is more direct. Critics have maintained that both of
these features are flaws. Thus, commenting on the first feature, Paul Churchland
writes (Churchland 1979, p. 90) as follows:
[That the] correlations discovered in one’s own case also characterize the cases of every other
human can therefore be inferred only by way of an inductive leap from the necessarily solitary
instance of one’s own (possibly atypical) case. The possible justification available for beliefs
about the minds of others is therefore at the vanishing point. One is limited to an incorrobor-
able generalization from a single case. In this way do we arrive at scepticism with respect to the
content and even the existence of minds other than one’s own.

I believe that this line of thought is wrong. However, it has a considerable amount of
prima facie force, and I will therefore deal with it at some length. On the other hand,
it seems to me that the second objection can be dismissed. In effect, this objection
charges that the argument is shown to be inadequate by the fact that it is impossible
to justify its conclusion by giving a second argument of some other type. But this
charge is absurd. Why should strength of a justification of one type depend on the
possibility of finding a second justification of a different type? (Many important
propositions in number theory cannot be proved unless one uses Weak Mathemat-
ical Induction or one of its equivalents. Does this show that it is impossible to justify
the propositions in question by giving the standard proofs (i.e., by proofs involving
Weak Induction and its equivalents)?)7

7
In a valuable paper on AA, A. Hyslop and F. Jackson distinguish a number of versions of the second
objection and give persuasive replies to all of them. They also give a reply to the first objection that overlaps
with my reply. However, they seem not to appreciate the relevance of an idea that I believe to be crucial (the
idea of biological kind). (See Hyslop and Jackson (1972). I am indebted to Michael De Armey for this
reference.)
The reader will have noticed that I say nothing about the well known discussions of AA by Malcolm and
Strawson. (See Malcolm (1958) and Strawson (1959, chapter 3).) It seems to me that the objections raised
by these authors are best seen not as criticisms of AA itself but rather as criticisms of certain of the views
about consciousness that create a need for AA. At all events, their objections raise issues that lie outside the
scope of the present chapter.
112 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

The single case objection presupposes that it is always wrong to base an analogical
argument on a sample with a single member. It is clear that this presupposition
contains an element of truth: much of the time, a sample with a single member is
extremely misleading. However, there are situations in which samples with one
member are extremely reliable guides to truth. Here are two such:
(i) Looking at the philosophy shelf while browsing in my local bookstore, I find
that there are five adjacent copies of a book entitled Reference and Essence. I take one
copy from the shelf and discover that there is a picture of a top hat and a rabbit on the
front cover. Noticing that the spines of the remaining copies are exactly like the spine
of the copy in my hand, I wonder whether I can be sure that the same picture appears
on the other covers. I am aware that publishers occasionally arrange for two or more
different pictures to appear on the covers of different copies of the same book, that
cover pictures are sometimes blurred beyond recognition by printing miscues, and
that such pictures are sometimes ruined in the course of packing and shipping. But
I also know that occurrences of these three types are extremely rare, and that if an
occurrence of the second or the third type had taken place in the present case, the
books would probably (though not necessarily) have been returned to the publisher.
So I infer that very likely the remaining four covers have the same puzzling picture as
the one in my hand.
(ii) A new species of aquatic mammals is discovered. Scientists observe a number
of these mammals, and reach the conclusion that a certain member of the species,
named Max, is normal both with respect to behavior and to all external physical
characteristics. When Max’s body is dissected after his death, they find that it
contains two lungs that are roughly equal in size. They calculate that Max’s lung
capacity was perfectly suited to his patterns of behavior: that is to say, they calculate
that if Max had had a larger or a smaller capacity, his behavior would have been quite
different. They conclude that all normal members of Max’s species have two lungs of
roughly the same size.
The strength of these arguments is due in part to the nature of the types of
resemblance that are seen to link the samples to the larger populations to which
the conclusions refer, and it is due in part also to the relationship between these types
of resemblance and the properties that are projected over the populations by the
conclusions. Thus, in each case, the type of resemblance involved is known to be an
extremely reliable sign that objects share all of the properties that belong to a given
category, and the property projected by the conclusion is known to belong to the
category in question. In short, both arguments satisfy the Reliability Condition.
When this condition is not satisfied, larger samples are needed. But when it is
satisfied, small samples are entirely legitimate.
Although it seems that the relationship between the Reliability Condition and
sample size is not widely appreciated among philosophers of mind, there are occa-
sional references to it in logic texts. It is pretty clear, for example, that Copi had
 IN DEFENSE OF TYPE - MATERIALISM 113

something like it in mind while writing the following passage (Copi 1978, p. 338):
“An argument based on a single relevant analogy connected with a single instance
will be more cogent than one which points out a dozen irrelevant points of resem-
blance between its conclusion’s instance and over a score of instances enumerated in
its premises.” Further, there is a highly promising sketch of a theoretical account of
the relationship in a recent contribution to confirmation theory by Thagaard and
Nisbett (Thagaard and Nisbett 1982).
It is clear, I think, that the Reliability Condition is satisfied by the foregoing version
of AA. The population mentioned in the conclusion is known to resemble the sample
in that all the elements of both sets are known to be normal members of a single
natural kind. Moreover, the following is known to hold: the relation being normal
members of the same natural kind is a reliable sign that similar overt characteristics
are due to similar underlying mechanisms, provided that the characteristics are
shared by most of the other members of the kind in question. The situation is
somewhat more complex in the case of biological kinds, for there tends to be
considerable intraspecific variation even with respect to properties of basic mechan-
isms. However, while mechanisms of the same type may differ considerably from one
another, it is nonetheless true (and known to be true) that different individuals tend
to have mechanisms of the same type. Moreover, this is especially true in the case of
mechanisms that underlie capacities and behavioral patterns that are essential to
physical well being and survival. Since the contribution of consciousness to physical
well being and survival is on a par with the contributions of breathing, digestion, and
circulation of the blood, it is fair to say that projection of phenomenal qualities is
authorized by the Reliability Condition.8
Apparently, then, it is possible to know that beings with similar nervous systems
enjoy conscious experiences like one’s own. What about the converse? Can one know
that beings that fail to have nervous systems like one’s own also fail to have conscious
experiences? In answering this question, it is necessary to remember that we have
ample inductive evidence, obtained both from self-observation and observation of
others (here I presuppose the foregoing version of AA), that conscious experience is
diminished or extinguished by reduction in brain activity. It seems that if the
existence of such evidence is recognized, one cannot consistently reject the claim
that consciousness requires brain activity unless one also raises general objections
about the validity of induction.

8
My argument that the Reliability Condition is satisfied turns on the claim that the following
proposition is known to be true: essential internal states and mechanisms of beings that belong to the
same species tend to be similar in type. This may seem to imply that the foregoing version of AA can only
be used by those who have mastered a certain amount of modern biology, but in fact my claim implies
nothing of the sort. Notice, for example, that the proposition is strongly supported by the fact that the
process which brings any one member of a species into existence is virtually the same as the processes by
which all of the members come into existence. One can be fully and vividly aware of this fact without
knowing any of the laws that belong to biological theory.
114 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

VIII
Before concluding, I would like to examine briefly the two main empirical objections
to the Correlation Thesis. These objections are summarized by Block and Fodor
(1972, pp. 160–1) in the following passages:
First, the Lashleyan doctrine of neurological equipotentiality holds that any of a wide variety of
psychological functions can be served by any of a wide variety of brain structures. While the
generality of this doctrine may be disputed, it does seem clear that the central nervous system is
highly labile and that a given type of psychological process is in fact often associated with a
variety of distinct neurological structures . . . Physicalism, as we have been construing it,
requires that organisms are in type-identical psychological states if and only if they are in
type-identical physical states. Hence, equipotentiality (if true) provides evidence against
physicalism.
[Second, psychological] similarities across species may often reflect convergent environmental
selection rather than underlying physiological similarities. For example, we have no particular
reason to suppose that the physiology of pain in man must have much in common with the
physiology of pain in phylogenetically remote species. But if there are organisms whose
psychology is homologous to our own but whose physiology is quite different, such organisms
may provide counterexamples to the psychophysical correlations physicalism requires.

These objections will be called respectively the Equipotentiality Argument and the
Appeal to Other Species.
As it stands, the major premise of the Equipotentiality Argument is much too weak
to bear its share of the weight of the conclusion. The major premise claims that a
single phenomenal kind may be associated with neurological states that differ in one
or more fairly significant respects, and the conclusion claims that there are phenom-
enal kinds that are not correlated with any one neurological kind. In order to obtain
the conclusion, it is necessary to supplement the major premise with the thesis that
states which differ in one or more significant respects cannot belong to a single
neurological kind. That is to say, it is necessary to add the thesis that a neuroscientist
is ex officio restricted to fine-grained principles of individuation for natural kinds.
Unfortunately, when one considers the variety of concepts that neuroscientists
actually employ, and the variety of explanatory goals that they set themselves and
occasionally reach, it becomes clear that they sometimes (though not always) use
principles of individuation whose grain is extremely coarse. Neuroscience is con-
cerned to understand the structure and behavior of single cells and the interaction of
small assemblies of cells, but it also recognizes the existence of large-scale mechan-
isms that involve scattered and gerrymandered collections of cells. It uses highly
abstract structural and mathematical concepts in describing such mechanisms and in
analyzing their behavior, and it also uses concepts from systems theory, communi-
cation theory, and computing. A science that uses such concepts in picking out its
kinds is fully capable of assigning states to a single natural kind even if the states
differ in one or more significant respects. (A similar position about the conceptual
 IN DEFENSE OF TYPE - MATERIALISM 115

framework of neurology is stated and ably defended in a recent paper by Patricia

Churchland. See Churchland (1980, p. 193 and pp. 204–7).)
In order to obtain the conclusion of the Equipotentiality Argument, it is not enough
to appeal to a case in which a single phenomenal quality is associated with two or more
distinct neurophysiological states. One must go on to provide an exhaustive charac-
terization of the distinct levels of description and explanation that belong to neurosci-
ence and show that no such level harbors a kind under which all of the states in
question may be subsumed. I am not in a position to claim that it is impossible to
accomplish these tasks, and I must therefore grant that there may be a version of the
Equipotentiality Argument that is immune to the foregoing objection. I wish to claim
only that no such version has yet appeared in the literature.
The Appeal to Other Species suffers from two flaws, both of which have already been
noted by Jaegwon Kim (1972). First, it suffers from the same flaw as the Equipotenti-
ality Argument: as it stands, it fails to do justice to the fact that states can belong to a
single natural kind even if they are quite different. Second, it presupposes that we can
know that biologically remote organisms enjoy conscious experiences like our own,
and this presupposition is highly questionable. It seems likely that knowledge of other
minds depends upon the possibility of constructing an argument like the version of AA
that we considered earlier. However, the strength of such arguments varies inversely
with the biological distance between other organisms and ourselves.
These replies to the Appeal to Other Species are intended only to shift the burden
of proof. It may be possible to strengthen the argument in such a way as to block the
first criticism. Moreover, the second criticism stands or falls with the view that AA
provides our only route to knowledge of other minds. Although this view enjoys a
certain amount of prima facie plausibility, it cannot be accepted without a supporting
argument. No such argument has been given here.
This defense of type-materialism suffers from a number of other lucanae. For
example, I have claimed that introspection plays a key role in the acquisition of
concepts of phenomenal qualities, and this entails that the acquisition of such
concepts depends on something like private ostensive definitions. But there are
several reasons, not all of which are found in Wittgenstein, for thinking that the
concept of a private ostensive definition is highly problematic. Thus, my defense of
type-materialism needs to be supplemented by a theory of private ostensive defin-
itions. I hope, however, that enough has been said to show that type-materialism is
worthy of more attention than it has thus far received.

References
Armstrong, D. M. (1968). A Materialist Theory of the Mind. New York: Humanities Press.
Block, N. (1978). “Troubles with Functionalism.” In C. W. Savage (ed.), Minnesota Studies in
the Philosophy of Science vol. 9. Minneapolis: University of Minnesota Press, 261–325.
116 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

Block, N. and Fodor, J. (1972). “What Psychological States Are Not,” The Philosophical Review
81: 159–81.
Boyd, R. (1980). “Materialism without Reductionism: What Physicalism Does Not Entail.” In
N. Block, (ed.), Readings in Philosophy of Psychology I. Cambridge, MA: Harvard University
Press, 67–106.
Brindley, G. S. (1973). “Sensory Effects of Electrical Stimulation of the Visual and Paravisual
Cortex in Man.” In R. Jung, (ed.), Handbook of Sensory Physiology VII/3B. New York:
Springer-Verlag, 583–94.
Brindley, G. S. and Lewin, W. S. (1968). “The Sensations Produced by Electrical Stimulation of
the Visual Cortex,” Journal of Physiology 196: 497–3.
Churchland, P. M. (1979). Scientific Realism and the Plasticity of Mind. Cambridge: Cambridge
University Press.
Churchland, P. S. (1980). “A Perspective on Mind-Brain Research,” The Journal of Philosophy
77: 185–207.
Copi, I. M. (1978). Introduction to Logic. 5th edition. New York: Macmillan.
Davidson, D. (1970). “Mental Events. In L. Foster and J. W. Swanson (eds.), Experience and
Theory. Amherst: University of Massachusetts Press.
Dobelle, W. H. and Mladejovsky, M. G. (1974). “Phosphenes Produced by Electrical Stimula-
tion of the Human Occipital Cortex, and their Application to the Development of a
Prosthesis for the Blind,” Journal of Physiology 243: 553–76.
Harman, G. (1966). “The Inference to the Best Explanation,” Philosophical Review 74: 33–95.
Harman, G. (1968). “Knowledge, Inference, and Explanation,” American Philosophical Quar-
terly 5: 164–73.
Holmes, G. (1945). “The Organization of the Visual Cortex in Man,” Proceedings of the Royal
Society B 132: 348–61.
Hyslop, A. and Jackson, F. (1972). “The Analogical Inference to Other Minds,” American
Philosophical Quarterly 9: 168–76.
Kim, J. (1972). “Phenomenal Properties, Psychophysical Laws, and the Identity Theory,” The
Monist 56: 177–92.
Kripke, S. (1980). Naming and Necessity. Cambridge, MA: Harvard University Press.
Lewis, D. (1980). “Mad Pain and Martian Pain.” In N. Block, (ed.), Readings in Philosophy of
Psychology I. Cambridge, MA: Harvard University Press, 216–22.
Lycan, W. G. (1981). “Psychological Laws,” Philosophical Topics 12: 9–38.
Malcolm, N. (1958). “Knowledge of Other Minds,” Journal of Philosophy 55: 969–78.
McGinn, C. (1982). The Character of Mind. Oxford: Oxford University Press.
Penfield, W. (1959). “The Interpretive Coriex,” Science 129: 1719–25.
Penfield, W. and Rasmussen, T. (1950). The Cerebral Cortex of Man. New York: Macmillan.
Putnam, H. (1981). Reason, Truth and History. Cambridge: Cambridge University Press.
Shaffer, J. (1968). Philosophy of Mind. Englewood Cliffs, NJ: Prentice-Hall.
Smart, J. J. C. (1959). “Sensations and Brain Processes,” Philosophical Review 68: 141–56.
Strawson, P. F. (1959). Individuals. London: Methuen.
Thagaard, P. and Nisbett, R. E. (1982). “Variability and Confirmation,” Philosophical Studies
42: 379–94.
 7
Imaginability, Conceivability,
Possibility, and the Mind-Body
Problem*

In the early seventies Kripke unveiled a line of thought that was designed to
resuscitate dualistic philosophies of mind.1 This line of thought was greeted with a
chorus of objections, some of which have helped considerably to bring the relevant
issues into sharper focus. Still, despite the number and philosophical value of these
replies, the line of thought has continued to attract sympathetic attention. Indeed, it
figures prominently in three of the most original and provocative contributions to the
contemporary literature on the mind-body problem—W. D. Hart’s Engines of the
Soul, Stephen Yablo’s “The Real Distinction Between Mind and Body” (as amplified
in its companion piece “Is Conceivability a Guide to Possibility?”), and David
Chalmers’s The Conscious Mind.2
In addition to being provocative, Kripke’s line of thought was characterized by
broad relevance. In one version it challenged token-materialism, and in another,
type-materialism. Both versions continue to attract admirers. (Token-materialism
asserts that each concrete psychological particular, such as the pain I experienced at
10 p.m. last night, is identical with a concrete physical particular, such as the firing of
my C-fibers at 10 p.m. last night. Type-materialism comes in more than one variety.
The strongest version claims that every type of psychological state is identical with
some type of physical state. A much weaker version, which is substantially more
plausible, claims that every type of sensory state is identical with some type of
physical state.)
In this chapter I will try to show that those who continue to be favorably impressed
by Kripke’s line of thought have underestimated the resources that are available to
materialists. Specifically, I will maintain that the most widely accepted part of that

* This is an expanded version of a paper that was read at the 1994 meetings of the Pacific Division of the
American Philosophical Association. I thank W. D. Hart and Stephen Yablo for extremely useful com-
ments on that earlier paper. In writing the present paper I have benefited substantially from criticism and
encouragement from David Chalmers, Ivan Fox, and Stephen Yablo.
1
See Kripke. (1980). See also Kripke (1977).
2
Hart (1988), Yablo (1990), Yablo (1993), Chalmers (1996).
118 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

line of thought, the argument against type-materialism, admits of an answer that is

fully satisfactory.
Two last preliminary comments. First, I will be concerned only to defend type-
materialism about qualitative states, that is, type-materialism about states that are
like being a pain in that they are forms of sentience. Second, although it is of course
my intention to carry the discussion forward, I will not be aiming at radical
originality. Thus, I will be relying heavily on ideas that are found in earlier papers
by Thomas Nagel and myself. My aims are to clarify these ideas, to develop and
supplement them in certain ways, and to show that it is possible to embrace them
without incurring any questionable philosophical commitments.

I
I will begin with a brief review of the portion of Kripke’s line of thought that counts
directly against type-materialism. (I will follow Kripke in focusing on pain and the
physical state which in fact accompanies pain in the human brain. (As is usual, I will
refer to this brain state as “C-fiber stimulation.”) But the argument should be seen as
implicitly general: according to Kripke, what the argument claims about pain and its
accompanying brain state could also be claimed, with equal legitimacy, about any
other pair consisting of a type of sensation and its corresponding brain state.)

Kripke’s argument
First premise: It appears to be possible for there to be instances of the property being a pain that
are not instances of the property being a case of C-fiber stimulation. For we can easily imagine,
and easily conceive of, a disembodied person who is experiencing pain.
Second premise: It appears to be possible for there to be instances of being a case of C-fiber
stimulation that are not instances of being a pain. For we can easily imagine, and easily
conceive of, a zombie whose C-fibers are undergoing a high degree of electro-chemical activity.
Third premise: Where X and Y are any two properties, if it seems to be the case that X and Y are
separable, in the sense that it seems to be possible for there to be instances of X that are not
instances of Y, then, unless this appearance of separability can be explained away, it really is the
case that X and Y are separable.
Fourth premise: Where X and Y are any two properties, if X and Y are separable, then it is not
the case that X and Y are identical.
Fifth premise: In general, if X is a commonsense natural kind and Y is a scientific kind that can
plausibly be identified with X, then there is a tendency for it to appear to us that X is separable
from Y. However, it is possible in most cases to explain away the apparent separability of X and
Y by attributing it to a tendency to confuse X with a different property—a property that
normally guides us in recognizing instances of X, but that is only contingently connected with
X. (To elaborate: In most cases, if X is a commonsense natural kind, then there exists a
property Z such that (a) instances of Z are normally instances of X, (b) this connection between
Z and X is contingent, and (c) we normally recognize instances of X by identifying them as
 IMAGINABILITY, CONCEIVABILITY, POSSIBILITY, AND THE MIND - BODY PROBLEM 119

instances of Z (i.e., Z is the property that guides us in “picking out” instances of X). When X is
associated with a property Z that meets this condition, it is possible to explain away X’s
apparent separability from an associated scientific kind Y by saying that the appearance of
separability is due in part to a tendency to misconstrue possible situations that contain
instances of Z as situations that contain instances of X, and in part also to a tendency to
misconstrue possible situations that lack instances of Z as situations that lack instances of X. To
be more specific, when it appears to a subject that it is possible for there to be an instance of
X that is not an instance of the associated scientific kind Y, it is possible to explain this
appearance away by saying that the subject has (i) imagined or conceived of a possible situation
that contains an instance of Z that is not an instance of Y and (ii) misconstrued this situation as
one in which there is an instance of X that is not an instance of Y. Equally, when it appears to a
subject that it is possible for there to be an instance of Y that is not an instance of X, it is
possible to explain this appearance away by saying that the subject has (i) imagined or
conceived of a possible situation that contains an instance of Y that is not an instance of
Z and (ii) misconstrued this situation as one in which there is an instance of Y that is not an
instance of X.)
Sixth premise: The apparent separability of being a pain and being a case of C-fiber stimulation
cannot be explained away in the way indicated in the fifth premise. (Thus, there is no property
Z such that (a) instances of Z are normally instances of being a pain, (b) this connection
between Z and being a pain is contingent, and (c) we normally recognize instances of being a
pain by identifying them as instances of Z. Being a pain is itself the property that guides us in
recognizing instances of being a pain. (Kripke puts the point in the following way: “Pain, on the
other hand, is not picked out by one of its accidental properties; rather it is picked out by the
property of being pain itself, by its immediate phenomenological quality.”3))
Seventh premise: The paradigm described in the fifth premise is the only model for explaining
appearances of separability away.
Lemma: By the fifth, sixth, and seventh premises, it is not possible to explain away the apparent
separability of being a pain and being a case of C-fiber stimulation.
Conclusion: By the first premise, the second premise, the third premise, the fourth premise, and
the lemma, being a pain is not identical to being a case of C-fiber stimulation.

In addition to sketching this argument, Kripke takes pains to illustrate the para-
digm of explanation that is described in the fifth premise. Kripke begins this part of
his discussion by pointing out that, despite the fact that the commonsense natural
kind heat is known to be identical with the scientific kind molecular kinetic energy, it
seems possible for heat to exist without being accompanied by molecular motion.
Kripke then maintains that we are normally guided in recognizing heat by the
property being the external phenomenon that causes the sensation S, where “S”
names the sensation that heat normally produces in human observers in the actual
world. Next, he says that when it seems to us that heat can exist without being
accompanied by molecular motion, this is because we are misconstruing the situation

3
Kripke (1980, p. 152).
120 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

that we are imagining or conceiving of. In particular, he says, when we have an

impression of possibility of the sort in question, we have it because (a) we are
imagining or conceiving of a situation in which being the external phenomenon
that causes the sensation S is exemplified but there is no molecular motion, and (b)
we are misconstruing this situation as one in which heat is present but molecular
motion is not. That is to say, as Kripke sees it, we are confusing a situation in which
the property by which we normally identify heat is exemplified with a situation in
which heat itself is present.
Before concluding this review of Kripke’s line of thought, we should recall that
Kripke nowhere provides any defense of the assumption that I have listed as the
seventh premise, the assumption that the explanatory paradigm illustrated by the
heat example is the only paradigm for explaining appearances of possibility away.4 As
we will see, this premise is false. Indeed, it is irredeemably false, in the sense that there
is no true claim, or set of true claims, that is capable of taking its place in the
argument.

II
I will now sketch an explanation of the apparent separability of pain and C-fiber
stimulation that does not conform to the paradigm we have just been considering.
Instead of exploiting the distinction between a property and its mode of presentation
(that is, the distinction between a property X and the property that guides us in
recognizing instances of X), this alternative explanation exploits a distinction
between two types of imagination.
The explanatory strategy I have in mind is set forth in a little-noticed footnote in
Thomas Nagel’s wonderful “What Is It Like to Be a Bat?”.5 The relevant part of the
footnote runs as follows:
A theory that explained how the mind-brain relation was necessary would still leave us with
Kripke’s problem of explaining why it nevertheless appears contingent. That difficulty seems to
me surmountable, in the following way. We may imagine something by representing it to
ourselves either perceptually, sympathetically, or symbolically. I shall not try to say how
symbolic imagination works, but part of what happens in the other two cases is this. To
imagine something perceptually, we put ourselves in a conscious state resembling the state we
would be in if we perceived it. To imagine something sympathetically, we put ourselves in a
conscious state resembling the thing itself. (This method can only be used to imagine mental
events and states—our own or another’s.) When we try to imagine a mental state occurring
without its associated brain state, we first sympathetically imagine the occurrence of the mental
state: that is, we put ourselves in a state that resembles it mentally. At the same time, we

4
Kripke says that the model illustrated by the heat example is “the only model [he] can think of.” See
Kripke (1977, p. 101).
5
Nagel (1974).
 IMAGINABILITY, CONCEIVABILITY, POSSIBILITY, AND THE MIND - BODY PROBLEM 121

attempt to perceptually imagine the non-occurrence of the associated physical state, by putting
ourselves into another state unconnected with the first: one resembling that which we would be
in if we perceived the non-occurrence of the physical state. Where the imagination of physical
features is perceptual and the imagination of mental features is sympathetic, it appears to us
that we can imagine any experience occurring without its associated brain state, and vice versa.
The relation between them will appear contingent even if it is necessary, because of the
independence of the disparate types of imagination.6

According to Nagel, then, there are three different types of imagination; the symbolic,
the sympathetic, and the perceptual. We use the sympathetic imagination in imagin-
ing mental states and the perceptual imagination in imagining brain processes. It is
also part of Nagel’s story that we can use the perceptual imagination to form images
of the absence of brain processes, that is, images that represent situations in which no
brain processes exist. Nagel uses this story as the foundation of an account of how we
are able to imagine disembodiment. According to Nagel’s account, when we imagine
a mental state that is not accompanied by a brain process, what we do is to splice
together a sympathetic image of a situation that contains a mental state and a
perceptual image of a situation in which no brain process is present. The fact that
we can do this without incoherence has no tendency to show that it is objectively
possible for the imagined mental state to exist without being accompanied by a brain
process. Rather our ability to do it is due to the fact that there are two types of
imagination that operate independently of one another.
The key element of this explanation is a claim that can be expressed as follows: If
P is a property of which one can be introspectively or perceptually aware, then, when
one imagines an instance of P, what one does is to put oneself into a state which is
similar to the state one is in when one is experientially aware of an instance of P. This
claim is intuitively plausible, and it also seems to be supported by the beautiful
experimental results about images and the imagination that psychologists obtained in
the seventies and eighties. Thus, there are many results, such as the classic experi-
ments by Roger Shepard and his colleagues concerning the processes by which we
determine congruence and incongruence relations between geometric forms, which
strongly suggest that visual images of objects and situations have many of the same
properties as the corresponding perceptual states.7 Moreover, Nagel’s claim receives
additional support from the results which indicate that imagination can facilitate
perceptual tasks. Ronald A. Finke summarizes these results in the following passage:

6
Quoted from Nagel (1974, footnote 11). Oddly, the ideas in this footnote have not received any
attention from the contemporary defenders of Kripke’s argument, nor, as far as I know, from any of
Kripke’s other commentators.
7
See, e.g., the papers collected in Shepard and Cooper (1982).
Of course, the interpretation of Shepard’s results is a matter of controversy. Not everyone would agree
with the assessment offered above. For a quite different view, see Pylyshyn (1981).
122 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

[T]he process of forming a visual image can serve a perceptual anticipatory function: it can
prepare a person to receive information about imagined objects. Mental imagery may therefore
enhance the perception of an object by causing the selective priming of mechanisms in the
visual system.8

As Finke points out, this priming hypothesis provides strong support for the view
that the process of forming a mental image involves many of the neural mechanisms
that underline visual perception.9
This brings us to the question of whether Nagel’s explanation is genuinely
compatible with materialism. At first sight, at least, the answer appears to be “yes.”
Consider, for example, the doctrine that pain is the very same thing as an electro-
chemical process in one’s C-fibers. If this doctrine is true, one might ask, then how is
it possible for experiences that count as imaginative presentations of pains to be quite
different in character from experiences that count as imaginative presentations of
cases of C-fiber stimulation? Nagel can respond by saying that the experiences are
different because they are produced by different psychological mechanisms: experi-
ences that count as imaginative presentations of pains are produced by mechanisms
that serve the sympathetic imagination, and experiences that count as imaginative
presentations of C-fiber stimulation are produced by mechanisms that serve the
perceptual imagination. But is it possible for psychological mechanisms to produce
radically different experiences if those experiences are in fact presentations of the
same property? Well, yes; this happens all of the time in the case of perceptual
presentations. Compare a visual presentation of the surface of a piece of sandpaper
with a tactual presentation of the same surface. Any two such presentations will be
quite different in point of intrinsic character; but still, the properties that are presented
by the former will overlap with the properties that are presented by the latter.
This answer is plausible; but it seems that there is an aspect of the worry about the
compatibility of Nagel’s explanation with materialism that it fails to address. Let us
trade our hypothesis about pain in for the slightly more concrete hypothesis that pain
is the very same thing as electrochemical activity in a network of C-fibers that is
essentially G-shaped, where G is a certain geometric structure. Now it can seem that if
this hypothesis is true, then any presentation of pain, whether a genuine case of
awareness or an imaginative construction, should have features which indicate that
what is presented is activity in a network that essentially exemplifies G. In other
words, it can seem that there should be something G-ish about both real and
imagined pains, something that would present a foothold for the concept of G. If
G-ishness is part of the very essence of pain, shouldn’t we be able to detect at least a
hint of G-ishness by inspecting a presentation of pain?
I think that this intuition is quite natural, but I also think that it is due to a
confusion. If presentations of pain do not present a foothold for the concept of G, this

8 9
Finke (1986, p. 92). Finke (1986, p. 88).
 IMAGINABILITY, CONCEIVABILITY, POSSIBILITY, AND THE MIND - BODY PROBLEM 123

is because the mechanisms that control the applicability of the concept are exclusively
perceptual, in the sense that they are exclusively responsive to features of perceptual
presentations. Thus, like all other concepts with spatial significance, the concept of
G owes its existence to our need to have concepts that make it possible for us to
classify extramental entities that are given to us perceptually; and, as a result, the
presentations that are capable of triggering the concept of G are limited to veridical
presentations of instances of G and to other mental states that are similar to such
presentations. Presentations of pain, whether genuine states of awareness or imagina-
tive constructions, are quite different from veridical presentations of instances of
G. Hence, when we examine such presentations with a view to detecting a hint of
G-ishness, it is inevitable that we will come up empty handed.
It appears, then, that Nagel’s account of how we are able to imagine disembodi-
ment has a couple of important virtues. Unfortunately, it is also true that it has a flaw.
Nagel’s account presupposes that it is possible for us to perceive brain processes, and
this presupposition is highly questionable. To be sure, we are able to use the naked
eye to perceive whole brains and various parts of brains. Further, by focusing
microscopes on preparations of dead tissue, we are able to perceive certain aspects
of the structure of individual brain cells. But neither of these things count as
perceiving electrochemical activity in living neurons. Accordingly, it seems reason-
able to say that brain processes lie on the theoretical side of the fuzzy line that divides
theoretical entities from observable entities. Our access to brain processes is mediated
by theories. We cannot be said to perceive them.
If it is wrong to say that one can perceive a brain process, then it must also be
wrong to say that one imagines a brain process by putting oneself into a state which
resembles the state one is in when one perceives a brain process. So Nagel’s account
of what is involved in imagining a brain process is mistaken. We must replace it with
a more realistic picture.
Reflection suggests that there are two ways of imagining a brain process. First, it is
possible to do so by putting oneself into a state which resembles the state one is in
when one indirectly perceives a brain process, that is, by putting oneself into a state
which resembles the state one is in when the following conditions are satisfied: (a)
one is perceiving a piece of apparatus whose function it is to detect the presence of
various kinds of brain process; (b) the apparatus indicates that a brain process of a
certain type T is currently occurring; and (c) one is aware that the apparatus indicates
that a process of type T is occurring. Second, it is possible to imagine a brain process
by putting oneself into a state which resembles the state one is in when one is
perceiving a model of a brain process and one is perceiving it as a model, that is,
by putting oneself into a state which resembles the state one is in when the following
conditions are satisfied: (d) one is perceiving a series of events that are more
accessible to the senses than brain processes; (e) there is a structural isomorphism
between the series of events and the segments of a certain brain process; (f) one is
aware of this isomorphism; and (g) in perceiving the series, one makes use of one’s
124 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

awareness of the isomorphism, that is, one thinks of the series as a representation of
the brain process in question.
Assuming that this account is more or less correct, we can explain what is involved
in imagining the absence of a brain process by saying that one does this by putting
oneself in an imaginative state that fails to satisfy conditions (a)–(c) and that also fails
to satisfy conditions (d)–(g). (If it should turn out that this account is inadequate, we
can improve it by adding the requirement that the state must include the thought that
the situation one is imagining is devoid of brain processes.)
At this point, we should take note of the fact that the explanatory paradigm that is
developed earlier can be applied not only to explain intuitions about disembodiment,
but also to explain intuitions about the possibility of zombies. Just as intuitions of the
first kind can be explained as due to one’s sympathetically imagining the presence of
pain while “perceptually” imagining the absence of C-fiber stimulation, so also intu-
itions of the second kind can be explained as due to one’s sympathetically imagining
the absence of pain while “perceptually” imagining the presence of C-fiber stimulation.
(I put “perceptually” in scare quotes to remind the reader of the need for qualifications
of the sort described two paragraphs back—that is, qualifications based on conditions
like (a)–(c) and (d)–(g).) It is clear, I think, that the possibility of forming imaginative
representations of the latter sort is no less compatible with type-materialism than the
possibility of forming imaginative representations of the former sort.
Before concluding these Nagelian reflections, we must consider one further issue.
Although I have tried to show that Nagel’s footnote offers us an adequate explanation
of certain of our intuitions about disembodiment and zombiehood, I have not yet
indicated how it might be true that our Nagelian account explains these intuitions
away. One explains a set of intuitions by describing the mechanisms that produce
them. To explain the members of the set away one must in addition provide evidence
which calls the reliability of the relevant mechanisms into question, i.e., one must
produce evidence which makes it reasonable to doubt that the intuitions produced by
these mechanisms are quite likely to be true.
To see that there are grounds for doubting the reliability of intuitions about the
possibility of disembodiment and zombiehood, we should first take note of the fact
that other intuitions about separability can plausibly be claimed to be due to
mechanisms of the same sort as the mechanisms I have claimed to be responsible
for these “Cartesian” intuitions. Here is a description of one of the “Cartesian”
mechanisms: it produces intuitions to the effect that pain can occur without being
accompanied by C-fiber stimulation by splicing together images of the presence of
pain and images of the absence of C-fiber stimulation. Now consider intuitions to the
effect that heat can occur without being accompanied by molecular motion. It is
plausible, I think, to attribute such intuitions to a mechanism that operates by
splicing together images of the presence of heat and images of the absence of
molecular motion. Or consider intuitions to the effect that water can exist without
 IMAGINABILITY, CONCEIVABILITY, POSSIBILITY, AND THE MIND - BODY PROBLEM 125

being accompanied by H2O. It is plausible to attribute such intuitions to a mechan-

ism that operates by splicing together images of the presence of water and images of
the absence of H2O.
Summarizing, we can say that the mechanisms that are responsible for intuitions
about the possibility of disembodiment and zombiehood are members of a family
F of mechanisms that can be described as follows: where M is any psychological
mechanism, M is a member of F if M operates by splicing together images of the
presence (or absence) of a commonsense phenomenon (i.e., a phenomenon to which
we have access by a commonsense faculty of awareness, such as introspection or
visual perception) and images of the absence (or presence) of a theoretical phenom-
enon (i.e., a phenomenon to which we have access only via theory-construction and
laboratory apparatus). Just as it is plausible to say that intuitions about the possibility
of disembodiment and zombiehood are due to the Nagelian mechanisms that I have
described earlier, so also, I wish to claim, it is plausible to say that our other intuitions
about the separability of commonsense phenomena and theoretical phenomena are
due, or at least tend to be due, to other members of F.10,11
We are in a position to assess the reliability of many of the members of F without
begging any questions concerning the mind-body problem, and when we make such
assessments, we find that the results are quite negative. Intuitions to the effect that
heat is separable from molecular motion are false, as are intuitions to the effect that
water is separable from H2O, as are intuitions to the effect that light is separable
from electromagnetic radiation. And so on. Since it is plausible to say that the
Nagelian mechanisms that are responsible for intuitions about disembodiment and

10
In effect, I am here proposing a new explanation of such intuitions as the intuition that heat is
separable from molecular motion—that is, an explanation that is quite different than Kripke’s explanation.
According to Kripke, it seems to us that heat is separable from molecular motion because we take ourselves
to be imagining a situation in which heat is present but molecular motion is absent when we are actually
imagining a situation in which the sensation of heat is present but molecular motion is absent (or because
we take ourselves to be imagining a situation in which molecular motion is present but heat is absent when
we are actually imagining a situation in which molecular motion is present but the sensation of heat is
absent). As I see it, this explanation is fundamentally misguided; for as I see it, in non-pathological
circumstances introspection gives us pretty accurate access to the contents of our own states of imagin-
ation. Accordingly, I prefer to explain the intuition about the separability of heat and molecular motion,
and all related intuitions, such as the intuition that water is separable from H2O, by saying that they are due
to image-splicing mechanisms of the sort described in the text.
11
As indicated in the previous note, I here mean to be proposing an explanation of imagination-based
intuitions of separability. There is one respect in which the present sketch of the explanation is incomplete.
As I see it, the “splicing” mechanisms operate to produce an intuition to the effect that it is possible that p
only when a subject is not already in possession of a “defeater” for that intuition—that is, only when a
subject is not already in possession of reasons for believing that it is necessary that not-p. Nothing is said
about the inhibiting effects of such defeaters in the text.
I am abstracting from considerations having to do with defeaters in the present section because I want
there to be one place in the chapter in which the main idea is presented without being accompanied by a
forest of qualifications. However, when we consider conceivability-based intuitions of separability in
Section III, I will be at pains to acknowledge the role of defeaters in inhibiting intuitions of separability.
See especially the penultimate and ante-penultimate paragraphs of Section III.
126 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

zombiehood are fundamentally akin to the mechanisms that are responsible for other
intuitions about the separability of commonsense and theoretical kinds, and since we
know that the latter mechanisms are fundamentally unreliable, the type-materialist
can claim to be in possession of an argument which makes it rational to doubt the
reliability of Nagelian mechanisms.
To be sure, there is a significant difference between Nagelian mechanisms and the
other members of F—Nagelian mechanisms make use of the sympathetic imagin-
ation as well as the perceptual imagination, but presumably this is not true, for
example, of the mechanisms that are responsible for intuitions about heat and
molecular motion. (Perhaps we could describe the non-Nagelian members of F as
mechanisms that splice together images provided by the sympathetic imagination
and images provided by the “perceptual” imagination.) It would be implausible to
maintain, however, that this difference makes the Nagelian mechanisms more
reliable than the other members of F.

III
We have thus far considered only intuitions about separability that are due to the
imagination. Now we must take note of the fact that there are also intuitions about
separability that are due to the faculty that is responsible for our ability to conceive of
possible situations (hereafter called the faculty of conception). These two sources are
quite different: the representations constructed by the imagination are largely quali-
tative in character, whereas the representations constructed by the faculty of con-
ception are largely conceptual and propositional. (In saying that the representations
constructed by the imagination are largely qualitative, I mean that, like sensory states,
they have defining characteristics that are qualitative and that, again like sensory
states, their ability to represent things is largely a function of these qualitative
characteristics.) Because of this difference, our explanatory task is not yet complete.
We have succeeded in explaining away intuitions concerning the separability of
sensory states and brain states that are due to the imagination, but what about the
corresponding intuitions that are due to the faculty of conception? Can they be
explained away as well? If so, how?
To prevent misunderstanding, I should acknowledge that the imagination and the
faculty of conception are importantly connected, in the sense that each tends to draw
upon the resources of the other in constructing representations of possibilities.
Suppose I imagine my daughter reading a book about ancient Mayan civilization.
My representation of this situation will have a central qualitative component, depict-
ing a young woman in a chair with a book in her hands, but it seems inevitable that it
will also have an important conceptual component. How else could it invoke the
highly abstract characteristic being a book about the ancient Mayans? Or suppose that
while reading The Warden I form a conceptual representation of one of the scenes—
say, a representation of a conversation between Mr. Harding and Archdeacon
 IMAGINABILITY, CONCEIVABILITY, POSSIBILITY, AND THE MIND - BODY PROBLEM 127

Grantly. Here my representation will consist largely of internal counterparts of the

sentences that Trollope uses in his description of the conversation. But it seems likely
(though not inevitable) that my representation will have a significant qualitative
dimension as well. My representation will very likely include a qualitative state which
is like the perceptual state I would be in if I were observe two gentlemen who actually
possess the physical features that Trollope attributes to his characters.
Thus, I do not wish to deny that the imagination and the faculty of conception are
closely aligned. Rather, my point is this: However closely aligned these two powers
may be in practice, it is in principle possible to use the faculty of conception to
construct representations that are largely or entirely without a qualitative dimension.
Because of this in-principle-independence of the faculty of conception, it is inappro-
priate to claim to have a basis for explaining a set of intuitions that are due to this
faculty simply in virtue of having a basis for explaining the corresponding intuitions
that are due to the imagination.
It appears, then, that it really is necessary to supplement the explanation in
Section II with a second explanation, a separate explanation which seeks to account
for the intuitions of separability that are due to the faculty of conception. As a first
step, I will introduce a quasitechnical notion—the notion of a sensory concept.
Sensory concepts have two features which collectively distinguish them from all
other concepts. First, trivially, if X is a sensory concept, then X denotes a type of
sensory state. And second, if X is a sensory concept, and S is the sensory state that is
denoted by X, then, in an important range of contexts in which one is justified in
ascribing X to an instance of S (specifically, the contexts in which one is led by
introspection to ascribe X to one of one’s own sensory states), the sense experience
that provides one’s justification for the ascription is identical with the instance of S to
which X is applied.
It follows from this characterization that the concept of pain is a sensory concept.
This follows because, in the first place, the concept of pain denotes a sensory state,
and in the second place, it is constitutive of the concept of pain that when one is
justified in ascribing the concept to one of one’s own sensory states, one is so justified
in virtue of the fact that one is currently in pain.
It also follows that sensory concepts are importantly different from neuroscientific
concepts—even if, as type-materialism maintains, sensory concepts coincide with
certain neuroscientific concepts in point of denotation. Assuming that type-materi-
alism is correct, the concept of pain denotes the same state as the concept of C-fiber
stimulation. But it is nevertheless true that these concepts are quite different in virtue
of having different justification conditions. As we just noticed, ascriptions of the
concept of pain to one’s own states are paradigmatically justified by experiences that
count as instances of pain. But the justification conditions for paradigmatic ascrip-
tions of the concept of C-fiber stimulation have nothing to do with pain, even when
the ascriptions are self-ascriptions. Rather, the experiences that are paradigmatically
relevant to justifying ascriptions of the concept of C-fiber stimulation are of the
128 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

following two kinds: first, experiences of the sort that confirm the neuroscientific
theories in which the concept of C-fiber stimulation is embedded, and second,
experiences of the sort that one has when one is observing an apparatus which
indicates that there is an instance of C-fiber stimulation in one’s current environ-
ment. (Experiences of the second sort might include, for example, the experiences
that are associated with observing images generated by PET-scan devices.)
A qualification. It is somewhat tendentious to say, as I have said earlier, that
ascriptions of concepts like the concept of pain to one’s own states are justified by the
occurrence of instances of the states to which the concepts refer. A number of
philosophers prefer to say that ascriptions of the concept of pain to one’s own states
are not justified by sensory evidence of any kind. Thus, some hold that such
ascriptions are self-justifying, and others think that they are beyond justification—
that they neither admit of justification nor require it. Fortunately, my present
purposes can be achieved even if the presuppositions of the foregoing characteriza-
tion of sensory concepts are rejected. All that is strictly necessary for present
purposes is that the reader grant two things; first, that concepts like the concept of
pain may legitimately be said to have denotations (and in particular, that they may
legitimately be said to denote sensory states); and second, that the justification
conditions associated with these concepts have peculiarities that distinguish them
from the justification conditions that are associated with neuroscientific concepts. If
these two assumptions are granted, then everything else that I wish to say will go
through.
We are now in a position to address the main question of the present section: Is it
possible to explain the intuitions of separability regarding pain and C-fiber stimula-
tion that derive from the faculty of conception in a way that is compatible with type-
materialism? I will try to show that the answer should be “yes.”
The explanation I have in mind can be obtained by pruning, qualifying, and
supplementing a construction that I presented in an earlier paper.12 It consists of
the following four propositions:
(1) There are no substantive a priori ties between sensory concepts and neuros-
cientific concepts. That is to say, apart from various principles that are
logically true (e.g., “For all x, if x is in pain, then either x’s C-fibers are
stimulated or x in pain”), there are no principles involving both sensory
concepts and neuroscientific concepts that can be known a priori to be true.
(2) If there are no substantive a priori ties between two concepts, then it is possible
to conjoin either of the concepts with the negation of the other without
producing an inconsistency. That is to say, it is possible to use the concepts

12
See Hill (1981). There is a related discussion in Hill (1991, pp. 90–5 and 96–8).
Like Nagel’s footnote, my discussions of Kripke’s argument appear not to have been answered by any
of the contemporary defenders of the argument.
 IMAGINABILITY, CONCEIVABILITY, POSSIBILITY, AND THE MIND - BODY PROBLEM 129

to conceive coherently of situations (i.e., to construct internally coherent

descriptions that purport to represent situations) in which there are particu-
lars that fall under one of the concepts but do not fall under the other concept.
(3) If it is within a subject’s power to conceive coherently of its being the case that
p, then, unless the subject has a good a posteriori reason to think that the
proposition that p could turn out to be inconsistent with a set of propositions
that have been shown independently to be necessary, the subject will come to
believe that it is possible—or at any rate, come to believe that it is prima facie
possible—for it to be the case that p.
(4) If type-materialism is true at all, then, since it is in effect a set of identity
statements involving rigid designators, it is necessarily true. However, due to
the fact that type-materialism is not part of our commonsense metaphysics,
we do not normally regard ourselves as having a good a posteriori reason to
think that propositions about disembodiment and the existence of zombies are
incompatible with necessary truths that have been independently established.
(Only the enlightened regard themselves as having such a reason!13)
All four of these claims are plausible. Thus, consider (1). Among other things, (1)
asserts that there are no substantive a priori ties between the concept of pain and the
concept of C-fiber stimulation. This is surely correct; it is in principle possible to
master either of these concepts fully without having mastered the other. Moreover, as
with (1), so also with (2)–(4)—reflection on them serves only to endear them to us
further.
It is reasonable to view (1)–(4) as providing an explanation of the sort we are
seeking. Thus, one of the components of (1)–(4) (viz., (3)) is a lawlike generalization,
and two of the other components (viz., (1) and (4)) can be seen as descriptions of
initial conditions. Moreover, (1)–(4) collectively imply the proposition that the
faculty of conception is disposed to generate intuitions to the effect that sensory
states and brain states are separable. Because of these features, (1)–(4) count as a
deductive-nomological explanation of the intuitions that we have been concerned to
understand.
Having found an explanation of these intuitions, we must ask whether a type-
materialist can use this explanation as a basis for explaining the intuitions away. Now
in order to do this, a type-materialist must provide reason to doubt that the
mechanisms cited in the explanation are fully reliable, i.e., reason to doubt that the
intuitions produced by the mechanisms in question are quite likely to be true. I will
now try to show that this can easily be accomplished.
In effect, the explanation based on (1)–(4) claims that there is a class H of
psychological mechanisms whose members work as follows: where M is any member

13
Several arguments that have been known to lead to enlightenment are collected in Hill (1991,
chapters 2–4).
130 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

of H, M takes two concepts as inputs, and then, provided that it is possible to conjoin
each of the concepts with the negation of the other without generating an inconsist-
ency, and provided also that there is no available a posteriori reason to think that the
two concepts are necessarily coextensive, M delivers an intuition of possibility as an
output. In other words, the explanation claims that there is a class H of mechanisms
such that, where M is any member of H, and A and B are any two properties,
M produces an intuition to the effect that A is separable from B, provided only,
first, that the concept of A is not connected by any analytic ties to the concept of B,
and second, that there is no available a posteriori reason to think that the former
concept is necessarily coextensive with the latter concept. Now it is clear that many
mechanisms that belong to H are highly unreliable. Thus, for example, (2) and (3)
imply that there is a member M of H such that M is capable of producing intuitions to
the effect that heat is separable from molecular motion. This member of H is
extremely unreliable; for as we know, all such intuitions are erroneous—it is impos-
sible to have a situation in which heat is present that is not also a situation in which
molecular motion is present.
To amplify: Strictly speaking, of course, (2) and (3) do not by themselves imply that
there is a member of H that is capable of producing intuitions to the effect that heat is
separable from molecular motion. Rather, (2) and (3) yield this implication when
they are combined with two additional claims: first, the claim that there are no
analytic links between the concept of heat and the concept of molecular motion; and
second, the claim that we are capable of lacking a posteriori reasons for thinking that
these two concepts are necessarily coextensive—that is, the claim that human beings
can fail to be in possession of the standard scientific reasons for affirming the
necessary coextensiveness of the concepts (and also, fail to be in possession of any
other reasons for affirming this identity).14 I take it that neither of these claims is
controversial.
We have observed that the mechanisms that are responsible for conceivability-
based intuitions about disembodiment and zombiehood belong to a certain category
of mechanisms, H, and we have also observed that a number of the mechanisms that
belong to H are highly unreliable. In view of these observations, it is reasonable for us
to have doubts about the reliability of the mechanisms that are responsible for

14
Actually, there are two ways in which one can fail to be in possession of the standard a posteriori
reasons for affirming the necessary coextensiveness of the concept of heat and the concept of molecular
motion. First, as noted in the text, one can be ignorant of the scientific account of the nature of heat—that
is, of the empirical theory which implies that heat is identical to molecular motion. Second, one can be in
possession of this theory, but not be fully aware that the theory gives one grounds for asserting that heat is
necessarily identical to molecular motion. That is to say, it is possible to be in possession of all of the
relevant scientific facts, but to fail for one reason or another to appreciate fully and non-confusedly
the metaphysical implications of those facts. (As I see it, there is a fairly strong tendency to fail to appreciate
the metaphysical implications in question—a tendency that can only be fully eradicated by considering the
Marcus/Kripke argument for the necessity of identity. This is why twentieth-century philosophers have
sometimes experienced intuitions to the effect that heat is separable from molecular motion.)
 IMAGINABILITY, CONCEIVABILITY, POSSIBILITY, AND THE MIND - BODY PROBLEM 131

conceivability-based intuitions about disembodiment and zombiehood. Moreover,

this conclusion can be strengthened. For these “Cartesian” mechanisms belong to a
sub-category of H whose members are uniformly unreliable—specifically, the sub-
category which consists of mechanisms that are responsible for conceivability-based
intuitions to the effect that commonsense kinds are separable from the theoretical
kinds that are correlated with them. (H includes all mechanisms which take pairs of
concepts as inputs and deliver intuitions of separability as outputs. The sub-category
includes only those mechanisms that take certain restricted pairs of concepts as
inputs, specifically, pairs whose first component is a concept of a commonsense kind
and whose second component is a concept of the theoretical kind that is correlated
with the given commonsense kind.) We are committed to holding that all of the non-
“Cartesian” members of this sub-category are extremely unreliable: that is to say, we
are committed to holding that if M is a mechanism that produces conceivability-
based intuitions to the effect that a physical commonsense kind is separable from its
correlated theoretical kind, then that mechanism is highly unreliable. It is clear,
I think, that because of these commitments, we are also committed to holding that
there is a strong inductive argument for the unreliability of the “Cartesian” members
of the sub-category.

IV
So far so good; but we must now take note of a difficulty. Proposition (1) asserts that
identity statements linking sensory concepts and neuroscientific concepts cannot be
known a priori to be true. Type-materialists must embrace this assertion—both
because it is clearly correct and because it is integral to the foregoing explanation
of intuitions of separability. But now we must ask: “Are type-materialists in a
position to explain the a posteriority of the identity statements in question? Can
they explain why mastery of the relevant concepts is not a sufficient condition of
coming to see that the statements are true?” There is a persuasive line of thought
which indicates that the answer is “no.”15
The argument I have in mind runs as follows: “Compare the statement that pain is
identical with C-fiber stimulation with any other identity statement concerning a
commonsense kind and a scientific kind. Compare it, say, with the statement that
heat is identical with the kinetic energy of molecules. Both of these statements are a
posteriori. However, the latter statement differs from the former in that its a
posteriority seems unproblematic—indeed, inevitable. When we are aware of heat

15
The line of thought presented here is in some ways similar to the “property dualism argument” that is
presented in White (1986). (See in particular pp. 351–3.) By the same token, the answer to this line of
thought that I give in the remainder of Section IV can be taken as a reply to White’s argument. (Also,
I believe that readers of Section IV will be able to see that it provides type-materialists with a basis for
answering Frank Jackson’s knowledge argument. See Jackson (1982) and Jackson (1984).)
132 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

from a commonsense perspective, we are not directly aware of heat itself. We are not
in touch with its intrinsic nature. Rather, we are aware of it as a phenomenon that has
a certain characteristic effect on human observers; we are aware of it via a certain
familiar sensation. In view of this fact about heat, it is no surprise that we cannot see a
priori that heat involves molecular motion. On the other hand, when we turn to
consider pain, we find that the situation is quite different. There is a commonsense
perspective (namely, the perspective afforded by introspection) which enables us to
be directly aware of pain. When we are aware of pain from this perspective, we are
aware of it as it is in itself. We are presented with the intrinsic nature of pain.
Accordingly, if it is true that pain is the very same thing as C-fiber stimulation, should
we not be aware of it as C-fiber stimulation? Further, should it not be possible for us
to see, simply in virtue of having mastered the use of the concept of pain and the
concept of C-fiber stimulation, that these two concepts are coreferential? Suppose
that Jones has fully mastered the use of both concepts. In order to have attained full
mastery of the concept of pain, Jones must at some point have been introspectively
acquainted with pain. Since introspective acquaintance with pain is always direct
acquaintance, Jones must be fully apprised of the intrinsic nature of pain. Now
suppose it is true that the concept of pain and the concept of C-fiber stimulation
are coreferential. Given that Jones is fully apprised of the intrinsic nature of pain, and
given also that the concept of C-fiber stimulation picks out its referent by describing
its intrinsic nature, should it not be possible for Jones to appreciate this coreferenti-
ality? Moreover, should it not be possible for him to appreciate it purely on the basis
of reflection?”
This line of thought brings two facts to our attention. First, it shows that anyone
who thinks that “[p]ain is identical with C-fiber stimulation” is true should find its a
posteriority problematic, and should therefore feel an obligation to explain this a
posteriority. And second, it shows that the task of providing an explanation is non-
trivial. In particular, it shows that it will not do to invoke the explanatory paradigm
that works in other, superficially similar cases, such as the case of heat and molecular
motion.
Fortunately, there is a way out. The question before us now is closely related to one
that we considered in Section II—viz., the question “If pain is the very same state as
the firing of a G-shaped network of C-fibers, then why don’t our experiences of pain
provide a foothold for the concept of G?” We answered the earlier question by
reminding ourselves of the fact that the concept of pain serves different purposes
than the concept of G. Whereas the concept of pain is designed to make it possible for
us to classify particulars on the basis of information that is available via introspection,
the concept of G is designed to make it possible for us to classify particulars on the
basis of information that reaches us via the senses. Because of this difference, we
observed, it is inevitable that the concept of pain should be responsive to different
sensory presentations than the concept of G. That is to say, it is inevitable that the
experiences that are causally responsible for applications of the former concept
 IMAGINABILITY, CONCEIVABILITY, POSSIBILITY, AND THE MIND - BODY PROBLEM 133

should be different from the experiences that are causally responsible for applications
of the latter concept. As I see it, the present difficulty admits of a similar solution.
Setting third person ascriptions of pain aside, we can, I think, make the following
observation: When someone applies the concept of pain to a particular, the experi-
ence that causes him or her to make the application is identical with the particular to
which the concept is being applied, and therefore, in the case of a non-erroneous
application, the particular in question is a pain. On the other hand, like the concept
of G, the concept of C-fiber stimulation is designed to classify particulars on the basis
of information that reaches us via the senses. Because of this, the experiences that
guide us in applying the concept of C-fiber stimulation are experiences whose
occurrence is restricted to contexts in which we are using the senses to observe
certain external phenomena. To be a bit more specific, the experiences that are
germane to the concept of C-fiber stimulation are of two kinds. First, there are the
experiences that motivate and confirm the neuroscientific theories in which the
concept of C-fiber stimulation is embedded. And second, there are the experiences
that lead us to apply the concept in particular contexts. (This second group includes,
for example, experiences of the sort that occur when we are using the senses to
observe images produced by PET scanners and MRI devices.)
We may conclude, I think, that there is a materialistically acceptable explanation of
the fact that the identity of pain and C-fiber stimulation can only be known a
posteriori. In full dress, this explanation runs as follows: “The concept of pain and
the concept of C-fiber stimulation serve classificatory purposes of quite different
kinds. As we saw, the former concept is designed to enable us to classify phenomena
on the basis of information that comes to us via introspection, and the latter concept
is designed to enable us to classify particulars on the basis of information that comes
to us via the senses. Because of this difference, the experiences that control our use of
the former concept are inevitably quite different from the experiences that control
our use of the latter concept. But now, when the use of one concept is controlled by
experiences that are altogether disjoint from the experiences that control the use of a
second concept, it is in general impossible for us to determine, simply in virtue of
having mastered the use of the concepts, whether they are coreferential. It should
come as no surprise, therefore, that it is possible to master the use of the concept of
pain and the concept of C-fiber stimulation without appreciating their coreferenti-
ality. It is inevitable that their identity be knowable only a posteriori.”
I close the present section by pointing out that there is a second way of expressing
the difference between the purposes that are served by the concept of pain and the
purposes that are served by the concept of C-fiber stimulation. Instead of expressing
this difference by appealing to the distinction between information that comes to us
via introspection and information that comes to us via the senses, we can express it
by invoking the closely related distinction between being IN a state and OBSERVING
a state. Thus, we can say that the concept of pain is designed to enable us to classify
states that we are aware of in virtue of being IN them, and that the concept of C-fiber
134 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

stimulation is designed to enable us to classify states that we are aware of in virtue of

OBSERVING them, where “observing” is understood broadly, in such a way as to be
applicable to states of awareness that are mediated by perceptions of such apparatus
as PET scanners and MRI devices.

V
There is an important issue that has not yet been addressed. I have maintained that
the mechanisms that are responsible for a number of our modal intuitions are
unreliable. This might be taken to show that the line of thought in the present
paper presupposes a general skepticism about our ability to obtain knowledge of
modal facts. And, by the same token, it might be taken to show that the line of
thought in question suffers from a disabling flaw. No argument that calls the general
run of our claims to have modal knowledge into question can be fully satisfactory, for
we feel sure that we possess a fairly large amount of such knowledge.
Fortunately, there are two features of the argument that prevent it from leading to
a radical modal skepticism. First, my claims about unreliability have been restricted
to intuitions that have implications concerning a posteriori questions about matters
of fact. (Hereafter I will call such intuitions a posteriori modal intuitions.) Thus,
consider the intuition that heat is separable from molecular motion. When it is
combined with the true thesis that properties are necessarily identical if they are
identical at all, this intuition leads to a negative answer to the question of whether
heat and molecular motion are identical—a question that is a paradigm of a poster-
iority. Accordingly, the intuition is an a posteriori modal intuition. Second, I have not
assumed that we are incapable of forming a posteriori modal intuitions that are
correct, but only that our a posteriori modal intuitions tend to be incorrect when we
form them without being fully apprised of the relevant empirical facts. Thus, for
example, in my account of conceivability-based intuitions concerning the separabil-
ity of heat and molecular motion, I claimed that the mechanisms that are responsible
for such intuitions operate only when we lack the scientific information which
indicates that the concepts of heat and molecular motion are necessarily coextensive.
In sum, while it is true that the lines of thought in earlier sections call a class of
modal intuitions into question, this class is highly restricted, consisting as it does only
of a posteriori modal intuitions that are formed independently of information con-
cerning the relevant empirical matters. Accordingly, the foregoing lines of thought
have no tendency to promote a general skepticism about the possibility of obtaining
modal knowledge. Moreover, it seems to me to be good to be skeptical about the
intuitions that belong to the given restricted class; for as I see it, it follows from their
description that such intuitions could not possibly be worthy of our trust!16

16
The question broached in this final section deserves a great deal more attention than I am able to give
it in a paper that is primarily concerned with other matters. I hope to return to it on another occasion.
 IMAGINABILITY, CONCEIVABILITY, POSSIBILITY, AND THE MIND - BODY PROBLEM 135

References
Chalmers, D. J. (1996). The Conscious Mind: In Search of a Theory of Consciousness. Oxford:
Oxford University Press.
Finke, R. A. (1986). “Mental Imagery and the Visual System,” The Scientific American 254:
88–95.
Hart, W. D. (1988). Engines of the Soul. Cambridge: Cambridge University Press.
Hill, C. S. (1981). “Why Cartesian Intuitions Are Compatible With the Identity Thesis,”
Philosophy and Phenomenological Research 42: 254–65.
Hill, C. S. (1991). Sensations: A Defense of Type-Materialism. Cambridge: Cambridge Univer-
sity Press.
Jackson, F. (1982). “Epiphenomenal Qualia,” Philosophical Quarterly 32: 127–36.
Jackson, F. (1986). “What Mary Didn’t Know,” Journal of Philosophy 83: 291–5.
Kripke, S. A. (1977). “Identity and Necessity,” in S. P. Schwartz (ed.), Naming, Necessity, and
Natural Kinds. Ithaca, NY: Cornell University Press, 66–101.
Kripke, S. A. (1980). Naming and Necessity. Cambridge, MA: Harvard University Press.
Nagel, T. (1974). “What Is It Like to Be a Bat?” The Philosophical Review 83: 435–50.
Pylyshyn, Z. (1981). “The Imagery Debate: Analog Media Versus Tacit Knowledge,” in
N. Block (ed.), Imagery. Cambridge, MA: MIT Press, 151–205.
Shepard, R. N. and Cooper, L. (eds.). (1982). Mental Images and Their Transformations.
Cambridge, MA: MIT Press.
White, S. (1986). “Curse of the Qualia,” Synthese 68: 333–68.
Yablo, S. (1990). “The Real Distinction Between Mind and Body,” Canadian Journal of
Philosophy, Supplementary Volume 16: 149–201.
Yablo, S. (1993). “Is Conceivability a Guide to Possibility?” Philosophy and Phenomenological
Research 53: 1–42.
 8
The Identity Theory*

I
The doctrine that mental states are identical with physical states was defended in
antiquity by Lucretius and in the early modern era by Hobbes. It achieved consid-
erable prominence in the 1950s as a result of the writings of Herbert Feigl (Feigl
1958), U. T. Place (Place 1956), and J. J. C. Smart (Smart 1959). These authors
developed reasonably precise formulations of the doctrine, clarified the grounds for
embracing it, and responded persuasively to a range of objections. More recently it
has been defended systematically by Hill (1991) and Papineau (2002). Other con-
temporary advocates include Loar (1990), McLaughlin (2004), Polger (2004),
McLaughlin (2010), and Balog (2012).
It is necessary to distinguish between theories of mental states that make claims of
token identity and theories that make claims of type identity. Theories of the first sort
maintain that concrete mental events, such as the pain I experienced at 3 p.m. today,
are identical with particular physical events, such as the processing of information
from peripheral nociceptors that took place in my brain at 3 p.m. Theories of the
second sort assert that mental properties are identical with physical properties.
A view of this sort might claim, for example, that the property being a pain is
identical with the property being a case of activity in regions R1, R2, . . . , and Rn of
the cerebral cortex. Now the proposition that a mental property ç is identical with a
physical property ł entails that all concrete tokens of ç are identical with tokens of ł,
but there is no entailment running in the other direction. In view of these facts about
entailment, it is clear that claims of type identity are logically stronger than claims of
token identity, and have correspondingly greater power to unify and simplify our
theories of human nature and the universe. Largely because of this greater power, the
former claims have received much more attention than the latter. Indeed, the
contemporary literature focuses almost exclusively on type identity.
Discussions of type identity generally presuppose the notion of a qualitative
characteristic or a quale. There is no universally accepted definition of this notion,

* Much of the material in Sections II and III is excerpted from the Introduction that Simone Gozzano
and I wrote for Gozzano and Hill 2011. I haven’t listed Gozzano as a co-author of the present chapter
because the remaining sections either come from another source (Hill 2009) or are new. The reader should
not suppose that Gozzano is committed to the views that are expressed in Sections I and IV–VII.
 THE IDENTITY THEORY 137

but there is wide agreement that qualia include the introspectible characteristics of
bodily sensations, such as being a pain, the introspectible characteristics of emotions,
such as the feeling of awe, and the characteristics of perceptual experiences that we
keep track of by expressions like “looks yellow,” “tastes sweet,” and “smells sulfur-
ous.” It is often said that a mental state counts as qualitative if there is something it is
like to be in that state.
Most advocates of type identity have seen it as concerned exclusively with mental
states that have a qualitative dimension. To be more specific, they have maintained
that qualitative properties are identical with neural properties of brain states, and
have denied that this is true of representational or intentional properties like being a
belief about Cleopatra and being a visual experience of Cleopatra. Roughly, the reason
for the denial is that representational states generally have representational contents
that involve extra-cranial objects and properties. For example, to believe that Cleo-
patra was of Macedonian descent is to be in a state that essentially involves relations
to Cleopatra and to ancient Macedonia. One could not be in this state if Cleopatra
and ancient Macedonia had not existed. Indeed, the very existence of the state
depends on the existence of Cleopatra and Macedonia. In contrast, it is generally
held that qualitative properties are purely internal. Bodily sensations and other
qualitative states are often caused by extra-cranial items, but they do not depend
logically on those items for their existence.
I will be concerned here with the version of the psychophysical identity thesis that
we have just been considering—the version which claims that every qualitative state
is type-identical with a neural state. My goal is to assess the current status of this
view, which I will call type-materialism, I will begin by reviewing the main grounds
for accepting the view, and I will then consider three attempts to call it into question.
In the end I will conclude (i) that the arguments in favor of type-materialism have
considerable merit, and (ii) that the standard objections to it can be answered in ways
that are prima facie persuasive. These conclusions suggest that type-materialism will
continue to be seen as an important option. I will, however, also note a few worries
about it that are different than the standard objections, and that may in the end turn
out to be more difficult to criticize effectively.

II
There are three principal reasons for accepting the view that qualitative properties are
identical with neural properties. First, as J. J. C. Smart emphasized in his early papers
on the topic, it is simpler than alternatives, because it sees only one category of
properties where other theories see two. Here is Smart’s well known formulation of
this point:
138 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

If it be agreed that there are no cogent arguments which force us into accepting dualism, and if
the brain-process theory and dualism are equally consistent with the facts, then the principles
of parsimony and simplicity seem to me to decide overwhelmingly in favor of the brain-
process theory. (Smart 1959, p. 156)

Second, the identity thesis has more explanatory power than the various forms of
dualism. Unlike dualism, it can reductively explain the large array of laws in which
psychological properties are involved, including the correlation laws that link psy-
chological properties to neural properties. Third, it does a better job of honoring our
intuitions about the casual powers of psychological states than do other theories. We
believe, for example, that pains are causally responsible for much of our thought and
talk about pain, and also for such forms of behavior as wincing, crying out, and
taking steps to secure relief. Type-materialism sustains all of those intuitions, and
does so in an especially simple and straightforward way. I will say a bit more about
each of these three considerations.
The simplicity argument invokes Occam’s Razor, which advises that entities are
not to be multiplied beyond necessity. This principle is widely thought to provide a
rationale for preferring type-materialism to property dualism. Even dualists are
inclined to agree that if one theory is more complex than another, then its advocates
bear the burden of proof. It remains to be seen, however, whether Occam’s Razor
provides an epistemic reason for accepting type-materialism or a reason of some
other kind. One might think it obvious that the Razor provides an epistemic reason.
After all, the difference between dualism and type-materialism is just that the former
goes beyond the latter in its existential commitments, making all of the existential
claims that type-materialism makes and many others as well. Or so it can seem. On
this view of the matter, it appears that dualism has a larger existential burden than
type-materialism, and that one therefore takes more of a risk in believing it. (Since
dualism has a larger existential burden, it is less likely to be true.)
Reflection shows, however, that these observations neglect an important dimen-
sion of the relationship between the two theories. It is true that dualism claims that
there is an irreducible category over and above the irreducible categories posited by
type-materialism, but it does not follow from this that type-materialism makes a
weaker claim than dualism. Dualism and type-materialism are alike in asserting that
reality contains a category consisting of psychological properties and also a category
consisting of neural properties. If dualism seems to be a more ambitious theory than
type-materialism, this is because, after making this claim about categories, dualism
goes on to assert that the two categories are mutually irreducible. But type-
materialism goes on to make an additional claim of its own—specifically, that one
of the categories is reducible to the other. It is not at all clear that a claim of
reducibility is weaker than a claim of irreducibility, and by the same token, it is not
at all clear that one would take less of a risk in accepting materialism. Accordingly, it
may be a mistake to see the simplicity argument for materialism as fundamentally
 THE IDENTITY THEORY 139

epistemic in character. Perhaps it should be seen as an aesthetic argument instead. In

explaining his commitment to simplicity, Quine once said that he had a taste for
desert landscapes. This can’t be all that there is to the matter, for the appeal of
simplicity is more universally appreciated than the beauty of the desert. But it may be
necessary to think of simplicity, in the relevant form, at least, as more closely related
to beauty than to probability or truth. (For further discussion, see Hill 1991,
pp. 29–40.)
As noted, the second argument for type-materialism emphasizes the explanatory
power of the doctrine. This argument has two versions. The first version begins with
the assumption that there are strong correlations between psychological states and
certain neural states. (Accordingly, it presupposes that the multiple realization
argument, which is discussed a bit later on, can be answered.) It then claims that
type-materialism provides the best explanation for these correlations. Thus, for
example, it claims that the best way of explaining the correlation between pain and
a certain brain state is to say that pain is identical with that state. Finally, it invokes
the best explanation principle, which asserts, roughly speaking, that one is entitled to
believe a theory of X if the theory provides the best explanation of all of the data that
are relevant to X. The conclusion is of course that we are entitled to believe type-
materialism. (For further discussion, see Hill 1991, pp. 22–6, and McLaughlin 2010.)
The second version of the argument is like the first; but instead of invoking correl-
ations between qualitative states and neural states, it invokes laws linking qualitative
states to other phenomena, such as behaviors of various kinds. Consider, for
example, the generalization that pain causes one to withdraw reflexively from
aversive stimuli. It would be nice to be able to explain this generalization. According
to the second version of the argument, we can provide such an explanation if we
suppose that pain is identical with the brain state that is the neural cause of reflexive
withdrawals. Indeed, it is claimed, we can provide the best explanation of the
generalization if we make this supposition. But if this is so, then the best explanation
principle authorizes us to accept the supposition. (See Block and Stalnaker 1999.)
Both of these versions of the explanatory power argument have been criticized.
I will not discuss the objections to them systematically. Instead, I will just reply briefly
to an objection to the first version that has often appeared in the literature.
(McLaughlin responds to a number of other objections in McLaughlin 2010.)
According to the objection I have in mind, it is a mistake to say that correlations
can be explained by saying that the correlated items are identical. The idea here is
that if X is the very same thing as Y, it is a logical error to say that X and Y are
correlated. There cannot be a correlation unless the correlated items are distinct. This
objection rests on a misunderstanding. The first version of the explanatory power
argument is concerned with correlation laws—that is, with propositions of the form
“An instance of X occurs when and only when an instance of Y occurs.” There is no
doubt that propositions of this form can be fully meaningful, and fully true, even if it
should turn out that the property to which “X” refers is identical with the property to
140 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

which “Y ” refers. Now a philosopher of mind is confronted with the question of

whether it is possible to derive certain propositions of the given form from more
fundamental propositions. It appears that the answer to the question must be “yes,”
because the propositions in question can be derived from the corresponding pro-
positions of the form “X is the very same thing as Y.” Also, they are more funda-
mental than the former propositions, because, if true at all, they are necessarily true,
and are therefore not in need of explanation. That is, they are more fundamental
because they can bring a chain of explanations to an end. (Perhaps it is worth
observing in this connection that identity propositions can have the status of laws
of nature. This is true, for example, of Newton’s second law, and of Einstein’s
observation that e = mc2. Presumably claims like “Pain is identical with brain state
B” can share this status.)
The third argument for type-materialism, which was brought to the fore of
contemporary discussions by Kim (1998), is based on the perception that mental
causation is robustly real—more specifically, the perception that psychological states
play essential roles in the causal histories of various forms of behavior, and also in the
causal histories of other psychological states. It seems prima facie that any sound
metaphysical theory should sustain this intuition. But it is not clear that theories
other than type-materialism are capable of sustaining it. Consider the generalization
that pain causes one to cringe. Because we have general inductive grounds for
believing that behavioral phenomena always have neural causes, and because crin-
ging is a form of behavior, it is reasonable to suppose that cringing is caused by a
certain brain state—say, B. Now if pain is identical with B, there will be nothing
mysterious about the fact that both pain and B cause one to cringe. Saying that pain
causes one to cringe and that B causes one to cringe will just be two different ways of
saying the same thing. But if pain is distinct from B, then we will have to say that
pain’s causal efficacy with respect to cringing merely duplicates that of B. This would
be unfortunate for three reasons. In the first place, if the causal contribution of pain
merely duplicates the contribution of B, we can’t say that the contribution of pain is
essential. B would continue to cause cringing even if God did away with pain. Second,
the claim that cringing has two distinct causes runs counter to Occam’s razor. And
third, if pain is distinct from B, it seems that we will have to say that it brings about
cringing without a continuous intervening process. That is, there will have to be some
point P in the physical process running from B to cringing at which pain acts directly,
without benefit of there being an intermediate process linking it to P. Accordingly, its
causal power will be mysterious, like that of telekinesis. Why should pain act on the
physical world at P rather than some other point in the physical process? Why
doesn’t it act at multiple points, or at none? There appears to be no way of answering
these questions.
This is an appealing argument, but reflection shows that it isn’t as powerful as it
initially appears. Thus, it isn’t clear that we have the right to suppose that qualitative
states play essential roles in the causation of physical phenomena. We have an
 THE IDENTITY THEORY 141

intuition to the effect that they are essential, but really, all that induction can show us
is that they are causes of certain events. It can’t show us that there aren’t additional,
over-determining causes of the same events. Hence, it isn’t at all clear that the
intuition of essentiality must be respected. Moreover, insofar as the causal argument
depends on an appeal to Occam’s Razor, it isn’t independent of the simplicity
argument. It doesn’t add much to appeal to simplicity a second time. This brings
us to the point about telekinesis. That point has strong shoulders and can carry some
weight, but it is closely related to the best explanation argument, and may not have
much independent force.

III
Type-materialism was seen as the standard solution to the mind-body problem in the
fifties and early sixties. Then, a few years later, it was summarily abandoned. The
principal reason for this change of opinion was the multiple realization argument,
which was originally propounded in a series of papers by Hilary Putnam. (See, e.g.,
Putnam 1967.)
It is intuitively plausible, Putnam claimed, that the members of many different
species have mental lives that are qualitatively similar, at least in part, to those of
human beings. It is natural to suppose, for example, that rodents and octopi can
experience pain, and also that certain birds and fish share some of our color qualia.
Now the brains of these creatures are structurally quite different than human brains,
and also quite different from each other. Because of these structural differences, it
seems unlikely that the correlation laws that link pain and color qualia to neural
properties in human brains prevail throughout the animal kingdom. That is to say, it
seems unlikely that there are any universal correlation laws. But multiple realization
of this sort precludes identity: unless a qualitative property is universally correlated
with a neural property, it cannot possibly be identical with a neural property. It
follows that type-materialism is wrong.
Putnam’s reasoning can be summarized as follows:
Premise 1: Where P is any psychological kind, there is a wide variety of creatures that can
possess P, including members of other species and complex androids like C3PO.
Premise 2: If there is a wide variety of creatures that can possess P, then there is no one physical
kind by which P is realized—at best, it is realized by different kinds in different creatures.
Premise 3: If P is realized by different physical kinds in different creatures, then P cannot be
identical with any specific physical kind.
Conclusion: No psychological kind is identical with any physical kind.

By way of illustration, Putnam maintained that it is very unlikely that pain is realized
by any one physical kind, because pain is common to animals, reptiles, and mollusks
(“octopuses are mollusca, and certainly feel pain” (Putnam 1967, p. 154)), and the
142 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

brains of these creatures differ radically in point of physical structure. Of course, if

there is no one kind that realizes pain, it is true a fortiori that there is no one kind
with which pain can be identified.
Putnam’s argument is rightly viewed as one of the most significant contributions
to the philosophy of mind of the twentieth century. Among other virtues, it provides
the main motivation for functionalism, which maintains that psychological states are
individuated by their casual roles, not by the structural or compositional features of
brain states. It seems likely that most philosophers of mind think that functionalism
provides the correct account of a broad range of psychological states. Still, impressive
as it is, there is reason to think that the multiple realization argument has been
overrated. I will briefly note a few objections.
First, Putnam seems to have overstated the case for the first premise. He main-
tained that members of a number of radically different species certainly feel pain, and
he seems to have held similar views about the distribution of other mental states, such
as feelings of thirst and hunger. What led him to these conclusions? We know that he
was not influenced by similarities between human brains and reptile brains, or by
similarities between human brains and mollusk brains, for his second premise
implies that any such similarities are less important than the differences between
brains of these types. So he must have been relying on behavioral similarities. But we
know today that behavioral similarities can be an untrustworthy guide to psychological
similarities. Consider visually guided action. In normal human beings it seems to
depend to at least some degree on conscious visual experiences. Thus, for example,
one’s ability to determine which way a pencil is tilting depends on conscious experi-
ences, and this is also true, to an even higher degree, of one’s ability to negotiate
complex landscapes. But we know today that blindsight patients are able to recognize
tilt, and that victims of visual form agnosia are capable of very complex endeavors,
such as hiking over difficult terrain. The fact that a creature can navigate a room
without bumping into furniture, or reach out and shake your hand, is not an adequate
basis for drawing conclusions about the contents of the creature’s visual consciousness.
Second, it’s possible to explain away the appeal of the first premise. It appears that
there is a heuristic for attributing mental states that leads us to make provisional
attributions on the basis of simple movements. To see this, recall the classic video
made by the social psychologists Fritz Heider and Mary-Ann Simmel (Heider and
Simmel 1943). In this film, two dark triangles and a dark circle engage in various
behaviors that are naturally interpreted as aggression, pursuit, flight, observation,
hiding, and bonding. Initially, at least, viewers find it almost impossible refrain from
interpreting the movements of the figures in terms of psychologically pregnant
descriptions of this sort, and they tend to posit purely psychological underlying
causes, such as affection, hostility, covetousness, fear, the desire to control, and the
desire to escape. But all that happens on the screen is that abstract geometrical shapes
move in various suggestive ways. Why is it so natural to interpret simple movements
in terms of these complex mentalistic concepts? The obvious answer is that there is a
 THE IDENTITY THEORY 143

more or less hard-wired heuristic for attributing mental states that takes account only
of motions. This explains why viewers are drawn to mentalistic interpretations of the
figures. Moreover, the thought that the attributions are due to a heuristic is defla-
tionary, implying that they are provisional and subject to correction as information
increases. And in fact, it seems that observers are inclined to withdraw the attribu-
tions when it is emphasized that the figures are simple two-dimensional shapes,
without an internal organization of any kind. Evidently, the initial attributions are
hostage to discovery of an appropriate internal complexity. There seems to be a
general pattern here. Thus, for example, we are willing to attribute pain to ants, but
we tend to withdraw these attributions when we come to appreciate that the ants
have none of the neural structures that support experiences of pain in human beings.
(Cf. Hill 1991, pp. 220–5.) In sum, it seems that many of the intuitive attributions of
qualitative states to which the multiple realization argument appeals derive from a
heuristic that is quick but also very dirty. They are “System 1” judgments, and as
such, they should not be allowed to place binding constraints on theories. (For a
discussion of System 1 heuristics, see Kahneman 2011.)
Third, the multiple realization argument seems to commit us to the functionalist
account of qualitative states—that is, to the view that such states are individuated by
their causal roles. But this is a problem: there are grounds for doubting that
qualitative states like pain can be identified with causal roles, for it is very difficult
to find a set of causal powers that is present in all and only those cases in which a
given qualitative state is present. Consider pain. Paralytics can experience pain, as
can babies, masochists, and those with the disorder known as pain asymbolia.
(Patients with this disorder insist that they continue to feel pain, but they maintain
that their pains no longer bother them. Their testimony is confirmed by imaging
studies, which point to lesions in the centers that are known to be responsible for the
emotional dimension of pain experience.) Paralytics cannot engage in pain behavior;
babies cannot form desires or beliefs about their pains; masochists differ from the rest
of us in that they actively seek painful experiences; and asymbolia patients see pains
as on a par with uninteresting tingles—they neither mind them nor find them
especially worthy of attention. (For discussion, see Grahek 2001.) In view of these
facts, it seems unlikely that there is a set of causal powers that is both necessary and
sufficient for the existence of pains (Hill 1991, pp. 73–6).
Fourth, contrary to what Putnam presupposes, it seems unlikely that we have
intuitions to the effect that members of remote species share highly determinate
qualitative properties with human beings. Rather, our intuitions indicate only that
they share highly determinable or generic properties. But this means that the shared
properties may after all be universally correlated with neural properties, for in
addition to sharing our determinable qualitative properties, members of remote
species also share many of our determinable neural properties.
Fifth, philosophers of neuroscience have emphasized a closely related point: if the
multiple realization argument appears sound, this may be because we are using
144 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

mismatched principles of individuation for psychological kinds and realizing neural

kinds. In a well known discussion of this view, William Bechtel and Jennifer Mundale
summarize it as follows:
[O]ne diagnosis of what has made the multiple realizability claim as plausible as it has been is
that researchers have employed different grains of analysis in identifying psychological states
and brain states, using a coarse grain to identify psychological states and a fine grain to
differentiate brain states. Having invoked different grains, it is relatively easy to make a case
for multiple realization. But if the grain size is kept constant, then the claim that psychological
states are in fact multiply realized looks far less plausible. (Bechtel and Mundale 1999, p. 202)

In general, in considering multiple realization, we must keep it vividly in mind that

neuroscience operates at a number of different levels, and that these different levels
are associated with principles of individuation that have different degrees of inclu-
siveness. To illustrate, suppose that an individual is in a qualitative state Q on two
different occasions. It will no doubt be possible to find measurable differences
between the neural state N1 that corresponds to the first instance of Q and the neural
state N2 that corresponds to the second instance. We can summarize this by saying
that Q is realized by two different neural states. But this has no tendency to show that
Q does not also correspond to a single, more abstractly individuated neural state.
Indeed, assuming that the instances of Q are intrinsically similar, we can expect that
neuroscience will be able to discern a common structure in N1 and N2. (Here I
diverge sharply from Aizawa and Gillett 2009.)

IV
Before bringing this discussion of the multiple realization argument to a close,
I should say something about a special class of Putnamian intuitions. I have main-
tained that we tend to withdraw attributions of mental states to a class of creatures if
we learn that the creatures in question have internal architectures that differ signifi-
cantly from ours. That is, we are committed to a “same insides” principle. Among
other things, this principle leads us to withdraw attributions to insects, mollusks, and
reptiles. But there are cases in which the brains of other creatures differ substantially
from ours along some dimensions, but are quite similar in other respects. I am
thinking here of imaginary androids like Commander Data. I think we are supposed
to imagine that Data’s brain is composed of synthetic material, and is in general quite
different than ours in point of the organization of the smallest units (it has chips
instead of neurons), but that we are also supposed to imagine that his brain is
fundamentally akin to ours in point of higher level structure. To fix ideas, let us
just stipulate that this is so in Data’s case—that his brain that is fully isomorphic to a
human brain at all levels other than that of the smallest units. Given this stipulation,
we have a strong inclination, I believe, to attribute the full range of human mental
states to Data. At first sight, at least, he comes close enough to satisfying the “same
insides” principle. So what should we conclude? Does this case vindicate Putnam’s
 THE IDENTITY THEORY 145

line of thought, even though cases involving members of other biological species fail
to do so?
No. In the first place, whatever our final assessment of Data may be, he has little
tendency to confirm the functionalist account of mental states. Functionalism asserts
that mental states are individuated by causal powers. Data’s mental states are in fact
similar to ours in point of causal powers, but it is Data’s structural similarity to us
that engages our intuitions at the deepest level. A creature who behaved just like Data
but who turned out to have a brain that was structurally quite different than ours
would be a much less robust candidate for the attribution of mental states.
Second, it is not at all clear that Data really satisfies the same insides principle. My
specification of the ultimate building blocks of Data’s brain has been extremely
vague. Would we continue to be willing to view him as a subject of qualitative mental
states after the details had been filled in? The fact that people are disinclined to view a
Blockean homunculi-head as a subject of qualitative mental states shows that there
are ways of filling in the details that forestall attributions of such states. (A Blockean
homunculi-head is a creature whose brain is composed of tiny human-like beings
linked by complex communication channels (Block 1980).) My guess is that people
would also be disinclined—or at any rate, less inclined—to attribute qualitative states
to Data if they were given full descriptions of the composition and behavior of chips.
But at the present point, that is just a conjecture. We need X-phi studies of people’s
reactions to Data when given various kinds of information about his insides before
we can draw firm conclusions.
Third, and most importantly, even if we were to conclude that Data has the full
range of human mental states, it would seem that we could still embrace a type of
identity theory. Specifically, we could identify qualitative mental states with activities
of certain kinds in systems with a certain sort of structure. Unlike the most familiar
forms of the identity theory, this one would not individuate the reductive base in
terms of material composition, but it would be akin to the familiar forms in that its
reductive base would consist of properties that can be exhaustively studied from the
perspective of neuroscience. Structural properties of populations of neurons are no
less in the domain of neuroscience that the embodiments of those properties in
neural tissue.

V
As we saw, the best explanation argument presupposes that there are correlation laws
linking qualitative mental states to brain states, and maintains that such laws are best
explained by type-materialism. We have now found that the leading attempt to call
correlation laws into question is badly flawed. Thus, it appears that the best explan-
ation argument is on firm footing. In the present section I will seek to buttress the
argument further by showing that the three main alternatives to type-materialism
146 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

aren’t in as good a position to explain psychophysical correlation laws as it is. These

alternatives are property dualism, functionalism, and representationalism.
Property dualism denies that qualia are identical with physical properties, and in
addition, it denies that the former properties supervene on the latter. A set of
properties S1 is said to supervene on another set S2 if there are no possible worlds
that are exactly alike with respect to the members of S2 but different with respect to
S1. Since property dualism denies supervenience, it allows that there are worlds that
coincide with the actual world with respect to all physical phenomena, but diverge
from it with respect to qualitative phenomena. From the perspective of property
dualism, it is a kind of accident that qualitative phenomena are arranged the way they
are in relation to physical phenomena. Indeed, it is an accident that there are any
qualitative phenomena at all.
Functionalism and representationalism agree with type-materialism in asserting
that the qualitative supervenes on the physical, but they deny the stronger thesis that
qualitative properties are identical with neural properties. Where type-materialism
entails that qualitative properties are ultimately individuated in the same way as
neural properties, in terms of the organization of neural tissue and its material
composition, functionalism asserts that the individuation is causal. Because of this,
it allows that qualitative states can be realized by a wide variety of physical systems,
for there can be many physical systems that have the same causal organization. Thus,
for example, computers with different physical structures can run the same program.
In doing so, they display the same causal organization, for the whole point of a
program is to confer a set of computational dispositions on a machine. This has an
important consequence. It implies that if functionalism is true, then there are no
correlation laws that link types of qualitative state to types of brain state. Brain states
can uniquely determine qualitative states, but no qualitative state can uniquely
determine a brain state. There is always more than one physical state that can
accompany a given qualitative state.
The third alternative, representationalism, has two forms. In one version it claims
that qualitative states are identical with certain representational states. On this view,
pain is a kind of perceptual or quasi-perceptual representation, though one that is in
the service of an internal perceptual modality. It represents actual or potential bodily
damage. According to the other version, qualia are the properties that are represented
by certain perceptual or quasi-perceptual representations. Pain, for instance, is said
to be a kind of bodily disorder, usually involving stress or damage, because it is
plausible that the representations that realize pain have the function of encoding
information about such disturbances.
Now as we saw, property dualism implies that the relationship between the
physical world and qualitative states is a metaphysical accident. It follows from this
that if property dualism is true, then psychophysical correlation laws are contingent
facts that are fundamental—that is, contingent facts that can have no deeper explan-
ation. It is a trivial consequence of this that type-materialism is in a better position
 THE IDENTITY THEORY 147

than dualism to explain the laws. Further, since functionalism denies that there are
bidirectional correlation laws, there is of course no chance at all of its explaining why
such laws obtain. Theorists with their heads buried in the sand are in no position to
explain the regularities that obtain above ground. This brings us to representation-
alism. The first version of this view maintains that qualitative states are identical with
representational states. Now a neural state can be nomologically related to a repre-
sentational state, but it cannot be identical with one, and it cannot bear any other
relation to a representational state that involves metaphysical necessity. This is because
neural states are individuated exclusively by their intrinsic properties. On the other
hand, representational states are individuated by their representational contents, which
means that they are individuated by relational properties. Hence, if qualitative states
are identical with representational states, they can only be tied to neural states by
nomological relations. No deeper relationship is possible. It follows that any correlation
laws linking qualitative states to neural states will be metaphysically fundamental, in
the sense that there will be no hope of deriving them from deeper facts that are
metaphysically necessary. Like property dualism, this version of representationalism
may recognize the existence of psychophysical correlation laws, but it will be unable to
explain them. As for the second version of representationalism, which maintains that
qualitative states are the phenomena that are represented by perceptual or quasi-
perceptual representations, it resembles functionalism in that it must deny that there
are universal correlation laws linking qualitative states to neural states. Consider pain,
for example. According to the second form of representationalism, pains are peripheral
disturbances that usually involve stress or damage. But brain states are in principle
dissociable from peripheral disturbances: a brain state that is normally caused by a
bodily phenomenon can also occur as the result of a neuroscientist’s probe, and a
bodily phenomenon can occur without sending its usual signal to the brain, as when a
spinal block has been administered. Thus, the states that representationalism identifies
with qualitative states are not perfectly correlated with brain states. Intuitively, though,
the psychophysical correlation laws hold even when the brain is decoupled from the
periphery. One can continue to feel pain in a limb when the limb has been amputated,
and one can honestly declare oneself free from pain even though one has just sustained
a serious injury. Or so it seems.

VI
There are a number of objections to type-materialism besides the multiple realization
objection. I will review two of the most important of these.
The first one derives from the fact that we do not normally distinguish between its
appearing to one that one is in a qualitative state and its being the case that one is in
that state. Thus, for example, when it seems to an agent that he is introspectively
aware of a pain, we conclude that he is aware of a pain. Because of this, it is natural to
suppose that that when an agent is introspectively aware of a pain, he must be aware
148 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

of its essential nature. After all, if there is no distinction between the appearance of
pain and the underlying reality, the agent is in direct epistemic contact with the pain
itself. Having made these initial points, the objection goes on to point out that one is
not aware of any of the structural characteristics that are definitive of neural states in
being introspectively aware of a pain: one could contemplate a pain for an eternity
without getting any inkling of the fact that it is accompanied by such things as the
firing of action potentials and the release of neurotransmitters. Now clearly, when we
combine the initial points with this additional one, we arrive at the conclusion that
the essential nature of pain does not include the structural features that are definitive
of neural states. But of course this implies that pain is not identical with a neural
state. (This line of thought is closely related to the so-called grain argument that was
originally devised by Wilfrid Sellars (Sellars 1962).)
The knowledge argument is another important objection to type-materialism. This
argument can be summarized as follows.
First premise: It is possible to have detailed theoretical knowledge of the physics and neurosci-
ence of color vision without thereby knowing anything at all about what it is like to experience
color—that is, about color qualia. A vision scientist who had been color blind from birth would
be in exactly this situation.
Second premise: If it is possible to know all of the physical facts about color vision without
knowing any of the qualitative facts, then the qualitative facts cannot be identical with physical
facts.
Conclusion: Facts involving color qualia are not identical with any facts involving physical
properties. (See Jackson 1982.)

The Cartesian modal argument is another important objection to type-materialism.

I will set it aside here because I have discussed it at length in other writings,
maintaining that it is vulnerable to multiple objections.
Many philosophers have been persuaded by these objections, but advocates of
type-materialism have developed replies that appear to have considerable dialectical
force. Thus, type-materialists have evolved a strategy for neutralizing the grain
argument. If we fail to apprehend the organizational complexity of neural states in
attending introspectively to pains, it is urged, this is because (a) introspection
involves conceptualization, and (b) the concepts we deploy in introspection fail to
register the complexity of these states, much as our commonsense concepts of plants
fail to register facts about their cellular organization or DNA. In other words, type-
materialists maintain that it is after all possible to draw an appearance/reality
distinction in connection with pains, though it is one that must be explained in
terms of the distinction between conceptualized reality and reality as it is in itself,
rather than, as is usual, in terms of the distinction between perceptually represented
reality and reality as it is in itself.
 THE IDENTITY THEORY 149

Type-materialists also appeal to conceptualization in responding to the knowledge

argument. More specifically, they appeal to the doctrine of conceptual dualism—the
doctrine that the concepts we use to represent pain and other qualia have special
features that distinguish them from the concepts that we use for physical properties.
Conceptual dualism is developed in different ways in Hill (1984), Hill (1991), Hill
(1997), Loar (1990), Papineau (2002), Balog (2009), and Balog (2012). On all of these
interpretations, it implies that the use of our qualitative concepts is epistemically
autonomous: it is not in any way responsible to the factors that govern the use of
physical concepts. Accordingly, on all of the interpretations, conceptual dualism
predicts and fully explains the fact that it seems possible to conceive of phenomena
in terms of physical concepts without deploying, or being in any way committed to
deploying, qualitative concepts. Now knowledge requires conceptualization. Because
of this, conceptual dualism is able to predict and fully explain the fact that it seems
possible to know all of the physical aspects of color vision without knowing any of the
qualitative aspects. Since conceptual dualism predicts and explains the fact in
question, and conceptual dualism is compatible with type-materialism, advocates of
type-materialism conclude that their view can accommodate the fact. Moreover, since
their view has the epistemological and metaphysical advantages we noticed earlier,
they conclude that it provides the best way of accommodating it. This conclusion calls
the knowledge argument into question. (Specifically, it undercuts the intuitive motiv-
ation for the second premise of the argument.)
Type-materialists tend to hold that it is possible to use similar lines of thought to
neutralize the remaining objections to their view. They maintain that conceptual
dualism can account fully for all of the phenomena involving qualia to which the
objections appeal. To account for these phenomena, they maintain, it is enough to
distinguish between neural properties as conceptualized in terms of physical concepts
and neural properties as conceptualized in terms of qualitative concepts. It is not
necessary to countenance a dualism of properties.

VII
It appears that type-materialism will be in the cards for some time to come. It is
extremely tempting to believe that neuroscience is in the business of discovering laws
that link qualitative states to brain states, and that these laws imply that the
qualitative realm is perfectly mirrored by certain patterns of activity in the cortical
realm. Advocates of the multiple realization argument reject this view; but we have
found reason to think that the argument is flawed. Of course, if the view is correct,
type-materialism is in a strong position, for it is uniquely well qualified to explain
correlations between qualitative states and brain states. Indeed, prima facie, at least,
alternative views seem incapable of explaining such correlations at all.
There are, however, several considerations that counsel caution. First, type-
materialism presupposes that qualitative states would be correlated with brain states
150 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

even if those states did not realize representational states. This presupposition can be
challenged. Our talk of qualia is essentially bound up with our talk of such things as
awareness, experience, and the ways in which external phenomena appear to obser-
vers. On a representationalist view of the mind, such talk can get a foothold only
when the relevant domain consists of representational states. To speak of awareness,
for example, is necessarily to speak implicitly of representation, for any state of
awareness is necessarily constituted by a representation of the item or items of which
the subject is aware. If this view is right, qualitative states could not possibly be
correlated with brain states in circumstances in which brain states fail to realize
representational states. Indeed, the whole framework of the mind would be missing
in such circumstances.
Second, although the strategy of using conceptual dualism to counter objections to
type-materialism is prima facie appealing, it remains to be seen whether it really
works. In applying the strategy, it is necessary to assume that our most fundamental
form of cognitive access to qualia involves conceptualization. Unfortunately, it is not
at all clear that this assumption is correct. After all, it is natural to suppose that
infants and animals can be aware of qualitative states like pains even though they are
not in a position to apply concepts to those states. Their awareness of qualitative
states seems to be entirely experiential and perceptual. Moreover, it seems that at the
most fundamental level, our awareness of qualitative states is closer to the paradigms
of perception than to conceptual states like belief and judgment. Thus, for example,
our awareness of pain is similar to perception in that it seems to represent its objects
as having spatial locations, and attention to pain seems to enhance awareness in the
same ways it enhances perceptual awareness. In view of these considerations (which,
I should add, can be bolstered by others—see the Introduction to this volume), it
seems that there is room for concern about conceptual dualism—and therefore about
type-materialism, since conceptual dualism is the best idea for defending type-
materialism that has thus far appeared.
Third, the possibility of adopting a perceptualist account of awareness of pain and
other qualitative states suggests that it is possible to draw a distinction between
appearance and reality with respect to such awareness. Thus, when awareness is
perceptual, it is possible to distinguish between an item’s seeming to be present,
because it is represented as present by a perceptual state, and the item’s actually being
present. Now if a distinction of this sort could be drawn with respect to awareness of
qualitative states, there would be grounds for doubting that such states are perfectly
correlated with brain states. To see this, consider a case of phantom limb pain—say, a
case in which a subject claims to feel a pain in his right arm despite the fact that the
arm has been amputated. There is no doubt that the subject seems to feel a pain, and
it is very plausible that this appearance of pain is due to activity in a certain region of
his brain. In general, it is very plausible that appearances of pain are correlated with
brain states. But does the subject really feel a pain in his right arm? It can be
maintained that pain is actually a disturbance of a certain sort in the body, and
 THE IDENTITY THEORY 151

that our subject is therefore only experiencing a hallucination of pain. I will not
attempt to justify this view here. (For arguments see the chapters “OW!” and
“Locating Qualia.”) But if it is true, then the correlation between qualitative states
and brain states is not perfect. The brain state that normally registers pain in the right
arm can occur without being accompanied by pain itself. This suffices to show that
the case for type-materialism is not airtight, for type-materialism stands or falls with
the psychophysical correlation theses.

References
Aizawa, K. and Gillett, C. (2009). “Levels, Individual Variation, and Massive Multiple Real-
ization in Neurobiology,” in J. Bickle (ed.), The Oxford Handbook of Philosophy and
Neuroscience. Oxford: Oxford University Press.
Balog, K. (2009). “Phenomenal Concepts.” In A. Beckermann, B. McLaughlin, and S. Walter
(eds.), The Oxford Handbook of Philosophy of Mind. Oxford: Oxford University Press, 292–312.
Balog, K. (2012). “Acquaintance and the Mind-Body Problem.” In S. Gozzano and C. S. Hill
(eds.), New Perspectives on Type-Identity: The Mental and the Physical. Cambridge: Cam-
bridge University Press, 16–42.
Bechtel, W. and Mundale, J. (1999). “Multiple Realizability Revisited: Linking Cognitive and
Neural States,” Philosophy of Science 66: 175–207.
Block, N. (1980). “Troubles with Functionalism.” In Block (ed.), Readings in the Philosophy of
Psychology, Vol. I. Cambridge, MA: Harvard University Press, 268–305.
Block, N. and Stalnaker, R. (1999). “Conceptual Analysis, Dualism, and the Explanatory Gap,”
The Philosophical Review 108: 1–46.
Feigl, H. (1958). “The ‘Mental’ and the ‘Physical’.” In H. Feigl, M. Scriven, and G. Maxwell
(eds.), Concepts, Theories and the Mind-Body Problem. Minneapolis, MN: University of
Minnesota Press, 370–497.
Gozzano, S. and Hill, C. S. (eds.) (2011). New Perspectives on Type-Identity: The Mental and the
Physical. Cambridge: Cambridge University Press.
Grahek, N. (2001). Feeling Pain and Being in Pain. Oldenburg: Bibliotheks- und Informa-
tionssystem der Universität Oldenburg.
Heider, F. and Simmel, M. A. (1943). “Heider-Simmel Demonstration,” <http://www.youtube.
com/watch?v=sZBKer6PMtM>.
Hill, C. S. (1984). “In Defense of Type Materialism,” Synthese LIX: 295–320.
Hill, C. S. (1991). Sensations: A Defense of Type Materialism. Cambridge: Cambridge Univer-
sity Press.
Hill, C. S. (1997). “Imaginability, Conceivability, Possibility, and the Mind-Body Problem,”
Philosophical Studies 87: 61–85.
Hill, C. S. (2009). “The Identity Theory.” In T. Bayne, A. Cleeremans, and P. Wilson (eds.), The
Oxford Companion to Consciousness Oxford: Oxford University Press, 359–63.
Jackson, F. (1982). “Epiphenomenal Qualia,” Philosophical Quarterly 32: 127–36.
Kahneman, D. (2011). Thinking Fast and Slow. New York, NY: Farrar, Straus, and Giroux.
Kim, J. (1998). Mind in a Physical World. Cambridge, MA: MIT Press.
Loar, B. (1990). “Phenomenal States,” Philosophical Perspectives 4: 81–108.
McLaughlin, B. (2004). “Color, Consciousness, and Color Consciousness.” In Q. Smith and
A. Jokic. (eds.), Consciousness. Oxford: Clarendon Press, 97–154.
152 PART II : A TYPE - MATERIALIST THEORY OF EXPERIENCE

McLaughlin, B. (2010). “Consciousness, Type Materialism, and Inference to the Best Explan-
ation,” Philosophical Issues 20: 265–304.
Papineau, D. (2002). Thinking about Consciousness. Oxford: Clarendon Press.
Place, U. T. (1956). “Is Consciousness a Brain Process?” British Journal of Psychology XLVII:
44–50.
Polger, T. (2004). Natural Minds. Cambridge, MA: MIT Press.
Putnam, H. (1967). “Psychological Predicates.” In W. H. Capitan and D. D. Merrill (eds.), Art,
Mind, and Religion. Pittsburgh: University of Pittsburgh Press, 37–48.
Sellars, W. (1962). “Philosophy and the Scientific Image of Man.” In R. Colodny (ed.), Frontiers
of Science and Philosophy. Pittsburgh, PA: University of Pittsburgh Press.
Smart, J. J. C. (1959). “Sensations and Brain Processes,” The Philosophical Review 68, 141–56.
PART III
A Representationalist
Theory of Experience
 9
OW! The Paradox of Pain*

I
It is generally possible to distinguish between the appearance of an empirical
phenomenon and the corresponding reality. Moreover, generally speaking, the
appearance of an empirical phenomenon is ontologically and nomologically inde-
pendent of the corresponding reality: it is possible for the phenomenon to exist
without its appearing to anyone that it exists, and it is possible for it to appear to exist
without its actually existing. It is remarkable, therefore, that our thought and talk
about bodily sensations presupposes that the appearance of a bodily sensation is
linked indissolubly to the sensation itself. This is true, in particular, of our thought
and talk about pain. Thus, we presuppose that the following principles are valid:

(A) If x is in pain, then it seems to x that x is in pain, in the sense that x has an
experiential ground for judging that x is in pain.
(B) If it seems to x that x is in pain, in the sense that x has an experiential ground
for judging that x is in pain, then x really is in pain.
There are alternative ways of expressing these principles. For example, (A) can be
expressed by saying that it is impossible for x to be in pain without x’s being
experientially aware that x is in pain, and (B) can be expressed by saying that it is
impossible for x to have an experience of the sort that x has when x is aware of a pain
without its being the case that x really is aware of a pain.
(A) appears to hold quite generally—even in cases that are somewhat outré. To
appreciate this, recall that soldiers and athletes often sustain serious injuries but show
no sign of being in pain, continuing to display normal behavior until the end of the
battle or the athletic contest, and perhaps even longer. They also deny that they are in
pain. We may feel confused in trying to describe situations of this sort, but we realize
that we have an obligation to concur with the person who has the injury. If he or she
denies that there is pain, and we have no reason to think that the denial is insincere,
we will acquiesce in it, and indeed insist on its truth. We feel that it would be deeply
absurd to override the testimony of the injured party.

* I have benefited considerably from discussions with Richard Fumerton, Jack Lyons, Thomas Senor,
Timothy Schroeder, and Lee Warren.
156 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

We can appreciate the plausibility of (B) by reflecting on cases of phantom limb

pain. When someone complains of pain in a limb that no longer exists, his or her
testimony is taken as fact, provided only that there is no reason to suspect insincerity.
To be sure, it will not seem correct, either to the person suffering from the pain or to
external observers, to say that the pain is in the part of the body where it appears to be
located, for by hypothesis there will be no such part. Because of this, it is true to say
that there is a certain discrepancy between appearance and reality in such cases. But
this is the only discrepancy. Thus, while we are prepared to say that the victim’s
perception that the pain is in the right leg is an illusion, we will allow, and in fact
insist, that the pain is in all other respects as it appears to the victim. In particular, we
will insist that it is a pain, thereby manifesting our commitment to principle (B).
It is important that we regard it as absurd to say that an agent is in pain in
circumstances in which the agent is not aware of a pain, and that we regard it as
absurd to say that an agent is not in pain in circumstances in which it seems to the
agent that he or she is in pain. This suggests that we think of (A) and (B) as necessary
truths—that is, as holding either because of deep metaphysical facts about pain and
our awareness of pain, or because of the a priori structure of our concept of pain.
There is one aspect of our thought and talk about pain that seems not to be fully in
keeping with (A) and (B). We all know that pain has causal powers with respect to
attention and awareness. When we are aware of a very intense pain, it seems to us
that it has the power to keep attention focused on itself—indeed, a power to so engage
attention as to make it difficult or impossible for us to honor other concerns and
interests. Equally, as a pain increases in intensity, it seems to us to increase propor-
tionally in its power to attract attention. Because of these phenomenological facts, we
have a tendency to suppose that pain can exert an influence on attention even prior to
our becoming aware of it. Thus, we say that pains can wake us up. On one very
natural construal of this claim, it presupposes a picture according to which pain can
exist prior to and therefore independently of awareness. According to this picture,
there are pains that are so “small,” so lacking in intensity, that they are incapable of
attracting attention, and exist only at a subliminal level. But a pain of this sort can
increase in intensity, thereby increasing also in its power to attract attention, and it
may finally attain a level of intensity that enables it to open the gates of awareness.
When this happens during sleep, we wake up.
Although this picture has a certain appeal, I think that its appeal is due to a
recessive strand in our commonsense conception of pain, a strand that pales in
comparison to the ones that find expression in (A) and (B). If we were fully
committed to the picture, we would be prepared to consider it epistemically possible
that an injured soldier actually has a severe pain, despite his professions to the
contrary, but that there is something wrong with the mechanisms in his brain that
support attention, and that this is preventing the pain from penetrating the threshold
of consciousness. When I have asked informants to assess the likelihood of this
scenario, however, they have all been inclined to dismiss it as absurd. The fact is that
 OW ! THE PARADOX OF PAIN 157

our discourse about pain is subject to divergent pressures. It serves a number of

purposes and is responsive to a complex phenomenology. As a result, it is not entirely
coherent. We resonate deeply to (A) and (B), but it should not surprise us that we are
occasionally led to say things, and to feel the appeal of pictures, that cannot be fully
squared with these principles.
At all events, I will go forward on the assumption that (A) and (B) strike us as
fundamentally correct, and in fact seem to us to enjoy a kind of necessity. Whether
the necessity in question has an ontological ground or is instead due to the a priori
structure of our concept of pain is a question that I will defer to Section III.

II
The fact that pains do not admit of a substantive appearance/reality distinction is
unfortunate, for it presents a serious challenge to an otherwise appealing account of
what it is to be aware of a pain. Indeed, we should view the absence of an appearance/
reality distinction here as extremely unfortunate, for the considerations that favor the
account in question actually seem to mandate it. Thus, we have an antinomy. On the
one hand, principles (A) and (B) appear not just to be true, but to be partially
constitutive of the concept of pain. And on the other hand, there are considerations
that seem to force us to embrace an account of awareness of pain that is flatly
incompatible with (A) and (B). This is the “paradox of pain” that is cited in the
subtitle of the present chapter.
The account of awareness of pain that I have in mind here is one that likens
awareness of pain to such familiar sorts of perceptual awareness as vision, hearing,
and touch.
When we consider the familiar forms of perceptual awareness from a phenom-
enological perspective, we find that they share a number of features, and this
impression is reinforced when we view them from the perspective of contemporary
cognitive psychology. I will mention several of these common features.1 It will be
clear as we proceed, I believe, that awareness of pain possesses the features as well.
First, perceptual awareness involves subconceptual representations of objects of
awareness. Reflection shows that we are able to perceive innumerable properties that
we are unable to name or describe. It is often said that we can use demonstrative
concepts, such as that shade of color, to specify these properties, but even if this
observation was fully correct, which is highly questionable, given that animals seem
to be capable of highly sophisticated perceptual representations while being very

1
Several of the seven features of perceptual awareness that I cite here are borrowed from Hill (2004).
Incidentally, the present paper is intended to extend the ideas and arguments presented in that earlier
effort. Hill (2004) makes a few observations about the main issue addressed in the present chapter, the lack
of an appearance/reality distinction for pain, but its treatment of that issue does not address its more
difficult aspects. Also, I was unable in that paper to discuss the concept of pain. I hope now to have filled
these lacunae in a reasonably satisfactory way.
158 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

limited conceptually, it would not diminish the case for the view that the properties
are represented subconceptually. This is because we must attend to a property
perceptually in order to demonstrate it. The representations that are employed by
perceptual attention when it provides grounding for demonstrative concepts cannot
themselves be conceptual.
Awareness of pain also involves subconceptual representations. To be sure, aware-
ness of pain often takes the form of a judgment to the effect that we are in pain, and
judgments put concepts into play. But what we are aware of in being aware of pains
can easily transcend the expressive powers of our conceptual repertoire. We have all
been aware of pains that have a significant degree of internal complexity. Perhaps it
would be possible to put together roughly adequate descriptions of such pains if they
were to last long enough, without changing in any way, or if we could remember their
particularities for a long enough time after they had disappeared. In actual fact,
however, the task of describing them fully is way beyond our powers. What we are
aware of is ineffable in practice if not in principle.
Second, all of the familiar forms of perception are associated with automatic
attention mechanisms, and also with attention mechanisms that are under voluntary
control. Among other things, such mechanisms can increase the resolution of our
experience of an object of awareness, and heighten the contrast between an object of
awareness and its background. The same is true of awareness of pain. We can attend
to pains, and when we do, there is a higher level of resolution and also a more salient
contrast between figure and ground.
Third, it is of the essence of perceptual representation to assign locations and other
spatial characteristics to its objects. The same is true of awareness of pain.
Fourth, there are a priori norms of good grouping that determine the ways in
which perceptual elements are organized into wholes. For example, we group visually
presented dots together if they are alike in some respect—that is, if they share a
neighborhood in space, or a shape, or a color, or a size, or a common fate. The same is
true of groups of pains. Suppose that you have three pains, two in your palm and one
on your wrist. The two in your palm will seem to form a unified whole of a certain
kind. Equally, two pains that are alike in intensity, or that begin to exist at the same
time, will seem to be members of a single “society,” even if they lie at some distance
from one another spatially.
Fifth, perceptual awareness always represents its objects as having highly deter-
minate forms of the properties that it attributes to them. I can believe that an object is
blue without believing that it is any determinate shade of blue. But I cannot visually
perceive an object to be blue without perceiving it to be navy blue or some other
highly determinate shade of blue. Equally, I cannot experience a pain as intense
without experiencing it as having a particular level of intensity.
Sixth, perceptual awareness is particularized. I can form a belief that there are three
books in a box without being en rapport with any of the particular books that make
my belief true. Equally, I can believe that someone has been eating my porridge
 OW ! THE PARADOX OF PAIN 159

without having any relevant beliefs about a specific individual. But perceptual
awareness is different. If I am perceptually aware of the presence of three books in
a box, I must be in some sense perceptually aware of each of the individual books.
Equally, if I am aware of the existence of a trio of pains in my arm, I must be aware of
each individual member of the trio.
Seventh, perceptual awareness has a certain mereological determinacy. I can form
a belief about an object without forming any belief about its parts. But unless an
object of perception is atomic, a minimum sensibile of some sort, I am inevitably
aware of a range of its parts in being perceptually aware of the object, and also of
certain of the structural relationships among the parts. (I mean to be using “part”
quite broadly here, so that it applies to temporal constituents of events and qualita-
tive constituents of complex properties as well as to spatial parts of physical sub-
stances.) This is also true of awareness of pains. Pains are normally experienced as
extended in space, and when they are so experienced, the parts are experienced
as well, as are a number of the structural relationships among the parts.
We have been reviewing certain of the common features of paradigmatic forms of
perception that are visible from the perspective of phenomenology and/or the
perspective of cognitive psychology. I have maintained that when we reflect, we
can see that these features are shared by experiential awareness of pain. I now wish to
remind the reader that there are other commonalities that are visible from the
perspective of cognitive neuroscience. More specifically, cognitive neuroscience
provides many data which suggest that awareness of pain is fundamentally akin to
haptic perception, thermal perception, and proprioception. It is widely held that the
latter forms of perception owe their character and indeed their very being to the
representational functions of certain structures in somatosensory cortex. Insofar as
we prescind from the emotional and behavioral phenomena that attend awareness of
pain, and think of awareness of pain as awareness of a purely sensory state, we find
that awareness of pain also owes its character and its being to representations in
somatosensory cortex. Moreover, in addition to this global commonality, there are
also many commonalities of detail—commonalities having to do with internal
organization, relationships to sensory pickup systems, and connections to higher
cognitive centers, such as those that are believed to be involved in attention. In view
of these similarities, it would be very uncomfortable to withhold the label “perceptual
system” from the structures that subserve awareness of pain while applying it to the
structures that subserve, say, haptic perception.2

2
The picture that I have sketched above is presented in much greater detail in Price (1999), chapter 5,
and in Basbaum and Jessell (2000). For discussion of the mechanisms that underlie other forms of bodily
awareness, see Gardner, Martin, and Jessell (2000), and Gardner and Kandel (2000).
Although the idea that activity in somatosensory cortex is primarily responsible for awareness of pain
appears to be widely favored by neuroscientists, there is also support for the view that activity in the insula
makes a significant contribution to such awareness. Even if this second view turns out to be correct,
however, it will still be true that the mechanisms that support awareness of pain are closely aligned with
160 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

As I see it, then, we are obliged to accept the view that awareness of pain is
constituted by representations that are fundamentally perceptual in character, and by
the same token, the view that awareness of pain is a form of perceptual awareness.
Alas, in accepting these views, we come into conflict with principles (A) and (B),
which deny that there is a substantive distinction between its seeming to one that one
is in pain and its being the case that one is in pain. The reason for this conflict is that
the relationship between perceptual representations and the items they represent is
always contingent. Suppose that a perceptual representation R represents a property P.
It is always possible for P to be instantiated without being represented by a corres-
ponding token of R, and for R to be tokened without there being a corresponding
instance of P. Because of this, if awareness of pain constitutively involves perceptual
representations, it must be possible, at least in principle, for there to be pains that one
is not aware of, and it must also be possible to be aware of pain, in the sense having an
experience just like the ones one has when one is aware of a pain, without one’s
actually being in pain. That is to say, there must be possible circumstances in which
one is in pain without seeming to be in pain, and possible circumstances in which it
seems to one that one is in pain without its actually being the case that one is in pain.
But (A) and (B) deny that such possibilities exist.
It appears, then, that when we try to combine the folk psychological picture of pain
and awareness of pain with various phenomenological and scientific facts, we arrive
at an antinomy. I will eventually propose a way of dealing with this conflict, but
before I can proceed to do so, it is necessary to take a closer look at its sources. This
effort will concern us in the next two sections.
It will be useful to adopt a couple of technical terms. In the future, when I wish to
speak of the somatosensory representations that appear to underlie sensory aware-
ness of pain, I will use the term “P-representation.” Further, I will use the term “P-
state” to stand for the bodily phenomena that P-representations represent. I will
eventually recommend a view about the commonsense concept of pain on which it is
true that “pain” and “P-state” are coreferential. But this view will also imply that
“pain” has a much more complex sense or meaning than “P-state.” “P-state” is meant
simply to be a name for bodily conditions involving actual or potential damage.
“Pain” is different. Its use is governed by (A) and (B), and perhaps by other a priori
presuppositions as well. It is not a mere name.
I will assume that P-representations represent P-states as having locations, inten-
sities, and various structural characteristics, such as that of having a central region
of high intensity and a surrounding border of low intensity. I will also assume that
P-representations assign P-states to such categories as burning, piercing, and throb-
bing. Further, I will assume that the contents of P-representations are determined by

perceptual mechanisms, for the considerations which count in favor of a role for the insula in awareness of
pain have counterparts which suggest that the insula plays a similar role in thermal perception and other
forms of bodily awareness. (For discussion see Craig (2002).)
 OW ! THE PARADOX OF PAIN 161

biology, and that the properties that serve as their contents are therefore properties to
which evolutionary pressures are sensitive. Thus, for example, while I wish to allow
for the possibility that a P-representation can represent a P-state as having the
property being located in the right hand, I wish to deny that a P-representation can
represent a P-state as having the property being located in a hand that was once
shaken by Bill Clinton. Equally, while I wish to allow that a P-representation can have
the content being of degree of intensity D, I wish to deny that a P-representation can
have the content being more intense than the pain I felt when I burned my hand
last week.

III
I will now argue that it is possible to explain our commitment to (A) and (B) without
making any deep metaphysical assumptions about the nature of pain or our aware-
ness of pain. Rather, (A) and (B) owe their plausibility entirely to the a priori
structure of our concept of pain. It follows from this view that there are no factual
obstacles to adopting a perceptual theory of awareness of pain. To create logical room
for such a theory, we need only make some adjustments in our conceptual scheme.
In developing these views, I will presuppose the familiar doctrine that concepts are
defined and/or individuated by core aspects of the roles that they play in cognition.
I will also borrow a term from Christopher Peacocke, who calls the principles that
describe such core aspects “possession conditions” (Peacocke 1992). Thus, where C is
a concept, the possession conditions for C are a set of propositions that describe core
aspects of the cognitive role of C. Each possession condition for C says, in effect, that
if one is to possess C, then one must possess a concept whose use has a certain core
aspect. Collectively, these individual possession conditions specify a sufficient con-
dition for an agent to possess C.
There are various ways of conceiving of possession conditions. On one of the
standard ways, they are concerned with dispositions to accept (and/or to reject)
propositions in various possible situations. It is sometimes supposed, for example,
that the possession conditions for a logical concept are concerned with dispositions
to infer propositions containing the concept from other propositions, and with
dispositions to infer other propositions from propositions containing the concept.
(Here of course I am conceiving of inference as involving acceptance.) Again, it is
sometimes supposed that the possession conditions for the concept of red are
concerned, among other things, with dispositions to infer propositions containing
the concept from perceptual experiences that represent redness as obtaining in the
world. A related view is that the possession conditions for mathematical concepts like
the concept of set and the concept of real number are concerned with dispositions to
accept the propositions that serve as axioms in the relevant mathematical theories.
The view that possession conditions are concerned with dispositions to accept
propositions has considerable intuitive appeal, but it is not inevitable. Many
162 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

philosophers prefer the view that possession conditions are concerned with norms of
acceptance. But those who hold this alternative view tend to acknowledge that
possession conditions are implicitly or indirectly concerned with dispositions, for
they generally maintain that an individual agent must have some tendency to
conform to the appropriate norms of acceptance in order to be credited with
possessing a concept. Moreover, those who prefer to think of possession conditions
as concerned primarily with dispositions generally allow that possession conditions
have a normative dimension. They recognize a normative dimension because they
believe that the dispositions that figure in possession conditions are held in place,
within the internal economies of individual agents, by various cognitive needs and
interests. They may also believe that the dispositions that govern concepts are
fostered and disciplined by social norms governing the use of words. After all,
words express concepts.
In the interests of definiteness, I will assume here that it is at least a rough
approximation to the truth to think of possession conditions as explicitly concerned
with dispositions to accept and reject propositions. I believe, however, that most of
my claims could be reformulated as claims about norms of acceptance.
Returning now at last to the concept of pain, I offer the suggestion that the
possession conditions for this concept include (1)–(4):
(1) In order to possess the concept of pain, one must be disposed to accept the
proposition that one is in pain at times when one has a P-representation.
(2) In order to possess the concept of pain, one must not be disposed to accept the
proposition that one is in pain unless one has a P-representation.
(3) In order to possess the concept of pain, one must not be disposed to reject the
proposition that one is in pain when one has a P-representation.
(4) In order to possess the concept of pain, one must be disposed to reject the
proposition that one is in pain at times when one does not have a
P-representation.
When I use the expression “has a P-representation,” I mean that there is currently an
active P-representation in the relevant agent’s somatosensory cortex, and that this
P-representation satisfies whatever further conditions, such as availability to other
cognitive modules, are necessary for it to constitute a conscious experience.
The foregoing possession conditions do not provide a full explanation of our
commitment to (A) and (B), but they go a long way toward providing such an
explanation. In particular, assuming that the condition of seeming to be in pain is
constituted by having an active P-representation in somatosensory cortex, they
explain why x will not accept a first-person ascription of pain unless it seems to x
that x is in pain, and why x will accept a first-person ascription in any circumstances
in which it seems to x that x is in pain. In general, (1)–(4) explain why first-person
judgments about pain tend to conform to (A) and (B). Now as I see it, first-
person judgments about pain are plausibly regarded as more fundamental than
 OW ! THE PARADOX OF PAIN 163

third-person judgments. (Thus, among other things, third-person judgments are

always trumped by first-person judgments in cases of conflict.) If this view is right,
it is reasonable to suppose that (1)–(4) answer the most fundamental questions about
conformity to (A) and (B). This is not to say that they answer all such questions. In
order to explain why third-person judgments about pain conform to (A) and (B) it is
necessary to supplement (1)–(4) with additional conditions. But discussion of the
additional conditions lies outside the scope of the present chapter.
According to the view I am recommending, the concept of pain has some
extremely unusual features. They are shared, I believe, by other concepts that stand
for bodily sensations, but most empirical concepts lack them. In the first place, (1)–
(4) have the effect of restricting acceptance of propositions of the form I am in pain
to circumstances in which one is having a perceptual experience of one specific sort—
specifically, an experience that involves a P-representation. The possession condi-
tions for empirical concepts normally envision a much broader range of acceptance-
promoting circumstances. For example, it is clear that the possession conditions for
the concept of red provide for acceptance of That object is red in circumstances in
which one’s experience does not attribute a definite color to the relevant object (say,
because it is night), but in which one has been told by a reliable associate that the
object is red, or in which one has inferred That object is red from other propositions,
such as the pair consisting of All of the objects on the table are red and That object is
on the table.
There is another peculiar feature that my proposal attributes to the concept of
pain. Possession conditions for perceptually grounded concepts generally make
provision for defeaters. Thus, for example, it is plausible that there is a possession
condition for the concept of red that can be formulated as follows:
In order to possess the concept of red, one must be disposed to accept That object is red in
circumstances in which (i) one is attending visually to an object o, (ii) one’s visual experience
represents o as red, and (iii) one believes no propositions that count as defeaters for That
object is red.

As is usual, I say that a proposition P is a defeater for a proposition Q, relative to a

believer x, if it is the case either (i) that x believes P and P is logically incompatible
with Q, or (ii) that x believes P and P implies that x’s evidence for Q is misleading in
the context at hand.3 Thus, where Q is the proposition That object is red, potential
defeaters for Q include That object is not red and That object is painted white but it
looks red because it is currently illuminated by red light. Each of these propositions
would become a defeater for Q, relative to x, if x were to come to believe it.
It is, I submit, extremely plausible that that there is a possession condition for the
concept of red that has a component corresponding to (iii). Otherwise it would be
appropriate for agents to accept That object is red even when they know that the

3
For further discussion of defeaters, see Pollock 1986.
164 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

relevant object looks red only because it is illuminated by red light. It is precisely
because it is not appropriate for agents to accept That object is red in such circum-
stances that the concept of red can be said to stand for the property of being red. The
property in question is after all an objective property, and therefore a property whose
comings and goings cannot be dictated by human perceptual experience. If the
possession conditions for the concept of red authorized acceptance of That object
is red in circumstances in which it is known that that property of being red is not
instantiated, then acceptance of the given proposition would not track the comings
and goings of the property, but rather merely the comings and goings of a certain
type of human perceptual experience.
Unlike the possession conditions for the concept of red, the possession conditions
that I am proposing for the concept of pain do not acknowledge the possibility of
defeaters. Thus, condition (1) tells us that if one possesses the concept of pain, then
the circumstance of there being an active P-representation in somatosensory cortex is
sufficient for being disposed to judge that one is in pain. It does not allow that one
might have a prior belief that could neutralize the epistemic force and the causal
efficacy of a P-representation, thereby modifying or suspending the disposition to
accept I am in pain. It gives perceptual experience full and final authority over belief.
Since (1)–(4) are unusual in the respects I have indicated, a defense of the view that
they succeed in capturing some of the possession conditions for the concept of pain
should include an account of why it might be useful for us to have a concept with the
features that they imply. What is it about P-representations and P-states that makes it
appropriate for us to adopt a concept whose possession conditions make no provi-
sion for first-person ascriptions that are not grounded in immediate experience, or
for the possibility that such ascriptions might be called into question by other
judgments?
I think that the answer has to do with the causal powers of P-representations.
P-representations constitute the condition of seeming to be in pain. Because of this
they have causal powers that are of great interest to us. They can cause all of the
emotions and attitudes that comprise the affective dimension of pain experience,
and, depending on such factors as persistence and intensity, they can induce us to
engage in all of the usual forms of pain behavior. Because these causal powers are of
deep and enduring interest to us, it makes sense for us to have a concept that can be
used to keep track of the states in which they inhere.
Now reflection shows that P-representations are not the only states that have the
powers in question. P-states possess them as well. It is clearly true that an active
bodily disturbance that constitutes actual or potential damage is capable of causing
fear, aversion, and all of the other emotions that constitute the affective dimension of
pain experience, and also that it is capable of causing such behaviors as nursing and
protecting an injured body part, at least if it is sufficiently intense. It is not true,
however, that P-states possess these causal powers autonomously. Rather, they
possess them only insofar as they are represented by P-representations. Disturbances
 OW ! THE PARADOX OF PAIN 165

that are not perceptually represented may occasion reflexive withdrawal, but they
have no tendency to trigger a conscious emotional response, and they are powerless
to produce intentional pain behavior. By the same token, it is a mistake to say that P-
states are proximal causes of emotional responses to pain and of intentional pain
behaviors. Rather, they are distal causes: they cause emotional responses and inten-
tional behaviors by causing P-representations. It is P-representations that count as
the proximal causes of the emotions and behaviors.
P-states possess the causal powers we have been considering, but they possess
them distally and at second hand, so they cannot be said to be the states in which the
powers most reliably inhere. It is P-representations that have that distinction.
Since the causal powers in question are of great interest to us, it makes sense for us
to have a concept that can be used to keep track of the states in which they most
reliably inhere. Hence, it makes sense for us to have a concept that can be used to
keep track of P-representations. What I wish to propose is that the concept of pain is
the concept that is used to play that role, and that it can play that role precisely
because its use is governed by (1)–(4). Those principles imply that our dispositions to
accept first-person ascriptions of pain are linked immediately, exclusively, and
indefeasibly to the presence of P-representations. Accordingly, insofar as we possess
the dispositions that they describe, we possess a concept that is tied to P-represen-
tations in the closest possible way.
It may be worth emphasizing, in this connection, that (1)–(4) easily accommodate
the fact that injured soldiers and athletes can deny that they are in pain without
qualification or acknowledgment that they might be apprised of considerations that
would lead them to change their minds, and that (1)–(4) also easily accommodate the
fact that victims of phantom limb pain can affirm that they are in pain with equal
confidence. Let x be an injured soldier or athlete of the sort in question. x is in a
P-state, but there is no accompanying P-representation. The P-state is not perceived.
Accordingly if x satisfies principle (3), x is disposed to deny categorically that x is in
pain. Moreover, this disposition will remain in force until such time as x’s P-state
comes to be represented by a P-representation. Suppose now that y is a victim of the
condition that we call phantom limb pain. y is not in a P-state, but y does have a P-
representation. Hence, if y satisfies principle (1), y is disposed to embrace I am in
pain unreservedly and without qualification.
I have explained why it is natural and appropriate for us to have a concept that is
governed by (1)–(4) in terms of our need to have a concept that can be used to keep
track of states that have deeply significant causal powers. Can the explanation be
generalized? Presumably our concepts of other bodily sensations, such as the concept
of itching, are associated with possessions conditions that are similar in structure and
content to (1)–(4). After all, we are not prepared to grant that there can be an itch in
one’s body without its seeming to one that there is an itch, or that it can seem to one
that there is an itch in one’s body without there actually being a sensation of that sort.
But is it true that when we experience bodily sensations other than pain, we are
166 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

generally in representational states that have significant causal powers? In most cases,
the answer is clearly “yes.” This is obviously true of the experience of itching, the
experience of bodily pleasure, gustatory experiences, olfactory experiences, and
experiences that inform us of higher levels of temperature and pressure. But reflec-
tion shows that it is also true, though true to a lesser degree, in other cases as well—
for example, in cases in which we experience very gentle sensations of warmth. Even
experiences of this sort have a tendency to trigger an emotional response, though of
course the response is generally quite mild.
In the present section I have proposed that the concept of pain is governed by
conditions (1)–(4). The proposal enjoys considerable plausibility, I feel, in virtue of
its ability to explain the most fundamental aspects of our commitment to (A) and (B).
Still, a defense of the proposal must do more than appeal to its fecundity, for its
claims about the possession conditions for the concept of pain represent those
conditions as fundamentally different in structure and content from the conditions
that govern most other perceptually grounded contents. It might be held that the
proposal is too outré to be acceptable. To counter this feeling, I have tried to show
that if our interest in the causal powers of P-representations is taken into account, it
is actually reasonable to predict that we have a concept that is governed by possession
conditions like (1)–(4).
Assuming that these efforts have been successful, we can conclude that the validity
of (A) and (B) is due to a priori features of our concept of pain. We do not need to
seek a metaphysical explanation. But this means that we are free to regard the form of
awareness that we call awareness of pain as a mode of perception. The problems that
arise when we try to combine this view with (A) and (B) can be solved by changing or
replacing the concept of pain.

IV
In the last section we were concerned exclusively with questions about the cognitive
role of the concept of pain. I would like to turn now to consider questions about its
semantic value or reference. What I believe, and what I will maintain here, is that the
concept of pain refers to P-states. In other words, as I see it, although it is built into
the cognitive role of the concept of pain that it is used to keep track of the comings
and goings of P-representations, it does not refer to P-representations, but rather to
the states that P-representations represent.
When it is combined with the foregoing claim about the cognitive role of the
concept of pain, this claim about the reference of the concept implies that the concept
suffers from a pervasive incoherence. To appreciate this, recall that there are occa-
sions when P-states occur without accompanying P-representations, and occasions
when P-representations occur without accompanying P-states. Accordingly, if it is
true that the concept of pain is governed by (1)–(4), and true also that it refers to P-
states, then there will be occasions when a proposition containing the concept is true
 OW ! THE PARADOX OF PAIN 167

but the possession conditions for the concept require that one have a inflexible
disposition to reject the proposition, and there will be occasions when the possession
conditions require one to be disposed to embrace a proposition containing the
concept irrevocably and without qualification even though the proposition is entirely
false. This amounts to a lack of cohesiveness between the possession conditions for
the concept and its semantic properties, or in other words, a lack of cohesiveness
between its cognitive role and its semantic role.
Of course, if it is true that the concept of pain is incoherent, then there is additional
motivation for the idea, floated in the last section, that the concept of pain should be
changed or replaced.
There are several reasons for thinking that the concept of pain refers to P-states.
One such reason comes to the fore when we reflect on what is involved in making
demonstrative reference to a particular pain. Suppose that you have been monitoring
the progress of two pains and have just noticed that one of them is increasing in
intensity. You may be inclined to register this development doxastically. If so, you
will probably focus your attention on the increasing pain and make a judgment with
the content That pain is getting worse. It is clear in this case that the demonstrative
concept that pain refers to the object of your attention. Now let us consider what is
going on here from the perspective of cognitive neuroscience. When we adopt this
perspective, I suggest, we see that you are perceptually attending to a P-state. (Or, in
other words, we see that you are deploying a perceptual representation with high
resolution and fairly narrow focus that represents a P-state.) Since you are attending
to a P-state, and the object you are attending to is the referent of that pain on the
occasion in question, the referent of the concept is a P-state. Further, what holds here
must hold generally. It must be true in general that demonstrative concepts involving
the concept of pain refer to P-states, for it is true in general that the reference of a
demonstrative concept is determined by the attention of the agent who is deploying
the concept, and, according to cognitive neuroscience, it is true in general that agents
attend to P-states when they attend to their pains.
Moreover, as I see it, anyway, this conclusion about demonstrative concepts
containing the concept of pain enjoins acceptance of the view that the concept of
pain refers to pains. How could that pain refer to a P-state if pain refers to something
else?
Other reasons for holding that the concept of pain refers to P-states emerge when
we reflectively consider the doctrine that is the only serious competitor of this view—
the doctrine that it refers to P-representations. I have been assured by a number of
philosophers that this doctrine is obviously right. As far as I can tell, this boils down
to the observation that the doctrine allows us to hold that the cognitive role of the
concept of pain coheres fully with its semantic properties, and to the claim that it is
desirable to embrace this optimistic perspective. I am in favor of optimism when it is
consistent with the facts, but in this case, I believe, it is not a live option.
168 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

Observe in the first place that the view is precluded by the semantic intentions
that one has when one attempts to describe a pain. In such cases, one attends to the
pain and tries to choose descriptive concepts that will do justice to its properties.
That is to say, one is concerned to describe the object to which one is attending.
But it is clear that one is attending to a P-state in such cases and not a
P-representation. Attention is something that we bestow on the objects that are
represented by perceptual representations, not on perceptual representations
themselves.
The view is also precluded by facts involving the locations of pains. To be sure,
P-representations have locations: they are located in somatosensory cortex. But of
course, pains can occur almost anywhere in the body but in somatosensory cortex. It
follows that if the concept of pain refers to P-representations, then it is not literally
true that pains have the locations that figure in our descriptions of them. Amazingly,
no one has ever made a true claim about the location of a pain! This seems
unacceptable.
It is sometimes maintained, however, that it is possible to bring the present
doctrine about the reference of the concept into harmony with our attributive
practices by embracing the idea that we use terms in different senses when we
mean to be characterizing pains than we do on other occasions. Thus, it is sometimes
urged, in effect, that when one says, for example, that a pain is located in one’s right
hand, what one is actually saying, without being aware of the fact, is that one has a
P-representation that represents a P-state as occurring in one’s right hand. More
generally speaking, the claim is that statements of the form “I have a pain in bodily
location L” have truth conditions that can be expressed by statements of the form “I
have a P-representation that represents its object as occurring in bodily location L,”
and that statements of the form “I have a pain of intensity level N” have truth
conditions that can be expressed by statements of the form “I have a P-representation
that represents its object as having intensity level N.” Does this suggestion work?
Does it protect the doctrine that the concept of pain refers to P-representations from
the objection we are considering?
No. The problems with the suggestion become evident when we consider argu-
ments like the following:

(I) I have a pain in my right hand.

My right hand is the only part of my body that has been touched by Bill
Clinton.
Therefore, I have a pain in the only part of my body that has been touched by
Bill Clinton.
 OW ! THE PARADOX OF PAIN 169

This argument is clearly valid; but if the theory of truth conditions for pain-state-
ments that we are considering was correct, the argument would be fallacious.
According to the theory, argument (I) is equivalent to argument (II):

(II) I have a P-representation that represents its object as occurring in my right hand.
My right hand is the only part of my body that has been touched by Bill Clinton.
Therefore, I have a P-representation that represents its object as occurring in
the only part of my body that has been touched by Bill Clinton.
It is easy to imagine a situation in which the premises of argument (II) are true and in
which the conclusion is false (especially since it appears to be the case that everyone
has shaken hands with Clinton at one time or another!). This is because the content
of P-representations is determined by biology. The selective pressures of evolution
are sensitive to properties like occurring in the agent’s right hand, but are altogether
incapable of bestowing contents like occurring in the only part of the agent’s body that
has been touched by Bill Clinton on perceptual representations.
It turns out, then, that the present proposal is incompatible with our intuitions
about the logical properties of sentences that ascribe locations to pains. But it also has
another flaw: it entails claims about linguistic processing that conflict with the
requirements of simplicity.
The present proposal should be understood as making two claims. First, it asserts
that all sentences of form ($) have truth conditions that are captured by sentences of
form (%):

($) I have a pain in bodily location L.

(%) I have a P-representation that represents its object as occurring in bodily
location L.
And second, it asserts that our language processors compute semantic representa-
tions of form (%) in interpreting sentences of form ($).
Now as this formulation makes clear, the present proposal is committed to the
proposition that our language processors somehow incorporate a complex meta-
physical picture of the nature of pain, and are guided by that picture in computing
deep structures that represent the truth conditions of such sentences. That is to say,
the proposal is committed to the following proposition: Under the influence of a
metaphysical picture, our language processors compute semantic representations for
sentences about pain that depart radically from the apparent logical structures of the
sentences themselves. It follows that the proposal is committed to a very large
amount of off-stage complexity—complexity of representations of truth conditions,
and complexity of the processes that compute such representations. Now when a
theory postulates hidden complexity, it has a special obligation to marshal evidence
from the relevant domain which shows that the postulate is well motivated. Here the
relevant domain is discourse about pain, so the evidence should consist of linguistic
data. But there is no linguistic evidence that supports the present proposal. On the
170 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

contrary, all of the motivation for it comes from a philosophical view about the
metaphysics of pain. We should conclude, then, that the proposal fails to honor the
requirements of simplicity.
As I have stated it here, this second objection is concerned only with the present
proposal. But reflection shows that it can be generalized. That is to say, if a proposal
resembles the one we are presently considering in that it construes sentences about
the locations of pains as equivalent to sentences about the representational contents
of P-representations, it will also resemble the present proposal in entailing extrava-
gant and poorly motivated claims about linguistic processing.4
It turns out, then, that the present proposal is wrong, and that the same is true of
all similar proposals. Now the proposal came to our attention because some such
claim is needed to shore up the view that the concept of pain refers to P-represen-
tations. Accordingly, the failure of the proposal means that the latter view should be
set aside. We are left, then, with the view that I wish to recommend—the view that
the concept of pain refers to P-states.
Let us review our findings. Assuming that the foregoing account of the possession
conditions for the concept of pain is correct, at least to a first approximation, the
cognitive role of the concept is closely bound up with P-representations. On the other
hand, we have now found reason to believe that the concept refers to P-states. Putting
these views together, we arrive at the conclusion that the state that determines the
cognitive role of the concept is altogether different from, and therefore metaphysic-
ally independent of, the state that serves as its referent.
But this is only part of the story. In addition to being metaphysically independent,
the two states are also nomologically independent. Thus, it sometimes happens that a

4
It may be helpful to illustrate this claim with an example. Consider the proposal that sentences of form (#):
(#) I have a pain in a bodily location that is F.
are equivalent to sentences of form (+):
(+) (ÓL)(L is a bodily location and I have a P-representation that represents its object as being in
L and L is F).
(Here “(ÓL)” is a substitutional quantifier. A substitutional quantifier is needed because “I have a
representation that represents its object as being in L” is an intensional context. (To be more accurate: it
is arguable that a substitutional quantifier is needed here.)) Applied to argument (I) in the text, this new
proposal yields the claim that (I) is equivalent to argument (III):
(III) (ÓL)(L is a bodily location and I have a P-representation that represents its object as being
in L and L is my right hand.)
My right hand is the only part of my body that has been touched by Bill Clinton.
Therefore, (ÓL)(L is a bodily location and I have a P-representation that represents its object as
occurring in L and L is the only part of my body that has been touched by Bill Clinton).
Unlike argument (II), argument (III) preserves the validity of argument (I). Thus, this new proposal
escapes my first objection to the proposal that is considered in the text. Reflection shows, however, that it is
vulnerable to a version of the second objection. It entails claims about linguistic processing that are poorly
motivated and at variance with standard principles of simplicity.
 OW ! THE PARADOX OF PAIN 171

P-state occurs when there is no corresponding P-representation, and it sometimes

happens that a state with the form of a P-representation occurs when there is no state
that answers to its content. Because these things can happen, the conditions under
which it is appropriate to be indefeasibly committed to a proposition containing the
concept of pain can diverge dramatically from the conditions under which the
proposition is true. In particular, it can be appropriate to for a wounded soldier to
deny intransigently that he is in pain even though the proposition that there is a very
intense pain in the soldier’s right leg is true. Equally, it can be fully appropriate for a
victim of phantom limb pain to commit herself fully and irrevocably to the propos-
ition that she is in intense pain even though the proposition is entirely false.
In view of these considerations, it seems appropriate to conclude that the cognitive
role of the concept of pain does not cohere fully with the semantic properties of the
concept. The cognitive dimension of the concept and the semantic dimension are
nomologically independent, and, moreover, there are occasions on which these
dimensions actually come apart.
Is it really credible that one of our most fundamental and frequently used concepts
is incoherent? In considering this question, please keep in mind the fact that none of
the foregoing remarks either imply or suggest that that the incoherence is readily
detectable. If anything, the lines of thought that we have followed suggest that the
incoherence should be virtually invisible from the perspective of common sense.
Thus, the argument for the existence of an antinomy in Section II is heavily indebted
to recent work in cognitive neuroscience, and the argument that has been developed
in the last two sections borrows heavily both from neuroscience and from contem-
porary philosophy. I would also remind the reader that there is good reason to
suspect other commonsense concepts of incoherence. The semantic paradoxes pro-
vide a strong rationale for the view that concepts like truth and reference are
governed by inconsistent principles, and it is arguable that Frege’s inconsistent
theory of sets is actually a successful explication of the commonsense concept of set.

V
Thus far I have attempted to do three things: to describe a paradox involving pain, to
explain the paradox by presenting a partial account of the possession conditions that
govern the concept of pain, and to maintain that the concept of pain is ultimately
incoherent. I would like now to recommend a way of dealing with the paradox. As
will be seen, the recommendation also addresses the incoherence.
According to the proposal that I have in mind, we should jettison the common-
sense concept of pain, in contexts in which our concerns are philosophical or
scientific, and replace it with two new concepts. One of the new concepts should
be a concept that refers to P-states, the other a concept that refers to P-representa-
tions. In each case, the concept should be endowed with possession conditions that
are consistent with its semantic properties. I will not attempt to set out appropriate
172 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

possession conditions here, for that would be a complex undertaking, and anyway, it
is generally possible in practice to acquire dispositions for using a concept that are
consistent with its intended semantic value by keeping that value vividly and fixedly
in view as one goes about the business of asserting and rejecting propositions that
contain it. In other words, in a case in which a reference has been stipulatively
assigned to a concept, it is generally possible to establish a practice of using the
concept by setting one’s sights on accepting true propositions containing the concept
and rejecting false ones. It seems to me that this is how theoretical concepts generally
acquire a pattern of use in scientific discourse.
At all events, even if I was in a position here and now to formulate two new sets of
possession conditions, my proposal would have no practical consequences. That is to
say, it is extremely unlikely that we would cease to use the concept of pain in everyday
life. The commonsense concept is deeply entrenched, and anyway, the paradox to
which it contributes is a theoretical problem—a problem that arises only when one
attempts to think systematically about pain, and to consider the picture of it that is
embedded in folk psychology in relation to the picture that is emerging in cognitive
science. The paradox of pain is not visible from the perspective of common sense.
Moreover, the incoherence of the concept does not inconvenience us in everyday life.
The point of the present proposal is not to reform our ordinary conceptual scheme,
but rather to point to a perspective from which the tasks of constructing an adequate
metaphysical account of pain and an adequate scientific account would be more
feasible.

VI
I have claimed that P-representations represent bodily disturbances that involve or
portend damage. There is a worry about this claim that may have occurred to the
reader. I will conclude by saying a few things about this concern.
When one considers the receptive fields of the nociceptive mechanisms in som-
atosensory cortex, and considers the way in which the mechanisms in question
respond to damage in those fields, it can seem that acceptance of my claim about
the representational contents of P-representations is obligatory. It should be acknow-
ledged, however, that there is other evidence that somewhat complicates the picture.
If a P-representation represents a disturbance in location L, then, one might suppose,
the intensity that the representation attributes to its object should correspond more
or less closely to the severity of the damage. But much of the time this is not the case.
Indeed, it seems that the intensity posited by a representation and the degree of
the represented damage are less likely to correspond than to diverge (Price 1999,
chapters 4 and 6).
The divergence occurs because signals indicating the amount of damage are very
often modified in the spinal cord and in the brain. Upward adjustments in signals can
be brought about by several factors. For example, it can happen that the signal is
 OW ! THE PARADOX OF PAIN 173

ratcheted up as a result of temporal summation in the spinal cord: the spinal cord
keeps track of how long a signal has been coming from location L, and after a
temporal threshold has been reached, it magnifies the signal before sending it on.
Downward adjustments can also be caused by a variety of factors. Speaking teleo-
logically, such adjustments probably occur in order to prevent the organism from
being overwhelmed by pain—that is, from being distracted by it in a way that would
preclude flight from danger or quiet recuperation.
In addition to the modulatory adjustments in the spinal cord and brain, there are
also adjustments at the sites where damage is located. Damage causes the release of
neuroactive chemicals at the site, and these chemicals can accumulate over time,
occasioning a commensurate increase in the rate of firing of peripheral nociceptors.
In short, there is a kind of peripheral temporal summation. And there is also
peripheral spatial summation.
In view of these considerations, if we embrace the view that P-representations
represent disturbances involving actual or potential damage, as I have recommended,
we will be obliged to say that P-representations systematically misrepresent the
disturbances with which they are concerned. In other words, we will have to say
that most of our sensory experience of pain is illusory. Or so it seems.
At first sight this conclusion can be unsettling, possibly prompting the thought
that it cannot be right after all to say that P-representations represent peripheral
bodily damage. It might occur to one, for example, that what they actually represent
is the level of nociceptive activity in the spinal cord.
On closer examination, however, it becomes clear that there are really no grounds
for worry. Thus, reflection shows that the foregoing line of thought has presupposi-
tions that may not be entirely correct. And, what is more important, it shows that
there is nothing really wrong with saying that a dimension of experience is system-
atically deceptive.
In formulating the foregoing line of thought, I have in effect presupposed the
following propositions: (i) there is a one-to-one correspondence between the levels of
intensity that are posited by P-representations and degrees of damage, and (ii) a P-
representation is accurate just in case the level of intensity that it posits corresponds
to the current degree of the actual or potential damage that the representation
represents. It is natural to assume that these presuppositions are correct, and
I believe that they are frequently taken to be correct in the literature. But (ii) might
be wrong. It could be that posited intensities are not meant to correspond to
concurrently existing degrees of damage, but rather to more complex magnitudes
that reflect a variety of factors. Thus, it might be, for example, that a given
P-representation was designed to be tokened either when there is concurrent damage
of level N, or there is damage of lesser level N* but the mechanisms responsible for
temporal summation have determined that the damage is persisting. If so, then the
representation would not simply signify the degree of the concurrently existing
damage, but would rather have the following disjunctive content: either there is
174 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

very recently incurred damage of level N or there is damage of degree N* that has
existed for the interval I. This suggestion can accommodate the existence of temporal
summation without implying that there is misrepresentation in such cases. And it
can easily be adapted so as to accommodate the existence of spatial summation. To be
sure, philosophers sometimes express reservations about views that assign disjunctive
contents to perceptual representations, but in this case, the motivation for the view
seems strong enough to outweigh those concerns.
Even if this suggestion for accommodating the effects of temporal and spatial
summation is right, it is probably true that our experience of pain misrepresents the
degree of damage in many cases. But we should not recoil from this observation.
Indeed, when we reflect, we come to appreciate that it would be surprising if the
representation of damage did not involve a substantial amount of misrepresentation,
for misrepresentation is actually quite common in perception, and in cognition
generally. There are many familiar perceptual illusions. A number of these occur
only in a highly restricted class of contexts, but others pervade a whole dimension of
perceptual experience. For example, Dennis Proffitt and his colleagues (Proffitt et al.
1995) have found that we tend to perceive hills as having steeper slants than they
actually possess, and that this effect is magnified when we are tired.5 They have also
determined that hills tend to look steeper when viewed from the top than when
viewed from the bottom. As Proffitt et al. point out, it is likely that we have evolved
with these tendencies because they are beneficial: the first one gives us a preference
for avoiding hills, which prevents unnecessary expenditures of energy, and the
second one gives us a preference for choosing slopes with lesser slant when we are
descending hills, which reduces the risk of accident. (I cite the beneficial character of
these illusions to highlight their similarity to illusions concerning the degree of bodily
damage. As we saw, the misrepresentation of damage can serve our need to focus our
attention elsewhere while escaping from danger, and our need to remain calm and
quiet while recuperating from damage.) There is also systematic misrepresentation in
other cognitive realms. This is shown, for example, by the work of Kahneman and
Tversky on assessments of frequency and probability (Kahneman and Tversky 1974),
and also by their work on assessments of utility (Kahneman and Tversky 1981).

VII
We started with the observation that there is no substantive distinction between the
appearance of pain and the corresponding reality. We then observed that there is
evidence from phenomenology, cognitive psychology, and cognitive neuroscience
which appears to mandate the view that awareness of pain is fully perceptual in
character. When we combined these observations we arrived at a paradox, for it is

5
I am indebted to Michael Pace for bringing Proffitt’s work to my attention.
 OW ! THE PARADOX OF PAIN 175

of the very essence of perception that there be a distinction between its seeming
to a perceiver that such and such is the case and its being true that such and such
is the case.
We then saw that it is possible to explain why there is no substantive appearance/
reality distinction in the case of pain in terms of features of the cognitive role of the
concept of pain. It follows that there are no empirical or metaphysical facts that
preclude a perceptual model of awareness of pain. It is possible to make the world
safe for a perceptual model by making appropriate revisions in our conceptual
scheme.
After considering these issues, we went on to notice that there is good reason to
believe that the concept of pain refers to bodily disturbances that involve actual or
potential damage. Evidently, then, it is true to say that pains are bodily disturbances.
But bodily disturbances are empirical phenomena, and therefore admit of an
appearance/reality distinction. Because of this, there is considerable tension between
two aspects of the concept of pain. Those features of its cognitive role that preclude a
substantive distinction between appearance and reality are at variance with its
referential dimension. This fact provides additional motivation for revising or elim-
inating the concept of pain.
Finally, we considered an argument which appears to show that our experience of
pains systematically misrepresents their quantitative character. After reflection, we
concluded that the argument has a questionable presupposition, and also that,
despite initial appearances to the contrary, the conclusion of the argument does
not pose a threat to the picture of pain and awareness of pain that is offered in the
present chapter.

References
Basbaum, A. J. and Jessell, T. M. (2000). “The Perception of Pain.” In E. Kandel, J. H. Schwartz,
and T. M. Jessell (eds.). Principles of Neural Science, 4th edition. New York: McGraw Hill,
472–91.
Craig, A. D. (2002). “How do You Feel? Interoception: The Sense of the Physiological
Condition of the Body,” Nature Reviews 3: 655–66.
Gardner, E. P. and Kandel, E. R. (2000). “Touch.” In E. Kandel, J. H. Schwartz, and T. M. Jessell
(eds.). Principles of Neural Science, 4th edition. New York: McGraw Hill, 451–71.
Gardner, E. P., Martin, J. H., and Jessell, T. M. (2000). “The Bodily Senses.” In E. Kandel,
J. H. Schwartz, and T. M. Jessell (eds.). Principles of Neural Science, 4th edition. New York:
McGraw Hill, 430–50.
Hill, C. S. (2004). “Ouch! An Essay on Pain.” In Rocco J. Gennaro (ed.). Higher Order Theories
of Consciousness. Amsterdam: John Benjamins, 339–62.
Kahneman, D. and Tversky, A. (eds.). (1974). “Judgments under Uncertainty: Heuristics and
Biases,” Science 185: 1124–31.
176 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

Kahneman, D. and Tversky, A. (1981). “The Framing of Decisions and the Psychology of
Choice,” Science 211: 453–8.
Peacocke, C. (1992). A Study of Concepts. Cambridge, MA: MIT Press.
Pollock, J. (1986). Contemporary Theories of Knowledge. Totowa: Rowman and Littlefield.
Price, D. D. (1999). Psychological Mechanisms of Pain and Analgesia. Seattle: IASP Press.
Proffitt, D. R., Bhalla, M., Gossweiler, R., and Midgett, J. (1995). “Perceiving Geographic Slant,”
Psychonomic Bulletin and Review 2: 409–28.

Postscript to “OW!”
As a result of recent work by Frank Durgin and my student Chaz Firestone (Durgin
et al. 2009, Firestone 2011), I would no longer wish to rely as heavily on the research
of Dennis Proffitt as I do in “OW!” and other writings; but it does appear that
Proffitt’s basic finding about geographic slant holds up under scrutiny. Thus, while
Durgin has failed to replicate some of Proffitt’s results, his research strongly confirms
Proffitt’s finding that we tend to overstimate slant (Li and Durgin 2010). Accord-
ingly, I still feel that that perception of slant is a good example of systematic
perceptual error. There are a number of other examples. Thus, e.g., Durgin has
shown that perceived rate of optic flow is also subject to systematic distortions.
(Durgin 2009)
Durgin, F. H. (2009). “When Walking Makes Perception Better.” Current Directions in
Psychological Science 18: 43–7.
Durgin, F. H., Baird, J. A., Greenburg, M., Russell, R., Shaughnessy, K., and Waymouth, S.
(2009a). “Who is Being Deceived? The experimental demands of wearing a backpack,”
Psychonomic Bulletin & Review 16: 964–9.
Firestone, C. (2011). “Against Visual Paternalism.” Brown University: Unpublished MA thesis.
Li, Z. and Durgin, F. H. (2010). “Perceived Slant of Binocularly Viewed Large-scale Surfaces:
A common model from explicit and implicit measures,” Journal of Vision 10: 1–16.
 10
Locating Qualia: Do They Reside
in the Brain or in the Body
and the World?

I. Introduction
I will be concerned to compare and evaluate two theories of qualitative states—that
is, states that have a proprietary phenomenological dimension. These theories pro-
pose quite different accounts of the metaphysical nature of qualitative states, and they
also offer quite different answers to the question of what is involved in being aware of
such states. In my view, both theories have great charm and are of great philosophical
interest. Indeed, I find their appeal so powerful that I have at different times
embraced each of them, accepting one for 17 years and the other for 10.
The first theory is sometimes known as type-materialism and sometimes as the
central state identity theory. It consists of three main claims: first, qualitative states
are identical with certain physical states; second, the physical states in question are
brain states—more specifically, states of the cerebral cortex; and third, we are aware
of qualitative states in virtue of judgments that subsume them under certain special
concepts that are generally known as phenomenal concepts. The notion of a phe-
nomenal concept is somewhat technical in character. It is meant to pick out concepts
that are applied to qualitative states directly, where this means that our use of them is
not guided by modes of presentation of any sort, and is not supported by any sort of
reasoning. This characterization is largely negative, but advocates of phenomenal
concepts have also offered positive accounts of them. (i) One view is that the use of
phenomenal concepts is governed by simple classificatory mechanisms that operate
directly on qualitative states, taking them as inputs and yielding judgments involving
the given concepts as outputs. (ii) It is also sometimes said that phenomenal concepts
have an indexical character (Loar 1990). On this view, a singular phenomenal
concept would perhaps have the form this sensation, and a general phenomenal
concept would perhaps have the form this type of sensation. (iii) Another account
maintains that phenomenal concepts literally have qualitative states as constituents,
and are therefore similar to terms that enclose the words to which they refer in
quotation marks (Papineau 2002). On this third view, a singular phenomenal concept
178 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

refers to the particular qualitative state that occurs in it as a constituent, and a general
phenomenal concept refers to the type of state that is exemplified by its constituent.
These views are of course quite different, but they are alike in implying that
phenomenal concepts are logically and materially independent of all physical con-
cepts. This enables them to explain how it is possible for us to think about qualitative
states without explicitly attributing any physical characteristics to them. To give them
a common name, the views are versions of conceptual dualism. Contemporary
advocates of the central state theory include conceptual dualism in formulations of
their position because it provides them with a tool for explaining away the intuitions
that favor the various forms of ontological dualism.
Early versions of the central state identity theory appeared in Place (1956) and
Smart (1959). The version I have just described has been advocated by a number
of authors, including Hill (1984, 1991, 1997); Loar (1990); Papineau (2002); and
McLaughlin (2003).
The second theory is representationalism. There are a number of versions of
representationalism in the literature, but I will focus here on a version that consists
of the following four doctrines. First, at the most fundamental level, awareness of
qualitative states does not involve or in any way depend on concepts, but is rather
purely experiential in character. In some cases (e.g., awareness of visual qualia) it is a
form of perceptual awareness, and in others (e.g., awareness of pain) it is a form of
quasi-perceptual awareness, in the sense that it has many or most of the characteristic
features of perceptual awareness. Second, like all awareness, awareness of qualitative
states is supported and indeed partially constituted by representations. In order to
become aware of a qualitative state, we must form a perceptual or quasi-perceptual
representation of it. Third, qualia are properties of external objects or of internal
states of the body.1 When we are aware of perceptual qualia we are aware of certain
relational, viewpoint-dependent properties of the objects that we see, hear, smell,
taste, or touch, such as the property of looking small and the property of sounding far
away. And when we are aware of the qualia of bodily sensations, we are generally
aware of properties of states and events that are located in extra-cranial regions of the
body. Thus, for example, when an agent is aware of a pain in her right foot, she is
aware of a disturbance in her right foot that involves actual or potential damage.
Fourth, qualitative properties are in some sense reducible to physical properties, but
it remains to be seen what form this reduction will take. As we will see a bit later on, if
the present representationalist picture is correct, the enterprise of specifying the exact
nature of qualitative properties depends to some extent on empirical questions that
have yet to be resolved. Without knowing exactly which properties count as quali-
tative, it is impossible to be sure whether a reducibility thesis involving qualitative

1
I will refer to the latter as peripheral states to emphasize their independence of the brain. Of course
there are plenty of pains that occur in the interior of the body.
 LOCATING QUALIA 179

properties should take the form of an identity thesis, or should instead claim that
qualitative properties are in some sense constituted by more basic properties.
The version of representationalism that consists of these four doctrines is pre-
sented and defended at some length in Hill (2009).2 Other versions are proposed in
Harman (1990), Dretske (1995), Lycan (1996), and Tye (2000).
When we compare these two theories we find that they have very different
advantages. The most compelling virtue of the central state theory is that it can do
justice to the fact that qualitative phenomena appear to be perfectly correlated with
cortical phenomena.3 Indeed, it accounts for this correlation in the simplest and most
straightforward of all possible ways. They co-occur because they are identical. The
correlations between qualitative phenomena and facts involving viewpoint-
dependent properties of external objects are much less robust, as are the correlations
between qualitative phenomena and facts involving intrinsic properties of bodily
states. Thus, as we will see, while there is arguably a perfect correlation between
experiences of pain and our brains being in certain states, the correlation between
experiences of pain and our bodies suffering actual or potential damage is less
impressive. Accordingly, if one wishes to construe qualia as viewpoint-dependent
properties of external objects and intrinsic properties of bodily states, one will have to
explain away a number of recalcitrant data.
On the other hand, the main virtue of representationalism is its account of
qualitative awareness. Reflection shows that that account is far superior to the one
that is built into the central state theory. Intuitively, awareness of qualia does not
require a complex conceptual apparatus. Can a vervet monkey construct judgments
about pains? More specifically, can a vervet construct judgments of the form This
sensation is a pain, or of the form This sensation is a sensation of this type? Can it
form complex concepts that are analogous to quotational singular terms? We are not
inclined to think so, but we feel pretty confident that vervets can be aware of pains.
Moreover, it seems wrong on introspective grounds to say that conceptualized
awareness provides our only form of access to qualia. Being experientially aware

2
The theory presented in Hill (2009) has two components that I have not presented here. One is a
distinction between the qualia that an experience represents and how it seems to one to have the experience;
and the other is the claim that how it seems to one to have an experience is determined by two factors—it is
determined in part by the representational content of the relevant representation, and in part by the
representation’s intrinsic and functional properties. Thus, for example, when one reflects on experiences as
of yellow it seems to one that when one has those experiences one is aware of a property that is perfectly
simple. But the experiences do not represent the property as simple. Rather, the property seems simple to
one because the representation that constitutes one’s awareness of it is a representational atom—it lacks a
constituent structure, and it therefore fails to articulate the internal complexity of the property it
represents, and also fails to support inferences to other mental states that would do justice to that
complexity. I think there are a number of aspects of how experiences seem to us that must be explicated
in this way in terms of “syntactic” and/or functional properties of representations.
3
I prescind here from the considerations which suggest that qualitative states are correlated with
different brain states in different species. I will just say that I do not find those considerations persuasive.
For discussion see Hill (1991, chapter 9), Hill (2009, pp. 30–5), and chapter 8 of the present volume.
180 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

that a lemon looks yellow is one thing, judging that it looks yellow is another. Equally,
there is a clear difference between being experientially aware of a pain and judging
that a pain is occurring. In general, introspection indicates that experiential aware-
ness is distinct from conceptualized awareness. Further, it seems that experiential
awareness is more fundamental: it plays a role in causing conceptualized awareness,
and it provides the content for certain forms of conceptualized awareness, as when
one fixes the reference of a demonstrative of the form that pain by focusing one’s
attention on a specific pain.4
I will focus on these apparent virtues of the two theories in the present chapter,
asking with respect to each whether it in fact confers a decisive advantage on the
theory that possesses it. I will discuss the first virtue in Sections II and III, and
the second in Section IV. Unfortunately, space limitations preclude my discussing the
virtues in a fully general way. Indeed, I will have to consider them in relation to a
single qualitative state. I think that pain is the state that is best qualified to serve as the
example here. It is, after all, the paradigmatic qualitative state.5 At all events, I will
focus on pain in the present chapter. I hope to consider additional examples on
future occasions.

II. Correlations
As we recently observed, pain is strongly correlated with activity in various regions of
the brain. In fact, generalizing from what is known, it seems likely that the correlation
between pain and activity in certain regions is perfect. Moreover, it is known that the
perceived intensities of pains are strongly correlated with levels or degrees of central
activity. This is shown, for example, by a well-known fMRI study conducted by Porro
et al. (1998). They found that perceived intensities correspond to levels of activity in
the primary somatosensory cortex, and also to levels in the anterior cingulate, motor,
and premotor areas. Similar results have been obtained by other investigators,
including Coghill et al. (1999) and Timmermann et al. (2001).
We find a quite different situation when we turn to consider correlations between
pains and peripheral disturbances. The correlations that exist are fragile, and fail to
hold over fairly broad areas. Moreover, even when pains co-occur with peripheral
disturbances, their perceived intensities fail to correspond to the intensities of
the disturbances with which they correspond. Or so it can seem. I will review the

4
I assume here that the relevant form of attention is a mode of experiential awareness. For discussion
see Hill (2009, esp. pp. 53–6, p. 93, and pp. 172–3).
5
I will be concerned here only with the sensory character of pain, not with its affective character. That is
to say, I will be prescinding from the badness or repugnance of pain, its ability to bother or distress an
agent. Please keep this restriction particularly in mind in assessing my claims about the neural seat of
awareness of pain. It is plausible that awareness of the sensory character of pain is supported principally by
mechanisms in the somatosensory cortex, but as a result of the work of Pierre Rainville and his colleagues
(Rainville et al. 1997, Hofbauer et al. 2001), we know that other areas of the brain are involved in awareness
of its affective character.
 LOCATING QUALIA 181

considerations that support this negative assessment in the present section, and will
then turn to consider how an advocate of the peripheral state theory might respond
to them in the next section.
First, as is well known, pains sometimes occur without being accompanied by
actual or potential damage. This is true of phantom limb pains and instances of the
thermal grill illusion (in which a combination of warm and cold stimuli give rise to
an experience of intense pain). It is often said to be true of ordinary headaches and of
lower back pains as well. Apparently investigators have thus far failed to find a form
of damage that might cause ordinary headaches, and in roughly 70 percent of the
cases, lower back pains are not accompanied by damage to the lower back (Melzack
and Wall 2008, p. 9).6
Second, conversely, damage sometimes occurs without being accompanied by
pain. This has happened in many recorded cases of battlefield injury. It also happens
with some frequency in athletic contests.
Third, there are a number of considerations indicating that the perceived intensities
of pains do not line up very well either with levels of noxious peripheral stimulation or
with levels of peripheral damage. I will mention six considerations of this sort.
(i) Temporal summation: when a noxious stimulus is repeated at short intervals,
the response to the stimulus in the spinal cord increases in magnitude, in a way that
reflects both the frequency of the stimulus and the length of the period during which
the successive applications have occurred. Some of the increase is occasioned by a
heightened peripheral response, but much of it is due to mechanisms in the spinal
cord itself. As a result, the pain accompanying each successive application of the
stimulus is perceived as more intense than its predecessor (Price 1999, p. 90).
(ii) Spatial summation: there is also considerable spatial summation of noxious
stimulation, much of it due to mechanisms in the spinal cord, and this contributes to
the perceived intensities of pain. Roughly, the larger the area to which a noxious
stimulus is applied, the more intense the pain that the stimulus occasions (Price 1999,
pp. 33, 89).
(iii) The modulatory influence of Αβ fibers: there are two kinds of fibers that carry
information about actual and potential damage from the periphery to the spinal
cord—fairly large myelinated Aä fibers that conduct impulses rapidly, and compara-
tively small unmyelinated C fibers that operate more slowly. In addition, surface
tissue is innervated by myelinated Αβ fibers, which are even larger than Αä fibers and
conduct impulses even more rapidly. These Aβ fibers are mechanoreceptors that are
activated by pressures falling within a broad spectrum, ranging from the very light

6
One possibility is that ordinary headaches are due to imbalances in neurotransmitters, and more
proximally to reactions to such imbalances by nociceptive nerves. If this should turn out to be correct,
headaches would correspond to a type of bodily disorder, though not perhaps to actual or potential damage.
Equally, it might turn out that lower back pains are generally due to disorders in processing centers in the
spinal cord.
182 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

pressures that come from gentle caresses to the massive pressures associated with
piercing cuts or bruising blows. Generally, when Aä fibers and C fibers are activated
by harmful pressures, the Aβ fibers are activated too. Intriguingly, however, when
signals carried by Aβ fibers arrive at the spinal cord, and they are combined with
signals arriving via the Aä fibers and the C fibers, they have the effect of reducing the
aggregated nociceptive signal. That is to say, the activity of the Aβ fibers actively
inhibits the neurons in the spinal cord that respond to signals from the Aä fibers and
the C fibers, decreasing the rate at which they fire. It follows that Aβ fibers have a
damping effect upon the perceived intensities of pains, in effect preventing the
intensities from doing full justice to the information that the damage-detecting
neurons convey (Basbaum and Jessell 2000, p. 475; Melzack and Wall 2008,
pp. 86–8; Price 1999, p. 90).
It is important that Aβ fibers provide us with a way of obtaining a certain amount
of relief from pain—of reducing it to manageable levels for short periods of time. We
are all familiar with the fact that the severity of a pain can be reduced by gently
stroking the area that surrounds a damaged area, and that it can be reduced even
more by shaking the relevant body part. These measures have the effect they do
because they activate Aβ fibers, thereby inhibiting the dorsal horn neurons that
register and aggregate information arriving from the Aä fibers and C fibers. It is
not known for sure that it is one of the functions of Αβ fibers to afford relief from
pain, but it is reasonable to conjecture that this is the case.
(iv) Long term sensitization: in cases of severe injury or repetitive noxious stimu-
lation of considerable duration, the impulses carried by C fibers can cause “wind up”
or long term sensitization, a condition that involves enduring changes in the struc-
ture and biochemical properties of neurons of the dorsal horn of the spinal cord.
Those afflicted by this condition often suffer from spontaneous pain, and generally
experience unusual or excessive levels of pain in response to mild levels of noxious
stimulation. Accordingly, perceived intensities are out of step with peripheral activity
(Basbaum and Jessell 2000, p. 479).
(v) The centrifugal influence of high-level psychological states: the perceived
intensities of pains are influenced significantly by various high-level psychological
factors, including attention (Bushnell et al. 2004), emotion (Rainville 2004), expect-
ation (Rainville 2004, Vase et al. 2004), learned associations, and stress (Rhudy and
Meagher 2000). To a large extent these factors operate centrifugally, by sending
facilitatory or inhibitory signals down pathways leading from the brain to various
segments of the dorsal horn of the spinal cord. These descending signals combine
with messages from the periphery to determine the strength and distribution of the
signals that will then ascend via afferent pathways in the spinal cord to the brain,
eventually producing an experience of pain. When one has an experience with an
etiology of this sort, the perceived intensity of the pain is of course influenced by the
relevant psychological factors, and is to that extent at variance with conditions on the
surface of the organism.
 LOCATING QUALIA 183

It should be noted that attention can influence perceived intensities in two quite
different ways. On the scenario I have just been describing, attention is captured by
incoming nociceptive signals. When this happens, the brain sends facilitatory signals
down the spinal cord, thereby augmenting the next wave of nociceptive signals.
This is in effect a processing loop that is initiated by exogenous factors. But there
is reason to believe that attention can also act directly on the perceptual mechanisms
in primary somatosensory cortex that are responsible for perceptual experiences
of pain, increasing the level to which these mechanisms are active, and thereby
ratcheting up the perceived intensity (Bushnell et al. 2004, p. 108). This might
happen, for example, if one should become interested in keeping track of the
frequency with which a noxious stimulus is applied, or decide to monitor the
qualitative nature of one’s pain, with a view to classifying it in terms of categories
like throbbing, shooting, and stabbing. In cases of this sort, attention is of course
controlled endogenously.
(vi) Response expansion: in a series of classic papers (e.g., Stevens 1962),
S. S. Stevens argued that the perceived intensities of pains are related to the intensities
of peripheral stimuli by power functions. Stevens further maintained that the expo-
nents of these functions are in some cases significantly greater than 1, and that this is
true, specifically, in the case of the perceived intensities of pains that are produced by
electric shocks. Thus, according to Stevens, the exponent of the function describing
the relationship between perceived intensities of pains and the corresponding electric
shocks is 3.5. Now there is room for doubt as to whether this exponent provides the
optimal interpretation of the relevant data. Moreover, there is reason to think that we
need much smaller exponents, and in fact exponents that are quite close to 1, to
describe the relationship between the perceived intensities of other forms of pain
(e.g., pains caused by heat) and the intensities of the corresponding causes (Adair et
al. 1967, Craig et al. 2001). It must be acknowledged, however, at least as a possibility,
that the perceived intensities of pain of certain types are not proportional to the
intensities of the corresponding stimuli. In a phrase, it may be that the perceived
intensities of certain pains are characterized by response expansion. If so, they pose an
additional challenge to the view that pains are peripheral states. (For an exceptionally
clear and detailed elaboration of this argument, see Pautz 2010.)

III. Six strategies for defending

the peripheral state theory
At first sight, anyway, these considerations amount to an impressive argument
against representationalism, which claims that pain is a peripheral state involving
actual or potential damage. But on reflection they seem less decisive. I will now
describe several defensive strategies that are available to advocates of what I shall call
184 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

the peripheral state theory of pain.7 I will not attempt to show that these strategies are
ultimately successful. At the present stage of knowledge, I don’t think it is possible to
arrive at a final assessment of the foregoing argument. Rather, I will just attempt
to show that the argument leaves the peripheralist plenty of logical room in which to
operate.
First, the peripheralist can account for phantom limb pains by saying that they are
hallucinations and thermal grill pains by saying that they are illusions. We already
know enough about the underlying mechanisms in these cases to be able to say that
the experiences in question arise from deviant or impaired somatosensory processing
(Craig and Bushnell 1994, Ramachandran and Blakeslee 1998, chapter 3). Going on
the etiologies of those experiences, we could predict a disruption of the reliable
connection between experience and damage that obtains in normal cases. Accord-
ingly, there is neuroscientific motivation for describing experiences of these two
kinds as hallucinations and illusions. Of course, in claiming that an experience of
pain is hallucinatory or illusory, one comes into conflict with folk psychology, which
in effect denies that it is possible to draw a distinction between appearance and reality
in the case of experiences of pain. As I have maintained elsewhere, however, it is
possible to make a strong case for setting the relevant portion of folk psychology
aside (Hill 2009, chapter 6, and chapter 9 of the present volume). The folk psycho-
logical account of pain is a theory, and like any theory, it must bow to alternative
theories that are superior to it in point of predictive and explanatory power. If we can
tell a story according to which certain experiences of pain are hallucinatory or
illusory, and that story has substantial predictive and explanatory power, we are
entitled to reject any folk psychological principles that conflict with it.
It may be possible to extend this way of treating phantom limb pains and thermal
grill pains to ordinary headaches and lower back pains. After all, given that pains of
these sorts are departures from the norm, it seems likely that it will someday be
possible to claim that they arise from deviant causes—that is, from causes that
disrupt normal processing and therefore make it likely that false positives will
occur (Melzack and Wall 2008, p. 188). There is a common practice of describing
false positives as hallucinations or illusions.
The second peripheralist strategy consists in making a relatively minor change
in the peripheralist theory of pain, replacing the claim that pains are bodily condi-
tions involving actual or potential damage with the claim that they are bodily
conditions involving actual damage, potential damage, or certain sorts of disorder.
This strategy would be useful if it should turn out that it is a mistake to try to explain
headaches and lower back pains in the way suggested in the preceding paragraph. On
that explanation, when we experience a headache or a lower back pain, it is because
deviant causal chains lead us to misrepresent relatively benign occurrences as cases of

7
See note 1.
 LOCATING QUALIA 185

actual or potential damage. This view is perhaps our best option at present, but it
could still turn out that the best way of accounting for headaches and lower back
pains is to classify them as pains of a third type (that is, as pains that are neither cases
of actual damage nor potential damage).8 If this should prove to be the case,
peripheralists could easily adapt by committing to the second strategy, which consists
in recognizing certain disorders as a third type of pain.
Third, with regard to cases in which damage occurs without being accompanied by
an experience of pain, it is open to the peripheralist to say simply that we are not
omniscient about our pains. Why should it be thought that every pain must be
registered by the relevant perceptual system? After all, no other perceptual system
provides us with exhaustive knowledge of its proprietary domain. To be sure, taking
this line brings one into conflict with folk psychology, which implies that all pains are
experienced. But this is just one of a number of areas in which folk psychology comes
into conflict with perceptualist models of awareness of pain. As we noticed a moment
ago, it is reasonable to reject folk psychological principles when they conflict with
theories that have greater predictive and explanatory power; and it can be shown
that peripheralist models succeed in predicting and explaining a broad range of
empirical facts.
Fourth, in evaluating considerations having to do with perceived intensity, it is
important to keep in mind the distinction between stimulation and damage. Noxious
stimulation consists principally of extremes of pressure, shock, and temperature, and
also of applications of certain chemicals. These stimuli are physical phenomena, and
they can therefore be measured by physical and chemical instruments. On the other
hand, while actual and potential damage are of course associated with noxious
stimuli, the degree of damage is not fixed by the magnitude of a physical or chemical
stimulus. Degree of actual or potential damage is degree of actual or potential harm—
that is, degree to which a stimulus jeopardizes the ability of the organism to function
in accordance with its biological design. Now it would clearly be wrong to maintain
that the perceived intensities of pains correspond closely to the magnitudes of
peripheral stimuli. This follows from several of the considerations that we noted
earlier. Insofar as perceived intensities depend on internal processing, they have a
marked tendency to diverge from the magnitudes of noxious stimuli. But it could still
be true that there is a significant correlation between perceived intensities and levels
of actual or potential damage. Indeed, it could be true that various forms of internal
processing, such as temporal and spatial summation, actually improve the corres-
pondence between perceived intensities and levels of damage. Thus, for example, it
could easily be true, and in fact probably is true in most cases, that damage increases
with the interval during which a noxious stimulus is continuously applied. If this is
the case, then a mechanism that computed sums of stimulation over sizeable intervals

8
See note 6.
186 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

would do a better job of keeping track of amounts of damage than a mechanism that
merely registered instantaneous levels of stimulation. Similarly, it is reasonable to
suppose that long term sensitization provides a kind of memory of continuing
damage, and that when it combines with a nociceptive signal caused by a new
noxious stimulus, augmenting the latter, it thereby does justice to the fact that the
threat posed by the new stimulus is greater than it would otherwise be, because it is
applied to an area that is already compromised. The distinction between the amount
of actual or potential damage and the magnitude of a stimulus may also be relevant to
the fact that the former increases as a power function of the latter. If an organism is
stimulated by two electric shocks, one of which is twice as great as the other, it may be
that the larger shock poses a threat to the integrity of the organism that is signifi-
cantly more than twice the threat that is posed by the first shock. Indeed, it seems
reasonable to suppose that this is the case.
We saw earlier that there are six factors that prevent perceived intensities from
being fully aligned with peripheral conditions. The peripheralist strategy we have just
been reviewing addresses four of these factors—temporal summation, spatial sum-
mation, long term sensitization, and response expansion. The strategy provides some
hope of explaining the discrepancies caused by these factors without conceding
anything to advocates of centralist theories. But the fourth peripheralist strategy
cannot stand alone. A peripheralist also needs a strategy for dealing with the
modulatory influence of Aβ fibers and a strategy for dealing with the centrifugal
influence of psychological factors. I will now describe a fifth strategy that can meet
both of these needs.
Like the fourth strategy, the fifth one presupposes that pains are bodily disturb-
ances involving actual or potential damage, and that intensities are properties of
these disturbances. Unlike the fourth strategy, however, which presupposes that
intensities are intrinsic properties of bodily disturbances, this one maintains that
they are relational properties—more specifically, relational properties that disturb-
ances have in virtue of contributing to nociceptive activity in the spinal cord. This
identification is plausible because it is natural to think of the representations that
support awareness of pains as standing for various levels of this activity.
The case for this claim has two parts. First, there is a very strong correlation
between perceived intensities and levels of nociceptive activity flowing upwards from
the dorsal horn of the spinal cord. Generally speaking, when a pain seems to an agent
to have a certain degree of intensity, the nociceptive activity in the spinal cord is at a
corresponding level; and the converse is also true. Second, as I will now show, it is
arguable that it is biologically important for the organism to keep track of this
activity. Nociceptive activity in the spinal cord is shaped by four factors—aggregated
information from the periphery concerning the scope and amount of peripheral
damage, information about pressure that is carried by the Αβ fibers, psychological
states whose job it is to induce the organism to modify its behavior so as to promote
healing and avoid further damage, and psychological states whose job it is to ensure
 LOCATING QUALIA 187

that the organism can pursue highly important agenda items, such as fleeing and
fighting. The dorsal horn in effect computes a weighted average of these factors and
then sends a signal whose strength corresponds to this average upward along various
pathways in the spinal cord, until it is ultimately registered by the relevant perceptual
system in the brain. In effect, the signal represents a “best guess” as to how important
the damage is, given the organism’s long term needs and current projects. Now it
follows from this description of the signal that it is very much to the advantage of
the organism to monitor its levels, and adjust its actions in accordance with them. If
the signal in effect represents a balanced, broadly based estimate of the significance
of the current damage, the organism stands to gain by keeping track of it.
To summarize, the representations that support awareness of pain do in fact keep
track of levels of nociceptive activity in the spinal cord, and there is a case for the view
that the organism needs to keep track of them. Now when it is reasonable to think
that a system of perceptual states does in fact keep track of the values of a variable,
and it is also reasonable to think that the organism needs to keep track of those
values, it is reasonable to suppose that the states in question represent the values.
Accordingly, there are grounds for the view that the states that support awareness of
pain represent levels of nociceptive activity in the spinal cord, and by the same token,
grounds for thinking that perceived intensities are identical with levels of that
activity. If we prescind from the factors that promote misrepresentation, the intensity
that one perceives a pain to have just is the corresponding level of nociceptive activity.
In the interests of simplicity I have developed this line of thought without
observing the distinction between intrinsic properties of nociceptive signals in the
spinal cord and relational properties that pains (i.e., certain bodily disturbances) have
in virtue of the large role they play in causing such signals. So I now emphasize that it
is relational properties that are represented. To see why this is true, observe first that
we are obliged to say that the representations that support awareness of pain are
concerned with peripheral damage. This view is broadly and deeply motivated by
several important facts, including the fact that the somatosensory representational
system that registers pain incorporates a map that is naturally interpreted as repre-
senting the peripheral sites where damage occurs (that is, the homuncular map that is
laid out along the post-central gyrus), the fact that representations of bodily locations
on this somatosensory map are linked with visual and motor representations of the
same locations, and the fact that the brain desperately needs information about
peripheral damage.9 Could it be true that the somatosensory representations in

9
Here is another version of the argument: It is pretty clear that somatosensory representations do
encode information about bodily disturbances involving actual and potential damage, and it is also pretty
clear that the brain needs to have representations that encode such information. Putting these facts
together, we have evidence for the conclusion that the representations have the function of encoding
information about bodily disturbances. But if nociceptive somatosensory representations have the function
of encoding information about bodily disturbances, then it is plausible that they represent those disturb-
ances. (For a more fully developed version of this line of thought, see Hill 2009, p. 180.)
188 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

question also stand for nociceptive signals in the spinal cord? That is, could it be true
that they represent entities of two different kinds? No. For one thing, introspection
teaches that one is aware of only one entity when one is aware of a pain—namely, the
pain itself. But also, it is beyond the powers of the relevant representational system to
present two entities simultaneously. This system cannot distinguish between infor-
mation arriving from the periphery and information arriving from the spinal cord,
since both kinds of information are carried by the same ascending pathways. The job
of the system is simply to register incoming information and interpret it as infor-
mation about the periphery. Given that this is its job, it can only represent activity in
the spinal cord by referring it to peripheral disturbances, and it can do this only by
representing certain relational properties of those disturbances—specifically, rela-
tional properties that they have in virtue of giving rise to activity in the spinal cord.
It is possible to be a bit more specific concerning the relational properties that the
sixth strategy identifies with perceived intensities. There are several ascending path-
ways in the spinal cord that carry nociceptive information to the brain. The pathway
that is most relevant to our present concerns is the spinothalamic tract, which
connects dorsal horn neurons with areas of the thalamus that are in turn directly
connected with the primary somatosensory cortex (Price and Bushnell 2004, p. 10).
Since I am supposing here that the primary somatosensory cortex is the seat of
awareness of pain, and therefore the seat of perceived intensities, the spinothalamic
tract can be described as the ascending pathway that carries the signals that corres-
pond most closely to perceived intensities, and have the most immediate causal
influence upon them. Because the tract has these properties, it is natural to suppose
that perceived intensities are relational properties that have the following form:
causing nociceptive activity of level L in the spinothalamic tract. (The view that
primary somatosensory cortex is the seat of awareness of pain is an oversimplifica-
tion, but there is reason to think that it does the lion’s share of the relevant
representational work. See Hofbauer et al. 2001.)
This choice of relational properties is to some extent arbitrary. The fifth strategy
could with equal justification have identified perceived intensities with relational
properties of the form causing nociceptive activity of level L in the lateral nuclear
group of the thalamus and with relational properties of the form causing nociceptive
activity of level L in the neurons that project from the lateral nuclear group to the
primary somatosensory cortex. (The lateral nuclear group of the thalamus contains
the areas of the thalamus where the spinothalamic tract terminates.) The nociceptive
activity in the spinothalamic tract is more or less the same as the activity in the
others, so it would make sense to invoke any one of them in giving an account of
perceived intensities. Perhaps it is best to think of the representations that support
awareness of intensities as standing for small bundles of relational properties, or as
standing ambiguously for each member of a bundle.
As I see it, however, it would be a mistake to equate perceived intensities with
relational properties of the form causing nociceptive activity of level L in the primary
 LOCATING QUALIA 189

somatosensory cortex. In order for a perceptual representation R to stand for a

property P, it must be true that R encodes information about P—or at least, that R
encoded information about P at some time in the past. But for R to encode
information about P, it must be true that instances of P cause tokens of R, and
moreover, it must be true that instances of P cause tokens of R in virtue of being
instances of P. These conditions on representation make it impossible for a repre-
sentation R to stand for a relational property of the form causing tokens of R. It
doesn’t make sense to say that tokens of R are caused by instances of the property
causing tokens of R in virtue of being tokens of that property.
So much for the fifth strategy. The sixth peripheralist strategy is concerned
specifically with endogenous attention. As we saw earlier, endogenous attention
can play a role in determining which representation of peripheral damage is tokened,
by acting directly on the representational system located in somatosensory cortex,
and it can thereby help to determine what level of intensity a pain is perceived to
have. Generally speaking, when one voluntarily attends to a pain, the perceived
intensity of the pain increases, perhaps by a significant amount. Prima facie, at
least, this poses a problem for peripheralist theories, for it causes perceived intensity
to diverge from extra-cranial intensities. Moreover, this is true whether extra-cranial
intensity is thought to be an intrinsic property of peripheral damage, or is instead
thought to be a relational property involving activity in the spinothalamic tract.
Endogenous attention may bring other features of a pain, such as location and
frequency of throbbing, into sharper focus. It may also help an agent to make
accurate judgments about whether a pain is increasing, decreasing, or remaining
the same, and it may make it possible for an agent to make more accurate compara-
tive judgments about the intensities of pairs of pains (provided that the agent is
attending equally to both members of a pair). But it prevents perceived intensities
from matching the “objective” intensities of individual pains, whether objective
intensities are taken to be degrees of peripheral damage or relational properties
involving spinothalamic activity. The sixth peripheralist strategy addresses this
problem by appealing to misrepresentation. Pointing out that endogenous attention
in effect interferes with the normal flow of information about objective intensities, by
increasing perceived intensities even though objective intensities remain the same, it
urges that the influence of endogenous attention on perceived intensities leads to
distortions. But of course, representations with distortions are misrepresentations.
It is independently plausible that attention can promote misrepresentation. Thus,
as we have recently learned from the work of Marisa Carrasco and her associates
(Anton-Erxleben et al. 2007, Carrasco et al. 2000, Carrasco 2009, Fuller and Carrasco
2006, Liu et al. 2009, Turatto et al. 2007), visual attention can change the appearances
of seen objects by increasing such features as apparent color saturation, apparent
contrast, apparent speed, and apparent size. By changing appearances in these ways,
attention enhances certain aspects of one’s epistemic position with respect to objects.
But it also worsens one’s epistemic prospects in other respects.
190 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

To illustrate, we know from Carrasco’s work (see, e.g., Carrasco 2009) that
attention can increase the apparent contrast of a grid of alternating dark and light
lines. This has certain cognitive benefits. For example, it increases one’s ability to
classify items with respect to perceptual categories whose associated criteria include
contrasting lines. But it also has cognitive disadvantages. Thus, as Carrasco’s work
also shows, when one heightens the apparent contrast of one item by bestowing
attention on it, one may be at a disadvantage in comparing the contrast of that item
with the degree of contrast of another item that is perceived concurrently. In giving a
boost to the apparent contrast of the first item, attention may cause it to look the
same as another item whose objective contrast is actually greater. In causing the two
items to look the same, even though their respective degrees of contrast are object-
ively different, attention increases one’s tendency to make inaccurate judgments.
Since the seminal work of Posner in the 1970s (Posner et al. 1979), it has been
widely held by philosophers and psychologists that attention always improves one’s
epistemic prospects. As a result of Carrasco’s work, however, we are beginning to
appreciate that the situation is much more complicated than that. The very import-
ant epistemic benefits of attention are accompanied by non-trivial epistemic costs.
In sum, the sixth peripheralist strategy explains the discrepancy between the
perceived intensities of attended pains and the “objective” intensities associated
with cases of bodily disturbances by maintaining that attention causes misrepresen-
tation of objective intensities. And it claims that recent work on visual attention
provides support for this account.
This completes my exposition of the six peripheralist strategies. As reflection
shows, all of these strategies carry substantial empirical presuppositions. For
example, the first and third strategies presuppose that the peripheralist theory of
pain is more successful than the central state theory in predicting and explaining the
relevant empirical data. Of course, whether this presupposition is true is an empirical
question. Most of the strategies also have substantial theoretical presuppositions—for
example, about the nature of representation, and about the value to the organism of
representing certain magnitudes. Now it is clear that we are not at present in a
position to pronounce with any finality on either the empirical presuppositions of the
strategies or on their theoretical presuppositions. Accordingly, to some degree,
anyway, it is an open question whether the strategies can be fully successful. But
I hope I have said enough to show that, at present, anyway, they are live options.
The prospects of defending peripheralism against objections based on correlations
are good.

IV. The experiential character of qualitative awareness

According to the central state identity theory, all awareness of qualitative states takes
the form of judgments, and therefore necessarily involves conceptualization. To be
aware of a qualitative state, one must represent it conceptually. Representationalism
 LOCATING QUALIA 191

denies this, maintaining instead that awareness of qualia is perceptual or quasi-

perceptual in character. In Section I we took note of some considerations that
speak in favor of this representationalist doctrine, including the fact that animals
who have limited powers of conceptualization can be aware of qualitative states (the
same is true of infants, by the way), the fact that we can discriminate introspectively
between conceptual and experiential awareness of qualia, the fact that conceptual
awareness of qualia seems to depend causally on experiential awareness, and the fact
that various forms of conceptual awareness inherit their contents from experiential
states in which one attends to particular qualitative phenomena. In addition to these
general considerations, there are also more specific reasons for thinking that aware-
ness of pain is experiential. Thus, awareness of pain is similar to perceptual awareness
in a variety of respects, including the fact that it has a phenomenological dimension,
the fact that it makes it possible for us to grasp extremely intricate patterns in a
flash, the fact that the information it provides tends to be highly determinate, rather
than abstract and general, and the fact that it is associated with a form of attention
that works in much the same way as perceptual attention, increasing the resolution of
awareness and the contrast between figure and ground. These points of similarity
involve psychological features of awareness, but there are also similarities that are
revealed by cognitive neuroscience. Thus, the neural mechanisms that support
awareness of pain are fundamentally akin to the mechanisms that support thermal
awareness and touch. Indeed the similarities are so numerous that it would be very
awkward to classify the former systems as perceptual while withholding this label
from awareness of pain. I think we are pretty much committed to regarding aware-
ness of pain as at least quasi-perceptual.10 The question is just what this fact tells us
about the prospects of the central state identity theory.
As a first step towards answering this question, let us consider whether it is
possible to interpret conceptual dualism in such a way that it is compatible with
the perceptual character of qualitative awareness. Does it make sense to claim that
concepts are at least partly perceptual in character? It seems that the answer should be
“yes,” for there are a number of theories that treat concepts as essentially perceptual in
character, or at least as having an important perceptual dimension. Consider the
theories of concepts that were developed by the classical empiricists. Hume,
for example, maintained that concepts (“ideas”) are in effect pale copies of percep-
tual states (“impressions”). In his view, concepts are literally built out of perceptual

10
A recent book that I very much admire (Burge 2010), Tyler Burge takes the view that since awareness
of pain is not governed by constancies like the ones that govern vision and hearing, there is no reason for
thinking that it counts as awareness of an objective phenomenon, located in three-dimensional space. This
leads in turn to a denial that awareness of pain is perceptual. These claims seem quite wrong to me.
Whether something is represented as being in three-dimensional space is a function of the contents of the
relevant representations. There are strong introspective grounds, and also strong neuroscientific grounds,
having to do with maps in somatosensory cortex and parietal cortex, for thinking that our representations
of pains assign them to bodily locations.
192 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

material. Views of this sort have been updated by contemporary writers like Barsalou
(1999) and Prinz (2002), and turned into detailed proposals. Like many others,
I doubt very much that efforts of this sort can be fully successful. To mention only
one problem, they fail to do justice to abstract concepts like truth and complex
number. But they nonetheless show that it may be possible to treat some subsystems
of our conceptual scheme as largely perceptual in point of format and functional
character. Moreover, Susan Carey’s recent book (Carey 2009) has similar implica-
tions. As Carey shows, there are powerful empirical reasons for thinking that there is a
realm of cognition that is intermediate between the perceptual realm and the realm of
beliefs and concepts. This is the realm of what she calls core cognition—a cognitive
module that is innately determined, provides our primary cognitive resources in
infancy, and continues to play a role in cognition throughout adulthood, providing
essential support for enterprises like interpreting perceptual inputs, identifying causes,
and counting. As Carey visualizes it, core cognition makes use of representations that
resemble perceptual representations in being analog and iconic, and also resemble
concepts in possessing inferential roles. Thus, for example, a core representation of an
object might resemble a perceptual state in the way that it represents the object’s
shape, but be like a concept of the object in that it figures in inferences concerning the
object’s future trajectory. It is possible, I believe, to adapt this theory of Carey’s to
provide a perceptual model of awareness of pain, thereby accommodating the con-
siderations adduced earlier.
It seems, then, that conceptual dualism is compatible with the perceptual character
of qualitative awareness. The crucial question is whether a perceptual model of
qualitative awareness can be appropriately combined with the claim that qualitative
states are identical with brain states. And here I think the answer is pretty clearly
negative: it just isn’t plausible that awareness of pain is perceptual awareness of a
region of the brain. There are several reasons for this view. For one thing, we are
obliged to take perceptual experience at face value, insofar as we can, and it is quite
clear that we experience pains as having peripheral locations. Second, this consider-
ation is reinforced by the psychological analogies with touch and thermal perception.
Since it is clear that touch provides us with information about what is happening on
the surface of the body, and that the same is true of thermal perception, the analogies
make it very plausible that awareness of pain is directed outwards, toward peripheral
regions that are being subjected to noxious stimulation, or in which damage has
already occurred (Hill 2009, chapter 6). Third, if we consider the regions of the brain
that are active during experiences of pain, we find that none of them have structural
or functional properties that would make it appropriate to view them as supporting
perceptual awareness of other regions of the brain. Of the regions that are highly
correlated with experiences of pain, the one that is most naturally interpreted as
perceptual (specifically, the strip of damage-sensitive neurons in primary somato-
sensory cortex) can easily be seen to have a peripheral orientation (Hill 2009,
chapter 6). Fourth, there is a good reason for thinking that the contents of the
 LOCATING QUALIA 193

representations that support experiences of pain are concerned with actual and
potential peripheral damage. Thus, it is plausible that the contents of the represen-
tations that support experiences of various kinds are determined by the informa-
tional needs of the organism, and that this is true, in particular, of the representations
that support our experiences of pain. Moreover, it is clear that the organism has a
pressing need for information about actual and potential damage at peripheral
locations. Putting these observations together, we have a case for the view that the
representations that support experiences of pain have contents that involve actual
and potential peripheral damage. And by the same token, we have a case for the view
that our experiences of pain are experiences of such damage. Finally, we most
definitely do not want to suppose that paradigmatic forms of perceptual awareness
like vision and hearing are directed on brain states. There are many reasons for not
wanting to take that route, such as the transparency of visual experience that is
emphasized in Harman (1990). But if awareness of pain is fundamentally akin to
perceptual awareness, as I have maintained, then if we were to view awareness of pain
as focused on brain states, we would be pretty much committed to saying that the
other forms of perception are focused on brain states as well.
In the end, I think we can see that it’s no accident that advocates of central state
identity have chosen to combine their main thesis with a conceptualist account of
awareness of pain, meaning thereby to invoke systems of concepts that are symbolic
and perhaps even word-like in character. There is no immediate problem with saying
that we have a system of concepts that makes it possible for us to keep track of brain
states. To be sure, difficulties arise in connection with particular proposals about how
conceptually supported awareness of pain might actually work. Thus, for example,
advocates of the thesis that the relevant concepts are demonstratives have never
managed to explain how the alleged demonstratives get their referents. Other
demonstratives get their reference from perceptual attention, but this is not an option
for the theorists I am considering here. It remains true, however, that when concep-
tual dualism is formulated in fairly abstract and generic terms, it can fit together with
a claim of central state identity reasonably well. The problems I have been reviewing
arise only when we try to make conceptual dualism more concrete by combining it
with the claim that qualitative awareness is fundamentally perceptual.

V. Conclusion
We have been mainly concerned with two arguments—one that favors the central
state identity theory and another that favors representationalism. Initially, at least,
both arguments appear to be quite strong. We have found, however, that there is
much that the representationalist can say in response to the first argument. To be
sure, all of the available options carry substantial empirical and theoretical presup-
positions, and it may turn out in the end that a number of these presuppositions fail.
But for the time being, it seems fair to say that the representationalist has a promising
194 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

line of defense. On the other hand, it seems unlikely that the central state theorist can
formulate an adequate defense against the second argument. To do this, it would be
necessary to explain away the facts which suggest that awareness of pain just is quasi-
perceptual awareness of a bodily condition. The prospects of success in such a
venture seem quite slender.

References
Adair, E. R., Stevens, J. C., and Marks, L. E. (1967). “Thermally Induced Pain, the DOL Scale,
and the Psychophysical Power Law,” American Journal of Psychology 81: 147–64.
Anton-Erxleben, K., Henrich, C., and Treue, S. (2007). “Attention Changes Perceived Size of
Moving Visual Patterns.” Journal of Vision 7(II), Article 5.
Barsalou, L. W. (1999). “Perceptual Symbol Systems,” Behavioral and Brain Sciences 22:
577–660.
Basbaum, A. I. and Jessell, T. M. (2000). “The Perception of Pain.” In E. R. Kandel,
J. H. Schwartz, and Jessell, T. M. (eds.), Principles of Neural Science, 4th edition. New
York: McGraw-Hill, 472–91.
Burge, T. (2010). Origins of Objectivity. Oxford: Oxford University Press.
Bushnell, M. C., Villemure, C., and Duncan, G. H. (2004). “Psychophysical and Neurophysio-
logical Studies of Pain Modulation by Attention.” In D. D. Price and M. C. Bushnell (eds.),
Psychological Methods of Pain Control: Basic Science and Clinical Perspectives. Seattle: IASP
Press, 99–116.
Carey, S. (2009). The Origin of Concepts. Oxford: Oxford University Press.
Carrasco, M. (2009). “Attention, Psychophysical Approaches.” In P. Wilken, T. Bayne, and
A. Cleeremans (eds.), The Oxford Companion to Consciousness. Oxford: Oxford University
Press, pp. 78–84.
Carrasco, M., Penpeci-Talgar, C., and Ekstein, M. (2000). “Spatial Covert Attention Increases
Contrast Sensitivity across the CSF: Support for Signal Enhancement,” Vision Research 40:
1203–15.
Coghill, R. C., Sang, C. N., Maisog, J. M., and Iadarola, M. J. (1999). “Pain Intensity Processing
within the Human Brain: A Bilateral Distributed Mechanism,” Journal of Neurophysiology
82(4): 1934–43.
Craig, A. D. and Bushnell, M. C. (1994). “The Thermal Grill Illusion: Unmasking the Burn of
Cold Pain,” Science 265: 252–5.
Craig, A. D., Krout, K., and Andrew, D. (2001). “Quantitative Response Characteristics of
Thermoreceptive and Nociceptive Lamina I Spinothalamic Neurons in the Cat,” Journal of
Neurophysiology 86: 1459–80.
Dretske, F. (1995). Naturalizing the Mind. Cambridge, MA: MIT Press.
Fuller, S. and Carrasco, M. (2006). “Exogenous Attention and Color Perception: Performance
and Appearance of Saturation and Hue,” Vision Research 46: 4032–47.
Harman, G. (1990). “The Intrinsic Quality of Experience.” Philosophical Perspectives 4: 31–52.
Hill, C. S. (1984). “In Defense of Type Materialism,” Synthese 59: 295–320.
Hill, C. S. (1991). Sensations: A Defense of Type Materialism. Cambridge: Cambridge Univer-
sity Press.
 LOCATING QUALIA 195

Hill, C. S. (1997). “Imaginability, Conceivability, Possibility, and the Mind–Body Problem.”

Philosophical Studies 87: 61–85.
Hill, C. S. (2009). Consciousness. Cambridge: Cambridge University Press.
Hofbauer, R. K., Rainville, P., Duncan, G. H., and Bushnell, M. C. (2001). “Cortical Represen-
tation of the Sensory Dimension of Pain,” Journal of Neurophysiology 85(1): 402–11.
Liu, T., Abrams, J., and Carrasco, M. (2009). “Voluntary Attention Enhances Contrast
Appearance,” Psychological Science 20(3), 354–62.
Loar, B. (1990). “Phenomenal States,” Philosophical Perspectives 4: 81–108.
Lycan, W. G. (1996). Consciousness and Experience. Cambridge, MA: MIT Press.
McLaughlin, B. P. (2003). “Colour, Consciousness, and Colour Consciousness.” In Q. Smith
and A. Jokic (eds.), Consciousness: New Philosophical Perspectives. Oxford: Clarendon Press.
Melzack, R. and Wall, P. D. (2008). The Challenge of Pain, updated 2nd edition. London:
Penguin Books.
Papineau, D. (2002). Thinking about Consciousness. Oxford: Oxford University Press.
Pautz, A. (2010). “Do Theories of Consciousness Rest on a Mistake?” Philosophical Issues 20:
333–67.
Place, U. T. (1956). “Is Consciousness a Brain Process?” British Journal of Psychology 47:
44–50.
Porro, C. A., Cettolo, V., Francescato, M. P., and Baraldi, P. (1998). “Temporal and Intensity
Coding of Pain in Human Cortex,” Journal of Neurophysiology 80(6): 3312–20.
Posner, M. I., Nissen, M. J., and Ogden, W. C. (1979). “Attended and Unattended Processing
Modes: The Role of Set for Spatial Locations.” In H. L. Pick and B. J. Saltzman (eds.), Modes
of Perceiving and processing Information. Hillsdale, NJ: Erlbaum, 137–58.
Price, D. D. (1999). Psychological Mechanisms of Pain and Analgesia. Seattle: IASP Press.
Price, D. D. and Bushnell, M. C. (2004). “Overview of Pain Dimensions and their Psychological
Modulation.” In D. D. Price and M. C. Bushnell (eds.), Psychological Methods of Pain
Control: Basic Science and Clinical Perspectives. Seattle: IASP Press, 3–17.
Prinz, J. J. (2002). Furnishing the Mind: Concepts and their Perceptual Basis. Cambridge, MA:
MIT Press.
Rainville, P. (2004). “Pain and Emotions.” In D. D. Price and M. C. Bushnell (eds.), Psycho-
logical Methods of Pain Control: Basic Science and Clinical Perspectives. Seattle: IASP Press,
117–41.
Rainville, P., Duncan, G. H., Price, D. D., Carrier, M., and Bushnell, M. C. (1997). “Pain Affect
Encoded in Anterior Cingulate but Not Somatosensory Cortex,” Science 277: 968–71.
Ramachandran, V. S. and Blakeslee, S. (1998). Phantoms in the Brain: Probing the Mysteries of
the Human Mind. New York: HarperCollins.
Rhudy, J. L. and Meagher, M. W. (2000). “Fear and Anxiety: Divergent Effects on Human Pain
Thresholds,” Pain 84: 65–75.
Smart, J. J. C. (1959). “Sensations and Brain Processes,” Philosophical Review 68: 141–56.
Stevens, S. S. (1962). “The Surprising Simplicity of Sensory Metrics,” American Psychologist
17: 29–39.
Timmermann, L., Ploner, M., Haucke, K., and Schmitz, F. (2001). “Differential Coding of Pain
Intensity in the Human Primary and Secondary Somatosensory Cortex,” Journal of Neuro-
physiology 86(3): 1499–503.
196 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

Turatto, M., Viscovi, M., and Valsecchi, M. (2007). “Attention Makes Moving Objects be
Perceived to Move Faster,” Vision Research 47: 166–78.
Tye, M. (2000). Consciousness, Color, and Content. Cambridge, MA: MIT Press.
Vase, L., Price, D. D., Verne, G. N., and Robinson, M. E. (2004). “The Contribution of Changes
in Expected Pain Levels and Desire for Pain Relief in Placebo Analgesia.” In D. D. Price and
M. C. Bushnell (eds.), Psychological Methods of Pain Control: Basic Science and Clinical
Perspectives. Seattle: IASP Press, 207–32.
 11
Visual Awareness and
Visual Qualia*

“It is of the essence of mental paint to be something of which we are aware.”

Ned Block
“No awareness without representation!” Anonymous sage

I will argue for an account of awareness of qualia that promises to bring qualia into the
physical fold. More particularly, I will present and defend a version of the view that has
come to be called “representationalism.” I will focus on the task of developing a
representationalist theory of visual awareness, but it will be evident, I think, that the
theory can be generalized so as to apply to experiential awareness of other kinds as well.
After explaining this theory of visual awareness, I will urge that it provides a satisfactory
answer to the metaphysical problems that qualia pose. I will then discuss the question of
where exactly qualia are to be located within our catalog of physical properties.

I
Although there are differences of opinion as to how many types of qualia there are,
there is wide agreement that qualia are associated with bodily sensations, emotions,
and perceptual experiences. Thus, one category of qualia includes such characteristics
as being a pain, being an itch, and being a surge of nausea. Another category consists of
the properties like the ones that we invoke when we speak of the icy hand of fear and
the glowing coals of anger. And a third category consists of properties that are associa-
ted with the ways that objects appear to us when we perceive them. As Jaegwon Kim

* An early draft of this paper written in 2006. It was intended for a Festschrift for Jaegwon Kim, but due
to publishing delays, the present, expanded version is the first to appear. I have benefited from conversa-
tions with Uriah Kriegel and James Van Cleve, and from discussions following talks at ANU, Monash
University, and the University of Sydney. A remark by John Bigelow following my talk at Monash was
particularly helpful. The content of the paper overlaps with that of several chapters in Hill 2009, but its
formulations tend to be simpler and more straightforward, and therefore more accessible. In some cases,
I believe, they are also more persuasive.
198 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

says, perceptual qualia “are, by definition, the ways that things look, seem, and appear
to conscious observers” (Kim 2006, p. 225).
In order to characterize this third category of qualia adequately it is necessary to
distinguish between two senses of the terms that we use to describe appearances.
I will focus here on “looks” since I will be concerned primarily with visual qualia in
the present paper, but my remarks will apply, mutatis mutandis, to “seems,”
“appears,” and a number of other appearance words.
There is a sense of “looks small” in which it can be correctly applied both to a toy
car that one holds in one’s hand and to a real car that one sees on the road far ahead.
In this sense, the expression can also be applied to a tall building that one sees from a
plane, and even to an immense star that one sees from the Earth. When one says that
an object looks small to an observer, using “looks small” in this phenomenological
sense, one is not claiming that the observer’s perceptual experience supports the
judgment that the object really is small. One is not saying that the observer’s
experience represents the object as small. Rather one is drawing an analogy between
the observer’s current visual experience and the visual experiences he has when is
viewing objects that are reasonably close at hand and really are small.
The phenomenological sense of “looks” is also to be found in claims about
apparent shape and apparent color. It is permissible to apply “looks elliptical” both
to an object that really is elliptical and is perpendicular to the observer’s line of sight,
and to a round coin that is slanted away from the observer. Equally, it is permissible
to apply “looks dark brown” both to a piece of chocolate and to a portion of a tan wall
that is cloaked in shadow.
In addition to the phenomenological sense of “looks,” there is also what is often
called its epistemic sense. When we say that an item looks small to an observer, using
“looks small” in this second sense, we mean that the observer’s current visual
experience provides adequate evidential support for the belief that the object is
small. When we have this second sense in mind, we would not be willing to say
that a car looks small to an observer if the car is at an appreciable distance from the
observer, for when a car is at an appreciable distance from an observer, the observer’s
visual experience presents him with “pictorial cues” that are indicative of distance.
Thus, for example, when an object stands at some distance from an observer, the
features of the object seem indistinct. In a case of this sort, the observer’s experience
supports the belief that he is seeing a car of normal size, but a car that is rather far
away. Accordingly, using “looks” in its epistemic sense, it is correct to say that the car
looks to be of normal size, despite the fact that it is also correct to say, using “looks”
in its phenomenological sense, that the car looks small to the observer.
We can also use “looks” in its epistemic sense to talk about appearances of other
kinds. Thus, it is quite appropriate to apply “looks round” to a coin that is tilted away
from an observer, and to apply “looks tan” to a portion of a wall that is poorly
illuminated, provided that the observer’s visual experience attests to this fact about
the lighting.
 VISUAL AWARENESS AND VISUAL QUALIA 199

Putting the distinction between these two senses of “looks” to use, we can define
visual qualia as characteristics that we are aware of in virtue of the ways that objects
lookp to us. (Here and hereafter, I use “looksp” to represent the phenomenological
sense of “looks.”) In general, perceptual qualia can be defined as the characteristics
we are aware of in virtue of the ways in which objects appear to us, where “appear” is
used in its phenomenological sense.
I have mentioned three categories of qualia—sensory qualia, emotional qualia, and
perceptual qualia. What do the members of these categories have in common? Why
do we group them together under a common label? One reason is that they seem to
us to share a number of epistemological properties. We believe that it is impossible to
know qualia adequately without experiencing them. We also believe that our grasp of
them is direct and immediate. Moreover, we are strongly inclined to think that our
awareness of qualia is not governed by an appearance/reality distinction. Thus, we
are inclined to think, it is impossible for it to seem to one experientially that one is in
pain without one’s actually being in pain. Equally, we suppose, it is impossible for it
to seem to one that an object looksp red to one without its actually being the case that
an object looksp red to one. Finally, we are inclined to think that experiential
awareness provides us with full access to the essential nature of qualia. Our grasp
of them is not perspectival or limited in any way. They do not have a hidden
dimension that experience fails to reveal.
These perceptions concerning our awareness of qualia would by themselves
provide a sufficient reason for viewing qualia as special, but we are also strongly
inclined to suppose that qualia have unusual metaphysical properties independently
of our awareness.
We view qualia as intrinsic properties—as properties that things have independ-
ently of their relations to other things. Also, we are inclined to think that there is
something metaphysically fundamental about qualia: when qualia are not themselves
are simple and unanalyzable, we are inclined to suppose, they are resolvable into
more basic qualia that are simple and unanalyzable. Further, we hold that qualia are
responsible for relationships of qualitative similarity and qualitative difference. This
last claim is explained in different ways by different authors, but it seems to be
common ground that qualitative similarity is a form of similarity that cannot be
analyzed in terms of shared causal powers or shared spatio-temporal relations. It is a
form of similarity that arises directly from the intrinsic natures of objects. Moreover,
since the qualia that are responsible for it are simple and unanalyzable, qualitative
similarity is simple and unanalyzable as well. In particular, qualitative similarity is
not analyzable in terms of shared microphysical structure.
Here, then, is a list of properties that we conceive of qualia as possessing. Now we
must ask: if qualia really do have these properties, can they be identified with or
otherwise reduced to physical properties of any kind? Reflection indicates that the
answer should be negative.
Consider, for example, the characteristic we are aware of when something looksp
yellow to us. From the perspective of experience, we are strongly inclined to say that
200 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

this characteristic is simple, in the sense that it cannot be resolved analytically into
other characteristics that are in some sense more basic. But the physical character-
istics that figure in scientific accounts of color vision are all complex. Science
attributes complexity to the enduring physical characteristics of the external sub-
stances that serve as the objects of perception, to the viewpoint-dependent charac-
teristics that objects possess in virtue of reflecting structured light to various vantage
points, and to the characteristics that various parts of the visual system come to
exemplify during the various stages of visual processing.
Further, when we grasp qualia experientially, they seem to us to have individual
natures that are distinct from the individual natures of all of the characteristics that
science investigates. Consider again the characteristic that we are aware of when
something looksp yellow to us. Call it C. Are we aware of C when we visually explore
the brain, or when we visually imagine the microprocesses in the brain that science
describes? No. Are we aware of it when we examine the retina, perhaps aided by a
powerful microscope? No. Are we aware of it when we visualize in imagination the
interactions of light waves and the surfaces of external objects? No. Are we aware of it
when we visualize the enduring spectral reflectances of such objects? No. In view of
facts of this sort, we are strongly inclined to think that there is no experience we can
have of physical characteristics, or can imagine ourselves having, that would put us in
touch with C.
Now if there were an appearance/reality distinction associated with visual qualia,
we could simply say that we have trouble locating C in the physical world because the
appearance it presents to us fails to reveal its true nature. That is, we could explain
away the apparent difference between C and all physical characteristics. Unfortu-
nately, however, it seems inappropriate to distinguish between C as it appears to us
and C as it is in itself. C is the ostensibly qualitative characteristic that is affiliated
with facts of the form x looksp yellow to y. Since there is no appearance/reality
distinction that is associated with these facts, how could there be an appearance/
reality distinction that is associated with C?
In general, it seems, when an object looksp a certain way to us, we are aware of a
characteristic that we cannot be aware of in any other way. No other form of
experience will reveal it to us. Moreover, when we experience it, we are not aware
of it as having a microphysical nature of any sort, or as determining relationships of
similarity of the sort that arise from microphysical structure. Accordingly, we sense
that it is different than the characteristics that science describes. And we are unable to
explain this sense of difference away by invoking an appearance/reality distinction.
Alas, these reflections seem to lead inexorably to the conclusion that there is an
unbridgeable gulf between qualia and the physical world.
Here then is the crux of the problem that qualia present to us. We feel obliged to
embrace a form of qualia realism. After all, it seems that we are aware of qualia. But
there is nothing in the physical world that answers to our conception of qualia.
Hence, qualia cannot be physical characteristics. The physical world does not exhaust
reality. In Jaegwon Kim’s apt phrase, there is a “mental residue” (Kim 2005, p. 170).
 VISUAL AWARENESS AND VISUAL QUALIA 201

II
As far as I can tell, there is only one theory of experience that provides a satisfactory
way of dealing with this problem—representationalism. Representationalism main-
tains that our awareness of the characteristics we call qualia essentially involves
representations, and that the representations in question are different in a variety
of respects than the representations that are involved in other forms of cognition.
Because of the distinctive features of these representations, it maintains, the proper-
ties they represent seem to us to have special features like intrinsicness and simplicity,
and seem to us to have individual natures that are not captured by scientific accounts
of experience. Despite our impressions to the contrary, representationalism asserts,
our awareness of qualia is governed by an appearance/reality distinction of a certain
sort. There is the way that the properties we call qualia are in themselves, and the way
that these properties seem to us in virtue of being represented by the representations
that subserve experiential awareness. It is only insofar as they are represented by such
representations that they take on the aspect of qualia.
Representationalism maintains, in other words, that the properties we call “qualia”
do not really have the properties that we take to be constitutive of qualitative
character. They only seem to us to have such features because of the peculiarities
of the representations in virtue of which we are aware of them. Moreover, if they
seem to us to have individual natures that are different from the natures of all
physical characteristics, this is because, and only because, they are represented in a
unique way. Representationalism adds that we are unable to see beyond these
apparent differences, and to appreciate the ultimate identity of the properties we
call “qualia” with certain physical properties, because it is not apparent to us, from
the perspective of common sense, that our awareness of them involves representa-
tions. Folk psychology does not reveal that representations are constitutively
involved in facts of the form x looksp F to y. Accordingly, it does not occur to us
that our awareness of the properties we call “qualia” might be governed by an
appearance/reality distinction. We think that it is necessary to take our experience
of qualia at face value.
This view of qualia can also be expressed, albeit somewhat opaquely, by saying that
qualia are not properties that exist independently of our awareness of them. They are
properties as seen from the perspective of the systems of representation that enable
experiential awareness. They are physical properties qua experientially represented.1

1
I mean to be stating a philosophical position here, not explicating the pre-theoretical use of the term
“qualia.” I think there is theoretical motivation for using “qualia” in two different ways, both of which have
affinities to the pre-theoretical use, but which diverge from it in being more precise. One use is linked to the
characteristics we are aware of in virtue of participating in facts of the form x looksp to y. I think it is natural
and appropriate to use the term “qualia” to refer to these characteristics, and I often so use it here and in
other writings. That is, as I see it, it is natural and appropriate to use “qualia” to refer to certain properties
that are objects of experiential awareness, and therefore, properties that are represented by the represen-
tations that constitute experiential awareness. But as we will see as we proceed, if one wishes to explain why
202 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

In developing this story, representationalism begins by calling attention to some of

the ways in which experiential awareness differs from the forms of awareness that
involve conceptualization and judgment. Among other things, it points out that
experiential awareness always provides us with access to highly determinate forms
of properties (when a thing looksp blue to one, it always looksp a highly determinate
shade of blue), that experiential awareness generally provides more information
about objects of awareness than can easily be captured by linguistic or conceptual
descriptions, that it automatically assigns the properties it represents to locations in
quality spaces, that it enables us to make extremely fine discriminations among the
members of domains, thereby giving us a sense of the density of the orderings that
obtain in the domains, and that it is associated with a special category of attention
mechanisms—mechanisms that enable us to adjust certain aspects of experiential
awareness, such as resolution and figure/ground contrast.
According to representationalism, it is possible to explain these and other distin-
guishing features of experiential awareness by invoking features of the systems of
representation that such awareness involves. Of course, at present this has the status
of a largely unsubstantiated hypothesis. We know very little about the representa-
tional systems that are involved in experiential awareness. We have but a meager
grasp of their formal properties, and views about their semantic properties are largely
conjectural. It is therefore fortunate that we do not need to have a well worked out
account of experiential representation in order to see that it is in principle possible to
provide representation-based accounts of the foregoing features. Suppose, for
example, that the systems of representation that are involved in experiential aware-
ness are analog in character. On this assumption, we can easily visualize a represen-
tation-based explanation of why experiential awareness always puts us in touch with
highly determinate forms of the properties it represents, and also an explanation of
why it generally provides us with access to the density of the orderings that govern
the domains with which it is concerned. The mere possibility of giving explanations
of this sort is sufficient to make representationalism an attractive theory. Advocates
of representationalism at least have the right to hope that that the distinguishing

qualia have seemed to pose a problem for physicalism, it is necessary to take account of intrinsic or
“syntactic” features of the relevant representations. Accordingly, when one thinks of qualia not just as
objects of experiential awareness, but as having features that pose a philosophical problem, it is best to
consider them as properties qua represented by experiential representations.
I should add here that my use of the term “representationalism” differs from the use that one finds in
such authors as Alex Byrne, Fred Dretske, and Michael Tye. (See, e.g., Byrne 2001, Dretske 1995, and Tye
2000.) The main tenet of these authors is that the qualitative character of an experience supervenes on its
representational content. I read this as committing them to the view that it is possible to explain all of the
features of the qualitative character of an experience, including those that pose philosophical problems, in
terms of the content of the experience, and therefore in terms of the properties that the experience
represents. As will become clear a bit later on, I think this view is false. In order to explain why the
qualitative character of an experience is philosophically problematic, it is necessary to appeal to such things
as the analog character of experiential representation, and the fact that certain experiential representations
fail to articulate the complexity of the properties they represent due to lack of internal structure.
 VISUAL AWARENESS AND VISUAL QUALIA 203

features of experiential awareness can be explained. It is not clear that any other
theory of experiential awareness confers this right.
Thus far we have been considering the general motivation for representationalism.
I will try now to say a bit more about the relevance of representationalism to the
metaphysical problem that arises when we suppose that there really are characteris-
tics answering to our conception of qualia. In these remarks, and also in later parts of
the discussion, I will focus on visual qualia—that is, on the qualia we are aware of in
virtue of participating in facts of the form x looksp F to y. I believe that what I will say
about visual qualia generalizes to qualia of other types, but I will not attempt to
defend this belief here.
Our awareness of visual qualia is experiential in nature. Now as we saw a bit
earlier, there is independent motivation for supposing that there is a distinctive
system of representation that subserves experiential awareness. In view of this fact,
representationalism contends, we have the right to assume that our awareness of
visual qualia is essentially representational, and that the representations involved in
such awareness belong to a distinctive system.
Applying this assumption, we can explain why certain properties seem to us to be
simple and unanalyzable, despite having an internal complexity of the sort that
science reveals. Thus, it is a familiar fact that a representation can encode informa-
tion about a property without encoding information about the details of its internal
organization. We know, for example, that a simple line drawing can represent a
house without indicating whether the house is composed of wooden planks, or red
bricks, or field stones, or cinderblocks. A fortiori, it doesn’t tell us anything about the
physical microstructure of the house. Nor does it tell us anything about the insides of
the house. It represents the house as having walls, a roof, a chimney, some windows,
and a door, period.
To develop this thought a bit, suppose that visual representation has an iconic
dimension. Now we know it to be a law of iconic representation that complexity in
the represented object is represented by complexity in the representation itself.2
Thus, for example, if a line drawing of a house lacks rectangular components
corresponding to bricks, then it does not represent the house as composed of bricks
or brick-like elements. Applying this law of iconic representation to the case of visual
representation, if an iconic visual representation fails to reflect the complexity of a
physical structure, it doesn’t represent that complexity. It fails to articulate it, and it

2
In a system of iconic representation, the representations articulate the structures or internal organ-
izations of the items they represent. To say that R articulates the structure of P, where having P consists in
having parts x1, . . . , xn that bear a relation ç to each other (or in having properties X1, . . . , Xn that bear a
relation å to each other), is to say (i) that R has parts r1, . . . , rn that respectively represent x1, . . . , xn and (ii)
that those parts bear a syntactic relation to each other that represents ç. I.e., saying that representations
articulate structure is to be explained in terms of homomorphisms between representations and the
internal structures of the things they represent.
204 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

therefore fails to represent it. This doesn’t mean that it represents the structure as
simple. Failing to represent as complex is not the same thing as representing as
simple. But from the perspective of a subject whose cognitive command of a physical
entity derives entirely from a visual representation, R, R’s failing to represent the
complexity of the entity will come to the same thing as attributing simplicity to the
entity. Because the subject’s cognitive commend of the entity derives entirely from
the representation, he will find it natural to attribute simplicity to the entity even if
the representation does not.3
I have been talking here about mereological simplicity, which has to do with
whether substances have parts, but there is a similar explanation of why experienced
qualities can seem simple. For instance, it is plausible that yellow seems simple to us
because, at the most fundamental level, our awareness of it derives from a simple
representation—a representation lacking both mereological and qualitative complex-
ity. Since the representation is in effect an atomic component of the system of
representation with which it is associated, it will provide the subject who is using it
with no ground for attributing complexity to yellow. Accordingly, it will be natural
for the subject to form the impression that the represented property is simple.
What is much more important, in addition to this explanation of why visual qualia
can seem simple and primitive, there is a representationalist explanation of why there
seems to be an unbridgeable gulf between the property we are aware of when
something looksp yellow to us and the properties that are revealed by the scientific
investigation of vision.
When things lookp yellow to us, we are deploying a representation that is different
than all of the representations that we deploy when things lookp other ways to us,
different than all of the representations that are involved in non-visual experiential
awareness, and different than all of the conceptually structured representations that
science makes available. Let us say that this representation represents the character-
istic phenomenal yellow. Now of course, the mere fact that we use a special repre-
sentation to keep track of phenomenal yellow could not by itself give rise to an
abiding impression that we are aware of a characteristic that is distinct from all other
characteristics when we are aware of phenomenal yellow. This merely creates the
possibility of such an impression. The impression itself arises from our sense that it
would be inappropriate to identify phenomenal yellow with a characteristic that we
grasp via some other representation.
We are all familiar with the fact that there can be multiple representations of a
single characteristic, and we all avail ourselves, from time to time, of the option of
identifying a characteristic that we initially grasp via one representation with a
characteristic that we initially grasp via a different representation. Thus, for example,

3
The line of thought of the last two paragraphs undercuts both versions of Sellars’ grain argument—the
one that is concerned with the complexity of phenomenal objects (e.g., pains), and the one that is
concerned with the complexity of phenomenal properties. (See Sellars 1967.)
 VISUAL AWARENESS AND VISUAL QUALIA 205

we all identify the liquid that is presented to us in visual experience when we look at
lakes and rivers with the substance that experiment and theory lead us to conceptu-
alize as H2O. But as this case reminds us, when we identify a directly experienced
characteristic with one that science posits, we are obliged to appeal to the fact that the
characteristic is represented by distinct representations in explaining and justifying
the identification. Only in this way is it possible to account for the fact that it is
necessary to go beyond what experience itself reveals about the characteristics in
order to appreciate the real nature of their relationship.
Now we must ask: Is it appropriate to invoke the difference between a property-
qua-represented and a property-as-it-is-itself in determining whether phenomenal
yellow is identical with some other property? From the perspective of common sense,
the answer is a resounding “NO!” We are aware of phenomenal yellow in virtue of
participating in facts of the form x looksp yellow to y; and when we view such facts
from the perspective of common sense, we find no reason to suppose that they
constitutively involve representations. That is to say, folk psychology affords no
glimpse of the representations that are constitutively involved in our experience of
phenomenal yellow. We can only appreciate the existence of those representations
from a highly theoretical perspective. (Representationalism is a highly theoretical
position.) Accordingly, we do not see how it could be true that phenomenal yellow
might have a nature that is not revealed by experiential awareness. Since there is no
apparent distinction between phenomenal yellow-as-represented-by-experience and
phenomenal yellow-as-it-is-in-itself, it cannot possibly be appropriate to go beyond
what experience reveals about phenomenal yellow in assessing phenomenal yellow’s
relationship to other characteristics.
It may be useful to recast this line of thought in a somewhat different form.
Suppose that X is a characteristic that is revealed by experience, and suppose also
that for some reason it is deemed desirable to identify X with a prima facie different
characteristic Y. In order to explain and justify the identification, which involves
going beyond the impression of X that experience itself provides, it is necessary to
explain how it is possible to grasp X experientially without appreciating its identity
with Y. It is normally possible to provide such an explanation by invoking some sort
of appearance/reality distinction. Thus, we might distinguish between X itself and a
property that serves as the mode of presentation for X. Or, if there is no other
property that serves as the mode of presentation for X, as will be the case if our
awareness of X is direct, then we might distinguish between X-as-it-is-in-itself and
X-as-it-is-represented by an experiential representation. But, to repeat, we must
invoke some such contrast in order to explain why the identity of X with Y is not
revealed by experience itself.
Now let us turn to consider the special case of the characteristic phenomenal
yellow. Is it possible to identify phenomenal yellow with some other characteristic,
say C? If we are to do so, there must be a way of explaining how it is possible to grasp
phenomenal yellow experientially without appreciating its identity with C. This
206 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

means that we must invoke an appearance/reality distinction of some sort. But folk
psychology does not recognize a distinction between appearance and reality in this
case. It fails to register the representational character of our awareness of phenom-
enal yellow, and by the same token, it fails to support any ambitions that we might
have to identify phenomenal yellow with another characteristic.
This completes my account of how representationalism explains our impression that
phenomenal yellow is distinct from all other characteristics. Of course, in addition to
explaining that impression, it offers a perspective from which it is appropriate to reject
the impression as illusory. Unlike folk psychology, representationalism affirms the
representational nature of our awareness of phenomenal yellow.
Thus far I have made six claims. It may be useful to summarize them. First, I have
claimed that when an object x looksp F to an observer y, y is aware of x as having a
certain property, a property that is invoked by the locution “looksp F.” I will
henceforth speak of this form of awareness as experiential awareness, and I will say
that the properties that are objects of experiential awareness, the properties that are
invoked by predicates of the form “looksp F,” are appearance properties. Second,
I have claimed that experiential awareness is representational in character. It consti-
tutively involves a representation of an appearance property. Third, I have claimed
that experiential representations have distinctive properties that set them apart from
representations of other sorts. Thus, for example, unlike conceptual representations,
experiential representations are analog or quasi-analog in character. Fourth, I have
claimed that visual qualia are appearance properties, and that experiential awareness
is the form of awareness that puts us in touch with qualia. Fifth, I have claimed that it
is possible to explain the special metaphysical properties that qualia seem to us to
have, such as simplicity and intrinsicness, in terms of the special nature of experi-
ential representations. Thus, for example, it is possible to explain why it seems to us
that qualia are simple by appealing to the fact that experiential representations of
qualia do not represent them as complex—that is, by the fact that such representa-
tions do not encode information about their internal organization. Finally, I have
claimed that qualia are ultimately physical in character. They can be located within
our independently motivated catalog of physical properties.
It is clear that the task of explaining and defending these claims is quite large. In
the present chapter I will be concerned only with two of them—the claim that
awareness of qualia is essentially representational in character, and the claim that
qualia can be found within the space of physical properties. These are perhaps the
most important of the foregoing claims, but the others are important too. They are
defended elsewhere.4

4
See Hill (2009).
 VISUAL AWARENESS AND VISUAL QUALIA 207

III
There are three main alternatives to representationalist theories of experiential
awareness. One alternative maintains that experiential awareness is a matter of direct
acquaintance with facts involving objects and certain special properties. On this view,
experiential awareness is simple and unstructured. It is a primitive cognitive relation
that can obtain between observers and certain facts. It occupies the ground floor with
respect to explanation and analysis. All that can be said of it is that it is a form of
awareness, and that the properties that are given to us by this form of awareness
are qualitative. A second alternative to representationalist theories is adverbialism.
Roughly speaking, adverbialism claims that when an object y looksp F to an observer
x, what is going on is that x is aware of y in a certain way. To be a bit more specific,
according to adverbialism, x is aware of y F-ishly. Thus, while adverbialism allows
that qualia exist, it maintains that they exist only as forms of perceptual awareness—as
ways of being perceptually aware of non-qualitative phenomena. That is to say,
adverbialism maintains that qualia are adverbial qualifications of an underlying
generic relation of perceptual awareness, a relation that agents bear to external
objects. The third alternative to representational theories of experiential awareness
is what might be called the doxastic theory. According to this view, experiential
awareness is propositional and doxastic in character, and therefore necessarily
involves some sort of conceptualization. To be aware of a quale is to token a
representation of a property, but the representation in question is fully conceptual
in character. All knowledge of qualia takes the form of judgments.
I think there are strong reasons for preferring representationalist accounts of
awareness to accounts that are based on acquaintance. Thus, in the first place,
representationalist accounts give us some hope of being able to account for the
ways that objects of awareness appear to us. In particular, what is especially relevant
to our concerns, they afford some hope of our being able to account for the ways that
the properties that we call “qualia” appear to us. These properties seem to us to be
intrinsic, seem to be simple, or at least to be analyzable into simple components, seem
to support relations of qualitative similarity and qualitative difference, and so on. As
noted earlier, representationalist theories of awareness have some promise of
explaining why certain objects of awareness seem to us to have these characteristics.
In addition to providing a foundation for explaining our impressions concerning
the objects of awareness, representationalist theories provide a basis for answering
certain key questions about the nature of awareness itself. At any given time one is
aware of certain objects and characteristics and not of others. Why is one aware of the
items that one is in fact aware of? Why isn’t one aware of the others instead? Further,
why is the scope of awareness broad at some times and not at others? Why does the
resolution of awareness change when it does? What is the difference between
attentive awareness and more casual forms of awareness? We are not yet in a position
to give detailed answers to questions of this sort by appealing to properties of the
208 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

system of representation that supports experiential awareness, but elementary reflec-

tion shows that it is possible in principle to answer them in this way. Thus, for
example, it is clearly possible to explain differences in the scope of awareness by
appealing to differences in the contents of representations: the more inclusive the
content of the representations that are involved in awareness, the broader the scope
of awareness. It is also possible to explain differences in resolution in terms of
differences in the contents of representations. But theories that invoke acquaintance
in their accounts of awareness are ipso facto precluded from answering questions
about how awareness works. Acquaintance is supposed to be simple and fundamen-
tal. If this is true, then facts involving acquaintance are in an important sense brute
facts. They cannot be reductively explained.
In view of these considerations, it is clear that representationalist theories of
experiential awareness are overwhelmingly superior to those based on acquaintance
in point of explanatory power. This is a good reason for preferring them. Another,
related reason is that they provide us with a simpler picture of the mind. They enable
reductive explanations of otherwise recalcitrant facts, thereby substantially reducing
our ontological commitments and streamlining our account of nomological
relationships.
I turn now briefly to adverbial theories of experiential awareness. According to
these theories, it will be remembered, qualia exist only as forms of experiential
awareness—as ways of being experientially aware of non-qualitative phenomena.
I see no merit in this suggestion. Adverbialism denies that there is such a thing as
experiential awareness of qualia. Instead, it maintains, qualia are ways of being
experientially aware of other things. But this claim fails to acknowledge a key fact.
Surely we are aware of qualia. We are not simply ignorant of them, as we would be if
there was no such thing as awareness of them. Clearly, for example, we know of the
existence of pain, and we know of it by virtue of being aware of it. Hence, adverbialism
must allow that there is non-experiential awareness of qualia of some sort. Now this
form of awareness will have to be explained in some way. But it can’t be explained
adverbially, on pain of regress. An explanation will have to appeal to acquaintance, or
to non-conceptual representations, or to judgments. Thus, in the end, either adver-
bialism is unable to account for our knowledge of qualia, or it collapses into one of the
other theories of qualitative awareness we have distinguished.
This brings us to the doxastic theory of experiential awareness. This view allows that
experiential awareness constitutively involves representations, but it differs from the
view I am defending under the label “representationalism” in that it claims that
the representations in question are conceptual through and through. It maintains
that to be aware of a quale is to make a conceptually informed judgment of a certain
sort.5

5
In the present chapter I use “representationalism” as a name for the view that awareness of qualitative
properties constitutively involves experiential or perceptual representations. But there is also a more
 VISUAL AWARENESS AND VISUAL QUALIA 209

We have already taken note of some facts that call the doxastic theory into
question. Thus, as we observed earlier, experiential awareness has all of the following
properties:
(a) It provides us with access to highly determinate forms of properties. (When an
object looksp blue to me, it generally looksp a highly determinate shade of
blue.)
(b) It provides more information about objects of awareness than can easily be
captured by linguistic or conceptual descriptions.
(c) It automatically assigns the properties it represents to locations in quality
spaces.
(d) It enables us to make extremely fine discriminations among the members of
domains, thereby giving us a sense of the density of the orderings that obtain
in the domains.
(e) It is associated with a variety of attention mechanisms, including mechanisms
that enable us to adjust certain aspects of experiential awareness, such as
resolution and figure/ground contrast.
Awareness that involves conceptualization and judgment lacks all of these features. It
is not true, for example, that conceptual awareness generally provides us with access
to highly determinate forms of properties. I can judge that something is blue without
judging that it has a highly determinate shade of blue. Nor does conceptual awareness
have property (b)—indeed, it would be contradictory to say that it does. It cannot be
true that conceptual awareness provides us with information that cannot be captured
in conceptual terms. Nor does conceptual awareness have property (c). A blind
person can have the capacity to form judgments about various shades of color
without knowing how to locate those shades within the color solid. Nor does it
have property (d). Anyone who has experiential access to colors will appreciate that
similarity orderings of shades are dense, or are at any rate very finely graduated. On
the other hand, a blind person, even if he has somehow managed to acquire an
immense number of color concepts, will not automatically have a sense of how finely
graduated such orderings are. Nor does conceptual awareness have property (e). To
be sure, we do speak of attention in connection with conceptualization and judgment.
For example, it is possible to attend in thought to a theorem in number theory. But
this sort of attention is different in kind than the forms of attention that are involved

inclusive sense of the term—a sense in which it is true to say that “representationalism” stands for any view
which claims that awareness of qualitative properties involves representations. On this more inclusive
sense, it is not required that the relevant representations be experiential in character. Now beliefs and
judgments are representational states, though the representations from which they are constructed are
conceptual rather than experiential. Accordingly, since the third proposal claims that introspective
awareness of qualitative states involves beliefs or judgments, it counts as a form of representationalism,
when the term is used with its more inclusive sense. I emphasize, however, that it does not count as a form
of representationalism when the term is used in the sense that is operative in the present chapter.
210 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

in experiential awareness. Thus, in the case of doxastic attention, there is nothing that
is strictly analogous to increasing resolution and figure/ground contrast.
This list of differences could easily be extended.6 Experiential awareness must be
distinguished from doxastic awareness.
We have now found strong reasons for setting all of the main alternatives to
representationalist accounts of experiential awareness aside. I will henceforth assume
that some sort of representationalist account is correct. More specifically, I will
assume that it is correct to say that awareness of qualia constitutively involves
representations of a sort that can appropriately be called experiential.

IV
I have been concerned thus far only with the nature of awareness of qualia. Before
going on to consider the nature of qualia themselves, I would like to emphasize the
point that a representationalist theory of awareness of qualia is sufficient by itself to
solve the metaphysical problem of qualia that we considered at the outset. Or at least,
a representational theory is sufficient provided that it can be developed in a satisfac-
tory way. However desirable it may be to supplement the theory with a positive
account of the nature of qualia, it is not necessary to do so in order to block the main
arguments for property dualism, and to bridge the gulf separating qualia from the
physical world that we considered at the outset.
The metaphysical problem of qualia derives from the fact that qualia seem to us to
have properties that all of their physical correlates lack. Thus, it seems to us that they
are simple and intrinsic. The physical properties that are correlated with qualia lack
these characteristics. It appears that representational theories of experiential aware-
ness have the capacity to solve this problem. A representational theory makes it
possible to draw an appearance/reality distinction with respect to qualia. On the one
hand, there are qualia-as-they-are-represented-by-experiential-representations. On
the other hand, there are qualia-as-they-are-in-themselves. Because it is possible to
draw this distinction, we are not obliged to say that qualia really are simple and
intrinsic. Moreover, it is plausible that, given an appropriate account of the nature of
experiential representations (an account which claims, among other things, that such
representations are analog in character), it will be possible to give a detailed explan-
ation of why qualia seem to us to have these characteristics. It will be possible to close
the explanatory gap. My present point is that it is possible to draw a distinction
between the appearance of qualia and the corresponding reality, and to develop a
theory of experiential representations that explains the appearances of qualia, with-
out having a theory of the nature of qualia. The mind-body problem arises because folk

6
I discuss several other features that distinguish doxastic awareness from experiential awareness in Hill
(2005). The paper is reprinted as Chapter 8 of the present volume. For additional discussion see Hill (2009,
Section 3.6).
 VISUAL AWARENESS AND VISUAL QUALIA 211

psychology provides us with an incomplete and otherwise faulty account of qualitative

awareness. Accordingly, it can be dealt with by developing a better theory of awareness.
It is not necessary to supplement such a theory with an account that explains exactly
what kind of properties qualia are. Relative to the mind-body problem, any positive
theory of the nature of qualia is a luxury.

V
This is not to deny that it is independently desirable to understand the nature of
qualia. It is clear that this is an important goal. I turn now to the task of constructing
such an account. Thus far, all that has been said about visual qualia is that they are
appearance properties—or in other words, that they are the properties we are aware
of in virtue of participating in facts of the form x looksp F to y. The goal now is to
consider the issues that arise when one attempts to go beyond this starting point.
There are four views about appearance properties that seem, at the outset, anyway,
to be genuine options:
First view: Appearance properties are properties of internal entities of some sort, such as
sensations or neural states of the visual system.
Second view: Appearance properties are objective, physical properties of external objects—
properties like objective shape and objective size.
Third view: Appearance properties are causal properties of the form being an external cause of
an internal mental occurrence with intrinsic property Q.
Fourth view: Appearance properties are viewpoint-dependent physical properties of external
objects such as subtending a visual angle of A degrees, undergoing angular displacement at rate
R, reflecting light of absolute intensity I in direction D, and reflecting light of spectral composition
C in direction D. That is, they are properties that external objects have in virtue of their
relations to vantage points, where vantage points are not internal to observers, but are rather
positions in physical space that observers can occupy.

These are not the only views about the nature of appearance properties that have
appeared in the history of thought, but they are certainly among the most prominent.
It is plausible, initially at least, that one of them must be correct. Unfortunately, as
I will try to show, they are objections to all of them. None of them, it seems, can be
true—or at least, it is not at all obvious how any of them can be true.
I will be concerned here only to evaluate these traditionally important views. I will
not go on to present and defend a fifth view. The task of finding an appropriate
account of appearance properties seems to me to be one of the hardest, and also one
of the most important, of the problems facing the philosophy of perception. My goal
in this chapter is just to explain some of the issues that must be addressed in any
search for a satisfactory view. I put forward a positive account of visual qualia in
other writings.7

7
See Hill (2009, pp. 156–68) and “The Content of Visual Experience” in the present volume.
212 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

Could the first view be correct? That is, could appearance properties be properties
of internal, mental entities of some sort? I think we can see that the answer is “no” by
reflecting on the transparency of visual awareness. Visual awareness is transparent in
the sense that the only objects that are presented to us in visual awareness are
external, physical objects. We are not aware of internal objects of any kind. Now
this implies that any properties we are aware of in visual experience must be
properties of external objects. Since we are not aware of any internal objects, we
could not be aware of any properties of internal objects unless we were aware of them
without being aware of them as characterizing objects, that is, as instantiated.
It seems absurd to say that visual experience presents us with properties that are
floating free of all objects, as mere possibilities of instantiation. Thus, transparency
provides us with a reason to say that appearance properties, and therefore qualia, are
properties of external objects.
Could appearance properties be objective, physical properties of external objects—
that is, properties like objective shape and objective size? It seems that the answer
must be “no.” Appearance properties are properties that we are aware of in virtue of
participating in facts of the form x looksp F to y. Such facts are essentially perspectival.
The appearances that objects present to us are constantly changing, but the objective,
physical properties of objects are relatively constant. In view of this fact, it seems
wrong to identify appearance properties with objective, physical properties.
This leaves us with two possibilities. One is that appearance properties are
properties of a sort that I will call Lockean. They are properties of the form being
an external cause of an internal state of the visual system with intrinsic property Q.
The other possibility is that appearance properties are what I earlier called viewpoint-
dependent properties of external objects—that is, properties that objects have in
virtue of relations to physically determined perspectives or vantage points, properties
such as visual angle and the spectral composition of reflected light. I will now sketch
an argument which suggests that the Lockean option is more problematic than the
viewpoint option. Then I will mention a consideration that seems to call the
viewpoint option into question.
As we’ve just noticed, the Lockean hypothesis asserts that visual experiences
represent causal properties of the form being an external cause of an internal state
of the visual system of type Q. Now a causal property of this sort is instantiated only
when the corresponding property Q is instantiated. Indeed, this holds as a matter of
necessity. But this makes it somewhat puzzling that a representation should stand for
a causal property of the given sort rather than the intrinsic property that it involves.
Clearly, advocates of the Lockean hypothesis owe us an explanation of how this is
possible. But what can they say? It won’t help to invoke the fact that a visual
representation encodes information about a causal power of the given sort, for as a
matter of necessity, the representation will encode information about the corres-
ponding intrinsic property as well. Nor will it help to say that it is of ecological value
to encode information about the causal property, for it is also be useful to encode
 VISUAL AWARENESS AND VISUAL QUALIA 213

information about the intrinsic property. Indeed, it is the fact that the property being
an external cause of an internal state of the visual system of type Q implicitly contains
information about Q that makes it possible for it to play a distinctive cognitive role.
In general, it is hard to see how the Lockean could come up with an explanation of
the required sort. To be sure, there would be no problem if it was appropriate for the
Lockean to claim that the visual system has the ability to represent the property being
an external cause of an internal state, together with the ability to combine a
representation of that property with representations of specific internal properties.
But the property being an external cause of an internal state is quite abstract, and it
would require considerable sophistication to represent it. We can of course represent
the property conceptually, but it seems a stretch to suppose that it is represented by
the visual system. So there is a rather substantial lacuna in the Lockean theory. This
isn’t necessarily a disabling flaw, but it is a ground for concern.
This brings us to the view that we experientially represent viewpoint-dependent
properties, such as the visual angles that objects subtend with respect to the nodal
point of the eye. Among other virtues, this view is in a good position to explain how
higher level visual cognition can put us in touch with objective properties. Thus,
there are well-known trigonometric laws linking visual angles to such features as
objective sizes and objective distances. For example, it is possible to recover the size of
an object from information about the eye height of the observer (the distance from
the observer’s eyes to the ground), the visual angle that the object subtends with
respect to the observer’s eye, and the visual angle determined by a pair of rays that
run respectively from the observer’s eye to the horizon and from the observer’s eye to
the base of the object.8 There is reason to think that observers have access to facts of
all these kinds, so it is plausible that they make use of them in arriving at represen-
tations of objective sizes, at least in some circumstances.
So it’s reasonable to consider the option of saying that visual qualia are viewpoint-
dependent properties. Unfortunately, it is not at all easy to specify viewpoint-
dependent properties that are capable of playing the desired role. Thus, the visual
angles subtended by external objects are the most obvious candidates to play the role
of apparent sizes, but the view that apparent sizes are visual angles faces a serious
problem. Roughly speaking, the problem is that apparent sizes are not strictly
proportional to visual angles. To see this, hold one of your thumbs about a foot
away from your eyes, and hold the other at a distance of two feet. When your thumbs
are seen from these two distances, the visual angle subtended by the nearer thumb is
twice as great as the visual angle subtended by the thumb that is farther away. But the
nearer thumb does not lookp to be twice as big as the farther thumb. On the contrary,
the two thumbs are very similar in apparent size. Considerations of this sort count

8
For precise formulations of relevant principles concerning visual angles (including the so-called
“horizon ratio” principle that is cited in the text), see Palmer (1999, pp. 232–3 and p. 321). For illuminating
philosophical discussion and additional references to the empirical literature, see Bennett (2011).
214 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

pretty decisively against the visual angles theory. Is it possible to find other view-
point-dependent properties that are correspond more closely to apparent sizes?
I think the answer is “yes,” but it is not easy to identify such properties. A search
of the literature reveals very few candidates.9
To summarize: There are problems facing the main traditional views concerning
the nature of appearance properties. At present, there is no clear answer to questions
about the metaphysical nature of these properties—nor, by the same token, to
questions about the metaphysical nature of visual qualia.

VI
In this chapter I have been concerned to effect a rapprochement between the realm of
matter and the realm of mind. More particularly, I have tried to show that represen-
tationalism gives us reason to think that it is possible to bring qualia into the fold of
physical properties. Representationalism promises to solve the metaphysical conun-
drum that Kim has called the problem of “mental residue.” But while representa-
tionalism clears the way for a physicalist account of qualia, it does not itself provide
such an account, because it does not itself specify physical properties that can play the
role of qualia. Unfortunately, the task of figuring out just which physical properties
play this role presents some substantial challenges.

Appendix
In this Appendix I will consider an objection to representationalist accounts of qualitative
visual awareness that many people have found impressive. According to the objection I have in
mind, we can conceive of situations in which agents are aware of qualia but in which the agents
lack representational properties. The paradigm of such agents is Swampman, a creature
brought into existence by a very unlikely but nonetheless possible quantum fluctuation in
swamp gas.10 It is stipulated that Swampman is just like a normal human agent in all physical
respects, including all the respects with which neuroscience is concerned. Now intuitively, a
creature with a brain just like ours would also enjoy awareness of qualia. But representation-
alism may not be able to allow that this is so, for most of the standard accounts of perceptual
representation claim that representational content depends on there having been a correlation
between a representation and a represented property at some point in the past, whether during
the learning history of the individual or during the period when evolution was shaping the
visual system. In short, most standard theories of visual representation seem to entail that
Swampman has no states with representational content. If this is so, then, in combination with

9
See the references in note 7.
10
Swampman is part of the legacy of Donald Davidson. See Davidson (1987).
 VISUAL AWARENESS AND VISUAL QUALIA 215

representational theories of qualitative awareness, the theories in question yield the counter-
intuitive claim that Swampman lacks awareness of qualia.
Now there are theories of representation that differ from the one that I have just charac-
terized. Thus, for example, Michael Tye defends a theory according to which representation of
a property P by a state S depends only on there being a disposition on the part of S to
accompany P, and a disposition on the part of P to accompany S. That is, on Tye’s theory, it
need only be true that if either of P and S were instantiated, the other would be instantiated as
well. Tye explains this account of representation in Tye (2000), and puts it to use in answering
the Swampman objection. As he points out, the theory makes it possible for him to say that
Swampman is in representational states after all.
I think Tye’s theory of representation has some important virtues, but I prefer views like
those of Millikan (Millikan 1984) and Dretske (Dretske 1986), according to which the contents
of visual representations are part of our biological endowment, having been fixed by evolu-
tionary processes in the remote past. So I wish to ask: if this is indeed the right way to think
about visual contents, is Swampman a counterexample to representational theories of quali-
tative visual awareness? It is clear that if a biological account of visual content is correct, and we
accept a representational theory of qualitative awareness, then Swampman cannot be said to be
aware of qualia. It is also clear that this result conflicts with an intuition: we can clearly and
distinctly conceive of Swampman, and because of this, we have a vivid impression that Swamp-
man is objectively possible. Should this conflict lead us to reject representational theories?
I begin by noting that conceiving is not by itself a reliable test for objective possibility.11
To determine that a state of affairs is possible, one must form an internally coherent concep-
tion of it, but one must also check to make sure that it is compatible with the full range of
Kripkean a posteriori necessary truths. Hence, before we can affirm that Swampman is
objectively possible, we have to make sure that this claim is compatible with all of the necessary
truths about qualitative awareness. But of course, the representationalist believes that there are
necessary truths that preclude the claim. According to the representationalist, it is a necessary
truth that awareness of qualia is constituted by representations. Moreover, the representation-
alist warmly embraces various closely related doctrines, such as the thesis that experience of
qualia is necessarily constituted by representations, the thesis that seeming to be aware of qualia
is also necessarily representational in character, and the thesis that an object cannot lookp any
way to an observer unless the observer is visually representing an appearance property. Because
of a prior commitment to these doctrines, the representationalist will feel comfortable about
denying that Swampman is objectively possible. There could exist a creature with a human
brain and body that has no states with representational contents, but that creature would also
lack an immense range of psychological states, including all those that have to do with
awareness, experience, perception, and appearing.
At this point, one might be tempted to respond that it’s just obvious that Swampman can
have visual experiences like those of a normal human agent. But reflection shows that any
such claim would beg the question. Indeed, what is much worse, any such claim would conflict
with the conclusions of plausible arguments. As we have seen, there are arguments for

11
I argue for this view in “Imaginability, Conceivability, Possibility, and the Mind-Body Problem” and
“Conceivability and Possibility,” both of which are reprinted in the present volume.
216 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

representational theories of awareness. As long as those arguments stand, all claims that run
counter to such theories, including claims about possibility, are to be viewed with suspicion.
This is not to maintain that Swampman is a psychological zero. On the contrary, I wish to
acknowledge that Swampman is similar to a normal human agent in all psychological respects
that can be fully described in functional and information-processing terms. He is functionally
and informationally isomorphic to a normal human agent. It is just that the states that figure in
his functional architecture are devoid of representational content. They encode information
about the same properties and objects as the corresponding states in a normal agent, but unlike
the latter states, they do not represent those entities. (Encoding information about X is not
sufficient for representing X.) Accordingly, they cannot be said to support awareness of the
entities, or to constitute experience of them.
But here a question arises. Could the psychological similarities between Swampman and the
normal agent provide a basis for reinstating the worries about qualia with which the chapter
began? Prima facie, it might seem that the answer is “yes.” After all, if Swampman is
psychologically similar to a normal agent in point of information processing and functional
architecture, and can even be in the informational and functional states that support or realize
normal awareness, it would be appropriate to describe him as “visually registering” qualia, and
perhaps even as “visually discriminating” them. It might be thought that these cognitive or
quasi-cognitive relations to qualia would be sufficient to reinstate the mind-body problem.
Now I doubt that this is so. As I see it, registering information about a characteristic does
not offer enough of a foothold to our ordinary conceptual scheme to support the arguments
that provide motivation for dualism. But suppose that this is wrong. Suppose that the
similarities we have been reviewing provide an adequate basis for saying that Swampman is
“quasi-aware” of qualia, in some fairly robust sense, and that it is possible to couch arguments
for dualism in the language of quasi-awareness. Then, I suggest, it will also be possible to
construct a “quasi-representational” theory of quasi-awareness, with an accompanying dis-
tinction between “quasi-appearance” and reality. That is, in the unlikely event that the
psychological similarities are sufficient to allow us to reconstruct the poison, they will also
be sufficient to allow us to reconstruct the antidote.

References
Bennett, D. J. (2011). “How the World is Measured up in Size Experience,” Philosophy and
Phenomenological Research 83: 345–65.
Byrne, A. (2001). “Intentionalism Defended,” Philosophical Review 110: 199–240.
Davidson, D. (1987). “Knowing One’s Own Mind.” Proceedings and Addresses of the American
Philosophical Association 60: 441–58.
Dretske, F. (1986). “Misrepresentation.” In R. Bogdan (ed.), Belief: Form, Content, and
Function. Oxford: Oxford University Press, 597–610.
Dretske, F. (1995). Naturalizing the Mind. Cambridge, MA: MIT Press.
Hill, C. S. (2005). “Ow! The Paradox of Pain.” In M. Aydede (ed.), Pain: New Essays on its
Nature and the Methodology of its Study. Cambridge, MA: MIT Press, 75–98.
Hill, C. S. (2009). Consciousness. Cambridge: Cambridge University Press.
Kim, J. (2005). Physicalism, or Something near Enough. Princeton: Princeton University Press.
 VISUAL AWARENESS AND VISUAL QUALIA 217

Kim, J. (2006). Philosophy of Mind, 2nd edition. Boulder, CO: Westview Press.
Millikan, R. (1984). Language, Thought, and Other Biological Categories. Cambridge, MA: MIT
Press.
Palmer, S. R. (1999). Vision Science. Cambridge, MA: MIT Press.
Sellars, W. (1967). “Philosophy and the Scientific Image of Man.” In R. Colodny (ed.), Frontiers
of Science and Philosophy. Pittsburgh: University of Pittsburgh Press, 35–78.
Tye, M. (2000). Consciousness, Color, and Content. Cambridge, MA: MIT Press.
 12
The Content of Visual Experience*

I. Introduction
I am a representationalist about perceptual experience. I hold that perceptual experi-
ences are partially constituted by representations, and also by the representational
contents that those representations possess. Thus, as I see it, it is an essential fact
about a perceptual experience that it represents the world as being a certain way—as
containing objects that have certain properties and stand in certain relations. In this
chapter I will propose an answer to a basic question about the representational
contents of experiences—specifically, what is the nature of the properties that figure
in those contents? In other words, what is the nature of the properties that are given
to us in perceptual experience, the properties that are the most basic and immediate
objects of perceptual awareness? In discussing this issue I will focus on visual
experience, but much that I will say will generalize to other perceptual modalities,
especially hearing and touch.
Representationalism is not the only possible view about the nature of experience,
but it is supported by powerful considerations. One is that it provides the best
explanation for the fact that perceptual experiences purport to present external
objects and situations to us, and can therefore be said to be veridical or non-veridical.
One might initially doubt this, finding it more plausible to say that experiences do not
have proprietary contents but only vicarious contents—contents that they possess
derivatively because they give rise to beliefs and other propositional attitudes that
have conceptual contents. But this alternative explanation of the fact that experiences
have veridicality conditions is in the end much less persuasive than the hypothesis
that they have representational contents that they possess independently of their
causal and epistemic relations to concepts. To mention only one consideration,
perceptual experiences tend to have contents that are much more complex than
the contents of the conceptually based descriptions that the mind pairs with them,

* The chapter is an expanded version of a talk I gave at the 2010 meeting of the Eastern Division of the
American Philosophical Association. The talk was my contribution to a symposium entitled “Appearance
Properties.” Brian McLaughlin gave a related talk, and Susanna Siegel was the commentator. I’ve been
helped considerably by their presentations, and also by later discussions with them. Also, I have profited to
no small degree from conversations with David Bennett, Ned Block, Anil Gupta, and William Warren. The
chapter has been improved significantly by Justin Broackes’s comments on an earlier version.
 THE CONTENT OF VISUAL EXPERIENCE 219

and that are even much more complex than any conceptually based descriptions that
the mind is capable of pairing with them, given the operative constraints of time and
mental workspace. (It would, for example, take a prohibitively long time for the mind
to form a conceptual representation that articulates my current perception of the
visible surface of my thumb.) Another reason for thinking that perceptual experi-
ences have proprietary contents is that experiences have properties such as resolution
and scope that are best explained by supposing that they derive from the contents of
representations. If experiences have representational contents, it’s comparatively easy
to explain why my current experience of my thumb reveals the whorls of a thumb-
print. It reveals the whorls because it is constituted by a representation that repre-
sents the whorls. Other accounts of experience have trouble explaining facts of this
sort. Consider, for example, what a Russellian acquaintance theory would say about
experiential awareness of my thumb—that it has my thumb and perhaps myself as
constituents, but that it is otherwise fundamental, lacking representational structure
and internal structure of any other kind. How can this hypothesis explain the fact
that I am aware of the whorls of the thumbprint, but not of the microphysical
structure that is no less a property of my thumb? And how can it explain the fact
that if my thumb was farther away, or I was more nearsighted, I would not even be
aware of the whorls? The answer is that it can’t. Acquaintance theories attribute too
little structure to perceptual awareness to be able to explain even its most basic
features.1
I will not be able to say any more in defense of representationalism in the present
chapter. A reader who is skeptical about representationalism should view the chapter
as concerned with the following hypothetical question: If experiences had represen-
tational contents, how should those contents be described? What would be the best
account of their character?
As noted, my topic will be the properties that are represented by perceptual
experience. It may be useful to expand a bit on how I see this topic. There are a
number of quite different views about the nature of those properties. On some
accounts, they are viewpoint-dependent properties like looking small and looking
elliptical. On other accounts, they are basic physical properties like objective size and
objective shape. There are also ecumenical accounts on which representational
contents include properties of both of these kinds. And on still other accounts, the
properties in question include certain natural kind properties like pine tree, animate,
and human face. I will maintain here that the first of these views is the best: the
properties represented by experience are limited to viewpoint-dependent properties.

1
To be strictly accurate, acquaintance theories can explain properties like resolution and scope, but not
by appealing to the nature of awareness itself. Instead, they must explain facts about awareness by positing
irreducible causal laws that represent states of awareness as a function of psychological or neural structures.
I discuss this gambit at some length in chapter 3 of Hill (2009), maintaining that it vastly complicates the
ontology of cognitive science and multiplies psychophysical laws beyond necessity.
220 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

We represent objective physical properties as well, of course, and also natural kind
properties. But experiential representations are concerned only with viewpoint-
dependent properties. The states that represent objective physical properties and
kind properties are not experiential states. Unlike experiential representations, they
don’t have proprietary phenomenologies.

II. Appearance properties

Suppose that an object looks a certain way to you. In the interest of definiteness,
suppose that the object is an SUV, that it is far ahead of you on a highway, and that,
because of its being so far away, it looks quite small to you. I suggest that this complex
fact has an internal structure that can be captured by saying, first, that you are aware
of the SUV, and second, that you are aware of it as having a certain property, a
property I will call looking small. In other words, as I see it, in virtue of participating
in the fact the SUV looks small to you, you are aware of a property of the SUV that is
in some sense a size-related property, but that is distinct from its objective size. It is a
perspectival property, in the sense that whether it is perceptually available depends
on the perspective that an observer is occupying.2
What can be said in defense of this view? The short answer is that it seems to be
forced on us by the fact that the SUV’s looking small to you has a substantial
phenomenological dimension. The phenomenological dimension of an experience
is a dimension that is given to the subject in virtue of having the experience, so it is a
dimension that the subject grasps or appreciates. The subject is necessarily aware of
the phenomenological dimension of an experience. But what exactly is it that one is
aware of in virtue of being aware of the phenomenological dimension of an experi-
ence? Evidently it must be true that one is aware of an object of some sort and one or
more properties of that object. Now on the face of it, the only object that you are
aware of in virtue of the SUV’s looking small to you is the SUV itself, so it seems that
we must say that your being aware of the relevant phenomenology consists in your
being aware of a property of the SUV. But the property in question can’t be the
objective size of the SUV, because the SUV is objectively large. The natural thing to
say is that the property is a matter of how the SUV appears from your current
perspective; but to say this is to say that it is a property that the SUV has in virtue of
its relation to that perspective. It is a viewpoint-dependent property. If we try to be
more specific, we find it natural to say that the property in question is being small
relative to your perspective, or looking small from where you are.

2
As is well known, there are at least two different senses in which it can be true to say that an object
looks a certain way to a subject. Often, following Roderick Chisholm, philosophers call these two senses the
phenomenological and the epistemic senses of “looks.” In the present chapter, I will always use “looks” in its
phenomenological sense. (Readers not familiar with the distinction will find an explanation in Section I of
“Visual Awareness and Visual Qualia.”)
 THE CONTENT OF VISUAL EXPERIENCE 221

Generalizing, I would like to suggest that whenever a subject y participates in a fact

of the form x looks F to y, y ipso facto has an experience with a distinctive
phenomenology, where its having this phenomenology consists in y’s being aware
of a characteristic of the form looking F. I think of characteristics of this sort as
properties that external objects possess in virtue of their relations to the visual
systems of observers, and as depending on such factors as distance, lighting, angle
of view, and relative motion. As is natural, I call them appearance properties.
We should observe that there are several categories of appearance properties, one
for each of the perceptual modalities. Thus, in addition to visual perspectival
properties like looking small, there are also auditory perspectival properties, olfactory
perspectival properties, and so on. I will focus on the visual category in this paper.
We have just been considering an argument for appearance properties that makes
extensive use of the notion of phenomenology. It is both good and bad to rely on this
notion—good because the notion has a fairly solid basis in intuition and experience,
and bad because the notion is extremely vague. Because of this vagueness, it is
fortunate that it is possible to formulate a related argument for the same conclusion
that doesn’t invoke phenomenology. Instead it reminds us that there are three
different dimensions of visual awareness, each of which is concerned with a propri-
etary domain.
Suppose you enter a room that is painted ochre. Suppose also that the wall facing
you is illuminated by sunlight coming through a window, and that, due to the shape
of the room and the position of the window, the sunlight illuminates different
portions of the wall to different degrees. Perhaps parts of the wall seem yellow
ochre, others seem umber, and still others seem dark brown. An example of what
I have in mind is afforded by Hopper’s Sun in an Empty Room. (Images of this work
are readily available on the internet.)
Now in a case of this sort, it seems intuitively correct to say that you are aware of
properties of three kinds. First, you are aware of the objective color of the wall. That
is, you are aware that that the wall is yellow ochre. Second, you are aware of the
different degrees to which different portions of the wall are illuminated. And third,
you are aware of what it is natural to call apparent colors, where an apparent color is a
color that an item appears to have in virtue of the way in which lighting interacts with
the objective color. In the case of Hopper’s painting (which represents a yellow ochre
room in which the lighting is very uneven), the apparent colors include several
different shades of ochre, umber, and brown. I don’t want to insist here on any
particular analysis of apparent colors. All that is relevant to my current purpose is the
point that you are visually aware of them. They are transitory features of portions of
the wall that your experience reveals to you, features that you apprehend or notice,
though perhaps not explicitly and attentively.
It might be objected here that if an observer knows that a wall is uniformly ochre,
and also knows how much light is falling on each portion of the wall, then she
automatically knows what apparent shade of color any one portion of the wall is
222 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

manifesting. That is, awareness of apparent color is uniquely determined by aware-

ness of objective color and awareness of the distribution of light intensities. But if this
is so, then awareness of apparent color is implicit in the other two forms of
awareness. It isn’t an additional form.
To see that this objection is misguided, notice that the observer cannot know
whether the wall really is uniformly ochre, or know the location and intensity of the
light source, without knowing a great many purely local facts about apparent color.
Awareness of global facts about color and light derive from averaging information
about the apparent color of specific regions of the wall. If one is seeing a wall for the
first time, awareness of global facts could not be acquired in any other way. But of
course, if awareness of apparent color is more fundamental than awareness of
objective color and awareness of lighting, it cannot be reduced to those other
forms of awareness.
Further, even if one knew the relevant global facts, one couldn’t use them to derive
information about the apparent color of any specific region of the wall, for global
facts about the light source do not determine how much light is falling on a given
region. The region could be recessed more deeply than all of the other regions.
Equally, it could be that the light source is partially occluded, with the result that the
light falling on the given region is less intense than the light falling on other regions.
To arrive at knowledge of the objective color and degree of illumination of a specific
region, one has to combine one’s knowledge of the apparent color of that region with
hypotheses about the objective color and degree of illumination of other regions.
Again we find that knowledge of apparent color is more fundamental than know-
ledge of objective color and lighting.
Perhaps it will be useful to give another example. Suppose you are watching a disc,
perhaps a coin, which is rotating on an axis that is perpendicular to your line of sight.
In this case as in the preceding one, it is natural to say that you are aware of
properties of three kinds. First, you are aware of the objective shape of the disc.
You see that it is circular. Second, you are aware of the disc’s motion. You see that it is
rotating. And third, you are aware of changes in the shape that the disc presents to
you as it moves—that is, of changes in its apparent shape. At one point during each
rotation the disc presents the aspect of a circle; at another point it presents the aspect
of a straight line; and at every other point it presents an elliptical aspect, where the
eccentricity of the ellipse is determined by the degree to which the horizontal axis of
the disc is slanted away from your line of sight. I think we all find it natural to say that
these changes in apparent shape make a visual difference, and more specifically, a
difference that we appreciate or observe. It is plausible that we see the various
apparent shapes, just as we see the motion of the disc. Indeed, it is plausible to say
that we see the motion in virtue of seeing the various shapes.
In general, it seems that one is visually aware of properties of three different kinds:
first, one is aware of an objective, intrinsic property of an external object, such as an
objective color or an objective shape; second, one is aware of an environmental
 THE CONTENT OF VISUAL EXPERIENCE 223

condition that influences perception, such as lighting or angle of view; and third, one
is aware of a property of the object that depends on both the objective property and
the environmental condition, but that is not identical with either—it is color-related
but not an objective color, or shape-related but not an objective shape, or size-related
but not an objective size, et cetera. It would simply be wrong to say that our
perceptual awareness is exhausted by properties of the first two kinds. We must
recognize awareness of properties of the third kind in order to have an adequate
account of perception.
Now as the examples make clear, the properties in question are characteristics that
objects have in virtue of the ways things appear to us visually. In other words, they are
properties that we are aware of in virtue of participating in facts of the form x looks
F to y. Accordingly, they are the properties we have already agreed to call “appear-
ance properties.”
To summarize, we have been considering arguments for the claim that facts of the
form x looks F to y involve awareness of certain properties of the object x—specif-
ically, awareness of properties of the form looking F. We have found two such
arguments, one that is based on the changing nature of perceptual phenomenology,
and another that is grounded in the fact that our perceptual awareness of objects has
three different dimensions. Both arguments imply that appearance properties are
relational and viewpoint-dependent. That is, they are not intrinsic or objective
properties of objects, but rather relational properties that objects have in virtue of
the proximal stimuli that they occasion. They vary with distance, angle of view,
lighting, relative motion, and all of the other external and internal factors that have
been emphasized in traditional discussions of perceptual relativity.

III. A challenge from vision science

I rely heavily on perceptual relativity in the foregoing arguments, but I do not argue
for it systematically. In the present section I take a closer look at the grounds for the
doctrine, focusing on the following formulation of it:

(PR) The way that an object looks is always changing. Moreover, changes in the
way that an object looks tend to be more rapid and also more pronounced than
changes in the intrinsic, objective properties of the object. Accordingly, the
properties that an object looks to have are in most cases different than the
properties that it actually has. There is generally a contrast between the visual
appearance of an object and the corresponding reality.
Interestingly, (PR) seems to be held in higher esteem by philosophers than by vision
scientists. Many philosophers accept it, and those who don’t are normally willing to
allow that it has a certain plausibility. As far as I can tell, however, while vision
scientists will generally agree that large objects sometimes look small, and that
224 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

circular objects sometimes look elliptical, they are often reluctant to agree that
contrasts of these sort are pervasive, preferring to think that they are confined to
special contexts, such as situations in which one is learning to draw in accordance
with the principles of perspective, and situations in which one’s attention is grabbed
by a gross contrast between a novel appearance and familiar reality, such as the one
that occurs when one sees houses for the first time from an airplane.
We find a version of this view, for example, in a passage in Stephen Palmer’s
magnificent Vision Science. At the beginning of his discussion, Palmer joins a number
of his colleagues in distinguishing between two forms or modes of perception:
An interesting approach . . . is to posit the existence of two different modes of visual perception.
What we call the proximal mode reflects mainly the properties of the retinal image, or proximal
stimulus. What we will call the distal mode reflects mainly the properties of the environmental
object, or distal stimulus.3

To illustrate, someone who was operating in the distal mode might describe a truck
seen far down the road as quite large, while someone who had adopted the proximal
mode might report that it looks small. Now it is customary for vision scientists to
allow that the proximal mode can play a prominent role in certain special contexts,
and also that perceptual experience can sometimes be a blend of both modes, but it
seems to be widely held that the distal mode is dominant. The main reason for this
view is that the visual system generally applies constancy transformations to inputs,
thereby prescinding from the influence of such contextual factors as distance and
angle of view. Giving expression to the view, Palmer writes as follows:
[P]erception is dominated by the distal mode in most ordinary circumstances, such as locomo-
tion in the world, coordinating bodily interactions with objects, and making standard compari-
sons among objects. It makes sense to be strongly in the distal mode, for example, in trying to
choose the largest piece of cake on the desert tray. Otherwise you might simply choose the
smallest piece just because it is closest, thereby causing its retinal image to be the largest.4

In other words, since the success of our actions usually depends on the objective
layout of the world, it generally behooves us to commit to perceptual strategies that
are designed to provide information about the objective properties of external
objects. But to commit to such strategies is to operate in the distal mode.
To elaborate, I wish to consider the view that in most contexts, constancy
transformations operate prior to visual experience, and achieve full success in
converting representations of properties of retinal images into representations of
objective properties, leaving no residual trace of the effects of perspective. On this
view, awareness of viewpoint-dependent properties can occur, but it is restricted to
special contexts. It occurs when observers deliberately suspend the operation of
constancy transformations, because they are engaged in special tasks that require

3 4
Palmer (1999, p. 313). Palmer (1999, p. 314).
 THE CONTENT OF VISUAL EXPERIENCE 225

information about proximal stimuli, when there is not enough information about
such factors as distance and lighting to permit constancy transformations to do their
normal work, and when proximal stimuli grab one’s attention because they depart
significantly from the norm. As noted, it seems that this view has a certain following
within the vision science community. At all events, it enjoys a certain prima facie
plausibility, and it therefore presents a significant threat to (PR). In discussing the view,
I will focus on the following version of it, which it is natural to call the constancy thesis:

(CT) Due to the operation of constancy transformations, how an object looks

remains more or less the same across a broad range of contexts. Thus, large objects
tend to look large even when they are at some distance from the observer. Equally,
round objects tend to look round even when they are slanted relative to the
observer’s line of sight. To be sure, a large object can look small, and a round
object can look elliptical, but these things will happen only when an observer
interrupts normal processing to focus attention on proximal stimuli, or when the
information about such factors as distance is too impoverished for constancy
transformations to be effective.
When the view is formulated in this way, we can easily appreciate that it is incon-
sistent with (PR). By the same token, it is clear that it threatens to undercut any
argument for the importance of appearance properties in which (PR) plays a role.
Should we accept (CT)? I will argue that the answer should be “no.” It must of
course be granted that constancy transformations play a significant role in visual
processing, but that fact does not by itself provide much reason to accept (CT). Thus,
it could easily be the case that the transformations generally fall short of achieving full
constancy—it could be that underconstancy is the norm. I will maintain that there is
good reason to think that that this possibility is realized. Indeed, there are reasons of
two different kinds. Some are introspective, and others are experimental.5 I will
emphasize the introspective reasons here.
I am in my study, sitting in a chair. There is a window in the facing wall,
and through it I can see my neighbor’s chimney and parts of trees. Bookshelves
cover the rest of the wall; most of them are filled with books. In the immediate
foreground is my arm, which I have raised so that I can point out a book to my
daughter, who is also in the room. My feet are supported by a hassock that is a few
feet farther away from the plane of my face than my pointing hand. Now it is clear to
me that the apparent sizes of these objects are not what one would expect on the basis
of (CT). On the contrary, my hand looks much larger than the chimney; it
looks slightly larger than the books; and it looks about the same size as my feet. In

5
Here is Palmer’s summary of the relevant experimental data: “[P]erceptual constancy is seldom
complete. Indeed, most experiments show systematic deviations from accurate perception of objective
properties in the direction of proximal image matches.” (Palmer 1999, p. 314.) This assessment is
consistent with others by Irvin Rock, H. A. Sedgwick, and Mark Wagner. See Rock (1975, pp. 30–2),
Sedgwick (1986, sections 21.1–21.23), and Wagner (2006, pp. 130–1).
226 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

short, my current visual experience is very much like the one that Mach recorded in
his famous drawing of how the world looked to him from the vantage point of a chair
in his library.6

Further, it seems that what we find in this case holds as a general rule: the apparent
sizes of objects are influenced significantly by their distance from one’s eyes.
I emphasize that it is possible to accept this thesis while also recognizing the
importance of constancy transformations in the etiology of visual experience. Sup-
pose you are looking at two blocks of wood on a table. Both of the blocks are the same
height, but one is twice as far away as the other. The image that is projected on your
retina by the first block is twice as large as the image that is projected by the second
block. Accordingly, if the visual system made no attempt to smooth out differences in
input that are due to distance, the first block would look twice as big as the second.
But it doesn’t. It looks larger than the second block, but it doesn’t look twice as large.
To accommodate facts of this sort, we need to recognize that constancy transform-
ations operate prior to experience and account for a significant portion of its
character, but we also need to recognize that they fail to achieve full constancy. It
is clear that we can embrace both of these views without inconsistency.
An advocate of (CT) will no doubt object that if my hand looks bigger than
the books on a distant shelf, and looks about the same size as my feet, this is

6
Mach (1959).
 THE CONTENT OF VISUAL EXPERIENCE 227

only because I am actively attending to information about peripheral stimuli, thereby

reducing the amount of influence that the constancy transformations can exert. If
I had been attending to distal stimuli, as I normally do in everyday life, my hand
would have looked much smaller than the books, and would have also looked much
smaller than my feet. Further, an advocate of (CT) might try to enhance the appeal of
these contentions by reminding us that there are cells in the upper levels of the visual
system that respond to objects independently of the viewpoint from which they are
seen. A priori, it seems at least possible that the information encoded by these cells is
the only information that is access-conscious on normal occasions. It might be that
the cells in question have to be actively inhibited in order for information about
proximal stimuli to become accessible to the subject.
It must of course be conceded that I was attending to apparent sizes in preparing
the foregoing account of my experience of my room. But does it follow that the
apparent sizes would have been quite different if I had not been attending to
information about proximal stimuli? No, it doesn’t follow. Moreover, we can see
that this claim is very implausible by considering the phenomenology of attention.
Thus, introspection attests that the objects in my study looked to be pretty much the
same after I adopted what Palmer calls the proximal mode as they did when I was still
operating in the distal mode. There was no dramatic change in the appearances of
objects when I changed modes. It is always so. Consider what it is like to be driving
on a highway, and to decide suddenly to attend to the apparent sizes of the vehicles
around you. When this happens, is your experience of those vehicles radically
transformed? No. You experience them in a different way, to be sure, but it would
be clearly false to say that vehicles go from looking small to looking large. It would be
more correct to say that they go from inconspicuously looking small to vividly
looking small.7
In addition to being vulnerable to this fact about the phenomenology of attention,
(CT) is called into question by a continuity argument. Most advocates of (CT) will
grant that even a very large object, such as an SUV, looks small when seen from a
great distance. They will also allow that an SUV looks enormous when seen from six
inches away. Their view is that constancy transformations prevail only when objects
are at “normal” viewing distances. It is thought, however, that they reign supreme
within that middle range. Now if this view is correct, there will inevitably be two
transformations in the rate at which the visual appearances change when an SUV is
driven towards a viewer at a constant velocity. When the SUV is still far away, its
apparent size will increase as it approaches. Then its apparent size will remain
constant for a considerable interval. Finally, the apparent size will start to increase

7
We know from the work of Marissa Carrasco and her associates that attention can cause changes in
apparent size, but the changes revealed by her work are comparatively small. See, e.g., Gobell and Carrasco
(2005, pp. 644–51).
228 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

again, probably quite rapidly, as it draws near. I can speak only for myself, for I know
of no relevant experimental data, but it seems plain that there aren’t two discontinu-
ities of these sorts in my experience of moving objects (or in the experiences I have
when I move towards or away from objects that are stationary). It never happens that
the apparent size of an object increases gradually, then becomes fixed, and then starts
to expand rapidly. Rather, apparent size increases gradually and continuously as
objects approach. Moreover, this is true whether I am attending to apparent sizes or
not. (I can be sure that it holds when I am not attending to apparent sizes because my
attention is never “grabbed” by a sudden increase in apparent size. In particular,
contrary to what (CT) predicts, it never happens that my attention is grabbed when a
fast-moving vehicle exits the “normal” viewing range and starts to expand rapidly.)
(CT) is also challenged by an occlusion argument. Let D1 and D2 be the near and
far boundaries that define the region of space in which objects count as being within
normal viewing distance—that is, the region of space with respect to which (CT) is
thought to hold. As long as the distance between D1 and D2 is sufficiently large to
make (CT) an interesting thesis, it will be possible for smaller objects located in the
closer areas of the region to occlude larger objects located in the farther areas of the
region. This would not be the case if (CT) were true.
To review, we have found that when we attempt to describe the ways that things
look to us, we come to appreciate that visual experience is characterized by perceptual
relativity. (PR) is confirmed. The advocate of (CT) is apt to dismiss such evidence, on
the grounds that it is a predictable consequence of a shift of attention away from its
normal and proper objects, the intrinsic, objective properties of external entities. But
we have seen that this reply fails to do justice to the phenomenology of attention, for
attention seems to enhance visual experience in certain respects, but not to transform
it, at least not to the degree that would be required for (CT) to be true. We have also
seen that (CT) implies variable and discontinuous rates of change for visual appear-
ances, and that this consequence of the thesis fails to square with visual experience.
Finally, (CT) is vulnerable to a third argument based on the ability of closer objects to
occlude larger objects that are located farther away.
This completes my case for a view that I will call the appearance thesis. According
to this view, the properties that we are aware of in visual perception, at the level of
visual experience, are appearance properties, where appearance properties are rela-
tional, viewpoint-dependent properties of external objects. More specifically, they are
relational, viewpoint-dependent properties that correspond to ways in which basic
sensible characteristics like sizes, shapes, and colors appear to observers. In addition
to making this claim about visual awareness, the appearance thesis also implies that
appearance properties figure prominently in types of visual experience that cannot be
said to involve awareness of properties—specifically, in illusory and hallucinatory
experience. We cannot be said to be aware of appearance properties when we are
undergoing illusions and hallucinations, but the appearance thesis implies that we
represent such properties as instantiated in such cases. Indeed, the thesis claims that
 THE CONTENT OF VISUAL EXPERIENCE 229

we always represent appearance properties as instantiated. The difference between

veridical and non-veridical perceptual experience is just that in the former case, the
appearance properties that are attributed by visual experience really are instantiated
by external objects.

IV. Other properties

Visual experience represents appearance properties. Does it also represent properties
of other kinds? More specifically, does it also represent the objective correlates of
appearance properties—objective sizes, objective shapes, objective colors, and the
rest? Further does it represent perceptible natural kinds such as the property of being
a pine tree or the property of being a human face?
The answer to this question very much depends on what we mean by “experience.”
On one conception of experience, visual experience is the realm of appearance: to
have a visual experience of an object is for the object to look some way to you—
to look small, or to look oval, or to look red. To put it another way, on the first
conception of experience, experience is exhausted by phenomenology, where phe-
nomenology has to do with the qualitative character of experience, and each quali-
tative characteristic corresponds to a way of appearing. On a second conception of
visual experience, a visual state is an experiential state if there is something it is like to
be in that state. This conception is different than the first, because it can be true that
there is something it is like to be in a state even if the state has aspects that aren’t
constitutively linked to ways of appearing. Consider, for example, the state of seeing a
ball. Intuitively speaking, there is something that it is like to be in this state. But there
are aspects of the state that aren’t concerned with the current visual appearance of the
ball. Only the facing surface of the ball can be said to appear to you, but you will no
doubt have expectations concerning the ball that go well beyond what is true of the
facing surface. For example, you will expect that you would continue to experience
convex curvature if you were to move around the ball, and you will expect that the
ball would roll if you were to push it. Neither of these expectations corresponds to an
appearance property that the ball is currently instantiating. Neither of them has an
attendant perceptual phenomenology. But still, one might say, these expectations are
experiential states. There is something that it is like to be in each of them, and this
suffices for them to count as experiential.
While it is clear that the first conception of experience precludes experiential
awareness of any characteristics other than appearance properties, it seems that
the second conception allows representations of objective sensible characteristics
to count as experiential. It may allow representations of certain observable
natural kinds to count as experiential as well. On the first conception, experience is
exhaustively constituted by awareness of the ways in which objective sensible char-
acteristics like sizes, shapes, and colors appear to observers. This clearly leaves no
room for independent experiential awareness of objective characteristics themselves.
230 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

(It is not as if there are two visual phenomenologies, one constituted by the ways that
objective characteristics look to observers, and another constituted by a form of
awareness of objective characteristics that is independent of how they look.8) Nor, a
fortiori, does this conception of experience leave any room for experiential awareness
of natural kinds, for kinds are individuated by objective properties, not by mere ways
of appearing. Awareness of appearance properties like looking yellow can only provide
evidence of membership in kinds. On the other hand, since the second conception of
experience does not restrict experience to ways of appearing, it allows for experiential
awareness of properties of other kinds. In particular, given that expectations concern-
ing parts of objects other than visible surfaces count as experiential on the second
conception, it is natural to suppose that representations of objective sensible charac-
teristics should count as experiential as well. Should we also suppose that the second
conception allows awareness of kinds to count as experiential? It is more difficult to
answer this question, due to the extreme vagueness of the notion what it is like, the
notion that the second conception uses to explain experience. But at first sight,
anyway, there is room for an argument that a rational reconstruction of the second
conception would allow for experiential awareness of natural kinds.
I strongly prefer the first conception of experience because I regard the vagueness
of the second conception as disabling. The notion of an appearance property is also
vague, but not to the same degree. It is tied to other notions, such as the notion of an
object’s looking a certain way to a subject, that are deeply embedded in folk
psychology and folk epistemology. These notions are known to earn their keep in
our explanatory and predictive practices, and in our epistemic norms. The notion of
what-it-is-likeness isn’t anchored in folk theories in the same way. Suppose you are
listening to a symphony. There is something it is like to hear the symphony, but what
exactly does this what-it-is-likeness involve? Certainly it has a phenomenological
dimension: your experience of the sounds of the orchestra is part of it. So are your
expectations concerning the notes that will immediately follow the present ones. But
what about your appreciation of the fact that the piece will end soon? Or the fact that
you are poised to answer questions about the relationship between the current
performance and other performances of the same piece that you have heard? As
I see it, there are no answers to these questions. One could of course adopt measures

8
I am in agreement here with Jesse Prinz. Here is what he says about such double phenomenologies
in Prinz (2012, p. 74):
Some people insist that they have conscious experiences corresponding to categorical representations, and
I just can’t find them in my own experience. When I look at a chair, try as I may, I only see a specific chair
oriented in a particular way. When I look at a coin, I just experience an ellipse. There is something bizarre
about saying that one experiences circularity of a tilted coin in addition to the ellipse. If this were a visual
experience, one might expect it to involve two overlapping shapes, an ellipse in a circle, but vision is not like
that. Likewise, what would it mean to say that one experiences chairness? What kind of experience would
that be? A chair as seen from no vantage point? A chair from multiple vantage points overlapping? A shape
possessed by all chairs? Phenomenologically, these options seem extremely implausible.
 THE CONTENT OF VISUAL EXPERIENCE 231

that would reduce the vagueness of the notion of what-it-is-likeness, perhaps by

sharpening the notion so that it comes to the same thing as the notion of a state that
is grounded in phenomenology in some intimate way. But I see little reason to think
that in its present form the notion is sufficiently determinate to play a useful role in
folk psychology or folk epistemology. A fortiori, I can’t see much of a future for it in
cognitive science. Accordingly, I will set it aside here and focus on the first concep-
tion. As I see it, this is the only notion of visual experience currently in our
conceptual lexicon that is in reasonably good shape.
In making this decision, I am in effect endorsing the view that the properties
represented in visual experience are exclusively appearance properties. Given this
endorsement, what should be said about awareness of objective properties? If it is
not experiential, is it at least visual? Can we be said to be visually aware of the true
size of an SUV, or the true shape of a rotating coin? It seems very plausible that
objective sensible characteristics like objective sizes and objective shapes are rep-
resented within the visual system itself, for it is known that there are cell popula-
tions in the visual system that respond to the objective characteristics of objects
across large ranges of viewing conditions. The same cell population may respond to
a coin even though it is seen from different distances and from different angles.
Moreover, it is natural to suppose that the activity of such populations supports
conscious awareness, at least if consciousness is understood to consist in the
availability of representations to such higher order cognitive agencies as the ones
responsible for planning, decision-making, and the fixing of belief. What about the
view that there is visual awareness of natural kinds? We know that the visual system
contains an agency that is especially adept at processing information pertinent to
human faces. It remains controversial whether this agency is biologically dedicated to
processing information about faces, or instead has a more general function, but let us
suppose that the former is true. And, further, let us suppose that this agency is capable
of producing representations of kinds like male human face and smiling human face.
Would this lend any plausibility to the view that the kinds in question are represented
at the level of experience, as opposed to being represented at a higher visual level? I
think not. Faces are three dimensional entities that are characterized by objective
properties. Representation of them emerges only at levels capable of detecting invari-
ances across multiple appearances.9 On the other hand, each particular experience of a
face can be analyzed in terms of the ways in which certain combinations of colors,
shapes, and sizes appear to us at particular moments. If we subtract awareness of such
appearances from experiences of faces, nothing would be left. There is no additional
facial phenomenology. To be sure, it seems likely that the facial recognition agency
contains subagencies that contribute to the processing of the appearances that are
characteristically associated with faces. As far as I can tell, however, this fact doesn’t

9
I part company here with Susanna Siegel. See Siegel (2010).
232 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

change the larger picture. These subagencies may be components of a more inclusive
agency that recognizes facial kind properties, but they are themselves just special-
purpose size and shape processors, dedicated to abstracting certain types of complex
appearance property from the input. Representing facial structure qua facial structure
calls for a grander design.10
Earlier I distinguished between two notions of visual experience—one that is
defined in terms of how things look to us, and another that is defined in terms of
what it is like to be in a state that has a visual component. I will conclude this
section noting a second distinction—that between a visual state that is itself fully
experiential and a visual state that is causally triggered by visual experiences. The
main claim I wish to make in the present section is that appearance properties are
the only characteristics that are represented by visual states that are themselves
experiential. One could I suppose use “experience” more broadly, in such a way
that a state that represents an objective shape counts as experiential, provided that
the state is always caused by experiences. But that is not how I am using the term
here. I mean to restrict the use of “experiential” to states that have proprietary
phenomenologies.

V. The metaphysics of appearance properties

In the present section I will try to look a bit more deeply into the nature of
appearance properties, focusing on apparent sizes, which I take to be the simplest
and most tractable case.
We know that apparent sizes are not proportional to objective sizes. If a brick is
seen from a distance of ten feet, it looks a bit smaller than a second brick that is seen
from a distance of five feet, even if the bricks have exactly the same objective size.
Seeking to accommodate this fact, philosophers have sometimes identified apparent
sizes with angular sizes, where the angular size of an object is the visual angle that the
object subtends with respect to the nodal point of a subject’s eye.11 It turns out,
however, visual angles are no better than objective sizes as candidates for the role of
apparent sizes. Here too there is a failure of proportionality. The angular size of a
brick that is ten feet away is only half as large as the angular size of the brick that
stands at a distance of five feet, but it does not look half as big as the latter. This is
because constancy transformations have played a role in shaping the subject’s
experiences of the two bricks. To a large extent, the transformations have reduced
the discrepancy between the two angular sizes. In sum, apparent sizes are not
proportional either to objective sizes or angular sizes, but rather represent quantities
that are located somewhere between these two extremes. To borrow a term from

10
I have been helped considerably in thinking about visual awareness of kinds by conversations with
Ned Block.
11
See, e.g., Harman (1997), especially p. 667, Tye (2000, p. 78), and Huemer (2001, pp. 120–2).
 THE CONTENT OF VISUAL EXPERIENCE 233

Robert Thouless, an early twentieth-century Scottish psychophysicist, apparent sizes

are a “compromise” between objective sizes and angular sizes.12
There are two points here that we need to keep in view. One is that apparent sizes
are the result of computations that are applied successively to information about
angular sizes in combination with information about other relevant variables, per-
taining especially to distance from the observer. This can also be put by saying that
apparent sizes are produced by applying various constancy transformations to the
sizes of retinal images of objects (in combination with other image-based informa-
tion), for the sizes of retinal images are equivalent to visual angles. The other point is
that the constancy transformations never achieve true constancy. There is never a
true proportionality between apparent sizes and objective sizes. Now as we noticed in
Section III, there are psychologists who maintain that when objects are seen from
normal viewing distances, we are only aware of their objective sizes, unless we make a
special effort to attend to appearance properties. But the first point is accepted by
a large majority of the philosophers and psychologists who study vision. It is just a
special case of the more general thesis that visual processing consists in applying
computable functions to input data. This thesis was embraced by many early
psychophysicists, such as Thouless, and it became orthodoxy as a result of Marr’s
work in the 1970s.13 In addition to the constancy transformations, the computable
functions of inputs include the transformations performed by the opponent process-
ing system and the transformations that infer shapes from data about motion.
I call apparent sizes and the other appearance properties that are produced by
applying constancy transformations to input data “Thouless properties” to mark the
intermediate character of these characteristics that was described so clearly and
forcefully by Thouless. We are very far today from being able to characterize these
transformations in detail. We aren’t at all sure how to specify them mathematically,
nor are we even able to specify the types of information on which they operate.
Answers to these questions must await future experimental work. But there is an
important question about Thouless properties that we can profitably consider now.
If we obtain representations of Thouless properties by applying computable
functions to characteristics of retinal images, how can it be that Thouless properties
are themselves properties of external objects? To see the force of this question,
consider a retinal image of a pencil that is held vertically, and let the image have
the height H. Suppose that certain functions f1, . . . , fn are applied to H on the way to
producing an experiential representation of the apparent size of the pencil. How
exactly does it happen that the experiential representation stands for a relational
property of the pencil? It is, after all, easy to construct an alternative hypothesis. Let
the property P be the property that a retinal image possesses if the image has height H
and the result of applying f1, . . . , fn to H and other relevant quantities has the

12
See Thouless (1931) and Thouless (1932).
13
Marr (1982). For a brief updating, see Prinz (2012, pp. 51–3).
234 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

numerical value N. We can suppose that the visual system routinely commits the
error of binding representations of properties of images like P to representations of
objects that have been assigned locations in three dimensional space, thereby erro-
neously representing those objects as instances of such properties. In other words, we
can suppose that the visual system projects computed properties of retinal images
onto external objects by combining representations of them with representations of
external locations. This is at least a coherent story. Moreover, it is compatible with
the transparency of visual experience. Even though, according to the story, the
properties that we are aware of in visual experience are properly speaking properties
of retinal images, we are aware of no objects other than external objects because the
relevant binding process is systematically erroneous.14
This projection hypothesis has a certain appeal, but there is an alternative story
about the visual processing of size information that seems to fit the known facts much
more comfortably. On this second account, information about locations in three
dimensional space enters into the processing of size information at a very early stage.
It must do so, because information about distance is needed for the constancy
transformations. Since they are integrated with information about locations at every
phase of constancy processing, representations of size acquire external reference at an
early stage. The visual system never has to worry about securing an external import for
Thouless properties, for the assumption that they are instantiated by external objects is
built into every phase of the process of constructing representations of them.15
What then is the general character of a Thouless property? An example is the
property P that an external object possesses just in case it is producing a retinal image
with a height H such that the result of applying f1, . . . , fn to H and other relevant
quantities has the numerical value N. P is a Thouless size. Alternatively, a Thouless
size can be seen as a property P* that an external object possesses just in case it is
subtending a visual angle V such that the result of applying the computable functions

14
I have flirted with projection hypotheses of this sort on and off for some time, paying special attention
to the hypothesis that representations of the intensities of pains and other bodily sensations are involved in
systematic binding errors. As is discussed at some length in “Locating Qualia: Do They Reside in the Brain
or in the Body and the World?” the intensities of pains are not strongly correlated with activity in the places
where pains themselves seem to be located, but rather with activity in the spinal cord and brain. One way to
deal with this fact is to suppose that representations of intensities are erroneously bound with represen-
tations of the peripheral sites where aversive activity occurs. I haven’t developed this view at any place in
my writings about sensory representation, but there is an excellent exposition of it in Pautz (forthcoming).
15
In chapters 8 and 9 of Burge (2010), Tyler Burge argues forcefully for the claim that visual
representations owe their objective reference to constancy transformations. I disagree with this. One of
my reasons is the fact that Thouless emphasized—constancy transformations reduce the viewpoint-
dependency of perceptual inputs but they do not eliminate it. Another reason is closely related to the
objection to projectivism that we’ve just been considering. A number of the most important constancy
transformations require information about locations in objective space as inputs. Others require informa-
tion about other objective conditions such as the intensity of the light source. It could be argued that the
constancy computations of the sort Burge has in mind—shape, size, hue, brightness, etc.—are interdepend-
ent with computations concerning objective factors like location and lighting, but I see no case for the view
that the former computations are autonomously responsible for objective reference.
 THE CONTENT OF VISUAL EXPERIENCE 235

f1*, . . . , fn* to V and other relevant quantities has the numerical value N*. P and P*
are both legitimate physical properties of external objects, though they aren’t prop-
erties that a physicist would want to mention in giving an inventory of the charac-
teristics that play a role in physical laws.
Why would Mother Nature endow us with the ability to represent Thouless
properties? A possible answer is that Thouless properties can simultaneously play
two useful roles. One is that of simplifying categorization. Even though they fall
short of achieving perfect constancy, constancy transformations prescind away from
the idiosyncratic and volatile nature of peripheral stimuli, thereby reducing the amount
of work that the mechanisms responsible for categorization have to do. Extracting a
pattern from data is easier when instances of the pattern are more similar to each other,
and the same is true of recognizing new instances of the pattern. The other role is that
of incorporating information about the relationship between the viewing subject and
the object of awareness. Thus, for example, the respective Thouless sizes of a series of
objects that have the same objective size but that are located at different distances from
the observer provide the observer with information about relative distances. To be sure,
the visual system is in possession of absolute distance information prior to the
construction of experiential representations, but it doesn’t follow that that information
is fully represented in experience. It is possible that the conscious portion of the visual
system has to rely on “pictorial” information in judging depth.

VI. Looking backward

We have arrived at three conclusions. First, there is the appearance thesis—the thesis
that experience is concerned with appearance properties, where appearance proper-
ties are relational, viewpoint-dependent properties of external objects. Second, there
is the claim that experience is not concerned with any properties other than appear-
ance properties. The objective characteristics of objects are registered by post-experi-
ential, amodal representations. And the third conclusion is that appearance properties
can be partially characterized by saying that they are Thouless properties—properties
that are computed from intrinsic properties of peripheral stimuli, and that are
intermediate between such properties the objective properties of external objects.

References
Burge, T. (2010). Origins of Objectivity. Oxford: Oxford University Press.
Gobell, J. and Carrasco, M. (2005). “Attention Alters the Appearance of Spatial Frequency and
Gap Size,” Psychological Science 16: 644–51.
Harman, G. (1997). “The Intrinsic Quality of Experience.” In N. Block, O. Flanagan, and
G. Güzeldere (eds.), The Nature of Consciousness. Cambridge MA: MIT Press, 663–75.
Hill, C. S. (2009). Consciousness. Cambridge: Cambridge University Press.
236 PART III : A REPRESENTATIONALIST THEORY OF EXPERIENCE

Huemer, M. (2001). Skepticism and the Veil of Perception. Lanham, MD: Rowman and
Littlefield.
Mach, E. (1959). The Analysis of Sensation. New York, NY: Dover Publications.
Marr, D. (1982). Vision. San Francisco, CA: Freeman.
Palmer, S. E. (1999). Vision Science: Photons to Phenomenology. Cambridge, MA: MIT Press.
Pautz, A. (forthcoming). “The Real Trouble for Phenomenal Externalists.” In R. Brown (ed.),
Consciousness Inside and Out. Berlin: Springer.
Prinz, J. (2012). The Conscious Brain. Oxford: Oxford University Press.
Rock, I. (1975). An Introduction to Perception. New York, NY: Macmillan Publishing
Company.
Sedgwick, H. A. (1986). “Space Perception.” In K. R. Boff, L. Kaufman, and J. P. Thomas (eds.),
Handbook of Perception and Human Performance, Volume I: Sensory Processes. New York,
NY: Wiley, 21(1)–21(57).
Siegel, S. (2010). The Contents of Visual Experience. Oxford: Oxford University Press.
Thouless, R. (1931). “Phenomenal Regression to the Real Object I,” Journal of Psychology
21: 339–59.
Thouless, R. (1932). “Phenomenal Regression to the Real Object II,” Journal of Psychology
22: 20–30.
Tye, M. (2000). Consciousness, Color, and Content. Cambridge, MA: MIT Press.
Wagner, M. (2006). The Geometries of Visual Space. Mahwah, NJ: Lawrence Erlbaum
Associates.
PART IV
Knowledge
 13
Process Reliabilism and
Cartesian Skepticism*

I. Introduction
I will be concerned here with a theory of doxastic justification that was first proposed
by Alvin Goldman in 1979.1 This theory, which is generally known as process
reliabilism, has in recent years gone through a number of changes, and each change
has added significantly to its complexity and subtlety. However, for present purposes
it suffices to consider a skeletal version of the theory that is similar to the version that
was unveiled in Goldman’s original advertisement. This version runs as follows:
where S is any subject, and p is any proposition, S is epistemically justified in
believing that p if and only if (i) S believes that p, (ii) S has this belief as the result
of a cognitive process that is highly reliable, (iii) S is not in possession of any reliably
formed beliefs that imply either that p is false or that the process mentioned in (ii) is
unreliable, and (iv) S could not easily come into possession of any beliefs of the sorts
mentioned in (iii).2 As is usual, I am here using “epistemically justified” to stand for

* I have been helped enormously by Alvin Goldman’s comments on three (!) ancestors of this chapter,
by Stewart Cohen’s comments on the immediately preceding ancestor, and by conversations with Thomas
Senor. I have also benefited from the comments of audiences at Central Michigan University, Indiana
University, and the University of Pittsburgh. My research was supported by a N.E.H. Summer Stipend.
1
Goldman (1979).
2
As any reasonable theory of epistemic justification recognizes, it can happen that a belief seems prima
facie to be justified, but that the belief falls short of being fully or ultimately justified because its possessor
has other beliefs that call either the truth or the justification of the first belief into question. In short, a
prima facie justification can be annulled or “defeated” by other considerations. Clauses (iii) and (iv) in the
present formulation of process reliabilism are intended to accommodate this intuition about the defeas-
ibility of justification.
Now, according to (iii) and (iv), if the prima facie justification that is possessed by one belief is defeated
by another belief, then it must be the case that the second belief is produced by a reliable process. There are
philosophers, I think, who would prefer to weaken (iii) and (iv) so as to block this implication. That is to
say, they would prefer to replace (iii) and (iv) with weaker conditions that run as follows: (iii0 ) S is not in
possession of any beliefs (whether reliably formed or not) that imply either that p is false or that the process
mentioned in (ii) is unreliable, and (iv 0 ) S could not easily come into possession of any beliefs of the sorts
mentioned in (iii0 ).
This alternative view of defeasibility seems wrong to me, but I will not attempt to criticize it here. Suffice
it to say that nothing of substance in the present chapter turns on the claim that (iii) and (iv) are correct: if
they were replaced by (iii0 ) and (iv 0 ), it would be necessary to change the details of several of the arguments,
but the conclusions of the arguments would remain fundamentally the same.
240 PART IV: KNOWLEDGE

the form of justification that is most intimately bound up with knowledge. Further, as
is also usual, I am assuming that “reliable” is to be understood in terms of probability
of truth, where “probability” stands for some sort of objective likelihood. Thus, to say
that a cognitive process is reliable is to say that beliefs produced by that process are,
as a matter of objective, empirical fact, quite likely to be true.
Although process reliabilism enjoys considerable popularity at present, and is
much discussed in the contemporary literature, very little has been written about
the relationship between reliabilism and Cartesian skepticism. As a result, we have no
real grasp of whether reliabilism has anything of importance to teach us about
skepticism, nor of whether skepticism has anything of importance to teach us
about process reliabilism. The present chapter is my attempt to rectify this situation.
I will be concerned with several interrelated issues. After sketching a version of the
Cartesian skeptic’s argument in Section II, I will consider an objection to process
reliabilism that derives from certain properties of that argument. I will provide a
detailed formulation of this objection in Section III, and will attempt to answer the
objection in Section IV. Then, in Sections V and VI, I will discuss the claim that
considerations having to do with reliability are ultimately irrelevant to the skeptic’s
concerns. I will examine two arguments for this claim, and will endeavor to show that
both of these arguments fail. If my replies to the arguments are successful, then, by
the end of Section VI, the way will be clear for the reliabilist to insist that the skeptic
must address questions of reliability in conducting his investigations. In Section VII
I will conclude by arguing that the skeptic is unable to deal with questions of this sort
in a satisfactory way. If the skeptic is obliged to confront such questions, then, I will
maintain, the skeptical enterprise is bound to fail.

II. The skeptic’s argument

Let McX be a normal human agent, and let the Real World Hypothesis (RWH) be the
hypothesis that can be obtained by conjoining the following propositions: (i) McX
has always had a normal human body; (ii) the sense experiences that McX has
enjoyed in the past have been due to external phenomena of the sorts that McX
has believed to be causally responsible for them (that is, they have been due to such
common-sensical external causes as cats and dogs, and tables and chairs); and (iii)
the pattern specified in (ii) continues to obtain in the present. Further, let SH be the
skeptical hypothesis which claims (i) that unbeknownst to McX, McX’s brain has
been removed from his body and is being kept alive in a vat of fluids, (ii) that McX is
connected to an extremely powerful computer that monitors all of McX’s thoughts,
(iii) that the sense experiences that McX has enjoyed in the past have been due to
electrical pulses inside this computer, and (iv) that the pattern described in (iii)
continues to obtain in the present.3

3
As is well known, Hilary Putnam has developed an engaging and provocative objection to skeptical
hypotheses like SH. (See, for example, Putnam (1981, chapter 1.) Putnam’s objection lies outside the scope
 PROCESS RELIABILISM AND CARTESIAN SKEPTICISM 241

Having set the stage with these definitions, we can formulate an argument for
Cartesian skepticism as follows:
First premise: Where S is any subject and H1 and H2 are any two empirical hypotheses, S is not
justified in preferring H1 to H2 (i.e., in believing H1 and rejecting H2) unless S is in possession
of evidence that supports H1 more strongly than it supports H2.
Second premise: In determining whether McX is justified in preferring RWH to SH, it is
appropriate to set all non-sensory evidence aside, and to focus exclusively on facts involving
McX’s sense experiences and their purely sensory characteristics.
Third premise: Since RWH and SH differ only with respect to their claims about the causation
of McX’s sense experiences, and not at all with respect to their claims about the experiences
themselves, the purely sensory evidence that is available to McX supports RWH and SH to the
same degree.
Lemma: McX is not justified in preferring RWH to SH, and by the same token, McX is not
justified in believing RWH.
Fourth premise: Where S is any subject and p is any proposition, S does not know that p unless
S is justified in believing that p.
Conclusion: McX does not know that RWH is true.

This line of thought seems to capture the spirit of a number of standard formulations
of skepticism. For example, it seems to capture most of the intuitions that are
mobilized by the argument that appears in the First Meditation.
The first, third, and fourth premises of the argument all enjoy a considerable
amount of intuitive plausibility, but the skeptic has often felt obliged to say some-
thing in support of the second premise. The standard defense of this premise is a line
of thought that I will call the evidence argument. This line of thought runs as follows:
In order to count as evidence, extramental facts must be known to us. Now the senses are our
only avenues of cognitive access to extramental facts. Accordingly, extramental facts can count
as evidence only insofar as they are represented by the deliverances of the senses. It follows that
all of our extramental evidence is in a sense included within our sensory evidence. Thus, if we
are justified in holding a belief on the basis of extramental evidence, then we must also be

of the present chapter; but since I attempt in the present chapter to construct a quite different objection,
I would like to sketch very briefly my reason for being dissatisfied with Putnam’s objection.
It seems to me that Putnam’s objection commits a fallacy of relevance. As I understand his argument,
one of Putnam’s main premises is the causal theory of reference (which, very roughly speaking, asserts that
the reference of terms depends on their causal relations to extralinguistic phenomena), and the other main
premise is the disquotational theory of reference (which, again very roughly speaking, asserts that all
instances of the schema “ ‘T ’ refers to x if and only if x is (a) T ” are analytic). (Here I am following
Brueckner and Wright. (See Brueckner (1992) and Wright (1992).) As I see it, the claims that these theories
make about reference are radically different—so different that they could not possibly both be true unless
they involved different concepts of reference, different senses of “refer.” (As Edward Minar has put it to me
in conversation, while the causal theory implies that facts involving reference are deep facts, the disquota-
tional theory implies that such facts are superficial or trivial facts.) If this is right, then the argument
contains an ineliminable equivocation.
242 PART IV: KNOWLEDGE

justified in holding the belief on the basis of sensory evidence. And, by the same token, if we
wish to determine whether a belief is justified by extramental facts, it suffices to consider the
belief in relation to the available sensory evidence. It follows, of course, that in considering
whether McX is justified in preferring RWH to SH, the skeptic is entitled to set all of McX’s
extramental evidence aside. It is entirely appropriate for the skeptic to focus on McX’s sensory
evidence.

This argument will play an important role in later sections.

The present version of the skeptic’s reasoning is intended to do justice to the
common core of a large family of arguments, and it is therefore silent on many issues
that can arise in discussions of skepticism. Thus, for example, nothing is said about
the possibility that McX is justified in preferring RWH to SH by considerations
having to do with the supraempirical virtues (i.e., with such properties of hypotheses
as simplicity and explanatory power). The reader may feel that this omission limits
the plausibility of the argument; but even so, it seems likely that he or she will at least
grant that the argument makes a start, that it provides a foundation on which the
skeptic might naturally hope to build. After all, the version of the argument that
appears in the First Meditation says nothing of simplicity nor of any other supraem-
pirical virtue; but it is usually thought to be philosophically challenging.
As is common, I have refrained from using the terms “epistemically justified” and
“epistemic justification” in stating the argument. However, it is clear that epistemic
justification is the form of justification that the skeptic has in mind. Thus, the fourth
premise makes it explicit that the skeptic is concerned with a form of justification that
is intimately bound up with knowledge. Given the standard definition of “epistemic
justification,” it follows immediately that epistemic justification is the target of the
skeptic’s argument.

III. An objection to process reliabilism

There are prima facie grounds for thinking that process reliabilism is called into
question by the plausibility of the skeptic’s argument—that is, by the fact that the
skeptic’s reasoning is extremely seductive. Reliabilism implies that questions of
epistemic justification depend ultimately on contingent empirical facts having to
do with the reliability ratings of cognitive processes. Hence, it implies that questions
of epistemic justification are a posteriori. But the skeptic attempts to support a very
general claim about epistemic justification—viz., the claim that commonsense beliefs
about the external world are unjustified—by an argument that seems to be entirely a
priori. It appears, therefore, that if process reliabilism gives a correct account of
epistemic justification, then the skeptic has made an absurd mistake—a mistake that
should be apparent to anyone who has mastered the concept of epistemic justifica-
tion. But we all know that skepticism is not absurd. As we all know, generations of
philosophers and laymen have studied Cartesian skepticism, and it has won the
 PROCESS RELIABILISM AND CARTESIAN SKEPTICISM 243

grudging respect of almost all of these individuals. Usually, of course, they have
found it impossible to accept the skeptic’s conclusion, but for the most part they have
been profoundly impressed by the reasoning that leads to that conclusion.
I have found that this objection to process reliabilism does not always seem
persuasive when it is encountered for the first time, so I will repeat it here in
somewhat different language: According to reliabilism, the question of whether one
is epistemically justified in holding a belief is essentially an empirical question about
the reliability ratings of the cognitive processes that are responsible for the belief.
Hence, since the skeptic wishes to reject the claim that McX is epistemically justified
in believing RWH, he is under an obligation to confront the question of whether
McX’s cognitive processes are reliable. In particular, he is under an obligation to
assemble empirical data concerning actual degrees of reliability. However, as can be
seen by examining the foregoing argument, the skeptic makes no attempt to dis-
charge this obligation. Indeed, he says nothing whatsoever which suggests that he is
aware that he is dealing with an empirical issue.
Thus, if process reliabilism gives a correct account of epistemic justification, the
skeptic is guilty of an omission that is extremely damaging. Moreover, reliabilism
seems to imply that it is an egregious omission—an omission that would be noticed
by anyone who has grasped even the most basic facts about the nature of epistemic
justification. However, philosophers and laymen have studied the argument for
centuries without noticing the lacuna. Instead of complaining that the skeptic fails
to address the issue of reliability ratings, they have come away profoundly troubled
by the apparent strength of the argument. It appears, therefore, that process reliabi-
lism has a consequence that is sharply at variance with the facts.4

IV. Two replies to the objection

Although this objection can seem to be quite telling, careful reflection shows that it
can be answered. Indeed, there are two ways of answering it.
The first answer: This answer begins with an observation about the logical struc-
ture of the skeptic’s argument—specifically, the observation that, contrary to what
the objection maintains, the skeptic does address questions of reliability. To be sure,
he does not address such questions explicitly. However, the argument implies that
considerations of reliability can have no bearing on the points at issue, and that it is
therefore entirely appropriate to set them aside. Thus, the first and second premises

4
Here I am developing an argument that I presented earlier in Hill (1991, pp. 135–7 and 139–40).
As I have already indicated, I wind up retracting this objection to reliabilism in the present chapter.
Fortunately, this retraction does not seriously impair any of the main arguments of Sensations. Thus,
contrary to what I believed at the time of writing that earlier work, I was not there obliged to reject
reliabilism; the issues with which I was concerned turn out to be independent, to a very large extent, of the
controversy between internalists and externalists.
I hope to discuss this topic at greater length elsewhere.
244 PART IV: KNOWLEDGE

of the argument imply that McX’s justification for preferring RWH to SH must
derive exclusively from relations between beliefs and sense experience. Among other
things, this rules out the possibility that McX’s preference is justified by the reliability
of certain of his belief-forming processes.
If all goes well for the skeptic, then, the first and second premises of the argument
will neutralize the perception that considerations of reliability must somehow be
relevant to the enterprise of assessing the justificactory status of McX’s beliefs. But
does it really seem likely that this could happen? After all, if the concept of epistemic
justification really is fundamentally reliabilist in character, then we must have a fairly
strong disposition to see questions of justification from a reliabilist perspective. Does
it really seem that a disposition of this sort would be inhibited by the first and second
premises? Our answer to this question must depend on the degree of plausibility that
we suppose these premises to possess. Now as we saw, the first premise enjoys a
considerable intuitive appeal.5 However, as we also noticed, the second premise is at
first sight somewhat problematic. If it is able to play a role in neutralizing a fairly
deep-seated disposition, it must owe this ability to the evidence argument (that is, to
the line of thought that we considered in the third paragraph of Section II).
How about it? Does the evidence argument have an appropriate amount of
plausibility? I am not of course in a position to speak for the reader, but I myself
find the argument to be extremely persuasive. To my mind, it is quite capable of
causing one to think that the skeptic has defined the area under dispute, and the
resources available to the contending parties, in a fair and accurate way, and that it
would therefore be a mistake to attempt to introduce new considerations into the
discussion. The argument is capable of causing one to think that the skeptic’s game
must be played by the skeptic’s rules.
The second answer: This answer consists in citing a line of thought which can
reasonably be supposed to play a role in shaping the form of debates about

5
Actually, a contemporary philosopher might have deep reservations about the first premise. It implies
that all questions of epistemic justification are decided by considerations having to do with evidence. In
responding to this claim, a contemporary philosopher might maintain that questions of justification often
depend crucially on considerations having to do with the supraempirical virtues (i.e., on considerations
having to do with plausibility, simplicity, and explanatory power).
As against such philosophers, I hold that the connection between the supraempirical virtues and
skepticism can be challenged. Thus, as I see it, if we were to expand the area of our present concern so
as to include considerations having to do with the supraempirical virtues, the skeptic could respond by
questioning the relevance of such considerations to matters having to do with epistemic justification.
Consider the two virtues that skeptical hypotheses most conspicuously lack—plausibility and explanatory
power. (Both Descartes’ evil genius hypothesis and the contemporary brain-in-a-vat hypothesis are inferior
to the real world hypothesis in point of explanatory power. Neither of the former can be said to explain our
sense experiences. At most, they are mere gestures in the direction of explanations.) The skeptic might allow
that considerations having to do with either or both of these virtues are germane to questions of practical
or pragmatic justification. But he would maintain that there is no reason for saying that beliefs with greater
plausibility and/or explanatory power are thereby in a better position to count as instances of knowledge. For
reasons given in chapter 1 of this volume, I think this response ultimately fails, but it seems sufficient to shift
the burden of proof back onto the shoulders of the skeptic’s opponents—at least temporarily.
 PROCESS RELIABILISM AND CARTESIAN SKEPTICISM 245

skepticism. This line of thought, which I will call the circularity argument, runs as
follows:
SH implies that McX’s beliefs about extramental facts are false, and it purports to explain their
plausibility away. In view of these considerations, it is clear that McX’s beliefs about extra-
mental facts must be regarded as controversial in any context in which SH is up for assessment.
Because of this, if McX were to attempt to defend his preference for RWH over SH by
appealing to any propositions about extramental facts, he would beg the question. But a
defense that begs the question is not a real defense. Hence, propositions about extramental
facts cannot provide a justification for McX’s preference.

Of course, since propositions about reliability fall within the category of propositions
about extramental reality, the conclusion of this argument implies that propositions
about reliability can have no bearing on the question of whether McX’s preference is
justified. In effect, the conclusion tells us that the skeptic has every right to set
considerations having to do with reliability aside.
Like the evidence argument, I find the circularity argument to be highly persuasive.
Moreover, while I do not know of any published version of the skeptic’s reasoning in
which the argument figures explicitly, I have found that it has a strong tendency to
surface, sooner or later, in conversations about skepticism. My guess is that both the
skeptic and his audience have versions of the circularity argument implicitly in mind,
and that these implicit versions are partly responsible for the fact that there is a
tendency to ignore considerations of reliability in discussions of skepticism.
To summarize: According to the objection stated in Section III, the skeptic
completely overlooks the relevance of considerations of reliability to the question
of whether McX is justified in preferring RWH to SH. Reflection shows that this
claim is highly questionable. The skeptic may not make explicit mention of reliability;
but there are two plausible lines of thought which suggest that he is entitled to set
considerations of reliability aside, and it is reasonable to suppose that he is counting
on these lines of thought to neutralize the reliabilist perceptions of his audience.

V. Why the evidence argument fails

We must now turn to consider whether the evidence argument and the circularity
argument are ultimately successful. Do they show that considerations of reliability
are irrelevant to the skeptic’s main concerns, or do they fall short of establishing this
claim? I will discuss the evidence argument in the present section and the circularity
argument in Section VI.
Now as I see it, the evidence argument is free from internal flaws; as far as it goes, it is
a successful argument. Reflection shows, however, that it fails to serve the skeptic’s
purpose. Its conclusion is unable to bear the logical weight that the skeptic assigns to it.
The conclusion of the evidence argument can be represented by the following
proposition:
246 PART IV: KNOWLEDGE

Proposition A: If one wishes to determine whether McX is justified in preferring RWH to SH, it
is permissible to set all of McX’s extramental evidence aside, and to focus exclusively on McX’s
sensory evidence.

What the skeptic needs, however, is a proposition which implies the second premise
of the main skeptical argument. As the reader may remember, this premise runs as
follows:
Proposition B: In determining whether McX is justified in preferring RWH to SH, it is
appropriate to set all non-sensory evidence aside, and to focus exclusively on facts involving
McX’s sense experiences and their purely sensory characteristics.

Prima facie, it appears that Proposition B follows easily from Proposition A; but
closer inspection reveals that there is a substantial logical gap between the two
propositions. Proposition A tells us that we may focus on a certain set of evidentiary
phenomena, namely, McX’s sense experiences. On the other hand, Proposition B tells
us that we may focus on a certain set of facts involving those phenomena. Specifically,
it tells us that we may focus on facts whose only constituents are McX’s sense
experiences and purely sensory characteristics of such experiences. Thus, while
Proposition A leaves it open which facts involving McX’s sense experiences are
relevant to the task at hand (i.e., the task of determining whether McX’s perceptual
beliefs are justified), Proposition B makes a definite claim about which facts are
relevant to this task. In short, Proposition B is more restrictive than Proposition A. It
follows that Proposition B has more content than Proposition A, and by the same
token, it follows that Proposition B is not implied by Proposition A.
Perhaps it would be useful to characterize the gap between A and B in terms of an
example. Suppose that C is the characteristic being an experience of a sensory type
T such that experiences of type T are usually produced in normal human beings by
visual interactions with crows. Further, suppose that F is the fact that C is exemplified
by a certain one of McX’s sense experiences. Now Proposition A leaves it open
whether we are entitled to consider facts like F in determining whether McX is
justified in holding his perceptual beliefs. But Proposition B tells us that we may
ignore such facts. For Proposition B authorizes us to ignore all facts involving McX’s
sense experiences other than purely sensory facts, that is, facts that involve only
purely sensory characteristics. (C is a characteristic of sense experiences, but it is not
a purely sensory characteristic, because it is a characteristic that sense experiences
have in virtue of certain of their relations to extramental particulars. It is an
informational characteristic of sense experiences.)
But is the logical gap between A and B of any importance for our present concerns?
Is it connected in any way with the skeptic’s dialectical responsibilities? Does the
skeptic make life any easier for himself by obtaining an (invalid) authorization to set
aside all facts other than purely sensory facts? It is clear that the answer is “yes.” As
the example in the previous paragraph shows, Proposition B gives the skeptic leave to
 PROCESS RELIABILISM AND CARTESIAN SKEPTICISM 247

ignore facts involving the informational characteristics of McX’s sense experiences,

that is, the characteristics that McX’s experiences possess in virtue of their causal and
probabilistic relations to extramental objects and events. It also gives the skeptic leave
to ignore facts involving the truth-generating propensities of sense experiences, that
is, facts involving characteristics that sense experiences have in virtue of their
tendencies to trigger reliable cognitive processes. In general, Proposition B gives
the skeptic leave to ignore all of the sorts of facts that externalists believe to be
constitutive of the evidentiary status of sense experiences. It appears, then, that by
fallaciously passing from Proposition A to Proposition B, the skeptic is in effect
sidestepping the question of whether an externalist theory of evidentiary status might
be correct. But clearly, this question is of fundamental importance for the skeptic’s
project. If an externalist theory of evidentiary status is correct, then the skeptic
cannot hope to establish a skeptical conclusion by an a priori argument. If an
externalist theory of evidentiary status is correct, then, instead of proceeding along
an a priori path, the skeptic should set about amassing empirical data concerning the
informational characteristics and/or the truth-generating propensities of McX’s
sense experiences.6
It appears, then, that the skeptic’s reasoning is badly flawed. It involves a logical
error, namely, the error of thinking that Proposition B is a consequence of Proposition
A. Moreover, this error makes a difference. Because of the error, the skeptic is able to
sidestep a set of ideas that threaten to call his entire modus operandi into question.

6
Can the evidence argument be strengthened in such a way as to make its conclusion more useful to the
skeptic? I think not—not without strengthening the key premise in a way that would make it unacceptable
to externalists.
As it stands, the evidence argument contains a number of inessential (and possibly misleading)
embellishments. When these embellishments are stripped away, it becomes clear that the argument has
a purely empirical character. It begins with the claim that our information about extramental phenomena
comes to us by way of the senses, and it goes on to conclude that any information about extramental
phenomena that is available to the part of the mind that is responsible for perceptual judgments must
somehow be represented at the sensory level—that is, at the level of sense experience. These are both
empirical claims (or, if you like, metaphysical claims), and as such, they are silent on the question that is of
primary concern in discussions of skepticism. They carry no implications concerning the epistemological
question of whether it is the purely sensory characteristics of such experiences or their informational
characteristics and truth-generating propensities that determine whether perceptual beliefs are justified.
Now, if we were to change the argument in such a way that it did have a bearing on this question, and in
particular, in such a way that it resolved the question in the way the skeptic requires, then the argument
would in effect put forward an internalistic theory of epistemic justification. Any such argument would fail
to serve the skeptic’s purpose.
The skeptic wishes to give an argument that establishes the second premise of the main skeptical
argument. This means that he wishes to give an argument that has two characteristics: first, it should be
transparently valid; and second, it should be epistemologically neutral, in the sense of having premises that
would be found plausible when assessed from any sane epistemological perspective. This second charac-
teristic is essential. If the skeptic offered a line of thought that depended at any point upon a claim that is
rejected by a particular epistemological school, then, since the skeptic’s ultimate conclusion is extremely
counterintuitive, the line of thought would be seen, not as presenting a paradox, but rather as a reductio of
the given claim, and by the same token, as a defense of the given epistemological school.
Needless to say, it is not the skeptic’s intention to provide a defense of externalism.
248 PART IV: KNOWLEDGE

But now we must ask: Why is the error so seductive? This question can be partly
answered by appealing to the superficial linguistic similarities between Proposition
A and Proposition B. I think, however, that if we are to do full justice to the
seductiveness of the error, we must find an additional explanation. I will briefly
sketch a view that provides one.
According to the view that I have in mind, we are in possession of two concepts of
evidential support, one externalist and the other internalist, and we have a tendency
to confuse them. Very roughly speaking, the externalist conception can be explained
as follows: Where T is a sense experience and B is a belief, T can provide evidentiary
support for B if and only if there is a cognitive process P and a set S of experiences
such that (i) S includes T, (ii) S would cause P to produce B if its members were given
to P as inputs, (iii) the set that results from S by deleting T does not have this causal
power, and (iv) the outputs of P have a fairly high objective probability of truth.7 The
internalist conception is like this externalist conception in that it invokes a form of
probability, but it is different in that the form it invokes is “personal” or “subjective”
in character. According to this second concept, sense experiences (or propositions
about sense experiences) are capable of influencing subjective probabilities, and since
they are, it is appropriate to think of them as being capable of providing evidence for
beliefs. Now the capacities of sense experiences to influence the subjective probabil-
ities of beliefs are independent of relations between sense experiences and extra-
mental objects and events. (Thus, the subjective probabilities of McX’s beliefs would
be affected by his sense experiences in the same way in a world in which McX was a
brain in a vat as they are in the actual world, provided only that the experiences
themselves were the same in intrinsic nature and in order of occurrence.) Accord-
ingly, the second conception represents the evidentiary statuses of sense experiences
as independent of their informational characteristics and their truth-generating
propensities.
Let us assume that we really are in possession of two concepts of evidential
support, and that these concepts roughly fit the foregoing characterizations. Now
these concepts are similar in a fundamental respect: They both represent evidential
support as depending on the abilities of sense experiences to probabilify beliefs. To
be sure, the notions of probability to which they appeal are markedly different.
However, as is shown by the history of theories of probability, the human mind
finds it extremely difficult to distinguish between different ways in which beliefs can

7
This definition is only a first approximation to an adequate characterization of the externalist concept
of evidence. The main difficulty with the definition may be expressed as follows: While it explains what it
means to say that a sense experience provides evidential support for a belief, it does not explain what it
means to speak of degrees of evidential support. (By the same token, it does not explain what it means to say
that a sense experience increases the support for a belief, nor what it means to say that a belief receives more
support from a set of sense experiences than does another belief.) I believe that it is possible to characterize
a more serviceable notion of evidential support in externalist terms (and indeed in more or less reliabilist
terms), but the task of providing such a characterization lies outside the scope of the present chapter.
 PROCESS RELIABILISM AND CARTESIAN SKEPTICISM 249

be probabilified by experience. (It is arguable that the tools that are necessary for
drawing the appropriate distinctions in a fully adequate way did not become available
until 1931, when Ramsey’s “Truth and Probability” appeared.)8 Hence, it is reason-
able to suppose that there is a tendency to confuse the externalist conception of
evidential support with the internalist conception. But if we are disposed to confuse
the externalist conception with the internalist conception, then, surely, we are also
disposed to make logical errors in contexts in which the differences between these
two conceptions are relevant. Accordingly, since the context of appraising the
justificational status of McX’s empirical beliefs is a context of precisely this sort, it
is reasonable to suppose that we have a disposition that would cause us to fall readily
into the logical trap that is created by the superficial linguistic similarities between
Proposition A and Proposition B. In other words, on the view I am presently
sketching, it is reasonable to see the skeptic’s logical error as entirely natural.9, 10

8
Ramsey (1931).
9
Since the idea of subjective probability is internalist in character, it might be thought that it is at
variance with the spirit of reliabilism. Moved by this perception, the reader might come to feel that the line
of thought of the last two paragraphs actually has some tendency to undermine the reliabilist line that I am
defending in this chapter. Process reliabilism claims that our central concepts of epistemic appraisal can be
explained in reliabilist terms. How can this be true, it might be asked, if it is necessary to introduce concepts
that are at variance with the spirit of reliabilism in discussing the issues that are raised by the skeptic’s
argument?
In response to this worry, I will stress that I have claimed only that we have a concept of evidential
support that is grounded in the concept of subjective probability, and that we have a tendency to confuse
this concept with an objective, externalist concept of evidential support. Neither of these claims implies
that there is a connection between the concept of subjective probability and the concept of epistemic
justification. By the same token, neither claim poses any threat to process reliabilism. (Process reliabilism
offers objective, externalist analyses of the concept of epistemic justification and a handful of closely related
concepts of epistemic appraisal. Period. It is entirely possible to embrace these analyses while holding that
we are in possession of a concept of doxastic appraisal that is not objective and externalist, provided only
that it is not claimed that this concept has a logical connection with the practice of epistemic appraisal.)
10
It might be useful to restate the point of the last two paragraphs in somewhat different language.
It appears that the following proposition is conceptually true: Where E is any possible sense experience
and p is any proposition, E counts as evidence for p if and only if E is capable of contributing to the
probability of p. Further, it appears to be true that there are two concepts of probability: First, as Bayesians
have urged, there is a concept that stands for degrees of rational belief; and second, as objectivists have
claimed, there is a concept that stands for tendencies of stochastic processes to display reliable frequencies.
These observations suggest that there may well be two concepts of evidence—one having to do with degrees
of rational belief, and the other having to do with objective probabilities of truth.
I will hereafter assume that there really are two such concepts. I will call them the “subjective
conception” and the “objective conception,” respectively.
We must recognize, I believe, that human beings have a tendency to confuse these concepts. My
suggestion is that the plausibility of the skeptic’s reasoning is partly due to this tendency. If we could
fully appreciate the relevance of the objective conception of evidence to the skeptic’s claims, we would find
those claims to be quite unpersuasive. However, partly because of subtle linguistic traps, and partly because
of the forementioned tendency, our reaction to the skeptic’s reasoning is to a large extent determined by
the subjective conception of evidence. This is unfortunate, because the skeptic’s reasoning is much more
attractive when it is viewed from the perspective defined by the subjective conception.
Of course, in addition to claiming that we are in possession of two conceptions of evidence, and that the
plausibility of skepticism depends in part on a tendency to confuse these conceptions, I am also making a
third claim—the claim that it is the objective conception that is intimately bound up with the concept of
knowledge.
250 PART IV: KNOWLEDGE

VI. Why the circularity argument fails

Let us turn now to consider the circularity argument. Is there any reason to think
that, despite initial appearances to the contrary, this second argument is flawed?
Yes. The circularity argument depends crucially on a premise that can be stated as
follows: If S is obliged to beg the question in defending a belief, then the belief is
unjustified. After reflection, it becomes clear that this premise is false. Thus, accord-
ing to the standard definition of “begging the question,” one begs the question in
defending a belief if, in the course of stating one’s defense, one assumes the very point
that is to be established. This definition allows that it is possible to be justified in
holding a belief even though one would perforce beg the question in attempting to
defend it. To appreciate this, suppose that McV has a terrible headache, but that
McV’s employer, who has grounds for thinking that McV is a malingerer, doubts
this. Here it is natural to suppose that McV is fully justified in believing that he is in
pain; but, assuming that no machine for scanning McV’s brain is available, and that
he has no fever nor any other overt symptom whose force is uncontroversial, he is
incapable of getting his employer to share his belief. All he can do is assert that he is
in pain. Since any such assertion “assumes the very point that is to be established,”
McV has no choice but to beg the question in defending his belief.
Reflection shows that the notion of begging the question serves a rather different
set of interests than the notion of being unjustified. The former notion is an
essentially pragmatic concept: it is designed to help us with the task of evaluating
the persuasive force of arguments. Thus, when we deploy this notion, we do so
because we wish to criticize someone for defending a claim in a way that is bound to
be unpersuasive. We think it is bad to beg the question in defending a claim because a
defense that begs the question is inevitably as controversial as the point that the
defense is designed to support. On the other hand, questions of epistemic justification
are logically independent of questions of persuasive force. To be epistemically
justified in holding a belief is to be entitled to hold it. The question of whether one
is entitled to hold a belief is logically independent of whether one is in a position to
convince others that one is so entitled, and, a fortiori, logically independent of the
question of whether one is in a position to convince others to adopt the belief
themselves.
Any decent theory of epistemic justification will sustain this point, but it is made
especially vivid by process reliabilism. According to reliabilism, if a belief is produced
by a reliable process, then, normally, one is justified in holding this belief. But it is
clear that the question of whether a belief is reliably formed is independent of
whether one is able to persuade others that it is reliably formed, and, a fortiori,
independent of whether one is able to persuade others that the belief is true.
But is it genuinely plausible to say that the skeptic has in effect confused a question
of justificatory force with a question of persuasive force? I think so. After all, the
question of whether one is justified in holding a belief tends to arise only in situations
 PROCESS RELIABILISM AND CARTESIAN SKEPTICISM 251

in which the belief is at least somewhat controversial. Because of this, it is reasonable

to suppose that one’s sense of whether a belief is justified is largely determined by
one’s sense of whether one is in a position to dispel controversy about the belief.
Moreover, one knows from experience that one’s ability to dispel controversy
depends entirely on the persuasive force of the considerations that one is able to
adduce in defending the belief. In view of these facts, it seems that human subjects
would find it hard not to confuse questions of justificatory force with questions of
persuasive force.

VII. Additional problems for the skeptic

I urged in Section IV that the skeptic has two reasons for thinking it appropriate to
set considerations having to do with reliability aside—the evidence argument and the
circularity argument. We have found that neither of these reasons is ultimately valid.
Accordingly, it may well be the case that the skeptic is obliged to consider the
question of whether McX’s empirical beliefs are formed by reliable processes.
Let us suppose that process reliabilism is true,11 and that the skeptic does have an
obligation to consider this question. Does it seem at all likely that he could succeed in
showing that McX’s cognitive processes are not reliable? In particular, does it seem at
all likely that he could succeed in showing that McX’s perceptual processes are not
reliable? Well, since questions of reliability are empirical questions, the skeptic would
be under an obligation to appeal to empirical data. An appeal of this sort would of
course be something of an embarrassment to the skeptic, holding as he does that no
empirical beliefs are epistemically justified. But, what is worse, it seems that it would
be impossible for him to come up with empirical data of the required sort. Thus, pace
Sextus Empiricus, it seems that it would be impossible to find empirical data which
establish that perceptual processes are globally unreliable—or, at least, impossible to
find empirical data which establish that the perceptual processes of an average
individual like McX are globally unreliable. To be sure, there are occasional conflicts
between the deliverances of the senses; and science occasionally calls a prima facie
robust perceptual datum into question. But facts of this sort are hardly numerous
enough to establish the sweeping generalization that the skeptic needs. It seems that,
on the whole, the available evidence indicates that the perceptual processes of normal
individuals like McX are quite reliable.
It seems appropriate to conclude that reliabilism, so far from being refuted by
skepticism, as we initially feared, may actually help us to understand why skepticism
is ultimately unsatisfactory.

11
I offer some arguments in support of this assumption in Hill (1994).
252 PART IV: KNOWLEDGE

References
Brueckner, A. L. (1992). “Semantic Answers to Skepticism,” Pacific Philosophical Quarterly 73:
200–19.
Goldman, A. (1979). “What is Justified Belief?” In G. S. Pappas (ed.), Justification and
Knowledge. Dordrecht: D. Reidel, 1–23.
Hill, C. S. (1991). Sensations: A Defense of Type Materialism. Cambridge: Cambridge Univer-
sity Press.
Hill, C. S. (1994). “Two Cheers for Process Reliabilism,” Pacific Philosophical Quarterly 75:
12–28.
Putnam, H. (1981). Reason, Truth and History. Cambridge: Cambridge University Press.
Ramsey, F. P. (1931). “Truth and Probability,” in R. B. Braithwaite (ed.), The Foundations of
Mathematics and Other Logical Essays. London: Routledge and Kegan Paul.
Wright, C. (1992). “On Putnam’s Proof that We Are not Brains-in-a-Vat,” Proceedings of the
Aristotelian Society, n.s. LXXXXII: 67–94.
 14
Hawthorne’s Lottery Puzzle
and the Nature of Belief
Written with Joshua Schechter

Suppose that T is one of ten million tickets that have been sold in a fair lottery.
Clearly the probability that T will lose is extremely high. Suppose that I know these
facts. Suppose, too, that T will in fact lose. Given these assumptions, it is tempting to
conclude that I can know P, the proposition that T will lose. Certainly the present
authors are tempted to conclude this. In his brilliant book Knowledge and Lotteries,
however, John Hawthorne makes a strong case for the opposing view. He urges that
in everyday contexts speakers are reluctant, or even outright unwilling, to claim that
agents know propositions like P. And he buttresses this appeal to conversational data
with several powerful theoretical arguments.
Hawthorne uses the expression “lottery proposition” to stand for propositions that
satisfy two conditions: first, they have a very high degree of probability; and second,
there is an intuitive reluctance to say that they are known to be true. He maintains
that many propositions concerning lotteries meet these two conditions, and that a
variety of other propositions meet them as well, including the proposition that
Hawthorne’s car has not been stolen since he parked it in a certain lot this morning,
and the proposition that Hawthorne will not have a major heart attack in the near
future. It is clear that these propositions are highly probable. But also, according to
Hawthorne, there is considerable intuitive reluctance to claim that they are within
our ken. Moreover, Hawthorne maintains that this intuitive reluctance is well-
founded. In most cases, he says, it is wrong to claim that an agent knows a lottery
proposition.
We feel that Hawthorne’s work on lottery propositions is of the first importance.
He makes a case for his views about lottery propositions that is prima facie quite
persuasive. Moreover, his views have important consequences. For example, as he
shows, doubts concerning the knowability of lottery propositions provide motivation
for a fairly general skepticism about our ability to know propositions of other kinds.
There is also a further reason for valuing Hawthorne’s discussion. He develops his
views about lottery propositions in conjunction with some independently motivated
ideas about the metaphysical relationships between knowledge and such things as
254 PART IV: KNOWLEDGE

explanation, assertion, epistemic possibility, epistemic probability, practical reason-

ing, theoretical reasoning, and practical interests. These ideas have a strong appeal.
Moreover, if they are sound, then traditional epistemology has failed to appreciate
the metaphysical importance of knowledge. The ideas imply that knowledge plays
a metaphysically foundational role with respect to communication, action, and
reasoning.
Although we feel challenged and stimulated by Hawthorne’s discussion, and feel
grateful to him for bringing a wealth of new data and new forms of argumentation to
the fore, we find ourselves in disagreement with many of his conclusions. We doubt
that the intuitive reluctance to attribute knowledge of lottery propositions is as strong
as he supposes. We also think that it is possible to explain much of the reluctance that
actually exists in a way that neutralizes its probative value. Further, we have reser-
vations concerning his claims about the metaphysical relations linking knowledge to
other things. As we see it, these relations are both less systematic and less intimate
than Hawthorne maintains. In this chapter, we make a case for these dissenting
opinions. We also put forward several positive proposals, urging, among other
things, that assertion is governed by a Gricean constraint that makes no reference
to knowledge, and that practical reasoning has more to do with rational degrees of
belief than with states of knowledge.

I
In motivating the claim that thinkers ordinarily do not know lottery propositions,
Hawthorne relies on a line of thought that he calls parity reasoning (pp. 4–5).1 We
will begin by reviewing this part of his discussion.
Consider a fair lottery with 1,000 tickets. Suppose the ticket that will in fact win is
ticket #1000. Suppose that I possess ticket #1, and that tickets #2 through #999 are
salient to me for some reason, perhaps because each is owned by a friend of mine.
Suppose that I form a belief that my ticket will lose on the grounds that it has a 999/
1000 chance of losing. Suppose, too, that I have a lot of time on my hands, and that to
fill the hours I form beliefs about the tickets owned by my friends. In particular, for
each of the tickets #2 through #999, I form the belief that it will lose. My evidence for
each of these beliefs is exactly the same as my evidence for the belief that my own
ticket will lose. In each case, my belief is based on purely statistical grounds.
Given this scenario, suppose that I count as knowing that my ticket, ticket #1, will
lose. Since the grounds for each of my 999 beliefs are exactly the same, if one of these
beliefs counts as knowledge, presumably so do the rest. Thus, I know that ticket #2
will lose, I know that ticket #3 will lose, . . . , and I know that ticket #999 will lose. But

1
Unless otherwise indicated, page numbers in the text refer to Hawthorne (2004).
 HAWTHORNE ’ S LOTTERY PUZZLE AND THE NATURE OF BELIEF 255

if I know that each of these tickets will lose, then surely I am in a position to know
that all of these tickets will lose. This is an application of the principle that knowledge
can be extended by conjoining propositions that one knows. Moreover, surely I am
also in a position to know that ticket #1000 will win. This is an application of the
more general principle that knowledge can be extended by deduction.
However, as Hawthorne notes, these two consequences seem preposterous. We do
not think it possible to know by statistical reasoning that the first 999 tickets will
lose—after all, the chance that one of them will win is very high. We also do not think
it possible to know by statistical reasoning that ticket #1000 will win—ticket #1000 is
just as likely to lose as any of the rest. So something must have gone wrong in the
foregoing reasoning. The natural suggestion is that the problem lies with the sup-
position that I know that ticket #1 will lose. Thus, Hawthorne argues, there is
pressure to think that we do not ordinarily know such lottery propositions.
The argument can be generalized. For any lottery proposition—for example, for
the proposition that my car has not been stolen since I parked it this morning—an
analogous line of thought can be used to motivate the claim that I do not know it.
Indeed, Hawthorne suggests that similar reasoning is what explains the responses
ordinary thinkers have to lottery situations. When thinkers find it intuitively plaus-
ible that they do not know lottery propositions—and disavow such knowledge—this
is typically because they (perhaps implicitly) engage in reasoning similar to the
foregoing line of thought.
Hawthorne characterizes this general pattern of reasoning as follows:
Parity Reasoning. One conceptualizes the proposition that p as the proposition that one
particular member of a set of subcases (p1, . . . ,pn) will (or does) not obtain, where one has
no appreciably stronger reason for thinking that any given member will not obtain. Insofar as
one reckons it absurd to suppose that one is able to know of each of (p1, . . . ,pn) that it will not
obtain, one then reckons oneself unable to know that p. (p. 16)

According to Hawthorne, there are two reasons that thinkers may find it absurd to
suppose that they are able to know of each of (p1, . . . , pn) that it will not obtain: It may
be obvious that at least one of (p1, . . . , pn) will obtain, or it may be obvious that some
consequence of the individual claims (such as their conjunction) cannot be known.
Parity reasoning thus plays an important role in Hawthorne’s monograph. He
engages in it to motivate the claim that thinkers frequently do not know lottery
propositions. He also appeals to it to explain ordinary thinkers’ intuitions and
responses to lottery cases. If we can find a principled basis to reject this general
line of thought, that will go a long way toward allowing us to comfortably accept that
thinkers ordinarily know lottery propositions. And it will substantially decrease the
pressure to modify our general picture of knowledge.
Before we present our response to Hawthorne’s parity argument, however, we
should first note that there is something both plausible and illuminating in
Hawthorne’s discussion. It is appealing to think that parity reasoning is what explains
256 PART IV: KNOWLEDGE

the reactions thinkers sometimes have to lottery cases, when they find it intuitive that
they do not know lottery propositions and disavow such knowledge. However, this
reliance on parity reasoning does not show that such reasoning is correct. Indeed, we
believe that parity reasoning is fallacious.
It will not be surprising that we reject Hawthorne’s parity argument on the
grounds that both the principle that one’s knowledge can always be extended by
conjoining known propositions, and the more general principle that one’s knowledge
can always be extended by deduction, are false.
Hawthorne carefully states the general principle as follows:
Multi-Premise Closure (MPC). Necessarily, if S knows p1, . . . ,pn, competently deduces q, and
thereby comes to believe q, while retaining knowledge of p1, . . . ,pn throughout, then S knows
q. (p. 33)

The narrower principle can be stated analogously:

Conjunction Introduction (CI). Necessarily, if S knows each of p1 and p2, competently deduces
their conjunction, and thereby comes to believe their conjunction, while retaining knowledge
of each of p1 and p2 throughout, then S knows their conjunction.

The main difficulty with these principles is familiar: Conjunction—and deduction

from multiple premises more generally—can aggregate risk. A thinker may be
justified in believing each of a set of propositions to a high degree but not in believing
their conjunction to nearly as high a degree. For instance, using the probability
calculus as a model of graded belief—admittedly an imperfect one—if two proposi-
tions each have probability 0.9, their conjunction may have a probability as low as
0.8. If, as seems plausible, knowing p is compatible with p having an epistemic
probability lower than 1, MPC and CI will be subject to counterexamples.
The general phenomenon of risk aggregation is well-known from discussions of the
preface paradox. One version of this problem goes as follows: There are very many
propositions that I count as knowing. Such propositions include simple claims of
mathematics and logic; claims about myself, my environment, and my past experi-
ences; and so on. Consider the conjunction of all of these claims. It seems that even
were I cognitively able to competently deduce the conjunction from the individual
claims that I know, I would not be in a position to know the conjunction. For I know
that I sometimes—very rarely, perhaps—make mistakes. It should seem likely to me
that at least one of the relevant propositions is false. It would be the height of arrogance
to go on and infer the conjunction, knowing full well that there is a significant risk of
falsity. As it happens, since the conjunction is a conjunction of claims that I know, if
I were to draw the inference, I would infer a truth. But this would be a happy accident.
The belief would not count as genuine knowledge. CI and MPC therefore fail.
Of course, to address the parity argument, it is insufficient to simply reject MPC
and related principles. We must also explain their appeal. Why is it that we find it
intuitive that, as Hawthorne puts it, “one can add to what one knows by deduction
 HAWTHORNE ’ S LOTTERY PUZZLE AND THE NATURE OF BELIEF 257

from what one knows” (p. 46)? Absent an answer to this challenge, the response to
the parity considerations is incomplete.
It will not be possible for us to address this issue until after we have developed our
general picture of knowledge, communication, and reasoning. Let us therefore put
aside this question for now and consider some of the striking claims Hawthorne
makes about the relationships between knowledge, assertion, epistemic possibility,
and practical reasoning.

II
So far we have been concerned only with the initial stages of Hawthorne’s discussion
of his central claim, the doctrine that agents cannot ordinarily be said to know that
lottery propositions are true. As we have seen, he urges that there is direct, intuitive
support for this doctrine. He also maintains that the supporting intuitions are
explained and buttressed by a line of thought that he calls the parity argument. But
this is not the whole story. In addition to invoking our intuitions and considerations
of parity, he provides the materials for constructing three additional arguments.
Hawthorne does not state these arguments explicitly. Rather, he supplies the key
premises and leaves the rest to the imagination of the reader. We wish to engage with
these arguments because we would like our assessment of Hawthorne’s central claim
to make contact with all of the relevant considerations.
We will call the three additional arguments the assertion argument, the epistemic
possibility argument, and the practical reasoning argument.
The assertion argument begins with the claim that we are generally unwilling to
assert lottery propositions. “Despite having good reason to think that a lottery ticket
will lose, it is typically out of place to declare outright ‘He won’t win the lottery’ in
advance of the drawing and without special insider information” (p. 21). Hawthorne
also endorses the increasingly popular doctrine that knowledge is the norm of
assertion. That is, he maintains that it is plausible, quite apart from considerations
having to do with lotteries, that conversation is governed by the following principle:

(P1) A speaker ought not to assert that p unless he or she knows that p.
In Hawthorne’s words: “The practice of assertion is constituted by the rule/require-
ment that one must assert something only if one knows it. Thus if someone asserts p,
it is proper to criticize that person if she does not know that p” (p. 23).
Given (P1), it is possible to construct a best explanation argument for Hawthorne’s
view about knowledge of lottery propositions that runs as follows: “Suppose that we
don’t normally know that lottery propositions are true, and that speakers are
generally aware of this fact. In combination with (P1), this assumption enables us
to explain the fact that speakers are not normally inclined to assert lottery proposi-
tions: they don’t assert lottery propositions because they recognize that they don’t
258 PART IV: KNOWLEDGE

know that the propositions are true. Thus, the assumption that we don’t know lottery
propositions enables an explanation of a considerable body of data. Moreover, it is
the best explanation, since other explanations can be shown independently to be
inadequate.” (Hawthorne does in fact consider two other explanations critically. Both
are based on Gricean maxims.)
We turn now to the epistemic possibility argument. Hawthorne shares the familiar
view that there is a connection between epistemic possibility and knowledge that can
be expressed as follows:

(P2) It is possible that p for S at t iff p is consistent with what S knows at t. (p. 26)
Now Hawthorne thinks that there is a use of “There is a chance that p is true” on
which it is equivalent to “It is epistemically possible that p.” Because of this, he holds
that (P2) has a companion that can be formulated as follows:

(P3) There is a chance of p for S at t iff p is consistent with what S knows at t.

(p. 26)
In defense of these principles, Hawthorne claims that their validity is the best explan-
ation of the fact that sentences of the form “I know that p but it is possible that not-p”
and “I know that p but there is a chance that not-p” are not typically assertible.
In view of (P3), it seems to be possible to reason as follows: “If S knows the lottery
proposition that ticket #666 will lose, then, given (P3), it cannot be true that there is a
chance for S that ticket #666 will win. But it is clear that there is a chance for S that
ticket #666 will win. After all, we are assuming throughout that lottery beliefs are
based on probabilistic reasoning. S knows the relevant probabilities, at least to an
approximation, and therefore knows that there is a real though small chance that
ticket #666 will win.”
The practical reasoning argument is the most interesting of the three additional
lines of thought. Hawthorne opens by claiming that it is normally inappropriate to
use lottery propositions as premises in practical reasoning. He illustrates this claim by
the following argument, which he quite rightly finds unacceptable:

The ticket is a loser.

So if I keep the ticket I will get nothing.
But if I sell the ticket, I will get a penny.
So I’d better sell the ticket.
Intuitively, he says, this argument is absurd because it uses a lottery proposition as a
premise.
Having made this initial claim, Hawthorne goes on to endorse the view that
knowledge is the norm of practical reasoning. He spells this view out in terms of the
following principles, which he takes to be plausible independently of considerations
having to do with lotteries:
 HAWTHORNE ’ S LOTTERY PUZZLE AND THE NATURE OF BELIEF 259

(P4) If one knows that p, then it is acceptable to use the proposition that p as a
premise in one’s practical reasoning—i.e., in one’s deliberations about how to act.
(p. 30)
(P5) If one doesn’t know that p, then it is not acceptable to use the proposition
that p as a premise in one’s practical reasoning. (p. 30)
In combination with Hawthorne’s initial claim, these principles lead to two inde-
pendent but closely related arguments for the conclusion that we cannot normally be
said to know that lottery propositions are true.
The first argument is quite simple. It proceeds directly from the initial claim and
(P4) to the desired conclusion. It is deductively valid.
The second argument has the form of an inference to the best explanation. If we
assume that we do not normally know lottery propositions to be true, we can explain
the unacceptability of the foregoing argument about selling one’s lottery ticket by
appealing to (P5). That this explanation is better than alternatives is shown by the
fact that it accords with our intuitions about what is wrong with the argument: “It is
clear that if one asks ordinary folk why such reasoning is unacceptable, they will
respond by pointing out that the first premise is not known to be true” (pp. 29–30).

III
We turn now to the task of assessing these arguments. We will try to show that they
all have disabling flaws.
We begin by observing that the first premise of the assertion argument, the claim
that a speaker must know that p in order to have a conversational entitlement to
assert that p, appears to have counterexamples. If a speaker has good reason to believe
that p at the time of the assertion, then the speaker is not usually subject to rational
criticism even if it turns out that p is false. That is to say, as far as we can see, there is
no practice of criticizing such speakers for having spoken inappropriately, or for
having exceeded the bounds of conversational entitlement. The falsity of p prevents
agents from knowing p, but it does not by itself undermine entitlements to assert p.
Similarly, it seems that it is conversationally quite appropriate for someone who has
good evidence for p to assert that p, even if, unbeknownst to him, there are Gettier
factors that prevent his justified belief from counting as knowledge.
In proposing that knowledge is the norm of assertion, Hawthorne is in effect
endorsing (P1):

(P1) A speaker ought not to assert that p unless he or she knows that p.
The foregoing considerations suggest that (P1) is incorrect. It is not immediately
clear what should be adopted in place of (P1). It is worth noting, however, that it is
easy to find a principle that is at least immune to counterexamples of the foregoing
sort. This is true, for example, of (P1*):
260 PART IV: KNOWLEDGE

(P1*) A speaker ought not to assert p unless he or she believes p and is justified in
so believing.
We do not wish to suggest that (P1*) is the whole truth about assertion, or even that it
is wholly true. We think that any account of assertion must incorporate a number of
diverse components. We cite it only to underscore the point that it is easy to
accommodate the counterexamples to (P1). There is no need to try to “save” (P1)
by resorting to heroic measures.
Our objection to the assertion argument is an obvious one, and it seems likely that
Hawthorne is aware of it. He is probably presupposing the response to the objection
that is presented in Williamson’s Knowledge and its Limits, a book that Hawthorne
cites as an inspiration for his own treatment of assertion (p. 21). Williamson
acknowledges that we generally refrain from criticizing speakers who assert proposi-
tions that they do not know, provided that they are justified in believing that the
propositions are true, but he tries to explain this fact in a way that is ultimately
consistent with the knowledge account of assertion. In giving this explanation, he
begins by observing that a speaker who is justified in believing a proposition is
generally (though not always) justified in believing that he knows the proposition.
In view of this fact, Williamson claims, it is generally appropriate to see speakers as
trying to conform to the norm linking assertion to knowledge, and as therefore
deserving to be treated with indulgence when they fail. That is to say, it generally
makes sense for us to excuse speakers who fail to conform to the norm. This is the
core of Williamson’s explanation. But he tries to buttress it by pointing out that it is
often our way to allow minor violations of norms to go unpenalized. We tend to
reserve criticism for violations that are particularly salient or particularly serious.2
Williamson’s defense of the knowledge account is ingenious, but we feel that it falls
short of success. A practice that involves a rule linking assertion to knowledge but
allows minor violations of the rule is more complex than a practice that is based on a
less demanding rule, such as the foregoing rule linking assertion to justified belief. In
a practice of the former sort there will have to be a convention allowing excuses
together with one or more rules specifying the gravity of various forms of infraction.
Moreover, each participant in such a practice will have to keep track of the various
psychological and epistemic factors that determine whether particular infractions
should be excused. Thus, insofar as there is a distinction to be drawn between being
justified in believing a proposition and being justified in believing that one knows a
proposition, participants will have to discriminate between these two states in
assessing the performance of speakers. And they will also have to assess the gravity
of infractions. In view of these considerations, the justified belief account of assertion
enjoys a prima facie advantage over the knowledge account. It is significantly simpler.

2
Williamson (2000, p. 258).
 HAWTHORNE ’ S LOTTERY PUZZLE AND THE NATURE OF BELIEF 261

Why then are Williamson and Hawthorne so taken with the knowledge account? As
far as we can determine, they are moved largely or even entirely by two considerations.
First, they believe that the knowledge account provides the best explanation of a certain
variant of Moore’s Paradox—that is, of the fact that it would be odd or “paradoxical”
for someone to assert a proposition of the form “p but I don’t know that p.” Second,
they think (i) that speakers are generally unwilling to assert lottery propositions, and
(ii) that the knowledge account provides the best explanation of this fact.
We have strong reservations about this rationale. The knowledge account is not
the only explanation of the Moorean Paradox. There are other, competing explan-
ations that are at least as plausible.3 Moreover, while (i) receives a certain amount of
support from the data involving assertion, there is a large range of data that it fails to
accommodate. It is often entirely appropriate to assert a lottery proposition.
In particular, there are many contexts in which it is entirely appropriate for agents
to assert that particular lottery tickets will lose. It is also appropriate in such contexts
for agents to assert that they know that particular tickets will lose. Suppose, for
example, that Dan is trying to figure out whether he can afford to buy a sailboat. Dan
might say, “Well, I have a lottery ticket. I might win the lottery. So I’d better factor
that possibility in somewhere.” Here it seems entirely reasonable for Dan’s friend
Jerry to respond as follows: “Get serious. We both know that you’re not going to win
the lottery. You should just forget about that possibility.” It is clear that conversations
of this sort happen all the time.
By themselves, the considerations we have adduced do not amount to a knock-
down objection to the knowledge account, nor to a decisive reason for rejecting the
assertion argument. Perhaps it is possible to accommodate the complexities in the
data involving lottery propositions by introducing additional qualifications. It is
clear, however, that the foregoing considerations provide reasons for being dissatis-
fied with the knowledge account, and for hoping that it is possible to provide a better
explanation of the data involving assertion. In fact, we think that it is possible to do
better. With a view to making this plausible, we will propose a new account of
assertion—an account that explains the comparatively simple conversational data

3
For instance, it is possible to explain the inappropriateness of asserting “p but I don’t know that p”
using principle (P1*). Suppose it is true that one ought not to assert a proposition unless one believes the
proposition and is fully justified in believing it. Suppose also that a certain speaker, Max, believes and is
fully justified in believing the first conjunct of “p but I don’t know that p.” Will Max also believe and be fully
justified in believing the second conjunct? For this to be true, Max would have to believe and be justified in
believing at least one of the following four propositions: (i) Max does not believe that p; (ii) it is not true
that p; (iii) Max is not justified in believing that p; and (iv) Max fails to satisfy the Gettier condition with
respect to p. (We are here assuming the standard theory of knowledge.) But given that Max believes, and is
justified in believing p, it seems very unlikely that he will believe any of these propositions. For example, if
Max believed (iv) and was justified in so believing, then he would not be fully justified in believing that p.
To see this, observe that if one believes that he is in a Gettier situation with respect to a certain proposition,
then he believes that special factors are present which prevent his prima facie justification for believing
the proposition from being a satisfactory guide to truth. The latter belief is a defeater for one’s prima
facie justification for believing the proposition, so it precludes one’s having a full or all-things-considered
justification for believing it.
262 PART IV: KNOWLEDGE

involving ordinary propositions, and that also does a better job than the knowledge
account of explaining the complex conversational data involving lottery proposi-
tions. We begin the process of formulating this account in the next section.

IV
If agents are often willing to assert lottery propositions, why does it seem otherwise to
Williamson and Hawthorne? Perhaps it seems otherwise because there is a type of
situation in which it is clearly inappropriate to give the advice that Jerry gives in the
foregoing sailboat example. Situations of this sort are by no means the norm; but they
are familiar to everyone, and they have a certain salience. We will describe these
situations in the present section, and will also offer the beginnings of an argument
that they are best explained in terms of a Gricean principle governing the flow of
information. Our Gricean explanation is incompatible with the explanation favored
by Williamson and Hawthorne, which is of course based on the knowledge account
of assertion. Accordingly, we see our Gricean explanation as driving the final nail into
the coffin of the knowledge account.
The situations we have in mind arise when an agent A1 is involved conversation-
ally with an agent A2 who is considering a course of action that might lead to a
considerable gain or a considerable loss, depending on whether a certain lottery
proposition is true. Suppose, for example, that Carol is wondering whether to assert
that Mary’s lottery ticket will lose, and suppose also that Mary is trying to decide
whether to hold onto the ticket or to sell it to a third party for a penny. In this case,
Carol would very likely refrain from asserting the proposition, even though she has,
and recognizes that she has, an excellent reason for believing that the proposition is
true. The reason for this reluctance, we submit, is the Gricean principle that when
one asserts that p, one thereby implicates one’s belief that the proposition that p is
more relevant to the audience’s informational needs than any of the other proposi-
tions that one is justified in believing. Thus, if Carol were to assert the proposition
that Mary’s ticket will lose, she would be implicating that she believes that this
proposition is more relevant to Mary’s present informational needs than any other
proposition about Mary’s ticket that she believes, such as the proposition that there is
a very small but nonetheless positive probability that Mary’s ticket will win. In fact,
however, as we will explain in detail later on, the latter proposition is much more
relevant to Mary’s present informational needs than the former proposition, because
it is the latter proposition that she should be using as a premise in deliberating about
whether to hang onto her ticket or sell it for a penny. It can be assumed that Carol is
aware of the nature of Mary’s current deliberations. By the same token, it can be
assumed that she is aware that the proposition that there is a very small probability
that her ticket will win is more relevant to Mary’s present informational needs than
the proposition that her ticket will lose. Since she is aware of these things, she must
 HAWTHORNE ’ S LOTTERY PUZZLE AND THE NATURE OF BELIEF 263

also be aware that it would be misleading for her to say something that would
implicate that her most relevant belief is the belief that Mary’s ticket will lose.
In Section VII we will attempt to explain and justify our claim that it is the
proposition that there is a very small probability that Mary’s ticket will win, and
not the proposition that the ticket will lose, that Mary should use as a premise in her
practical deliberations, and also the attendant claim that it is the former proposition
that is most relevant to Mary’s present informational needs. It is enough for our
present purposes that the reader agree that if these claims are correct, then it would
be inappropriate for Carol to assert the proposition that Mary’s ticket will lose. We
hope that we have made this conditional claim plausible.

V
We turn now to the epistemic possibility argument. It can be formulated as follows:
First premise: For all p, S, and t, there is a chance of p for S at t iff p is consistent with what
S knows at t.
Second premise: Where A is an agent, L is the proposition that A’s lottery ticket will win, and
T is a time prior to the drawing, there is a chance of L for A at T.
Lemma: Hence, L is consistent with what A knows at T.
Third premise: If L is consistent with what A knows at T, then A does not know not-L at T.
Conclusion: A does not know not-L at T. That is, A doesn’t know that A’s ticket will lose.

The second premise holds in virtue of A’s being aware that there is one chance in ten
million that the ticket will win.
We claim that the argument is fallacious. To be specific, there is an equivocation
on the word “chance.” In our view, the first premise is a conceptual truth, a principle
that holds because it is constitutive of the concept of there being a chance for an agent
that a proposition is true. Accordingly, the relevant notion of chance has no content
over and above the notion of being logically consistent with a certain body of
knowledge. It has no special connection either with the concept of the objective
probability of an event or the subjective probability of a proposition. On the other
hand, “chance” has a quite different meaning in the second premise. As we observed
a moment ago, the second premise holds in virtue of A’s being aware of the objective
probability that his ticket will win. The notion of chance here is an essentially
probabilistic notion, a notion that has no special relationship with the notion of
logical consistency or the notion of an agent’s body of knowledge. It follows that the
argument involves a fallacy of equivocation.
Our rationale for viewing the argument in this way has two parts. First, in
explaining the notion of chance that figures in the first premise, Hawthorne tells us
that it comes to the same thing as the notion of epistemic possibility. Now as many
authors have pointed out, it is extremely plausible that the notions of epistemic
264 PART IV: KNOWLEDGE

possibility and knowledge are interdefinable. In particular, whether a proposition is

epistemically possible for an agent is a matter of whether the proposition is logically
consistent with the agent’s body of knowledge. In view of this, there is good reason to
think that this concept has no essential relation either to the concept of objective
probability or to the concept of subjective probability. Second, if the second premise of
the epistemic possibility argument made use of the same concept of chance as the first
premise, it would be necessary to know the composition of A’s body of knowledge in
order to determine whether the premise is true. More specifically, it would be
necessary to know whether the negation of L is a component of A’s body of knowledge.
But this means, in effect, that it would be necessary to know whether the negation of
L is known by A. Any claim to possess such knowledge would beg the question, for the
point of the argument is to establish that A does not know that L is false.
Thus far our evaluation of the epistemic possibility argument has presupposed a
rather simple picture of the notion of chance that figures in the first premise.
According to this simple picture, the relevant notion of chance is definable in
terms of the concept of knowledge and the concept of logical consistency. Reflection
shows that this picture fails to capture the full complexity of the notion. The notion of
chance that is definable in terms of knowledge is best seen as involving probability
rather than logical consistency. That is to say, when we speak of a proposition as
having a certain probability in relation to knowledge, what we mean is that the
proposition has a certain conditional probability relative to the class of propositions
that are known to be true. Now of course, when we are speaking of this form of
probability, it is true to say that propositions that are known to be true have the
highest degree of probability, and by the same token, it is true to say that the
negations of these propositions have a probability of 0. Accordingly, when we have
this form of probability in mind, it is inappropriate to assert a sentence of the form “S
knows that p but there is a (non-zero) chance for S that not-p.” But this should not
blind us to the fact that it would be natural and appropriate to assert a sentence of the
same form in circumstances in which it was clear that some other form of probability
was under discussion. For instance, it would be entirely appropriate to claim that
S knows that p is true while acknowledging both that it is rational for S to assign a
non-zero degree of belief to not-p and that not-p has a non-zero objective probability.

VI
The practical reasoning argument actually consists of two lines of thought. The first is
a deductive argument with two premises. One premise is the claim that it is generally
inappropriate to use lottery propositions as premises in practical reasoning. The
other is (P4):

(P4) If one knows that p, then it is acceptable to use the proposition that p as a
premise in one’s practical reasoning—i.e., in one’s deliberations about how to act.
 HAWTHORNE ’ S LOTTERY PUZZLE AND THE NATURE OF BELIEF 265

In combination, these premises entail that we generally lack knowledge of lottery

propositions. The second line of thought is a best explanation argument. Its first
premise is identical with that of the first line of thought. The second premise is (P5):

(P5) If one doesn’t know that p, then it is not acceptable to use the proposition
that p as a premise in one’s practical reasoning.
Once these two premises have been stated, we are invited to observe that it is possible
to explain why it is generally inappropriate to use lottery propositions in practical
reasoning by invoking (P5) and the hypothesis that it is generally impossible to know
lottery propositions. We are also told that this explanation is superior to all compet-
ing explanations.
There are problems with the premises of both arguments.
We have already considered a case, the sailboat example, in which an agent clearly
has a right to use a lottery proposition in his practical reasoning. It would be easy to
construct additional cases. Accordingly, the claim that serves as the first premise for
both arguments is false. The claim enjoys a certain initial plausibility, so presumably
some fraction of its content is correct. That is to say, there is presumably a circum-
scribed class of cases in which it is inappropriate to rely on lottery propositions in
practical reasoning. But there is a significant range of cases in which just the opposite
is true.
Moreover, it is clear that (P4) and (P5) are flawed. It is entirely possible for one to
feel extremely confident that one knows a proposition while recognizing that it would
be inappropriate to use the proposition as a premise in certain forms of practical
reasoning. Suppose that Jill has excellent reason for holding that Jack is not in New
York today—perhaps she saw someone who looks exactly like Jack on Benefit Street
in Providence a short while ago, and the person in question responded to her wave
with a broad smile of recognition. In these circumstances, Jill will feel that she has
every right to claim that she knows that Jack is not in New York. But will she be
willing to stake the lives of her children on this proposition? Of course not! Despite
feeling sure that she knows that Jack is not in New York, she will recognize that there
is a miniscule chance that she is wrong. (Here we remind the reader of the preceding
section, in which we maintain that knowledge of p does not preclude one’s allowing
that there is a chance that not-p.) Because of this, Jill will think it inappropriate to
accept certain bets involving the proposition—specifically, bets in which she stands
to lose a great deal. Moreover, the rest of us will applaud this attitude. So (P4)
is wrong. To see that (P5) is also wrong, observe that in certain cases, anyway, one is
entitled to use a proposition in one’s practical reasoning simply because one is
epistemically justified in believing it. The proposition need not be true, and one
need not satisfy whatever proposition it is that normally serves to block Gettier
counterexamples. For example, if weather.com has promised beautiful weather, and
Jane relies on this announcement in formulating her plan for the day, she will not be
266 PART IV: KNOWLEDGE

open to rational criticism if the announcement turns out to be false. People may say
that her decision to spend the day at the beach was unfortunate, or even inadvisable,
but they will not say that she went wrong in her practical reasoning.
We may conclude, then, that the practical reasoning argument does not work.

VII
But this should not be the end of our inquiry. It is clear that the practical reasoning
argument fails, but we should consider whether there is an argument that fares better.
After all, Hawthorne is quite right in claiming that the following argument is
unacceptable:
The ticket is a loser.
So if I keep the ticket I will get nothing.
But if I sell the ticket, I will get a penny.
So I’d better sell the ticket.

This shows that there is a type of situation in which it is inappropriate to use a

lottery proposition as a premise. What are the common features of situations of this
type? What distinguishes them from other situations? How frequently do they occur?
Also, do such situations have any implications concerning knowledge of lottery
propositions? Is it perhaps true that it is inappropriate to rely on lottery propositions
in such situations because there is something about them that precludes knowledge
of lottery propositions?
In considering these questions, it is important to keep in mind the distinction
between practical reasoning that is certainty-based and practical reasoning that is
probability-based—or in other words, the distinction between decision-making under
certainty and decision-making under risk. In a situation in which an agent is attempt-
ing to decide among a set of actions A1, . . . , An, it may be true that the agent is in a
position to predict the outcome of each action with certainty, or it may be that for
each Ai, the agent is forced to recognize two or more possible outcomes for Ai, and
can do no more than assign a probability to each of them. It is relatively simple to
choose an action in situations of the first sort: one need only perform a cost/benefit
analysis, adding up the benefits that would accrue from each Ai and subtracting the
costs, and then choose the action that would be the most beneficial. On the other
hand, practical reasoning that is probability-based involves much more complicated
calculations. In a case that calls for probability-based reasoning, it is necessary to take
information about probabilities into account. It is not known exactly what form these
probabilistic calculations take in actual human decision-making, though there are
some well motivated proposals, such as prospect theory. In the interests of definiteness,
we will suppose that probability-based practical reasoning involves calculations of
expected utilities and that choices are aimed at maximizing expected utility. This
assumption is no doubt unrealistic in a variety of respects, but it has the value of
 HAWTHORNE ’ S LOTTERY PUZZLE AND THE NATURE OF BELIEF 267

illustrating the point that a procedure that is sensitive to a range of probabilities will be
much more complex than a procedure that in effect treats all probabilities as 0 or 1.
Since practical reasoning that is certainty-based is much simpler than practical
reasoning that is probability-based, it behooves one to rely on the former whenever it
is possible to do so. Indeed, the savings in time, energy, and memory are so great that
it behooves one to act as if all probabilities were 0 or 1 in cases in which the actual
probabilities are close to those extreme quantities, provided that doing so would not
significantly change the final results. We will say that a situation in which it is
appropriate to act as if low probabilities were 0 and high probabilities were 1 is a
simplification-permitting situation. Thus, in a simplification-permitting situation,
there is a simplifying idealization of probabilities that (a) enables one to use a
certainty-based procedure for choosing an action rather than a probability-based
procedure, and (b) yields the same conclusion concerning which action to perform as
would a calculation that took information about actual probabilities into account.
We now wish to claim that when probabilities are close to 0 and 1, it is generally
true that one is in a simplification-permitting situation. One will be in a situation that
is not simplification-permitting only if the costs or benefits associated with one or
more of the outcomes are quite large. In a case of this sort, the difference between a
very low probability and 0, or the difference between a very large probability and 1,
will generally be such as to have an impact on calculations of expected utility. But in
other cases, the differences between realistic assumptions about probabilities and
idealizations will not be reflected in the end result. Cases of this second sort will be
the norm, for it is rare that an agent stands to gain or lose a great deal by the choice of
a single action.
Of course, for the distinction between simplification-permitting and non-
simplification-permitting situations to be of practical importance, it would have to
be possible to identify simplification-permitting situations as such in advance of
performing the calculations that probability-based reasoning requires. Fortunately, it
generally is possible to recognize simplification-permitting situations in advance.
This is illustrated by the foregoing sailboat case, and also by the Carol/Mary case. In
the sailboat case it is reasonably clear in advance that the probability of Dan’s
winning the lottery is so low as to be of virtually no consequence with respect the
overall probability of Dan’s having enough money to purchase a sailboat. In the
Carol/Mary case, on the other hand, both Carol and Mary should be able to see, in
advance of grinding through the actual calculations, that the expected utility of
hanging onto the ticket might exceed the expected utility of selling the ticket, given
that the latter quantity is just one penny.
To amplify: In the sailboat case Dan is considering whether to take his ownership
of a ticket into account in determining whether he will be able to afford to buy a
sailboat. It would clearly be expensive in terms of time and energy if he were to take
his ownership of his ticket into account in his calculations. Moreover, since the
probability of his winning is very low, it is reasonably clear in advance that the
268 PART IV: KNOWLEDGE

ultimate conclusion of his reasoning would not be materially affected by taking his
ownership of the ticket into account. In view of these considerations, the costs
associated with taking his ownership of the ticket into account in his calculations
exceeds any gain that might accrue from taking it into account. By the same token, it
makes sense for Dan to proceed on the assumption that his ticket will lose. In the
Carol/Mary case, on the other hand, it is definitely not clear in advance that taking
Mary’s ownership of the ticket into account would have no effect on the calculations.
Mary is considering the question of whether to hang onto her ticket or to sell it for a
penny. Even a cursory look at the relevant factors, the probability of her winning and
the amount of the prize, shows that the expected utility of her owning the ticket is
highly pertinent to this question. Or at least, this will be true if, as is usual, the lottery
is one in which the winner will receive a large amount of money. Suppose, for
example, that Mary’s ticket is one of 10,000,000, and that the winner will receive
$5,000,000. Here one need not calculate the actual expected utility of her hanging on
to the ticket to appreciate that it is probably greater than a penny, which is the
expected utility of selling the ticket. (The expected utility of hanging onto the ticket is
actually fifty times as large as the expected utility of selling it.) By the same token, it
would be inappropriate for Mary simply to assume that her ticket will lose. That
would be equivalent to assuming that the expected utility of continuing to own the
ticket is zero.
We have thus far been concerned to develop a general theory of when and why it is
appropriate to use lottery propositions as premises in practical reasoning, without
taking their probabilities into account. We now observe that the theory provides
answers to two of the questions about lottery propositions that arose at the beginning
of the present section. In the first place, it enables us to explain what it is that
distinguishes situations in which it is appropriate to use a lottery proposition as a
premise in practical reasoning, without taking its probability into account, and
situations in which it is not. The former situations are precisely the ones that can
be seen in advance to be simplification-permitting. To spell this out a bit, they are the
situations in which it can be seen in advance that the actual probabilities are
sufficiently close to 0 and 1, and the costs and benefits are sufficiently modest, that
certainty-based practical reasoning will produce the same answers as probability-
based reasoning. The theory also provides an answer the question of how frequently
situations of the sort in question arise. The answer is “quite frequently.” This holds
for two reasons. First, as we have already seen, cases in which it is safe to rely on
simplifying assumptions about the probabilities of lottery propositions are the norm.
And second, it seems to be generally possible to determine in advance whether a
situation is one in which it is safe to rely on simplifying assumptions. All one has to
do, to make such a determination, is check whether the probabilities are close to
0 and 1, and the costs and benefits are modest.
In Section IV we asserted the following conditional: If it is true that Mary should
use the proposition that there is a very small probability that her ticket will win in her
 HAWTHORNE ’ S LOTTERY PUZZLE AND THE NATURE OF BELIEF 269

practical reasoning, and not the proposition that the ticket will lose, and it is
therefore true that it is the former proposition that is most relevant to Mary’s present
informational needs, then it would be inappropriate for Carol to assert the propos-
ition that Mary’s ticket will lose. We hope that we have now succeeded in making it
plausible that the antecedent of this conditional is true. If so, we can claim to have
offered a reasonably complete diagnosis of what is wrong with Hawthorne’s assertion
argument. Contrary to what the argument contends, it is possible to explain a
speaker’s unwillingness to assert a lottery proposition without assuming that the
speaker fails to know the proposition. When an unwillingness of this sort exists, it is
possible to explain it by citing the Gricean duty not to assert propositions that are
irrelevant to the perceived informational needs of the audience.

VIII
The assertion argument, the epistemic possibility argument, and the practical reason-
ing argument are all ultimately unpersuasive. Let us now return to our earlier
discussion of parity reasoning. Recall that such reasoning relies upon Multi-Premise
Closure and related principles. We earlier rejected MPC on the grounds that deduct-
ive inference from multiple premises can aggregate risks. But we also noted that an
adequate response to the parity argument requires providing an explanation of why
we find MPC so intuitively compelling. A fully adequate response must also show
why Hawthorne’s response to aggregation of risk considerations fails, and why the
main motivation he puts forward for MPC is problematic. We are now in a position
to answer these challenges.
Let us begin by considering Hawthorne’s response to the aggregation of risk
considerations. Recall that we argued that MPC fails because deduction can aggregate
risk in a way that destroys knowledge. In responding to this objection, Hawthorne
relies on his discussion of chance. He suggests that knowing p is incompatible with
there being any chance that p is false. Since risk is just the chance of falsity, in cases
where thinkers know the premises of a deductive argument, there is no risk to
aggregate (p. 48).
As we have already seen, however, Hawthorne’s discussion of chance trades on an
equivocation on the word “chance.” Knowing p may be incompatible with there
being a chance that p is false, for the sense of “chance” definable in terms of
knowledge and logical consistency (or conditional probability). But it is not incom-
patible with there being a chance p is false, for the sense of “chance” relevant to
epistemic probabilities. And it is this latter notion that is relevant to risk aggregation.
Hawthorne’s response therefore fails.
Let us now turn to Hawthorne’s primary positive motivation for MPC. He suggests
that anyone who rejected MPC for a particular argument could be made to sound
270 PART IV: KNOWLEDGE

very much like Lewis Carroll’s foolish Tortoise. Such a person would presumably
answer yes when asked whether he accepted each of the premises of the argument. He
would also answer yes when asked whether he accepted that the premises jointly
entail the conclusion of the argument. But he would answer no when asked whether
he accepted the conclusion of the argument. And this seems to be an “exceedingly
strange” pattern of responses. Insofar as we should avoid any commitment to such a
pattern of responses, we should accept MPC (pp. 39, 49).
This argument relies on the claim that knowledge is the norm of assertion. As we
have already seen, that is an implausible view. What is more plausible is that in
typical conversational contexts, a speaker will (and should) assert a claim only if he or
she believes that the claim is relevant to the audience’s informational needs. This
Gricean principle can be used to explain why the foregoing pattern of responses
seems so strange, even assuming the falsity of MPC. Given the Gricean principle, it is
natural to expect that speakers will only assert a sequence of premises followed by the
claim that the premises jointly entail some conclusion if the speaker accepts the
conclusion and takes the inference to be one that the audience should draw. The
“exceedingly strange” pattern of responses violates this expectation. This explains
why it would feel uncomfortable to answer the sequence of questions in the given
way, and why it would sound so odd to hear such a pattern of responses. This
discomfort would obtain notwithstanding the fact that the answers to the questions
would all be true.
Finally, let us turn to the question of why MPC is such a tempting principle. Why
do we find it intuitively compelling that we can add to our knowledge by drawing
competent deductions from what we know? The general answer we want to propose
is that in the most salient cases of deduction from known premises, MPC obtains.
The mistake is one of overgeneralization. Let us explain.
In our thinking about knowledge, we find it very natural to focus on certain
paradigm cases. For example, we think about basic logical and mathematical know-
ledge, knowledge of simple conceptual truths, perceptual knowledge about the
immediate environment, knowledge based on memory of the recent past, knowledge
of well-known historical facts, and the like. These are the clearest and most central
cases of knowledge. Ordinary thinkers are most comfortable attributing knowledge
in such cases. Indeed, it is quite plausible that our understanding of the concept of
knowledge is tied to our attribution of knowledge in these paradigm cases.
In each of these paradigm cases of knowledge, we are extremely confident about
the truth of the relevant propositions. What could be more certain than our knowledge
of simple logical and mathematical truths, and our perceptual knowledge, about the
immediate environment? In our reflections about knowledge, then, we focus on
examples of known propositions with extremely high epistemic probabilities.
In our thinking about deduction, we find it natural to focus on deductive argu-
ments similar to the ones we actually employ. Such arguments are relatively simple,
with a small number of premises and inferential steps. This is because our minds are
 HAWTHORNE ’ S LOTTERY PUZZLE AND THE NATURE OF BELIEF 271

subject to significant cognitive limitations, including limitations on our short term

and working memory capacities. Indeed, we typically think only about particularly
simple examples of deductive arguments. This is because we typically consider
arguments that we can survey entirely in our heads. In our reflections about deduction,
then, we focus on deductive inferences with only a few premises and inferential steps.
Taken together, these considerations suggest the following moral: In our thinking
about knowledge and deduction, the cases of deductive argument from known prem-
ises that are salient to us are ones in which (i) the epistemic probability of each premise
is extremely high and (ii) there are only a few premises and inferential steps.
There is a general fact about probabilities that is relevant here. Given any deduct-
ively valid argument, the probability of the conclusion will be at least as high as the
probability of the conjunction of the premises. Therefore, if the probability of each
premise is extremely high and there are only a few premises, the probability of the
conclusion will be very high, too. For example, if there are two premises, each with
probability 0.98, the probability of the conclusion will be at least 0.96. (It will be at
least 0.9604 if the two premises are probabilistically independent.)
Using this general fact about probabilities, we can conclude that in cases of
deductive arguments from known premises that are salient to us, the epistemic
probability of the conclusion will be very high. It is plausible that knowing p entails
justifiably assigning a high epistemic probability to p, but is compatible with not
assigning 1 to p. Given this connection between knowledge and epistemic probability,
in all—or almost all—of the salient cases of deduction from known premises, the
conclusion of the deduction will count as knowledge. This is what explains the
intuitive appeal of MPC. Since MPC holds in the salient cases of deduction from
known premises, we find it natural to believe that it holds generally. And that is a
mistake.

IX
Hawthorne’s discussion of lottery propositions is rich and very stimulating. The
arguments he puts forth for the claim that thinkers do not ordinarily know lottery
propositions are arresting. However, as we have seen, they are ultimately unpersua-
sive. The direct, intuitive data are not nearly as compelling as Hawthorne supposes.
The parity argument relies on a fallacious principle connecting knowledge and
deduction. The assertion, epistemic possibility, and practical reasoning arguments
rely on incorrect claims about the metaphysical connections between knowledge on
the one hand, and assertion, chance, and practical reasoning on the other. As we have
seen, once we adopt an appropriately complex view of the nature of belief and
272 PART IV: KNOWLEDGE

reasoning4—and of the metaphysical connections between knowledge, communication,

action, and reasoning—it becomes evident that Hawthorne has not yet made a case for
his view of lottery propositions.

References
Hawthorne, J. (2004). Knowledge and Lotteries. Oxford: Oxford University Press.
Williamson, T. (2000). Knowledge and Its Limits. Oxford: Oxford University Press.

4
There are two ways in which the picture of belief and reasoning that we have been sketching is a
simplification of our actual views. First, we have been treating epistemic probabilities as though they were
beliefs whose contents involve the attribution of probabilities to propositions. But it may be preferable to
understand epistemic probabilities in terms of degrees of conviction or confidence. On this view, decision-
making under risk is practical reasoning that takes into account differences in the degrees to which various
propositions are believed. Second, we have been treating epistemic probabilities as though they obeyed (or
ought to obey) the probability calculus. Here, too, we have reservations. Thinkers do not believe logically
equivalent propositions to the very same degrees. Nor should they. For instance, it would not be rational to
have the highest degree of belief in a complex logical truth, at least in the absence of a proof.
 15
Conceivability and Possibility*

I. Introduction
It is widely recognized, both in philosophy and everyday life, that some propositions
are not just true but necessarily true. They are propositions that could not have been
false, that would have been true no matter how the world had been arranged. To give
some examples, it is widely acknowledged that the laws of logic are necessary, and
that the same is true of principles that are constitutive of concepts, such as the
proposition that green is a color, and the proposition that a fortnight is a period of 14
days. There is also wide agreement that the laws of mathematics are necessary.
In addition to recognizing a category of necessary propositions, most of us also
recognize that some propositions are possible. This second category includes all true
propositions, including all of the propositions that are necessarily true, but it also
includes many propositions that are false, such as “I am not typing right now” and “I
do not exist.” Even though I am in fact typing right now, I might not have been
typing, and even though I do exist, I might not have existed. Possibility can be defined
in terms of necessity: to say that a proposition is possible comes to the same thing as
saying that it isn’t necessary that the proposition is false.
Necessity and possibility are known as modal features of propositions. There are
two other types of modal status that should be noted. A proposition P is said to be
contingent if it is possible for P to be true and also possible for P to be false, and P is
said to be impossible if the denial of P is necessary.
There are many important philosophical questions about the nature of modality,
but there are also significant questions about knowledge of modality. This chapter is
concerned with the latter questions. How do we determine whether a proposition is
necessary or contingent? What procedures do we use for recognizing possibility?
Traditionally, philosophers have held that—to a very large extent, anyway—we
achieve modal knowledge by relying on the faculty of conceiving—that is, on our
ability to conceive of objects and situations. We know that a person might not have
existed by conceiving of a situation in which the parents of the person remain
childless, and we know that it is necessary that 3 is the immediate successor of 2 by

* I am indebted to my student David Hefer for helpful conversations. I have also been helped by
comments on an earlier version by two referees for Oxford University Press.
274 PART IV: KNOWLEDGE

trying to conceive of a number between 2 and 3 and realizing that it cannot be done.
Philosophers have also traditionally held that the imagination plays a significant role
in the acquisition of modal knowledge. In some authors, this view is closely related to
the view that modal knowledge derives from conceiving, for they use “imagine” so
broadly that states of conceiving are included in its sphere of application. But other
authors distinguish between conceiving and imagining, maintaining that the former
necessarily involves the use of concepts, and is therefore best understood as akin to
thought and judgment, while the latter always involves imagery of some kind.
According to these authors, there are at least two distinct ways of gaining modal
knowledge. Unfortunately, I will not be able to discuss this view here. There is room
only to discuss one form of modal knowledge, and conceiving is the natural choice.
Conceiving is a more comprehensive source of modal knowledge than the imagin-
ation, for we are able to represent more possibilities by using concepts than by using
images.
I will be concerned to characterize conceiving, to determine whether it is sufficient
to explain our modal beliefs, or is merely a component of a larger package, to
consider whether its deliverances are reliable, and to assess the prospects of employ-
ing it to settle an important metaphysical issue, the mind-body problem. These topics
will be the primary focus of the chapter, but to do justice to them, it will be necessary
to attend to some metaphysical questions. Hence, I will also be concerned to some
extent with modality itself, considering its different forms, how those forms might be
analyzed, and why they are important to us.

II. Conceiving
To conceive of a moose wearing a suit from Brooks Brothers is to form a complex
representation involving the concepts of a moose, Brooks Brothers, a suit, the relation
of being purchased from, and the relation of wearing. The representation will also
have a logical structure. Thus, for example, the concept of a suit and the concept of
Brooks Brothers will be bound respectively to the first and second argument posi-
tions of the concept of being purchased from. This example illustrates how conceiv-
ing works in general: to conceive of something is to form a representation of it that
has concepts as its building blocks and the structural relations of logical grammar as
its cement.
As this characterization implies, it is possible to conceive of anything that can be
represented by binding together concepts in a way that conforms to the rules of
logical grammar. Thus, for example, it is possible to conceive of a sample of water
that is not composed of H2O, a rectangle that is a perfect ellipse, and a barber who
shaves all and only those men who do not shave themselves.
It follows that simple, undisciplined conceiving is not a reliable test for possibility.
On the present account of conceiving, it is possible to conceive of anything, including
logical contradictions. If we are to have a reliable test for possibility, we must rely
 CONCEIVABILITY AND POSSIBILITY 275

instead on what I will call constrained conceiving—conceiving that is compatible with

the laws of logic, the principles that are constitutive of concepts, and any other
propositions that are assumed to be necessary in the relevant context. The need to
impose constraints on simple conceiving has not always been acknowledged in the
history of philosophy, but there are many anticipations of the point. Thus, for
example, Descartes maintained that we might easily be misled if we rely on conceiv-
ings that are confused or insufficiently articulated. This observation is correct, but it
does not go far enough. In addition to checking for clarity and distinctness, it’s at
least necessary to consider conceptual representations in relation to the laws of logic,
and in relation to the principles that spell out the contents of the relevant concepts,
with a view to determining whether the representations are compatible with these
principles. Another way to put this is to say that in order to show that X is possible,
where X is an object or a state of affairs, one must construct a model of X in thought,
and then determine whether the model satisfies the laws of logic and the principles
that are constitutive of the relevant concepts. Of course, given that our resources of
time, memory, and attention are quite limited, any argument that a model satisfies
the relevant principles will be fragmentary and therefore of heuristic value at best.
But conceiving must involve a certain amount of checking on constraints if it is to be
trustworthy even to a modest degree.

III. A priori modalities

To see how these ideas work in practice, let us focus on a conception of modality that
has arguably been the primary focus of a number of philosophers, including Des-
cartes, Hume, Kant, and the logical positivists. On this conception, the concept of a
necessary truth is coextensive with the concept of a proposition that can be known to
be true a priori. Propositions are regarded as possibly true if they are compatible with
the a priori truths.
Advocates of this conception have generally allowed that we can gain knowing of
possibility by conceiving, but this makes sense only insofar as conceiving is held to be
subject to a constraint. Suppose that one conceives of a situation by forming a
conceptual representation R. Then it must be true that R is compatible with all of
the categories of a priori propositions that are constitutive of necessity. In the
interests of simplicity, I will assume here that there are exactly two such
categories—the laws of logic and the principles that are constitutive of concepts
(hereafter the laws of concepts). Thus, to establish that a situation is possible, one
must construct a representation of the situation in thought and determine that it is
compatible with the a priori propositions of these two types.
It would be interesting to develop this approach at greater length, but I wish to
focus here on an epistemological problem that arises already in connection with the
present bare-bones summary. The summary offers an account of how we come to
know possibility, but it provides no explanation of how necessity is recognized. It is
276 PART IV: KNOWLEDGE

assumed at the outset that all and only a priori propositions are necessary. But how
can that assumption be justified? How can we come to know that it is true? Not by
constrained conceiving. Since the assumption is built into the definition of the
relevant form of constrained conceiving, any attempt to use that form of conceiving
to establish the assumption would be circular. Nor can we hope to establish it by an
argument based on informal characterizations of a priority and necessity. To say that
a proposition is a priori is to say that it is possible to know it without relying on
information provided by experience. On the other hand, to say that a proposition is
necessary is to say that it is true under all possible arrangements of facts, whether
empirical or non-empirical. As Kripke emphasized, there are no immediate logical
connections between these ideas (Kripke 1980, Kripke 2011). We can conceive of a
proposition’s being known independently of experience without conceiving of it
as true under all possible arrangements of facts, and we can also conceive of a
proposition’s being true under all such arrangements without conceiving of it as
knowable independently of experience (by any being who is remotely akin to us). To
say that the two concepts are coextensive is to make a substantive claim that cannot
be established by a simple argument. Moreover, given the differences between the
two concepts, it is very hard to see what a complex proof of coextensiveness would
look like.
Faced with this difficulty, one might be tempted to explain our knowledge of
necessity by saying that it arises from an explicit definition that reduces the concept
of necessity to non-modal concepts. Thus, one might propose to define necessity by
saying that a proposition is necessary just in case it is a law of logic or a law of
concepts. But this particular proposal is not very attractive, because, like all purely
disjunctive definitions, it provides no basis for explaining why its disjuncts should be
grouped together under a common label. Another thought is that we might define
necessity by saying that a proposition is necessary just in case it is known a priori.
This definition would significantly reduce the task of explaining modal knowledge,
for it is comparatively easy to see how one can be justified in believing a definition.
After all, a definitional belief merely records a fiat that arises from and expresses the
will of the believer. Unfortunately, reflection shows that this approach is unsatisfac-
tory. Any account of necessity must vindicate the intuition that necessity is a matter
of truth relative to all possible configurations of facts, and no stipulative definition in
terms of non-modal concepts can achieve that result. Thus, for example, the defin-
ition in terms of a priority implies that there is no more to the idea of necessity than
being knowable independently of experience. But this seems wrong. When one
asserts that a proposition that is known to be a priori is necessary, one isn’t just
making the uninteresting claim that it’s a priori. One is making a substantive claim
about the world. Or so it seems.
Here is another reason for doubting that necessity can be captured by a definition
in terms of a priority. We know that all of the laws of logic are necessary. But do we
know that all of the laws of logic are actually known a priori? No. No creature who is
 CONCEIVABILITY AND POSSIBILITY 277

subject to the resource constraints that govern actual human cognition can be said to
have an exhaustive knowledge of logic. To be sure, we can conceive of an ideal being
who is capable of knowing all the laws of logic, but we cannot be sure that such a
being exists in the actual world. The most we can be sure of is that it is possible for
such a being to exist. Hence, if we are to explain necessity in terms of a priority, we
must say something to the effect that a proposition is necessary just in case it is
possible for there to be a being who knows the proposition a priori. It is clear that the
definition is not reductive. Thus, it makes explicit use of a modal notion, and
moreover, a modal notion that belongs to the same family as the modal notion
that is being defined. (The notion of necessity that is being defined is the traditional
Cartesian notion, and it therefore satisfies the condition that it is possible to know all
necessary truths a priori. The notion of possibility that figures in the definiens must
satisfy the corresponding condition—that is, the condition that all knowledge of
possibility is a priori. Thus, if knowledge of the relevant form of possibility presup-
posed a posteriori knowledge, as it would, for example, if it was the notion of
nomological possibility, the definiens would not sustain the idea that our knowledge
of necessity is a priori. It follows that the notion of possibility in the definiens is of the
same species as the given notion of necessity.) There are related reasons for doubting
other attempts to explain necessity reductively.
Can we say that we acquire knowledge of necessity by some special cognitive
faculty that might be called modal intuition? No. Since we have no independent
reason to believe in modal intuition, such a move would be question-begging, and
anyway, it would provide us with no explanatory insight. It would be like saying that
we are apprised of the truth of mathematics by a special faculty of mathematical
intuition. That move has had its defenders in the history of philosophy, including
Plato and Gödel, but most philosophers have held it in low esteem.
At this point, one might be tempted to dismiss the classical package consisting of a
priori modality and the associated form of constrained conceiving as confused.
Perhaps it should be jettisoned in favor of some alternative package. But this proposal
is also unacceptable. For the significance of the problem is quite general. Let N be any
collection of truths that are thought to be necessary. We can test for possibility by
conceiving of situations that satisfy the members of N, but how will we show that the
members of N really do have the modal status that has been claimed for them? We
will not be able to argue that all efforts to conceive of counterexamples to the
members of N result in failure, for as we have seen, it is possible to conceive of
anything at all. Nor will we be able to establish that the members of N are necessary
by showing that all attempts to construct counterexamples via constrained conceiving
come to grief. The relevant form of constrained conceiving will presuppose that the
members of N are necessary, and this will preclude using that form to give a non-
circular justification of the necessity of those propositions.
To summarize, if we are not to wind up as skeptics about modal knowledge, we
must recognize that constrained conceiving is not the only way of determining the
278 PART IV: KNOWLEDGE

modal status of propositions.1 But we must also recognize that it would not help to
try to explain modality via reductive definitions based on non-modal concepts. Such
definitions can explain how we gain knowledge of necessity, but they inevitably fail to
do justice to its distinctive metaphysical nature, which has to do with truth under all
possible configurations of facts. Nor can we deal with the problem by positing a new
cognitive faculty. That approach has certain advantages, but they are the advantages
of theft over honest toil. Nor can we deal with it by putting the traditional conception
of modality aside in favor of an alternative conception.
With a view to solving this problem, or at least making some progress toward its
solution, let us return to the intuitive characterization of necessity, which says that a
proposition is necessary if it is true relative to all possible configurations of facts—or
in other words, true at all possible worlds. According to this characterization, modal
knowledge is either implicitly or explicitly knowledge of possible worlds, and modal
reasoning is reasoning that leads from some claims about possible worlds to other
such claims. Now in order to describe a possible world it is necessary to use concepts,
and by the same token, it is necessary to presuppose any propositions that may be
constitutive of the concepts one deploys. After all, there can be little point in
deploying a concept if one is not prepared to grant the truth of the principles that
determine its content and guide its use. It follows that in describing a world, one must
presuppose that the laws of logic hold in the world, for we will need to use logical
concepts in any such description, and the laws of logic are constitutive of our logical
concepts. Further, one must presuppose that all of the principles that are constitutive
of relevant non-logical concepts hold in the world. Suppose, for example, that we
want to consider a possible world that contains one or more spherical objects. It will

1
As a referee has pointed out, it might be useful to restate the argument for this point in somewhat
different terms.
To determine whether a situation is possible, we typically start by conceiving of the situation, where this
means that we combine concepts to form a representation of it. But this is not enough, because it is possible
to form a conceptual representation of virtually any situation, including one that has features that are
straightforwardly contradictory. (For example, it is possible to form a conceptual representation of this
coin’s being both round and square—as is illustrated by this very sentence.) To complete the exercise, we
must see whether our conceptual representation will pass through a filter that consists of the necessary
propositions. This means that we must have some way of determining which propositions are necessary.
Can we make such determinations by some sort of conceivability test? At first sight, it might seem that the
answer should be “Yes.” Thus, we might hope to determine whether a proposition P is necessary by trying
to conceive of a situation in which not-P is true, with failure in this enterprise being taken as grounds for
attributing necessity to P. On reflection, however, it becomes clear that this procedure is badly flawed. As
we’ve just observed, it’s possible to conceive of anything. Hence, even if P is necessary, an attempt to
conceive of a situation in which not-P is true would meet with success. Can we proceed by first forming a
conceptual representation of a situation in which not-P is true, and then attempting to pass this represen-
tation through a filter of necessary propositions? No. That would beg the question at issue, which is how we
can determine which propositions should go into the filter.
I conclude that no test based principally on conceivability can be adequate to determine the modal status
of all propositions. There must be some way of identifying necessary propositions that is largely or entirely
independent of conceivability.
 CONCEIVABILITY AND POSSIBILITY 279

be necessary to mention this fact in describing the world, and this means that the
concept of a sphere must figure in the description. Now the concept of a sphere is
governed by a number of principles, including the principle that a sphere is a three
dimensional figure, and the principle that every point on the surface of a sphere is
equidistant from its center. It would make no sense to describe an object as a sphere
unless one was prepared to see it is conforming to these principles. Accordingly, they
will be presupposed by any description of the world.
This line of thought can be summarized by saying that the laws of logic and the
laws of concepts must be partially constitutive of the concept of necessity, because they
will perforce be presupposed in any attempt to characterize the truth-makers of
modal claims. It follows from this that if we are ever epistemically entitled to believe
any propositions that are concerned with necessity, then we are entitled to assume
that the laws of logic and the laws of concepts are necessary. Of course, this
conditional falls short of implying the categorical proposition that we are epistem-
ically entitled to attribute necessity to the laws. Whether that holds depends on
whether we are epistemically entitled to believe any propositions involving the
concept of necessity. I will not be able to deal with this larger issue here. My present
ambition is more modest. I am assuming that knowledge of necessity is possible. My
concern is only to explain how it is possible. That is, my goal is identify ostensible
sources of modal knowledge, and to assess possible impediments to trusting them.
If we wish to develop these views systematically, we will begin by saying that the
following principle is the first component of an implicit definition of the concept of
necessity:
(a) If a proposition is necessary, then it is true relative to all possible configur-
ations of facts.
Next, we will offer the following principles as additional clauses of the implicit
definition:
(b) If P is a law of logic, then P is necessarily true.
(c) If P is a law of concepts, then P is necessarily true.
Other clauses of the implicit definition will include the axioms of a modal logic,
presumably S5. (There will be some redundancy here, since the first component of
the implicit definition implies the standard modal axiom to the effect that a propos-
ition is true in the actual world if it is necessary, and the second component is
equivalent to another standard modal axiom.) Our third step will be to point out that
all of the components of the implicit definition are epistemically available to anyone
who possesses the relevant concept of necessity. If any claims involving that concept
count as objects of knowledge, they will have that status as well, because a reflective
grasp of them is a precondition of any legitimate use of the concept. And fourth, we
will observe that despite being characterized by independent axioms, the concept of
necessity is unified by the fact that the laws of logic and the laws of concepts enjoy a
280 PART IV: KNOWLEDGE

common nature. They would be presupposed in any attempt to specify the possible
configurations of facts that serve as truth makers for modal claims.
Unlike the explicit definitions of necessity that we considered earlier, this implicit
definition makes no attempt to reduce the concept of necessity to non-modal
concepts. On the contrary, it makes explicit use of a modal concept in characterizing
necessity. (Cf. principle (a) and the S5 axiom that if a proposition is possible, then it is
necessarily possible.) In effect, on the present conception, the notions of necessity
and possibility mutually constrain one another, due to their interactions in the
clauses of the implicit definition.
It remains to be determined whether principles (b) and (c) are sufficient to specify
the main categories of necessary propositions. What about the class of mathematical
truths? I think we must draw a distinction in order to answer this question. On the
one hand, there are propositions that are partially constitutive of mathematical
concepts. Examples include the proposition that the successor function is one-one
and the proposition that sets are individuated by their members. On the other hand,
there are substantive claims about the existence of mathematical objects, such as the
principle that every natural number has a successor and the axiom of choice.
Members of the former class (which I will henceforth refer to as constitutive math-
ematical laws) can appropriately be grouped with the laws of logic and the laws of
concepts, for they too determine the contents of their constituent concepts and
provide guidance in their use. It makes sense to expand the implicit definition to
include a clause that represents them as necessary. But it is not at all clear that
members of the latter class are on a par with the laws of logic and the laws of
concepts. And by the same token, it is not at all clear that they should figure in an
implicit definition of the concept of necessity.
In sum, while simple, undisciplined conceiving is not a reliable source of modal
knowledge, it appears that constrained conceiving can be used with some confidence
to establish that propositions are possible, where constrained conceiving consists in
showing that a logically structured conceptual representation is compatible with the
laws of logic, the laws of concepts, and certain of the laws of mathematics. To be sure,
one cannot hope to establish conclusively that a representation is compatible with all
of these laws, for there can be no complete proof procedure for establishing logical
consistency. However, one can at least give heuristic arguments for compatibility
claims of the relevant sort. We have also found that knowledge of necessity must
come from some source other than constrained conceiving. Plausibly, it derives from
certain axioms that provide a conceptual framework for describing the possible
worlds that serve as truth-makers for modal claims. These axioms are clauses of an
implicit definition of the concept of necessity. Since they have this status, one is
entitled to believe them simply in virtue of possessing that concept, provided that one
is entitled to hold any beliefs in which the concept plays a role. It is perhaps the best
argument for this approach that most other ways of explaining our epistemic
entitlement to beliefs about necessity quickly come to grief. (For an exception, see
 CONCEIVABILITY AND POSSIBILITY 281

the Appendix. For more discussion of epistemic entitlements arising from implicit
definitions, see Boghossian 1997.)

IV. Metaphysical modalities

In a dramatically sharp break with the traditional view, Saul Kripke and Hilary
Putnam argued forcefully in the 1970s that necessity does not coincide with a priority
(Kripke 1980, Putnam 1975). There is a large class of necessary propositions that are
not a priori. In defending this assertion, Kripke and Putnam relied primarily on
appeals to intuition, presenting their readers with examples and urging that their
claims about the examples were vindicated by conceivability tests. It turned out that
their readers shared these assessments, as did the subjects in a range of imaginative
experiments conducted by the psychologists Frank Keil and Susan Gelman (Keil
1989, Gelman 2003).
Generalizing from the examples that Kripke and Putnam presented, we have the
claim that the true members of the following categories are both necessary and a
posteriori:
(i) Propositions consisting of proper names and the identity predicate
(ii) Propositions about the origins and ancestry of human beings
(iii) Propositions about the origins of artifacts
(iv) Propositions about the original material constitutions of physical objects
(v) Propositions attributing atomic numbers to atoms
(vi) Propositions attributing molecular structures to chemical kinds
(vii) Propositions attributing genetic structure to biological kinds.
To illustrate, this view entails that the proposition that Mark Twain was identical
with Samuel Clemens is necessary despite being a posteriori, and that the same is true
of the proposition that Queen Elizabeth II came from a certain sperm and egg, the
proposition that George the VI was the father of Elizabeth II, the proposition that a
certain rock was initially composed entirely of granite, and the proposition that water
is composed of H2O molecules.
In addition to arguing that there are necessary propositions that are not a priori,
Kripke maintained that there are a priori propositions that are not necessary. His
examples of the contingent a priori are controversial, but, inspired by his efforts, his
readers soon identified other examples that have won a wider following. Thus, it is
plausible that various indexical propositions fill the bill, including the proposition
that I am here now, the proposition that I exist, the proposition that I am thinking,
and the proposition that I am thinking that P, where P is a mathematical proposition,
or any other proposition whose constituent concepts can be acquired independently
of experience. I can know that I am here now simply in virtue of possessing the
relevant indexical concepts, but it is entirely contingent that any particular person is
in any particular place at any particular time.
282 PART IV: KNOWLEDGE

On one interpretation of their work, Kripke and Putnam should be seen as

criticizing the traditional view of modality—and in fact, as decisively refuting it.
On this view, traditional philosophers like Descartes were concerned with the same
type of necessity as Kripke and Putnam, but they had false views as to the proposi-
tions that enjoy that status, mistakenly holding that the propositions are limited to
ones that can be known a priori. This interpretation enjoys a certain amount of
plausibility, but there is also another interpretation that merits consideration.
According to this second construal, Descartes and other traditional philosophers
were concerned with a different form of necessity than Kripke and Putnam, a form
that is possessed by a narrower range of propositions. On this interpretation, the
Cartesian view is similar to the view put forward by Kripke and Putnam in that they
both explain necessity in terms of truth relative to all possible configurations of facts;
but this is merely a formal similarity, for the two views employ different conceptions
of possibility, the Cartesian conception being much more inclusive than the Kripke-
Putnam conception.
Now as I see it, there is ample motivation for both the Cartesian view of modality
and the more recent Kripke-Putnam view. Thus, we are often interested in knowing
what scenarios are mandated by the structure of our conceptual scheme, and what
scenarios are allowed by that structure, and we are also often interested in knowing
what scenarios are either mandated or allowed by the structure of our conceptual
scheme in combination with assumptions about the essential natures of objects.
Accordingly, I find it more plausible to suppose that we have two different concep-
tions of modality, and will presuppose this view in the sequel. In order to have names
for the two types of possibility, I will refer to them respectively as conceptual
possibility and metaphysical possibility. (I will refer to the two corresponding types
of necessity as conceptual necessity and metaphysical necessity.)
How do we recognize metaphysical possibility? The answer is that we recognize it
by constrained conceiving, but in this case, the constraints are more numerous. In
addition to the laws of logic, the laws of concepts, and the constitutive mathematical
laws, they include all propositions of types (i)–(vii), together with any other a
posteriori necessities that may exist.
This proposal is quite different than the one that Kripke appears to favor. More
specifically, as I read him, he appears to think that if we guard against certain possible
confusions, we can achieve modal knowledge by making use of simple, undisciplined
conceiving. Thus, for example, he appears to think that if we are careful not
to confuse the proposition that Hesperus is identical with Phosphorous with a
proposition that picks out Hesperus and Phosphorous by mentioning certain of
their contingent properties, such as the proposition that the brightest heavenly body
seen in the evening sky in the West is identical with the brightest heavenly body seen
in the morning sky in the East, we will not be able to conceive of a situation in
which the former proposition is false. Whether Kripke holds it or not, this view is
clearly wrong. To conceive of a situation in which Hesperus is not identical with
 CONCEIVABILITY AND POSSIBILITY 283

Phosphorus one need only construct a representation in which the concept of

Hesperus and the concept of Phosphorus are bound respectively to the first and
second argument positions of the concept of identity, and then prefix the concept of
negation to the result. Like many philosophers, Kripke appears to have a view of
conceiving that does not mesh with the psychological facts. To conceive is to form a
conceptual representation. Period. As we observed earlier, given this understanding
of conceiving it is possible to conceive of anything whatsoever, even including
contradictions. It is only disciplined conceiving that provides an adequate test for
possibility.
How do we recognize that an a posteriori proposition is a metaphysical necessity?
According to Kripke, we do so by a process that is partly a priori and partly a
posteriori. Focusing on the proposition that a certain table was not made of ice at its
origin, he argues that the proposition is both necessary and a posteriori, and then
offers the following account of how we recognize it as necessary (Kripke 2011):
So we have to say that although we cannot know a priori whether this table was made of ice or
not, given that it is not made of ice, it is necessarily not made of ice. In other words, if P is the
statement that the lectern is not made of ice, one knows by a priori philosophical analysis,
some conditional of the form “If P, then necessarily P.” If the table is not made of ice, it is
necessarily not made of ice. On the other hand, then, we know by empirical investigation that
P, the antecedent of the conditional, is true—that this table is not made of ice. We can conclude
by modus ponens:

P  ▱P
P
▱P
Generalizing from this case, we can conclude that for every proposition P that
belongs to one of the foregoing seven categories (that is, categories (i)–(vii)), Kripke
thinks that the conditional If P, then it’s necessary that P can be known a priori, and
that we come to know It’s necessary that P by inferring this proposition from the
conditional and the proposition that P, which we learn by normal a posteriori
methods.
Kripke’s discussion is illuminating, but it leaves us with the question, “How do we
know the major premises of these arguments?” How do we know conditionals of the
form If P, then it’s necessary that P, where P is a proposition of one of the types (i)–
(vii)? The answer I wish to recommend derives from one that was originally
proposed by Alan Sidelle (Sidelle 1989). According to Sidelle, we are entitled to
accept propositions of the given sort simply in virtue of possessing the concept of
metaphysical necessity, for those propositions are partially constitutive of the con-
cept. Building on this suggestion, I wish to propose that the concept of metaphysical
necessity is implicitly defined by a rather heterogenous class of propositions that
includes the laws of S5 modal logic, appropriate versions of principles (a), (b), and (c)
284 PART IV: KNOWLEDGE

from the preceding section (with the notion of conceptual necessity replaced by the
notion of metaphysical necessity), the principle that a proposition is necessary if it is
one of the constitutive laws of mathematics, and general principles that imply all
conditionals of the sort that we have just been considering. Since these propositions
are all components of an implicit definition of the concept, they are available to
anyone who possesses it. Moreover, anyone who possesses it is epistemically entitled
to believe them, provided that we are ever entitled to believe any propositions about
metaphysical necessity.
This proposal has a certain prima facie appeal, but before it can be accepted, we
must consider whether the clauses of the implicit definition have enough in common
to yield a concept of necessity that is internally coherent. What if anything ties
together the various a priori laws with propositions of types (i)–(vii)? On the face of
it, there is a vast difference between, say, De Morgan’s laws and the proposition that
George VI was the father of Elizabeth II. What could unify a concept that owes its
content to such different propositions?
In considering this question, we should recall that the various a priori laws are
unified by the fact that they are all constitutive of the concepts that we must use in
characterizing possible worlds. We presuppose them in describing the truth-makers of
modal claims, so it is appropriate that they figure in an implicit definition of necessity.
If a similar claim could be made about propositions of types (i)–(vii), there would be no
obstacle to seeing the concept of necessity as unified. So we must ask whether there is a
reason for thinking that propositions of the given types are presupposed in descriptions
of possible worlds. And in fact, it seems that there is such a reason, for it seems that we
presuppose propositions of types (i)–(vii) in assessing claims to the effect that actual
objects and actual kinds exist in other possible worlds. Consider, for example, the claim
that Elizabeth II exists in another world. Reflection suggests that we would not accept
this claim unless we were convinced that there are individuals in other worlds who have
exactly the same ancestors as Elizabeth. No matter how much a person in another
world resembles Elizabeth in other respects, we will not allow that the person is
identical with Elizabeth unless she has the same parents as Elizabeth, the same
grandparents, and so on. In general, we will not say that actual people exist in other
worlds unless the relevant propositions of type (ii) are satisfied. But this means that we
presuppose true propositions of type (ii) in describing other worlds. In other words,
those propositions are constitutive of the metaphysical modalities because we assume
that they are true in characterizing the facts that serve as truth-makers for modal
claims. Similar remarks apply to true propositions of type (i) and types (iii)–(vii).
This line of thought provides a rationale for supposing that all propositions of the
form If P, then it’s necessary that P, where P is a proposition of one of the types (i)–
(vii), are constitutive of the concept of metaphysical necessity. But it is best not to
think of them as separate constituents of an implicit definition of the concept, for on
that construal of the definition, it would be quite unwieldy. It’s better to suppose that
the constituents of the definition are generalizations from which all such particular
conditionals follow.
 CONCEIVABILITY AND POSSIBILITY 285

Here, then, is a theory that systematizes and explains the modal insights of Kripke
and Putnam, while enabling an account of how we acquire modal knowledge—an
account that is based on constrained conceiving and grasp of the clauses of an
implicit definition.
There is, however, a question about knowledge of the metaphysical modalities that
remains to be addressed. In contemporary philosophy the metaphysical modalities
are held in high esteem, and knowledge of facts involving them is highly prized. If we
know that a proposition is metaphysically possible or metaphysically necessary, then,
it is held, we know something very important about it. Now the foregoing theory
leaves it somewhat puzzling why this should be so. This is because the theory
represents propositions of types (i)–(vii) as merely stating necessary conditions of
existence in other possible worlds. They do not state sufficient conditions of trans-
world identity. This comes to the fore when we consider consequences of the implicit
definition like this one:

(d) If George VI was the father of Elizabeth II, then it is necessary that he was her
father.
In combination with the fact that George was indeed the father of Elizabeth, this tells
us that it is a necessary condition of Elizabeth’s existing in any other world W that
George be her father in W, but it does not imply that Elizabeth exists in W. Nor does
it imply that Elizabeth exists in any other alternative to the actual world. Why would
we want a metaphysical framework which tells us that actual entities can’t exist in
other worlds unless certain conditions are satisfied, but provides no basis for saying
that actual entities do exist in other worlds? It seems that the necessary conditions
would be of little value unless there were also sufficient conditions. If this is right,
then the foregoing implicit definition of metaphysical necessity is incomplete. In
addition to the clauses cited earlier, an adequate implicit definition should contain
clauses that state sufficient conditions of trans-world identity. I will not attempt to
formulate principles of this sort here, and in fact, as a review of the literature will
attest, it has proved quite difficult to come up with principles that can withstand
scrutiny. (See, e.g., Mackie 2006.) But it seems that if the metaphysical modalities
really do have the value that is usually claimed for them, such principles must exist,
and must be built into our modal concepts.
I find this line of thought persuasive, and will therefore presuppose in the sequel that
true propositions of forms (i)–(vii) owe their necessity to their being consequences of
principles that state necessary and sufficient conditions of trans-world identity.

V. Nomological modalities
Roughly speaking, a proposition is nomologically necessary if either it is metaphys-
ically necessary or it follows from the laws of nature. As with conceptual necessity
286 PART IV: KNOWLEDGE

and metaphysical necessity, we can suppose that this new modal notion can be
captured by an implicit definition. In this case, the definition will imply that all of
the following are necessary: the laws of logic, the laws of concepts, the constitutive
mathematical laws, any additional laws of mathematics that are required for natural
science, all true propositions of types (i)–(vii), and the laws of nature. It will also
contain the laws of an appropriate modal logic. Assuming that it is possible to be
epistemically justified in believing any propositions concerning nomological neces-
sity, we can be justified in believing the clauses of an implicit definition that has these
consequences simply in virtue of having the concept of nomological necessity in our
repertoire. As in other cases, it is possible to form justified beliefs about the corres-
ponding form of possibility by constrained conceiving, though now the constraints
must include some substantive portions of mathematics and the laws of nature.
In addition to the categorical nomological modalities, there are also relative
nomological modalities that pose additional metaphysical and epistemological ques-
tions. In considering questions of categorical possibility, we simply ask whether
propositions are compatible with the laws of nature; but in considering questions
of relative possibility, we ask whether propositions are compatible with the laws of
nature together with certain assumptions about the conditions that prevail in a given
context. It is categorically possible for a human being to run a mile in under four
minutes, but it isn’t possible for me to do so, given certain particular facts about my
body-type and level of training. Generally speaking, when we are considering oppor-
tunities and risks, we are focusing on the nomological possibilities that are presented
by specific contexts. That is, we are focusing on questions of the form, “What are the
nomological possibilities confronting agent A, given that A is in a situation of type
S?” Being able to answer questions of this sort is crucial to human welfare. But how
do we arrive at such answers? How do we gain knowledge of relative possibility?
Abstractly considered, the task of determining what is nomologically possible
relative to an agent A and a situation S involves constructing a model that satisfies
the laws of nature, a description of A, and a description of S, and then asking whether
scenarios of interest are also satisfied by that model. It seems likely, however, that
nature has endowed us with practical heuristics for answering specific categories of
questions about relative possibility. Thus, for example, instead of constructing a
model and showing that it satisfies certain relevant laws of nature, it seems that
our job is generally the easier one of searching memory for information about what
happened in situations similar to the present one that we have encountered in the
past. Information about these past outcomes is what makes it possible for us to draw
conclusions about the opportunities and risks that are presented by the current
situation. At a more fundamental level, as J. J. Gibson pointed out, we probably
have certain perceptual modules that are dedicated to recognizing such basic relative
possibilities (or “affordances”) as graspability, reachability, and climbability (Gibson
1979). We know that heuristics of both these types are reliable, for the plans and
decisions that they support are often successful.
 CONCEIVABILITY AND POSSIBILITY 287

In addition to providing a metaphysical framework that makes it possible for

questions about opportunities and risks to arise and be addressed, the nomological
modalities also provide a framework for counterfactual reasoning, and for the
enterprise of assessing counterfactual conditionals for truth or falsity. Thus, when
we ask what would have happened if certain circumstances had been different than
they actually were, we are in effect asking what would have happened if a situation as
close as possible to the actual situation, but different from it in the given respects, had
evolved in accordance with the actual laws of nature.
Questions of this type are important to us for a variety of reasons, including
especially the fact that such questions enable us to figure out the contributions that
various causal factors make to outcomes (by determining what an outcome would
have been if a certain factor had not been present), and the fact that they enable us to
assign responsibility to agents for the consequences of their actions (by asking how
things would have turned out if an agent had not performed an action, or had
performed it in a different way). How do we answer questions concerning counter-
factual processes and counterfactual courses of action? It seems that certain cognitive
modules are dedicated to counterfactual reasoning, and that science is beginning to
understand the ways in which such modules work. (See, e.g., Byrne 2007.) These
modules must of course be included in any catalog of the sources of modal know-
ledge. (As we will see in the Appendix, it is sometimes maintained that the modules
in question can be used to attain knowledge of the metaphysical modalities as well as
knowledge of the nomological modalities. Here, however, I mean to be focusing on
the simpler view that is concerned only with the latter sort of knowledge.)
It is sometimes maintained that the metaphysical modalities have little cognitive
value beyond that which is possessed by the nomological modalities, and there are
even philosophers who favor the more extreme view that the former coincide with
the latter (Edgington 2004). It is understandable that people would be tempted by
these ideas, since the practical importance of modality is largely due to its role in
planning, decision-making, causal analysis, and assigning responsibility to agents,
and as we have just seen, the modal requirements of these endeavors seem to be fully
met by the nomological modalities. Thus, we again face the question of whether the
metaphysical modalities are genuinely important, and also the question of why
knowledge of them should be prized. Fortunately, reflection shows that these ques-
tions admit of reasonably simple answers: even though the metaphysical modalitites
have little practical value, they offer a number of fairly straightforward theoretical
advantages. Thus, in the first place, we often find it theoretically interesting to
consider what the world would be like if the laws of nature were different in one or
more respects. This enables us to assess the degree to which the laws of nature are
interdependent, and the degree to which some are more fundamental than others.
We would of course lose these benefits if we focused narrowly on nomological
possibility. The metaphysical modalities also provide a framework in which it is
possible to frame definitions of important metaphysical concepts, such as
288 PART IV: KNOWLEDGE

supervenience, realization, and reduction. Among other things, these notions add
considerably to our understanding of relations between actual objects, properties,
and states of affairs. A third advantage is that the metaphysical modalities afford a
basis for intuitively plausible explanations of key concepts from semantics, such as
the notion of a truth condition.

VI. Cartesian modal arguments

As is well known, Descartes thought it possible to establish that his mind and body
were distinct by an argument based on a claim about the relationship between
conceivability and possibility (Descartes 1984, p. 54). Abstracting from some details
that need not concern us here, his argument comes to this:
A. First premise: We are able to conceive of Descartes’s mind existing without being accom-
panied by Descartes’s body. (Justification: Descartes’s evil genius scenario)
Second premise: If we are able to conceive of Descartes’s mind existing without being accom-
panied by Descartes’s body, then it is objectively possible for this to occur.
Lemma: It is objectively possible for Descartes’s mind to exist without being accompanied by
Descartes’s body.
Third premise: If it’s possible for x to exist without being accompanied by y, then x is distinct
from y.
Conclusion: Descartes’s mind is distinct from Descartes’s body.

Because they tend to doubt the Cartesian doctrine that the mind is a substance, on a
par with the body, contemporary philosophers tend to reject this argument; but they
often accept, or at least show respect for, closely related arguments that are concerned
with qualitative mental properties, like pain and the sensation of heat. Continuing to
prescind from irrelevant details, we can represent the view of these philosophers by
saying that they see merit in arguments like this one:
B. First premise: We are able to conceive of pains existing without being accompanied by any
physical events.
Second premise: If we are able to conceive of pains existing without being accompanied by any
physical events, then it is objectively possible for this to occur.
First Lemma: It is objectively possible for pains to exist without being accompanied by physical
events.
Second Lemma: It is objectively possible for instances of the property pain to exist without
being accompanied by instances of any physical property.
Third premise: If it is objectively possible for instances of a property P to exist without being
accompanied by instances of a property Q, then P is not identical with Q.
Conclusion: The property pain is not identical with any physical property.
 CONCEIVABILITY AND POSSIBILITY 289

The same philosophers also see merit in arguments that begin as follows, and then
proceed along the same lines as argument B.:
C. First premise: Where E is any type of physical event, we are able to conceive of events of type
E existing without being accompanied by pains.
Second premise: If we are able to conceive of events of type E existing without being accom-
panied by pains, then it is objectively possible for this to occur.

As the reader will have noticed, none of these arguments makes a general claim about
the relationship between conceivability and possibility. None of them asserts that
conceivability is always or even usually a reliable source of modal knowledge. Instead
they make comparatively modest claims concerning conceiving that are restricted to
mental phenomena. Those who favor the arguments would maintain, I believe, that
this restriction protects them from the strictures concerning conceivability that have
been put forward in earlier sections. I will be concerned in the present section to
evaluate this view.
Iwill focus exclusively on arguments like B. Ishare the general skepticism concerning
arguments like A. Moreover, while there are interesting differences between arguments
like B and arguments like C, what I will say concerning B can be reformulated so as to
apply to C. In discussing B, I will understand it to be concerned with metaphysical
possibility. This is in keeping with the intentions of contemporary advocates of such
arguments.
Now at first sight, B appears to have a substantial flaw. As we saw in Section IV,
Kripke and Putnam put forward persuasive reasons for thinking that knowledge of
the essential properties of empirical kinds is a posteriori. More specifically, they
maintained that natural science provides our only reliable access to the essential
properties of empirical kinds. Now B is concerned with pain, and on the face of it,
pain is an empirical kind. Our knowledge of pain comes from experience. But
B presupposes that it is possible to grasp the essential nature of pain simply by
deploying the faculty of conception. Thus, according to the second premise of B,
conception alone suffices to establish that pain can exist without being accompanied
by physical phenomena, and this implies that conception alone can be used to show
that physical phenomena are not implicated in any way in the essential nature of
pain. Hence, the second premise seems to be flatly opposed to the lessons we learned
from Kripke and Putnam. Either it is wrong or Kripke and Putnam were mistaken.
Although this objection was stated with considerable force by Kripke himself, he
did not see it as fatal, and in fact he formulated an ingenious reply to it. The central
thesis of his reply is that pain is a special case. Unlike other empirical kinds,
qualitative empirical kinds like pain are known to us directly. We are immediately
acquainted with them. As a result of this privileged access, we are intimately familiar
with their essential natures. In the case of pain, for example, acquaintance provides us
with an exhaustive knowledge of what it is. Further, in conceiving of pain, we rely on
concepts that are based squarely on immediate acquaintance, and that therefore
290 PART IV: KNOWLEDGE

reflect all of the features of pain that acquaintance reveals. It follows that in conceiv-
ing of situations involving pain, our conceiving is constrained by the essential nature
of pain. But this means that when we conceive of a situation involving pain, that
situation really is an objective possibility. In other words, the second premise of B is
acceptable.
In formulating this line of thought, Kripke relied heavily on the notion of picking
pain out. Thus, for example, he contrasted the way in which we pick heat out with the
way in which we pick pain out. Heat, he maintained, is picked out as the external
phenomenon that causes the sensation of heat. In other words, it is picked out as the
external phenomenon that gives rise to a certain sort of appearance. But there is no
appearance/reality distinction in the case of pain: “in the case of a mental phenomenon,
there is no ‘appearance’ beyond the mental phenomenon itself” (Kripke 1980, p. 154).
Hence, pain is picked out simply as an experience of a certain sort—specifically, as the
experience of pain: “Pain, on the other hand, is not picked out by one of its accidental
properties; rather it is picked out by the property of being pain itself, by its immediate
phenomenological quality” (Kripke 1980, p. 152). This immediate phenomenological
quality is the essence of pain; to grasp it is to grasp pain’s essential nature, and to grasp
that nature in its entirety.
What should we make of this set of views? It seems true enough that we are directly
acquainted with pain, if this means only that awareness of pain is not mediated by
awareness of something else. However, Kripke seems to think that if we are directly
acquainted with pain, in the given sense, then we somehow have full access to the
essential nature of pain. That is to say, he thinks that our epistemic relationship to
pain must allow us full access to the modal facts about pain—and in particular, that it
must allow us to answer such questions as whether it is possible for pain to exist
without being accompanied by any physical kind. When we reflect, however, we see
that the fact that a phenomenon is self-presenting does not imply that we have full
access to its essential nature. To say that we are directly acquainted with a property P
is simply to say that there is no other property Q such that we are aware of P in virtue
of being aware of Q. But even if we are directly acquainted with a property in this
sense, it can still be true that we do not have epistemic access to it that enables us to
fully grasp its essential nature. For it can still be true that our awareness of P
constitutively involves a representation of P, and it may be that this representation
fails to fully reveal the essential properties of P.
In other words, Kripke appears to be presupposing the following principle:
If we have direct epistemic access to pain, then when we imagine a possible situation that
contains a pain, we automatically have full access to the essential nature of the pain we are
imagining. Hence, since nothing physical presents itself to us when we are imagining the pain,
we can be sure that we are not imagining anything physical in virtue of imagining the pain.

This principle claims that there is a close relation between direct access to the
essential nature of a pain and complete access to its essential nature. It is far from
 CONCEIVABILITY AND POSSIBILITY 291

obvious that this claim is true. Indeed, it is closely related to a presupposition of

Descartes’ original argument that was challenged by Arnauld. (In his original version
of the argument, Descartes took it as a premise that if we are able to conceive of X
clearly and distinctly, then X is objectively possible. Arnauld objected to this principle
on the grounds that our conception of X might fail to represent all of X ’s essential
properties. His example involved conceiving of a right triangle. We are able to
conceive of a right triangle without conceiving of it as satisfying the Pythagorean
condition that the square of the hypotenuse is identical to the sum of the squares of
the sides, but it isn’t objectively possible for a right triangle to exist without satisfying
this condition (Arnauld 1984, pp. 139–42). Descartes saw the force of this criticism
and struggled to cope with it, but few readers have found his response satisfactory.)
To make this more concrete, suppose for a moment that pain can be identified
with a type of bodily disturbance, and that awareness of pain involves a perceptual
representation of a disturbance of the given type. On this view, it is possible to draw a
Kantian appearance/reality distinction with respect to awareness of pain: there is
pain as it is represented perceptually, and pain as it exists in itself. Moreover, as Kant
would insist, it may well be the case that the appearance of pain fails to do justice to
the reality. This is true even if the representation that registers pain represents it
directly. Thus, suppose that pain is in fact represented directly by some representa-
tion R—suppose that there is no property P such that R represents pain by virtue of
representing P. It could still be true that R fails to provide awareness of all of the
essential properties of pain, because it could still be true that R lacks the right internal
structure to articulate those properties. What is in fact a complicated physical
structure could appear, from the perspective of someone whose awareness of pain
is determined by R, as a simple, undifferentiated quality.
I am not claiming at the moment that pain is a bodily disturbance, or that our
access to pain is perceptual, though I believe there is abundant evidence that these
propositions are true (Hill 2009 and chapters 9–11 of this volume). I am just claiming
that they are epistemic possibilities, and that they therefore must be ruled out before
it can be maintained with confidence that our epistemic access to pain puts us in
touch with the essence of pain. One cannot defend the latter view simply by pointing
out that pain is self presenting, or that we are directly aware of pain. But that is what
Kripke tries to do.
It may be useful to state this point in a different way. For an a priori test to be reliable
in the case of “Pain is not identical with any physical kind,” it would have to be true
that we have an a priori grasp of the full essential nature of pain. But there are credible
stories about our epistemic access to pain which deny that this is so. In particular, a
story which claims that awareness of pain consists in entertaining a perceptual
representation of bodily damage may deny it. Before we can use conceivability to
determine whether the given sentence is metaphysically possible, therefore, it would be
necessary to rule out perceptualist accounts of awareness of pain on independent
grounds. Since it is very unlikely that this can be done a priori, it is a mistake to
think that it is possible to test the given sentence for metaphysical possibility in a purely
292 PART IV: KNOWLEDGE

a priori fashion. Unconstrained conceivability fails as a guide to objective possibility

even in the very special case involving qualitative mental states.
Cartesian modal arguments face additional difficulties. For example, they presup-
pose the erroneous thesis that it is possible for mere conception to afford us an
exhaustive grasp of the essential natures of physical phenomena. But I think enough
has been said to show that in their present form, at least, such arguments are
hopelessly simplistic.

VII. Conclusion
I have claimed that various conceptions of necessity are presented to us by implicit
definitions, and that these definitions constrain the use of conceivability in testing for
the corresponding forms of possibility. There are alternative ways of approaching
these topics that I have not considered here, but reflection shows, I believe, that the
present approach provides the most unified and least problematic account of the
various notions of necessity, and also the most plausible account of modal know-
ledge. I illustrate this claim in the following Appendix.

Appendix
In this Appendix I will say something about an alternative proposal concerning metaphysical
necessity that has seemed right to me at various times in the past, and that appears to be
enjoying a modest surge in popularity at present. According to the proposal I have in mind,
metaphysical necessity can be defined in terms of the subjunctive conditional. There are three
different forms that such a definition might take:

(a) It is metaphysically necessary that P = df if it were true that not-P, then it would also be
true that P.
(b) It is metaphysically necessary that P = df it would be true that P no matter what else was
true.
(c) It is metaphysically necessary that P = df if it were true that not-P, then a contradiction
would also be true.

Lewis focused on (a) in Counterfactuals (Lewis 1973); Williamson and I proposed versions of
(b) several decades later (Williamson 2005, Hill 2006); and in his recent work, Williamson has
tended to emphasize (c) (Williamson 2007).
To appreciate the plausibility of these proposals, recall the standard view concerning the
truth conditions of subjunctive conditionals:

(*) A subjunctive conditional “If it were the case that A, then it would be the case that C” is true
at a metaphysically possible world W just in case either (i) A is not true at any metaphysically
possible world, or (ii) there is a metaphysically possible world at which A and C are true that is
more similar to W than any metaphysically possible world at which A and not-C are true.
 CONCEIVABILITY AND POSSIBILITY 293

On the assumption that (*) is at least roughly correct, it is possible to argue for version (a) of
the theory as follows: “Suppose that P is metaphysically necessary. Then condition (i) of (*) is
satisfied, because not-P is false at every metaphysically possible world. Given (*), it follows that
‘If it were true that not-P, then it would also be true that P’ is true. Suppose now that this
conditional is true. Given (*), it follows that either (i) not-P is not true at any metaphysically
possible world, or (ii) there is a metaphysically possible world at which not-P and P are both
true. Now if (i) obtains, P is metaphysically necessary. But (i) must obtain, since (ii) is absurd.
Hence, P is necessary.” It is possible to give closely related arguments for proposals (b) and (c).
The subjunctive theory of metaphysical necessity is appealing for several reasons, one being
that it helps to explain the perceived importance of metaphysical necessity. Since the subjunct-
ive conditional plays a role in a variety of important cognitive processes, notions that can be
explained in terms of it have a reflected glow. Another advantage is that the theory holds out
the promise of being able to explain knowledge of the metaphysical modalities in a uniquely
appealing way. We know that there are powerful cognitive modules that process subjunctive
conditionals, for it is a datum that we are capable of evaluating subjunctives, even if they are
quite complicated, across a wide range of contexts. If the metaphysical modalities are reducible
to subjunctive conditionals, we could enlist these modules in explaining how we acquire
knowledge of the metaphysical modalities. Or so it can seem.
Unfortunately, however, these virtues of the subjunctive theory are balanced by difficulties.
A particularly serious difficulty arises from clause (i) of (*). As is illustrated by the line of
thought two paragraphs back, this clause will play a crucial role in any defense of the theory.
Now in effect, (i) claims that all subjunctive conditionals with impossible antecedents are
vacuously true. There are things that can be said in defense of this claim, but it is put into doubt
by pairs of conditionals like the following, all of which have metaphysically impossible
antecedents:

(1) If Obama had had different parents, he would have had different DNA.
(1’) If Obama had had different parents, he would have been 6 inches tall.
(2) If a higher grade of cement had been used in the construction of Murphy Tunnel, the
tunnel wouldn’t have collapsed.
(2’) If a higher grade of cement had been used in the construction of Murphy Tunnel, the
tunnel would have been five miles long.
(3) If classical logic was false, the problem would lie with the law of the excluded middle.
(3’) If classical logic was false, the problem would lie with modus ponens.

Propositions (1)–(3) strike us as true, or at least as reasonable things to say; but it would be
wrong to say that they strike us as vacuously true, for if they did, (1’)–(3’) would also strike us
as true, whereas they clearly strike us as false. Examples of this sort suggest that we should
explore alternative ways of explaining subjunctives. More particularly, they suggest that we
should do away with clause (i) of (*) and reformulate clause (ii) in such a way that it is based on
a system of worlds that includes impossible worlds, including worlds in which Obama had
different biological parents, worlds in which the Murphy Tunnel was constructed of different
material at its origin, and worlds in which classical logic is false. But of course, if we base our
account of subjunctive conditionals on a system that includes impossible worlds, we will not be
able to explain metaphysical necessity in terms of subjunctives, for it will not be true that
metaphysically necessary propositions are true in all of the relevant worlds. Where A is any
294 PART IV: KNOWLEDGE

proposition that counts intuitively as metaphysically necessary, there will be members of the
relevant space of worlds in which A is false.
There is another way of dealing with the problem posed by pairs of subjunctives like the
ones under consideration, but it is somewhat Byzantine compared to the solution that invokes
impossible worlds, for it commits us to a complex theory of the intentions of speakers
who assert propositions like (1), (2), and (3). On this view, while (1) is vacuously true, speakers
who assert it intend to communicate a quite different proposition, the proposition that
Obama’s biological parents are causally responsible for his DNA. Equally, speakers who assert
(2) intend to communicate the belief that the inferior quality of the cement was causally
responsible for the collapse of the tunnel, and speakers who assert (3) intend to communicate
the belief that the law of the excluded middle is less obviously trustworthy than the other laws
of classical logic. More generally, the thesis here is the Gricean idea that speakers recognize the
triviality of subjunctive conditionals with impossible antecedents, and exploit that triviality in
communicating more interesting propositions (Grice 1975). According to this idea, they assume
that their conversational partners will also be aware of the triviality of the propositions, and that,
being disinclined to believe that the speakers intend to make trivial claims, their partners will be
moved by this perception to consider other propositions that the speakers may have intended to
communicate, fixing finally on the ones that the speakers actually have in mind.
These claims have a certain plausibility, and at the end of the day, it may turn out that
Gricean ideas provide an acceptable solution to the present problem. It should be noted,
however, that the Gricean account of propositions like (1)–(3) is considerably more complex
than the account that follows from the preceding theory, according to which the truth
conditions of such propositions are best explained in terms of impossible worlds. An advocate
of the preceding theory can say that a speaker will assert (1) just in case he or she believes that
the truth condition of (1) is satisfied, but an advocate of a Gricean approach must say that
asserting (1) is a complex affair—an affair that involves determining that the truth condition
(1) is trivially satisfied, but that also involves fixing on another proposition that is true and that
it would be useful to convey, and working out that the speaker’s conversational partner will be
able to recognize his or her intention to communicate that other proposition. This additional
complexity appears to be a liability, for as far as I can see, there is neither introspective nor
experimental evidence that we actually have the complex Gricean intentions that the second
theory attributes to us. To be sure, if a speaker were to assert (1), she would no doubt expect her
conversational partner to infer that Obama’s parents are causally responsible for his DNA. But
she would expect this because the proposition is a natural inference from (1) itself, not because
she thinks that she can best get this proposition across by saying something too trivial to be taken
at face value. In other cases, there is introspective evidence that Gricean intentions underlie
assertions of propositions. A good illustration is Grice’s example of the professor who writes a
letter of recommendation saying only that a certain job candidate has excellent handwriting. We
all know that if we were to write such a letter, we would be counting on the recipient to notice the
triviality of our claim, and to infer from it that we do not have much confidence in the candidate.
It would be our intention to damn the candidate with faint praise. But it is not at all obvious that
we are guided by similar intentions in asserting propositions like (1)–(3).
A further point is that while the Gricean account presupposes that it is common knowledge
that subjunctives with impossible antecedents are vacuously correct, this is actually quite far
from the truth. If it was common knowledge that such subjunctives are vacuous, we would not
 CONCEIVABILITY AND POSSIBILITY 295

be so perplexed by the problem that is posed by propositions like (1)–(3). Indeed, the problem
of how best to interpret them would not arise. Even Lewis, who was on the whole a stalwart
champion of (*), saw the doctrine of vacuous truth as a conjecture, acknowledging that the
reasons in favor of it are “less than decisive” (Lewis 1973, p. 25).
I observe finally that the claim of the subjunctive theory to provide us with a uniquely
appealing account of knowledge of metaphysical necessity is spurious. According to the theory,
we come to know that a proposition P is metaphysically necessary by coming to know that a
subjunctive conditional is true—say, to invoke definition (b), by coming to know the condi-
tional “P would be true if any proposition were true.” But how do we arrive at knowledge of the
conditional? Presumably by showing that P can be obtained as the conclusion of a subjunctive
argument no matter what proposition is taken as premise. But it would be impossible to show
this unless there were a rule authorizing us to assert P at any stage in the construction of a
subjunctive argument. That is, there would have to be a rule telling us that it is permissible to
take P as a free assertion in any piece of subjunctive reasoning. Suppose, for example, that P is
the proposition that George VI is the father of Elizabeth II. How could it be true that this
proposition is derivable from any premise we might adopt in the course of subjunctive
reasoning? Reflection shows that this could be true only if there is a rule which says that if x
is a biological parent of y, then it is permissible to assert a proposition to that effect at any point
in any stretch of subjunctive reasoning. Now there is very little difference between a rule of this
sort and a proposition of the following form:

($) If x is a biological parent of y, then for any proposition P, if it were the case that P, then it
would be the case that x is a biological parent of y.

In adopting the rule, one is in effect adopting ($) as an axiom. So it is natural and appropriate
to think of the subjunctive theory of necessity as committed to a set of propositions like ($). But
this means that the subjunctive theory has essentially the same structure as the implicit
definition theory that we considered in Section IV, for the heart of that theory is a set
propositions like (#):

(#) If x is a parent of y, then it is metaphysically necessary that x is a parent of y.

Thus, while the subjunctive theory takes a range of propositions like ($) to be constitutive of
subjunctive reasoning, and therefore, to be constitutive of the subjunctive conditional itself, the
implicit definition theory takes the corresponding range of propositions like (#) to be constitutive
of metaphysical necessity. There is a deep structural parallel between the two theories. And by the
same token, there is a deep parallel between the ways in which the two theories explain
knowledge of metaphysical necessity. The subjunctive theory explains it in terms of our grasp
of definition (b) (or one of the equivalent definitions) and our acceptance of a host of proposi-
tions like ($), and the implicit definition theory explains it in terms of our acceptance of the
corresponding host of propositions like (#). If we evaluate the two theories simply on the basis of
their explanation of modal knowledge, it seems that there is very little basis for giving preference
to one over the other. It is wrong to suppose that the subjunctive theory has an explanatory edge.
In this most recent part of the discussion I have been assuming for the sake of argument that
the subjunctive theory of necessity is viable. But of course, the arguments of earlier paragraphs
call this assumption into question. One might have thought that there is at least reason to hope
296 PART IV: KNOWLEDGE

that the Gricean defense of the subjunctive theory can be made to work, because the subjunct-
ive theory offers an explanation of modal knowledge that is uniquely privileged. The last
paragraph is meant to show that this thought is misguided.

References
Arnauld, A. (1984). Fourth Set of Objections, in J. Cottingham, R. Stoothoff, and D. Murdoch
(eds.), The Philosophical Writings of Descartes, Vol. II. Cambridge: Cambridge University
Press, 138–53.
Boghossian, P. (1997). “Analyticity.” In B. Hale and C. Wright (eds.), A Companion to the
Philosophy of Language. Oxford: Blackwell Publishers Ltd., 331–68.
Byrne, R. M. J. (2007). The Rational Imagination. Cambridge, MA: MIT Press.
Descartes, R. (1984). “Meditations on First Philosophy.” In J. Cottingham, R. Stoothoff, and
D. Murdoch (eds.), The Philosophical Writings of Descartes, Vol. II. Cambridge: Cambridge
University Press, 3–62.
Edgington, D. (2004). “Two Kinds of Possibility,” Proceedings of the Aristotelian Society
Supplementary Volume, vol. 78: 1–22.
Gelman, S. (2003). The Essential Child: Origins of Essentialism in Everyday Thought. Oxford:
Oxford University Press.
Gibson, J. J. (1979). The Ecological Approach to Visual Perception. Boston: Houghton Mifflin.
Grice, H. P. (1975). “Logic and Conversation.” In D. Davidson and G. Harman (eds.), The
Logic of Grammar. Encino, CA: Dickenson, 64–75.
Hill, C. S. (2006). “Modality, Modal Epistemology, and the Metaphysics of Consciousness.” In
S. Nichols (ed.), The Architecture of the Imagination. Oxford: Oxford University Press,
205–35.
Hill, C. S. (2009). Consciousness. Cambridge: Cambridge University Press.
Keil, F. (1989). Concepts, Kinds, and Cognitive Development. Cambridge, MA: MIT Press.
Kripke, S. (1980). Naming and Necessity. Cambridge, MA: Harvard University Press.
Kripke, S. (2011). “Identity and Necessity.” In Kripke, Philosophical Troubles. Oxford: Oxford
University Press, 1–26.
Lewis, D. (1973). Counterfactuals. Oxford: Blackwell Publishing.
Mackie, P. (2006). “Transworld Identity.” In Stanford Encyclopedia of Philosophy. <http://
plato.stanford.edu/entries/identity-transworld/>.
Putnam, H. (1975). “The Meaning of ‘Meaning’.” In Putnam, Mind, Language and Reality.
Cambridge: Cambridge University Press, 215–71.
Sidelle, A. (1989). Necessity, Essence, and Individuation. Ithaca, NY: Cornell University Press.
Williamson, T. (2005). “Armchair Metaphysics, Metaphysical Modality and Counterfactual
Thinking.” Proceedings of the Aristotelian Society 105: 213–35.
Williamson, T. (2007). The Philosophy of Philosophy. Oxford: Blackwell Publishing.
 16
Concepts, Teleology, and Rational
Revision*

I. Introduction
Are all of our beliefs susceptible to empirical evaluation, and therefore at risk, in
principle at least, of being called into question by empirical evidence? That is, is every
belief fundamentally empirical in character? Or is there a special class of beliefs that
are immune to empirical revision?
As is well known, W. V. Quine argued forcefully in “Two Dogmas of Empiricism”
(Quine 1980) for the first of these two options. According to him, our grounds for
any belief are mainly empirical, though supraempirical virtues like simplicity, fruit-
fulness, and continuity with tradition may also play a role in determining whether a
proposition should be believed, at least in the case of highly theoretical hypotheses.
Moreover, every belief is in principle revisable in the light of experience.
Although revolutionary when they were first put forward, these views of
Quine’s have come to be widely accepted, even to the point of having some claim
to be considered the received position. As I see it, however, there is still a great deal
to be said for the other side. I will make a case for that side in the present chapter.
More specifically, I will argue for two main theses. One of these is the claim that a
number of our beliefs cannot be called into question by empirical evidence. This is
true, for example, of definitional beliefs, such as the belief that a fortnight is a period
of fourteen days, and also the laws of classical logic. Now it might seem at first that
this claim is all that one needs to secure immunity to revision. But even if a belief is
not at risk of empirical refutation, it might still be true that the motivation for
holding the belief could be undermined by empirical evidence. Thus, for example,

* Earlier versions of this paper were presented to a discussion group at Australian National University
in 2007, a conference honoring Ruth Millikan at the University of Connecticut in 2008, a conference on
thought experiments and the a priori in Forteleza, Brazil in 2009, a conference on imagination and the a
priori at the University of St. Andrews in 2010, and a meeting of the Society for Philosophy and Psychology
in 2011. I learned a great deal from the discussions on all those occasions. I have also been helped
considerably by the comments of Alex Byrne on one of the earlier versions, and by advice from Jamie
Dreier, Ivan Fox, Hilary Putnam, and Joshua Schechter. My deepest debts are to David Christensen, Anil
Gupta, and Vann McGee, who have placed their very considerable abilities at my disposal on a number of
occasions.
298 PART IV: KNOWLEDGE

even if there cannot be evidence that disconfirms the laws of classical logic, empirical
discoveries might still provide motivation for adopting an alternative to classical
logic, such as quantum logic. Or so it might seem. My second thesis will challenge
this view, maintaining that the motivation for embracing certain empirically irrefut-
able beliefs is non-empirical, at least in part, and that it therefore cannot be under-
mined or undercut by empirical discoveries. In sum, I shall argue that certain of our
beliefs are immune to empirical revision in two senses: they cannot be called into
question by empirical evidence, and they cannot be deprived of their cognitive value
by such evidence.

II. Quinean skepticism

Since Quine’s attack on immunity to revision is widely regarded as decisive, it will
behoove us to take a look at it before proceeding to consider the prospects of
constructing a positive account of this notion. I will also discuss Quine’s critique of
analyticity, because, on some accounts of analyticity, analyticity and immunity to
revision come to much the same thing. At one point Quine himself endorses such a
view, saying that analytic statements “hold come what may” (Quine 1980, p. 43).
“Two Dogmas of Empiricism” begins with a critique of various traditional
attempts to explicate analyticity. Quine considers the following proposals:
(a) A statement is analytic if it is true in all possible worlds.
(b) A statement is analytic if its denial is self contradictory.
(c) A statement is analytic if its predicate attributes to its subject no more than is
already conceptually contained in the subject.
(d) A statement is analytic if it is true by virtue of meanings and independently of
fact.
(e) A statement is analytic if it is logically true or it can be turned into a logical
truth by putting synonyms for synonyms.
(f ) A statement is analytic if it is true in virtue of definitions of its constituent
terms.
(g) A statement is analytic if it is true in virtue of the semantical rules that govern
the use of its constituent terms.
Roughly speaking, Quine’s view is that the ideas that figure in these definitions—
ideas like meaning, synonymy, and definition—are highly problematic because they
are not embedded in well articulated theories with determinate predictive content. To
put the point another way, although the ideas are all definable in terms of one
another, they do not have any robust definitional or doctrinal links to concepts
that have a clear predictive or explanatory role. (Here and elsewhere in the present
section, I follow Quine in framing the discussion in terms of words and statements
rather than concepts and propositions. This will reduce possible distortions in
reporting Quine’s views. I hope it will be clear that what is said in the present
 CONCEPTS , TELEOLOGY, AND RATIONAL REVISION 299

section can easily be reformulated so as to apply to the universe of discourse of other

sections.)
According to a familiar interpretation of the argument (Grice and Strawson 1957),
its main premise is the claim that it is necessary to appeal to analyticity itself in
explaining the various notions that figure in the traditional definitions of analyticity.
On this interpretation, Quine’s objection to analyticity boils down to the claim that
circular definitions are illegitimate. It is customary among those who read Quine in
this way to respond by saying that definitional circles can be perfectly acceptable.
They then illustrate this claim with examples. Reflection shows this interpretation is
quite misguided. Quine’s point is not that it is necessary to appeal to analyticity in
explaining key concepts in the definitions, but that those concepts are as lacking in
clarity as analyticity itself. They cannot be used to explain analyticity because they do
not have clearly defined roles in fruitful cognitive endeavors.
My objection to the argument is quite different than this standard objection. It is
that Quine’s list of explanations fails to acknowledge a number of promising pro-
posals. Thus, the list makes no mention of proposals that make use of teleological
notions, such as the proposal that a statement is analytic if acceptance of the
statement is necessary for one or more of its constituent terms to serve fundamental
cognitive interests. Nor does it address proposals that invoke epistemic notions, such
as the proposal that an analytic statement can be identified as one that we are justified
in believing simply in virtue of the most basic dispositions or commitments govern-
ing the use of the terms that serve as its constituents. (Cf. Boghossian 1997.) Quine’s
list also fails to take note of proposals that are based on the idea of explanation. An
example is Paul Horwich’s proposal (Horwich 1999) that the analytic truths involv-
ing a term are the ones that make a special, foundational contribution to the task of
explaining the overall use of the term. To give one more example, the list fails to
acknowledge the possibility that analyticity can be explained in terms of the notion of
correct use. According to this idea, a statement S is analytic if S contains a term T
such that an agent must accept S in order to count as knowing how to use T correctly.
To be sure, as things now stand, we are unable to say much in a theoretical vein about
what mastery of use consists in; but we have considerable knowledge of what correct
use involves in specific cases. Generally speaking, we are able to recognize mastery of
use when we see it. This is shown by the fact that we are able to correct language
learners when they go astray, by the fact that we are able to compile dictionaries that
command wide assent, and by the fact that we are in practice able to discriminate
between linguistic differences and factual disagreements. Mastery of use is an ability,
and so is the capacity to recognize mastery of use. As with the ability to play the piano
and the ability to recognize grammatical sentences, it is possible in principle for there
to be a scientific account of what these abilities consist in.
After delivering his critique of traditional accounts of analyticity, Quine begins an
extended argument for his revisability thesis, according to which it is in principle
possible for any statement whatsoever, including even a law of logic or a law of
300 PART IV: KNOWLEDGE

mathematics, to be called into question by empirical evidence. This argument then

provides the basis for a sustained attack on the notion of a priori knowledge.
According to the main traditional conception of the a priori, an agent A can be
said to know a priori that a statement S is true just in case the following three
conditions are satisfied: (i) A knows that S is true, and therefore has a true justified
belief it is true; (ii) A’s justification for believing S is independent of empirical
evidence; and (iii) A’s belief in S is immune to empirical revision, in the sense that
the course of experience cannot provide A with an epistemic ground for rejecting S.
Quine’s revisability thesis challenges clause (iii) of this account, implying that there
are no statements that satisfy it.
While it is not altogether easy to see how Quine proposed to defend his revisability
thesis, I think that the preponderance of the evidence shows that he had two
arguments in mind.
The first argument consists in an appeal to the history of science. Quine never
makes this argument fully explicit, but it is plausible that it can be formulated as
follows:
First premise: If any proposition is empirically unrevisable, then the laws of logic are empir-
ically unrevisable.
Second premise: The laws of logic have in fact been called into question by empirical
discoveries.
Conclusion: No proposition is empirically unrevisable.

As a passage in “Two Dogmas of Empiricism” makes clear, Quine would have

defended the second premise by claiming that certain logical laws have been chal-
lenged by quantum mechanics. (See Quine 1980, p. 43. Quine 1986, pp. 85–86, offers
a more nuanced treatment.)
In thinking about this first argument, it is important to distinguish between the
claim that it is rational to consider replacing classical logic with quantum logic, and
the claim that it is rationally obligatory to make such a replacement. It appears that
most contemporary philosophers of science reject the second claim. Reflection
shows, however, that Quine is only committed to the first claim. His revisability
thesis just asserts that it is in principle possible for there to be evidence that makes it
rational to revise logic. This can be defended by simply pointing to rational debate
among informed parties about whether to undertake a revision. There is no need to
point to cases in which a revision has actually occurred.
Quine’s second argument for the revisability thesis derives from a certain concep-
tion of rational revision—a conception that Quine takes to be implicit in scientific
practice, and also to be defensible by a normative argument (not that he distinguishes
sharply between these two endeavors). When a theory generates an experimental
prediction that turns out to be false, Quine points out, there are always a number of
different ways of revising the theory so as to cancel the prediction. Thus, where
X1, . . . , Xn is a list of the statements and rules of inference that play a role in the
 CONCEPTS , TELEOLOGY, AND RATIONAL REVISION 301

derivation of the prediction, one could in principle block the prediction by revising
any one of them. Which Xi will be chosen? Quine’s answer is that the revision will be
based, and should be based, on three global properties of the theory in which the Xi’s
are embedded—specifically, predictive power, simplicity, and continuity, where
continuity is taken to be a matter of similarity between the pre-revision theory and
the post-revision theory. One should try to make a revision that will maximize a
weighted average of these three variables. What exactly does this answer count
against immunity to revision? Because nothing has been said about preserving the
Xi’s that belong to logic or mathematics, or that count intuitively as immune to
revision for some other reason. If a particular Xi belongs to mathematics, say, it will
be preserved just in case it is a component of the revised theory that best satisfies the
three criteria. There is no guarantee that any given statement will pass this test. But is
it really rational for us to allow logic and mathematics to be susceptible to this sort of
revision? This is just the question of whether it is really true that predictive power,
simplicity, and continuity are the only three desiderata that we are rationally required
to consider. And Quine thinks we will in the end agree that they are. After all, he will
say, the principal goal of cognition is the efficient prediction and control of nature.
This is our primary cognitive interest. Moreover, he will add, it is clear that the three
desiderata are the properties of theories that best promote this interest.
As I see it, then, Quine’s second argument for the revisability thesis is based
ultimately on a doctrine that is sometimes associated with pragmatism—the doctrine
that efficient and accurate prediction is the main purpose of cognition. I believe that
it is this doctrine that leads him to describe his position as involving a “shift toward
pragmatism.”
Turning now to the question of whether the foregoing arguments are sound, I will
argue that they both suffer from substantial flaws.
In the first place, it is far from clear that the example involving classical logic and
quantum mechanics is a case of empirical revision in the relevant sense. A relevant
case would be one in which there were empirical grounds for rejecting a statement
that appears to be a priori. Instead of seeing quantum mechanics as providing a
reason to reject classical logic, and to replace it with quantum logic, it is possible, and
indeed natural, to see quantum mechanics as providing a reason to add the vocabu-
lary of quantum logic to our logical repertoire, and as providing a reason to use
concepts from this new logic in formulating and developing microphysics, while
continuing to use classical logic in formulating and developing such branches of
knowledge as mathematics, philosophy, and law. In other words, instead of seeing
quantum mechanics as providing reasons for replacing an old logic with a new one, it
is natural to see it as providing reasons for supplementing an old logic with a new one,
and for using these two logics in tandem. To be sure, it is not exactly a trivial matter
to combine logics—as J. H. Harris pointed out (Harris 1982), it can happen that the
wall between two logics collapses when they are combined to form a single system.
Accordingly, if one wished to combine two logics, it would be necessary to impose
302 PART IV: KNOWLEDGE

restrictions on the use of each of the logics to prevent them from collapsing into one
another. It appears, however, that relatively weak restrictions would suffice, and more
specifically, that it would suffice to adopt restrictions that slightly modify the scope of
classical logic without changing its substance. (If classical logic and quantum logic are
presented as natural deduction systems (i.e., as based on systems of rules of inference
like the rules (i)–(iii) that are cited in Section III), it suffices to adopt a restriction
forbidding one to apply rules belonging to one of the systems within derivations
constructed in accordance with rules that belong to the other system. For discussion
of the different case in which two logics are presented as axiomatic systems, see
Schechter (forthcoming). Schechter shares my view that it is possible to prevent
collapse by adopting restrictions that preserve the basic identity of classical logic.)
It seems, then, that there is a strategy for dealing with empirical challenges to logic
that has the effect of preserving the logical commitments that have traditionally been
seen as a priori. Instead of advocating wholesale replacement, the strategy recom-
mends that we supplement familiar vocabularies and systems of belief with new ones,
and continue to use the familiar items to serve the non-empirical purposes that they
have always served. It seems fair to say that any argument concerning the bearing of
developments in the history of science on questions of immunity to revision has an
obligation to take this strategy into account. But Quine’s brief discussions do not
acknowledge it, nor does the formal argument that those discussions suggest.
Accordingly, the argument cannot stand on its own. To make it work, it would be
necessary to show that the alternative strategy is unsatisfactory. It is far from clear
that this could be done.
There is also a problem with Quine’s second argument, which derives his revisa-
bility thesis from the idea that the efficient prediction of experimental data is our
dominant cognitive concern, and the only factor that should play a role in fixing
belief. Reflection shows that this idea is quite wrong. We have a number of cognitive
interests that cannot be reduced to efficient prediction. Thus, to give an especially
simple example, we have a legitimate interest in introducing abbreviations for
complex concepts. It is of course true that concepts that serve this interest tend to
simplify our predictive instruments and thereby make them more efficient; but our
interest in abbreviations is not reducible to our empirical concerns. It would be
reasonable to employ abbreviative concepts even if we were animated mainly by
religious purposes, and had little interest in predicting nature or improving our
understanding of it. For another example, consider our mathematical beliefs. It is
implausible that we hold these beliefs on the basis of empirical evidence. Even so,
however, it is evident that they are quite important to us. We regard it as entirely
rational to seek to obtain more of them, even when doing so precludes one’s making
additions to one’s stock of non-mathematical beliefs. Thus, for example, if someone
devotes her life to describing a mathematically interesting structure, we are more
likely to applaud her choice than to condemn it. Moreover, we are likely to applaud
the choice even if it prevents her from acquiring a certain range of predictive abilities.
 CONCEPTS , TELEOLOGY, AND RATIONAL REVISION 303

Perhaps, if she had not been so preoccupied with pure mathematics, she could have
made a major contribution to empirical science. Will this reflection cause us shake
our heads over the tragedy of a wasted life? Of course not! Obtaining knowledge of
mathematically interesting structures seems to us to be of fundamental importance.
A third example is afforded by our modal beliefs. There are many questions of
possibility and necessity that we would like to be able to answer, even though the
answers will do little if anything to augment the predictive power of empirical
theories.
It is clear, then, that we have some cognitive concerns that are independent of our
interest in the efficient prediction and control of nature. Since it fails to acknowledge
this important fact, Quine’s second argument falls far short of establishing its
conclusion.
To summarize: Although Quine’s objections to analyticity and immunity to
revision are widely regarded as devastating, or at least as shifting the burden of
proof to advocates of these two notions, reflection shows that they actually contain
quite sizeable holes. A positive account of analyticity has nothing to fear from them,
nor does a positive account of what it is for a proposition to be constitutive of a
concept. Moreover, they leave defenders of the a priori with plenty of logical room in
which to operate.

III. Concepts and cognitive interests

As the reader will recall, Wittgenstein’s Philosophical Investigations (Wittgenstein
2001, p. 6e) encourages us to think of linguistic expressions as having functions:
Think of the tools in a tool box: there is a hammer, pliers, a saw, a screw-driver, a rule, a glue-
pot, glue, nails and screws.—The functions of words are as diverse as the functions of these
objects.

Wittgenstein is here speaking about words rather than concepts, but what he says
about words seems to me to be true of concepts as well: they have functions, and
those functions are highly diverse, differing among themselves as much as the
functions of a hammer, a saw, and glue.
It must be acknowledged that in many cases, it is very difficult to figure out what
cognitive interests a concept serves. Consider, for example, the concept of truth. It is
arguable that we had no clue as to the cognitive function of this concept, despite
many centuries of speculation about the topic, until Quine pointed to its role in
generalized and indefinite assertion (Quine 1986, p. 12). Similarly, although great
strides have been made in recent years, it is clear that many generations failed to
appreciate the complex functions of conditional concepts. Even today we seem to
lack fully adequate accounts of these concepts.
Still, the fact that cognitive functions are elusive in some cases should not blind us
to the fact that they are frequently transparent. Among other things, it is not hard to
304 PART IV: KNOWLEDGE

see that perceptual concepts serve an interest in encoding information about such
observable properties as sizes, shapes, and colors, that theoretical concepts serve an
interest in formulating laws of nature, and therefore an interest explaining and
predicting observable phenomena, and that logical concepts serve an interest in
framing propositions that can play certain roles in inference.
Moreover, in many cases, philosophical reflection has extended commonsense
knowledge of functions considerably. Consider, for example, the concept of disjunc-
tion. Casual observation can lead to the conclusion that the value of disjunction
derives in part from its ability to participate in such inferences as reasoning by cases,
disjunctive syllogism, De Morgan’s laws, and the abbreviative inference that leads
from pairs of propositions of the forms if p then r and if q then r to propositions of the
form if p or q then r. But philosophical reflection has gone beyond these observations.
Specifically, it has shown that all of the useful deductive inferences involving dis-
junction can be captured by rules of the following forms (together with appropriate
rules for the other connectives):
(i) For any class of propositions , if p is derivable from , then (p or q) is derivable from
as well.
(ii) For any class of propositions , if q is derivable from , then (p or q) is derivable from
as well.
(iii) For any class of propositions , if (p or q) is derivable from , r is derivable from p
together with the members of , and r is also derivable from q together with the
members of , then r is derivable from .

Since all of the useful deductive inferences involving disjunction follow from (i)–(iii),
it is possible to identify the inferential interests that disjunction serves simply by
enumerating these rules. But more: we also know that disjunction serves an epistemic
interest that is independent of our interest in having a concept that conforms to (i)–
(iii). More specifically, it serves an epistemic interest in forming propositions that we
can be justified in believing even when we aren’t justified in believing either of their
constituents. Thus, for example, we can be justified in believing it will either rain
tomorrow or snow even if we are not justified in believing either of the propositional
constituents of this claim. This much is apparent to common sense. Philosophical
and mathematical reflection have gone beyond this casual observation by providing a
specific rule for assessing the degree to which epistemic support for simpler proposi-
tions is transferred to disjunctions that contain them. Expressed as a formula
governing the probabilities of disjunctions, the rule comes to this:
ðRÞ If PrðeÞ> 0; then Prðp or q=eÞ ¼ Prðp=eÞ þ Prðq=eÞ  Pr ðp and q=eÞ
In combination with the other rules of the probability calculus that govern disjunc-
tion, this rule enables us to give a quite concrete account of the epistemic functions of
the concept. Accordingly, it is arguable that when (i)–(iii) are combined with the
rules of the probability calculus that govern disjunction, we have a package that can
 CONCEPTS , TELEOLOGY, AND RATIONAL REVISION 305

account for all of the most fundamental functions of that concept. To put the point in
somewhat different terms, given that disjunction serves our interest in having a
concept that conforms to (i)–(iii) and rules like (R), we can see how it serves the
full range of the interests that are associated with disjunction, including, among
many other things, our interest in having a concept that enables us to form new
beliefs by disjunctive syllogism, and our interest in being able to form justified beliefs
in complex propositions even though the available evidence falls short of justifying
belief in their constituents.
It appears, then, that by building on common sense, philosophy has made it
plausible that in a number of cases, at least, concepts are associated with sets of
cognitive interests that are rich enough to account for their individuation.
The view I wish particularly to recommend, however, is that questions about the
cognitive functions of concepts are empirical, and that science must therefore play
the dominant role in identifying such functions and describing their interconnec-
tions. In conjunction with common sense, philosophy can throw light on global
functions such as contributing to predictive and explanatory success, and can also
play a role in identifying the more specific functions that are performed by especially
simple concepts, like the logical constants. But the primary responsibility for devel-
oping a theory of the cognitive functions of concepts must lie with scientific disciples
such as psychology, anthropology, and the history of science. Questions about
cognitive functions are often too delicate, and too complex, to be settled by causal
observation and armchair reflection. Fortunately, science is already making lengthy
strides forward in this area, as can be seen by considering recent work in experi-
mental philosophy, and recent psychological research on such topics as categoriza-
tion, cognitive development, and taxonomy. Thus, for example, in her recent book on
conceptual development, Susan Carey describes empirical studies that answer a
number of questions about the cognitive functions of the concepts of infants and
young children, including their object concepts, their numerical concepts, and their
concept of causation (Carey 2009). She also describes a number of findings that
throw light on the cognitive interests served by a number of adult concepts, such as
the notions of weight, mass, volume, and density.
Thus far we have seen that common sense and philosophy have succeeded in
identifying a number of the general functions that concepts perform, and have also
taken note of some plausible claims about the concrete functions of specific concepts.
We have also seen that psychology is often able to go beyond common sense and
philosophy in specifying functions. Indeed, given the successful empirical inquiries
into functions that Carey reports, it is reasonable to expect that various scientific
disciples will someday converge on a general taxonomy of concepts, with positions in
the taxonomy being determined by cognitive functions.
In addition to maintaining that concepts serve cognitive interests, and that it is in
principle possible for such interests to be identified, I wish to claim that concepts
often serve cognitive interests that lie far afield from our empirical concerns. We can
306 PART IV: KNOWLEDGE

appreciate the merits of this third claim by considering the concepts that belong to
logic and mathematics. It is true that concepts of these two kinds support empirical
inquiry in various ways, but it is plausible that they also serve non-empirical
purposes. Thus, for example, it is plausible that the laws of classical logic play
essential roles in such non-empirical disciplines as law, mathematics, and theology,
not to mention large parts of philosophy. Because they serve non-empirical purposes,
the laws of logic could survive discoveries that called into question their utility in
empirical endeavors. Euclidean geometry provides additional evidence for the third
claim. In the early twentieth century empirical scientists gave up on the view that
Euclidean geometry provides an adequate model of physical space, but they con-
tinued to regard it as an exceptionally important piece of pure mathematics. More-
over, as a glance at the mathematical offerings at any university will attest, it retains
this status today. This could not be true if the purposes that Euclidean geometry
serves were purely empirical. No doubt it is still taken seriously for a variety of
reasons, but one of them is surely that it articulates a structure that is exemplified in
some possible worlds, and that human beings resonate to representations of that
structure psychologically, independently of whether they believe it to provide a
model of the physical world.
Speaking more generally, it seems that mathematical concepts serve an interest in
articulating intuitions about structures that are possible and have a powerful aes-
thetic appeal. This is not an especially informative characterization of mathematics,
and it certainly does not succeed in capturing all of the interests that lead mathemat-
icians to pursue their subject, but it does seem to be true as far as it goes, and it
suffices to make the point that mathematics is not a slave to empirical purposes. I will
elaborate a bit in Section VI.
Now if the cognitive interests that serve as the raison d’être of a concept are
entirely or largely non-empirical in character, then the commitments that govern the
daily use of the concept can authorize beliefs and/or inferences that are non-empir-
ical as well, in the sense of authorizing us to hold beliefs and engage in inferences
independently of the direction that the course of empirical evidence happens to take.
Indeed, the commitments that govern the use of the concept should authorize beliefs
and/or inferences that are independent of the course of empirical evidence, because
beliefs and/or inferences generally play essential roles in the proper functioning of
concepts. A belief or inference that makes it possible for a concept to serve a non-
empirical cognitive interest should not be held hostage to the course of evidence.

IV. Abbreviative definitions

The purpose of an abbreviative concept is not to add to the predictions of an
empirical theory, but to achieve economy of belief and reasoning. More specifically,
an abbreviative concept makes it possible to form propositions that (i) are shorter
than counterpart propositions that contain a complex concept, and (ii) have the same
 CONCEPTS , TELEOLOGY, AND RATIONAL REVISION 307

epistemic properties as their more complex counterparts (except for those arising
from the mereological structure of the abbreviated concept).
It follows that when an agent introduces an abbreviative concept, he or she must
adopt rules of use which guarantee the coincidence of epistemic roles that is
envisioned in (ii). In the case of the concept of a fortnight, an agent could do this
by adopting (F1) and (F2):

(F1) Where P is any proposition containing one or more occurrences of the

concept period of fourteen days, P* is any proposition that is just like P except
for containing one or more occurrences of fortnight in place of occurrences of
period of fourteen days, and E is any body of evidence, the epistemic probability of
P* relative to E is to be derivative from, and exactly identical with, the epistemic
probability of P relative to E.
(F2) Where E is any body of evidence, the following proposition is to have the
maximum epistemic probability relative to E:
(D) A fortnight is a period of fourteen days.
It would be possible to derive (F2) from (F1) and the assumption that for any body of
evidence E, every proposition that is a logical truth should have the maximum
epistemic probability relative to E. This is so because the proposition a period of
fourteen days is a period of fourteen days is a logical truth, and because proposition
(D) is just like that proposition except for having an occurrence of fortnight in place
of an occurrence of period of fourteen days. For present purposes, however, it is useful
to be able to think of (F1) and (F2) as equally fundamental. (F2) plays a crucial part in
fixing the epistemic role of the concept of a fortnight, for it is that rule that authorizes
us to take proposition (D) as a free premise in any context whatsoever (except
perhaps for contexts involving propositional attitudes). In the interests of clarity, it
is desirable to keep our discussion of abbreviative definitions independent of more
complex questions about the epistemic role of logical concepts.
Is it legitimate for an agent to commit to using the concept of a fortnight in
accordance with (F1) and (F2)? Yes. When an agent acquires a new concept, it is up
to him or her to determine how the concept will be used, and doing this clearly
involves determining which bodies of evidence will count as supporting propositions
that contain the concept, and how much support each such body will provide. Where
could the epistemic properties of a concept come from, if not from the rules of use
that the agent adopts? To be sure, there are restrictions that an agent who is
introducing a new concept must observe: the rules of evidence must not authorize
contradictory beliefs, and they must be internally coherent, in the sense that they
must not imply conflicting claims concerning the evidential support concerning
particular propositions. Moreover, in addition to these two general restrictions,
there are others that govern rules of use of specific kinds. Of particular relevance
here are two additional restrictions that the logical tradition has recognized as
governing abbreviative definitions (Gupta 2008). One says that rules that introduce
308 PART IV: KNOWLEDGE

an abbreviative concept must be conservative, in the sense that they must not
authorize us to form any new beliefs other than ones that involve the given concept.
And the other says that an abbreviative concept must be eliminable—that is, it must
be true that every proposition containing the concept is equivalent to one that lacks
the concept. These restrictions are important. But as reflection will attest, they are
easily satisfied by (F1) and (F2).
I have claimed that an agent commits to using the concept of a fortnight in
accordance with (F1) and (F2) in adding the concept to his or her conceptual lexicon.
It is this commitment that makes it possible for the concept to serve the purpose of
abbreviating period of fourteen days. But what does it mean to say that an agent has
undertaken a commitment of this sort? There are two quite different ways of
answering this question. One option is to say that the agent forms an intention to
use the concept in accordance with (F1) and (F2)—that is, an intention to use it in the
ways that (F1) and (F2) specify. The other option is to say that the agent acquires a
disposition to use the concept in accordance with (F1) and (F2). I will not attempt to
decide between these options here. (On either option it will be true that the com-
mitment serves a cognitive interest of the agent.)
On the account I am proposing, the use of the concept of a fortnight is governed by
rules of evidential support, but some philosophers would prefer to say that it is
governed by rules that assign truth conditions to propositions containing the con-
cept. More specifically, they would prefer to say that when we add the concept to our
lexicon, we in effect stipulate that any adequate interpretation of the lexicon must
confer the same truth values on propositions containing fortnight as they confer on
the corresponding propositions containing period of fourteen days. In my view, this
account of the situation is confused. When we approach the task of introducing a
new concept, we are not envisioning a range of possible interpretations of our
conceptual scheme, and we do not have in mind a relational concept of truth. Indeed,
as I see it, the concept of truth under an interpretation is a late addition to our
conceptual scheme, one that we acquire from mathematical logic long after the basic
shape of the scheme has been determined. The content of the commonsense concept
of truth is fixed by the class of propositions that have the following form:
The proposition that P is true just in case P,

and we learn to wield the concept by acquiring a commitment (or a disposition) to

accept members of this set (Horwich 1998, Hill 2002, chapters 3 and 4 of the present
volume). This view precludes the idea that the concept of truth is relational. More-
over, the view implies that propositions and their conceptual constituents must have
fully determinate natures before it is meaningful to talk of truth and truth conditions,
for it implies that we presuppose propositions in explaining what it is for proposi-
tions to have truth values. It follows that the rules that fix the nature of concepts and
propositions cannot take the form of assignments of truth conditions.
To summarize, the main points so far are that the concept of a fortnight is used
mainly to abbreviate the more complex concept period of fourteen days, and that the
 CONCEPTS , TELEOLOGY, AND RATIONAL REVISION 309

way to provide for this use is, first, to arrange for the former concept to have the same
epistemic properties as the latter, and second, to arrange for proposition (D) to have
an epistemic status that makes it possible for us to use it as a free premise in any
context. I will turn now to consider the implications of these views for questions
about revisability and a priority.
It is clear that as long as (F1) and (F2) are in place, there will be an epistemic right
to believe proposition (D), because the proposition will be supported to the max-
imum degree by every possible course of experience. It follows that the proposition is
not vulnerable to disconfirmation by empirical evidence. But what about (F1) and
(F2) themselves? Can empirical evidence call them into question? No, because they
make no empirical claims. As we have already observed, they are conservative rules in
the sense that they do not add to our epistemic commitments: it is possible to adopt
them without becoming committed to any proposition not involving fortnight to
which one was not previously committed. This is true, in particular, of empirical
propositions. (F1) and (F2) do not add to the empirical content of our theory of the
world. They simply provide for alternative formulations of empirical commitments
that we have already made.
I will not go into the question of whether proposition (D) is known a priori,
because that would involve us in questions about the nature of knowledge, and also in
questions about permissible ways of arguing for the existence of knowledge, that are
best left aside in the present context. But we should consider the question of whether
our justification for believing (D) is a priori. A justification is a priori if it is
independent of empirical evidence. Now reflection shows that “justified independ-
ently of empirical evidence” can mean two quite different things. On the one hand, it
can mean that our justification for believing (D) does not depend on any particular
body of empirical evidence—that it will be justified by the available evidence no
matter what direction the future course of experience happens to take. And on the
other hand, it can mean that our justification for believing (D) is independent of all
empirical evidence. Now it is clear that if the foregoing reasoning is correct, (D) does
count as having a non-empirical justification in the first sense. (F2) implies that (D)
will be justified by any possible future course of experience. But it also seems possible
to argue that (D) has a non-empirical justification in the second sense. An agent who
is conforming to (F2) is in a position to reflect that she will be justified in believing
(D) no matter what form the future course of experience will take. It seems natural to
say that this reflection can provide a justification for believing (D) that is prior to all
future courses of experience, and that is therefore in an important sense independent
of all of them.
I have maintained that (D) has a special cognitive and epistemic status in virtue of
its relationship to the foundational rules of support concerning the concept of a
fortnight. Moreover, I have maintained that (D) will retain this special status
indefinitely—in fact, for as long as we continue to use the concept. This second
view is challenged by the following remarks by Quine, which urge that any special
310 PART IV: KNOWLEDGE

status that is originally possessed by propositions like (D) is transitory, and that such
propositions eventually acquire the status of empirical generalizations:
Legislative definition introduces a notion hitherto unused, or used only at variance with the
practice proposed, . . . so that a convention is wanted to settle the ambiguity. [“Legislative”]
refers to the act [of defining a term], and not to its enduring consequence . . . .This is because we
are taking the notion of truth by convention fairly literally and simple-mindedly, for lack of an
intelligible alternative. So conceived, conventionality is a passing trait, significant at the moving
front of science but useless in classifying the sentences behind the lines. It is a trait of events
and not of sentences. (Quine 1976, pp. 118–19)

According to Quine, then, while a proposition like (D) may enjoy a certain elevated
cognitive status at the moment at which it is initially adopted, this special status is
necessarily short lived. Elaborating, we can perhaps say that according to Quine,
one’s original intention to use a concept C* as an abbreviation for C is modified over
time, and that it eventually evolves into a generalized intention to use C* to advance
one’s main cognitive goals, such as predicting data, explaining data, and registering
information. At this point the proposition a C* is a C, which was originally accepted
on the basis of the rules of evidence that establish an abbreviative practice, has no
special status. It is on a par with the other propositions in one’s system of beliefs, and
as a result, like any other statement, it can become a casualty of revisionary activity.
This line of thought seems quite misguided to me. In particular, it seems quite
wrong to suppose that the intention to use C* as an abbreviation will be swamped or
eroded by the desire to use C* to advance more important cognitive concerns. The
desire to use C* as an abbreviation for C is perfectly compatible with the desire to use
it in framing predictions and explanations, and also with the desire to use it in
registering information. Indeed, using C* as an abbreviation for C can be a way of
furthering these other concerns. When one introduces C* as an abbreviation for C, C
already plays a role in various cognitive activities, including, we can suppose, the
activities of predicting data, explaining data, and registering information. One
introduces C* because one wants to have a more efficient way of conducting these
activities, not because one wishes to conduct some other sort of business that is
independent of them. It follows from this that one can continue to treat C* as an
abbreviation for C across time, without encountering eroding pressures from other
concerns. If it makes sense to use C* to streamline the business of predicting, etc. at
one point of time, it can make sense to continue to do so. Indeed, there is no reason in
principle why one might not continue to do so for an indefinitely long time.
I wish to claim, then, that there is no natural limit to the lifespan of an intention to
use a concept as an abbreviation. Contrary to what Quine seems to have thought,
there is no tendency in the nature of things for intentions of this sort to be swept
away by our pursuit of more fundamental cognitive goals.
Before moving on to consider other matters, I want to mention a distinction
between two quite different forms of revision. We have found reason to think that
 CONCEPTS , TELEOLOGY, AND RATIONAL REVISION 311

if the proposition a C* is a C is an abbreviative definition that introduces a concept

C*, then it is just not possible, even in principle, to find empirical evidence that makes
it appropriate for us to set the proposition aside as false. In this sense, there can be no
empirically motivated revisions of definitions. Reflection shows, however, that there
is another sense in which it can be said that abbreviative definitions are subject to
empirical assessment. This sense is brought to the fore by the following argument:
“When we introduce a new concept C* to abbreviate a more complex concept C, it is
generally true that C is a useful concept—an expression that plays a valuable role in
one or another of our cognitive practices. We introduce the defined concept simply
because we would like to have a more efficient way of conducting the valuable
business in which C is involved. Now consider a case in which C is involved in
empirical business. Suppose, in particular, that C enjoys a life in a branch of science.
Further, let this case be one in which the theory in which C is involved eventually
becomes discredited. Eventually we acquire evidence that leads us to lose our earlier
confidence in the theory, and we wind up with the view that C does not refer to
anything. After this revision, we would no doubt progressively lose interest in C, and
perhaps we would eventually reach a point at which it played only a very small role in
our mental life. It could then easily happen that we ceased to make any use of the
abbreviative concept C*, and that we ceased to affirm the definition by which it was
originally introduced. The best way to describe this example is to say that it is one in
which empirical evidence leads us to set a definition aside as useless. In other words,
while it is not an example in which evidence calls the truth of a definition into
question, it does seem to be an example in which evidence undermines commitment
to a definition.”
Thus, in addition to revisions that are occasioned by empirical disconfirmation,
there are revisions that occur when the motivation for an abbreviative concept is
undercut by empirical developments. It is clear that revisions of the second sort do
sometimes occur, but it can also happen that the motivation for an abbreviative
concept cannot be undercut in this way. To see this, it suffices to take another look
at Euclidean geometry, which as standardly developed, contains many concepts
that serve to abbreviate concepts that are more complex. In the early twentieth
century physicists lost their motivation for deploying those abbreviative concepts in
formulating laws of nature, but the concepts have been retained by the mathemat-
icians who continue to study Euclidean structures. In general, the motivation for
using abbreviative concepts is often quite complex, involving a priori components
as well as empirical ones. To the extent that this is true, discoveries that undercut
the empirical motivation for abbreviations have no tendency to reduce our con-
ceptual lexicon.
In sum, it appears that in at least one case, that of abbreviative definitions, it is
appropriate to view propositions as having an interesting form of immunity to
empirical revision, and that it is also possible to view them as justified a priori.
312 PART IV: KNOWLEDGE

V. Logical concepts
According to the picture I sketched earlier, logical concepts play two main roles in the
cognitive life of an agent—an inferential role and an epistemic role. I will not be able
to discuss this claim in full generality here. Rather I will have to focus on a single
example. The natural choice is classical disjunction, for the stage has already been set
for a discussion of that concept. I will focus on the question of whether the rules that
govern the use of classical disjunction, and the logical beliefs to which they give rise,
are immune to empirical revision. As we will see, there is reason to answer to this
question in a way that is more in line with traditional rationalist doctrines than with
Quinean empiricism. I hope it will be clear that what I will say about this one case can
be adapted so as to apply to others.
Now as we saw earlier, it is possible to account for all of the inferential properties
of classical disjunction by supposing that it is governed by the introduction and
elimination rules of classical logic (i.e., rules (i)–(iii) of Section III). As for the
epistemic role of the concept, its most important feature is that it enables us to
form justified beliefs in complex propositions in circumstances in which we would
not be justified in believing simpler ones. To be more specific, it is part of this role
that we are in some circumstances justified in forming a belief of the form P or Q even
though we are not justified in believing P or in believing Q. It is also part of the role
that we are in some circumstances justified in believing a proposition of the form if
P then Q or R in circumstances in which we are not justified in believing if P then Q or
in believing if P then R. It is possible to account for these epistemic properties of
classical disjunction by supposing that it is governed by the following epistemic rule:
ðRÞ If PrðeÞ> 0; then Prðp or q=eÞ ¼ Prðp=eÞ þ Prðq=eÞ  Pr ðp and q=eÞ:
The importance of this rule, and the disjunctive beliefs that it authorizes us to form,
become evident when the beliefs in question are considered in relation to the
inferential role of disjunction. (R) gives us the right to form beliefs that we can
then exploit in various ways in the inferences that are directly authorized by rules
(i)–(iii), and in the inferences that are authorized by (i)–(iii) in combination with the
rules that govern the other connectives. (The beliefs with which (R) is concerned
come in degrees—they are graded beliefs rather than the “full-strength” or “binary”
beliefs. The question of how graded beliefs are related to binary beliefs is complex,
and I cannot address it here, though I should explicitly acknowledge that I am
assuming that there are principles that permit transitions from the former to the
latter.)
To summarize: if it is true, as I contend, that the main cognitive functions of
classical disjunction are inferential and epistemic, it is possible to describe the main
contributions that classical disjunction makes to cognition by listing the classical
introduction and elimination rules and the laws of the probability calculus that
govern disjunction, including principally (R).
 CONCEPTS , TELEOLOGY, AND RATIONAL REVISION 313

Is the epistemic role of classical disjunction on a par with its inferential role, or is
one more fundamental than the other? Is it possible to explain its inferential
properties in terms of its epistemic properties, or the latter in terms of the former?
No. There is no chance of a reductive explanation in either direction, for deductive
reasoning makes use of different abilities than the numerical computations that are
required for probabilistic updating. It should be noted, however, that there is an
important relationship of another kind between the roles of inference for disjunction
and the laws of epistemic probability.
Thus, to put the point somewhat impressionistically, it can be shown that the
probability that the conclusion of a disjunctive inference is false can never be greater
than the probability that at least one of the premises is false. To be more precise, it is
possible to establish (1)–(3):

(1) For any two propositions p and q, the probability that (p or q) is false can never
be greater than the probability that p is false.
(2) For any two propositions p and q, the probability that (p or q) is false can never
be greater than the probability that q is false.
(3) For any three propositions p, q, and r, the probability that r is false cannot be
greater than the probability that at least one of the following propositions is false:
(p or q), (p à r), and (q à r).
(Here à is the material conditional and (p à r) and (q à r) represent the inferences
p├ r and q ├ r respectively. I am simplifying by focusing on the case in which the
premises of disjunctive inferences are single propositions rather than classes of
propositions, but the argument can easily be extended to cover the latter case
involving finite classes.)
Each of (1)–(3) is a special case of the general rule that the probability of the
conclusion of a classical argument is false is always less than or equal to the
probability of at least one of the premises is false. (For a proof, see Bennett (2003,
sections 21 and 53).)
Is it legitimate for us to adopt the package of rules that consists of the introduction
and elimination rules and the family of probabilistic rules that includes (R)? Yes. As
we observed earlier, it is permissible to adopt whatever rules of use one wants in
introducing a new concept, provided that the relevant restrictions are satisfied. And
the relevant restrictions are satisfied in the present case. Thus, it is clear that the
package does not authorize contradictory beliefs, or assign conflicting properties to
individual propositions. Moreover, in keeping with the requirements of the logical
tradition, the rules of inference for classical disjunction are conservative, in the sense
that they do not authorize us to believe any new propositions other than ones that
contain disjunction. That is, apart from propositions that contain disjunction, our
doxastic state after adopting the introduction and elimination rules will be exactly the
same as it was before.
314 PART IV: KNOWLEDGE

To be sure, when the introduction and elimination rules are combined with (R), we
acquire an epistemic right to hold new beliefs that do not involve disjunction. Thus,
(R) authorizes us to form new disjunctive beliefs, and the rules of inference for
disjunction authorize us to derive non-disjunctive conclusions from disjunctive
premises. Inevitably, when these rules are combined, they authorize us to form
new non-disjunctive beliefs. To see this, suppose that r is a proposition that does
not contain disjunction, and suppose also that in combination with (R), the available
evidence entitles us to believe (p or q). Suppose further that r is derivable from p and
is also derivable from q. In these circumstances, the elimination rule for classical
disjunction authorizes us to infer the non-disjunctive proposition r. It appears, then,
that even though the rules of inference for classical disjunction are conservative when
taken by themselves, the package consisting of the rules together with (R) is not
conservative. Is this a problem?
No. There is nothing intrinsically wrong with rules that authorize new beliefs. We
subscribe to many such rules. Moreover, the package consisting of the rules of
inference for classical disjunction and (R) is particularly innocuous, As we have
already seen, if an inference is performed in accordance with one of the classical rules
for disjunction, then the probability that the conclusion is false cannot be greater
than the probability that at least one of the premises is false. It follows that when we
obtain the non-disjunctive proposition r in the way described in the previous
paragraph, we are not taking any new epistemic risks. We will never be in a situation
in which our evidence requires us to accept all of (p or q), (p à r), and (p à r) while
demanding that we reject r.
There is, however, a question pertinent to revisability that remains to be addressed.
It is clearly possible for there to be an empirical theory that enjoys great predictive
success on the whole, but that in some cases leads via the classical rules of inference to
conclusions that are false. Now if the theory is sufficiently successful, it seems that it
might be rational, in some reasonably robust sense, for the relevant cognoscenti to
consider changing the theory by replacing the classical connectives with connectives
that are governed by less powerful rules of inference—that is, by rules that do not
permit one to deduce the false conclusions. Arguably this is what happened at one
point in the debate about the proper formulation of quantum mechanics. Birkhoff
and von Neumann proposed a change in the underlying logic, and Putnam and
various others endorsed their proposal. More recently, a number of philosophers
have objected to this approach. It appears that all parties to this dispute were rational,
perhaps ideally so. Doesn’t this show that the classical rules of inference are in
principle vulnerable to revision?
Although I cannot hope to do full justice to this question here, I wish to suggest
that we have already taken note of an adequate strategy for dealing with situations
like the one presented by quantum mechanics. The main idea is that such situations
can always be handled by supplementing classical logic with an alternative. There is
no need to replace it. To be sure, this strategy presupposes that there will always be a
 CONCEPTS , TELEOLOGY, AND RATIONAL REVISION 315

reason for hanging on to classical logic. But the presupposition can be defended. It is
recognized on all sides that classical logic gives us inferential options that such
alternatives as intuitionistic logic and quantum logic cannot provide. The alternatives
to classical logic are constitutionally weaker, which is why the question of replacing
classical logic arises in the first place—since classical logic is more powerful, theories
framed in terms of it can have false consequences while theories based on the
alternatives do not. Because of this extra strength, classical logic serves more cogni-
tive interests than the alternatives. It has additional cognitive value. It would be
rational for us to want to continue to avail ourselves of this additional value in
reasoning about non-empirical domains, even if it proved inconvenient to use it in
formulating and developing empirical theories.
It may be useful to consider this situation in somewhat greater detail. We should
begin by taking note of a small significant difference between the foregoing rule for
classical disjunction and the corresponding rules for quantum disjunction, which
I will represent by quor. Here again are the classical rules:

(i) For any class of propositions , if p is derivable from , then (p or q) is

derivable from as well.
(ii) For any class of propositions , if q is derivable from , then (p or q) is
derivable from as well.
(iii) For any class of propositions , if (p or q) is derivable from , r is derivable
from p together with the members of , and r is also derivable from q together with
the members of , then r is derivable from .

And here are the corresponding rules for quantum disjunction:

(iv) For any class of propositions , if p is derivable from , then (p quor q) is

derivable from as well. (Here and elsewhere, I use quor to express quantum
disjunction.)
(v) For any class of propositions , if q is derivable from , then (p quor q) is
derivable from as well.
(vi) For any class of propositions , if (p quor q) is derivable from , r is derivable
from p, and r is also derivable from q, then r is derivable from .

Now inspection reveals a difference between these two sets of rules. (iii) allows that the
derivations running from p to r and q to r may make use of statements that belong to ,
but (vi) authorizes no such use of . To put the point in positive terms, (vi) requires that
the derivation running from p to r make use only of p itself, and that the derivations
running from q to r make use only of q itself. It is easy to miss this difference in the two
formulations, but it amounts to an immense difference in generative power. It is precisely
because the rules differ in this way that it is possible to use the rules of inference governing
classical disjunction to prove the distribution law (viz., (p and (q or r)) iff ((p and q) or
(p and r))), but it is not possible to prove a distribution law using the rules that govern
quantum disjunction. (For discussion of these points see Gibbins (1987, chapter 9).)
316 PART IV: KNOWLEDGE

Thus far we have seen that classical disjunction and quantum disjunction play
different roles in inference, and we have also seen that this difference is captured by,
and in fact derives from, the difference between (i)–(iii) and (iv)–(vi). I would like
now to emphasize that the difference between classical disjunction and quantum
disjunction is important. Classical disjunction has inferential powers that quantum
disjunction lacks, and as a result, the former serves cognitive interests that the latter
cannot serve.
Suppose you have adequate grounds for believing a proposition p and also for
believing a disjunctive proposition (q or r). Suppose also that you do not have any
reason that is independent of your reasons for believing p and believing (q or r) for
believing the complex proposition ((p and q) or (p and r)). Suppose further that
another proposition s can be derived from each of (p and q) and (p and r), but that it
cannot be derived from any one of p, q, and r in isolation from the others. Suppose
finally that you would like to have a right to believe s because doing so would enable
you to settle a question that is of some interest. Clearly this is a possible epistemic
situation. Now in these circumstances, classical logic authorizes you to infer ((p and
q) or (p and r)) from your beliefs p and (q or r), and it also authorizes you to infer s
from ((p and q) or (p and r)). Thus, it gives you the right you seek—the right to
believe s. On the other hand, since quantum logic does not authorize you to infer ((p
and q) quor (p and r)) from p and (q quor r), it does not give you the right to believe s.
It is easy to think of concrete examples of this scenario. Suppose you believe that
Ed will come to Lee’s party, and also that either Suzy or Carly will come as well.
Suppose further that you believe both of the following conditional propositions: if Ed
and Suzy come to the party, then everyone at the party will have fun; and if Ed and
Carly come to the party, then everyone at the party will have fun. Given that or here
expresses classical disjunction, you have the right to infer from these beliefs that you
will have fun if you go to Lee’s party. This could be a very useful thing to know! It is a
mark against quantum logic that it would prevent us from deriving it. (I have given
an empirical example here in the interests of concreteness, but it is easy to see that
there are countless similar cases in which premises and conclusions are concerned
with non-empirical domains.)
It appears, then, that classical logic provides advantages that quantum logic cannot
match. It may be the case that large portions of our conceptual scheme can be recast
in quantum logical terms without significant loss, but there are definitely situations in
which a thoroughgoing acceptance of quantum logic would place us at a disadvan-
tage. We could not formulate beliefs that have the same logical consequences as our
classical beliefs, and accordingly, many rational beliefs would be lost to us. Accord-
ingly, it seems advisable to go forward with a lexicon that contains two concepts of
disjunction, one that we use in formulating the laws of quantum mechanics, and
another that we use in other domains.
I have maintained that the rules governing the use of classical disjunction are
immune to empirical revision, and also that classical disjunction serves purposes that
 CONCEPTS , TELEOLOGY, AND RATIONAL REVISION 317

are (i) important and (ii) independent of its role in empirical theories. I turn now
briefly to the case of logical laws that involve disjunction. There are two points that
are relevant to our present concerns. First, the laws in question are immune to
empirical disconfirmation. This is a special case of the following general principle:

(E) Where p is any law of classical logic, and e is any body of evidence, Pr(p/e) = 1.
As with (R), I am thinking of (E) as a fundamental rule concerning the allocation and
evaluation of evidence. In effect, it helps to define the classical connectives by
stipulating that certain propositions containing them will never be challenged by
empirical discoveries. Second, there is no risk that the motivation for embracing the
classical connectives will be undercut by empirical discoveries, such as the discovery
that it would be desirable to use different connectives in formulating a particular
empirical theory. By the same token, there is no risk that we will be led by an empirical
discovery to abandon the classical laws, and allow principle (E) to lapse into irrele-
vance. This is true for the same reason that empirical discoveries could not undercut
our motivation for the classical rules of inference. Even if they were to cease to guide us
in our reasoning about an empirical domain, there would be other domains in which
they continued to play a useful role. That is, like the classical rules of inference, they
would earn their keep by contributing to non-empirical endeavors. Hence, they are
immune to empirical revision in the strongest possible sense. They will be available to
guide reasoning for as long as human beings continue to think.

VI. Additional examples of immunity

We have found that a number of propositions and rules of inference are immune to
empirical revision in two senses. They cannot be called into question by empirical
evidence, and the motivation for accepting them can survive changes, including even
radical changes, in empirical theories. The first form of immunity is epistemic
immunity, and the second form is teleological immunity. According to our earlier
reflections, both of these forms of immunity are possessed by abbreviative defin-
itions, and also by the laws and rules of inference of classical logic. (This isn’t true of
all abbreviative definitions, but as the example of Euclidean geometry shows, it is true
of some.) I will now ask whether there are propositions or rules beyond the ones that
we have been considering that are either epistemically or teleologically immune to
revision. What are the scopes of these two conceptions of immunity? Are they
narrow or broad?
Arguably, they include the laws of mathematics. In many cases, our reasons for
accepting such laws include the fact that they are empirically fruitful, but there are
grounds for thinking that there are other, deeper reasons for accepting them. At the
deepest level, it seems, mathematicians believe the laws of mathematics because it is
aesthetically reinforcing to do so—because in accepting the laws they gain access to a
318 PART IV: KNOWLEDGE

realm of contemplation and activity that is aesthetically rewarding. It also seems

likely that mathematicians are interested in their subject because they believe that it
describes non-empirical entities of some sort—perhaps systems of abstract objects, or
perhaps patterns or properties that are instantiated in other possible worlds even if
not in ours. These two non-empirical interests are intimately related. The activity of
describing abstract entities would have little appeal if it lacked aesthetic value, and the
aesthetic value of mathematical activity would be lessened, perhaps considerably, if
mathematicians became convinced that there are no entities answering to their
descriptions. This picture of the reasons for being interested in mathematics receives
strong support from two considerations. First, as we noticed earlier, mathematicians
continue to be interested in Euclidean geometry despite the fact that it is no longer
seen as a description of physical space. And second, many mathematical systems
have been invented and pursued without there being any empirical applications in
sight. This is true, for example, of hyperbolic geometries. In view of these consider-
ations, it seems appropriate to conclude that acceptance of mathematical axioms is
due at the most fundamental level to a combination of aesthetic reinforcement and a
sense that mathematics describes a non-empirical domain, and therefore, that our
interest in maintaining them cannot be undercut by empirical discoveries. By the
same token, it seems appropriate to say that they are immune to teleological revision.
Moreover, a case can be made that the laws of mathematics are immune to
epistemic revision. Given that there are non-empirical reasons for accepting them,
we would not want to adopt rules of evidence which implied that their acceptability is
conditional on our having certain types of evidence and not others. Accordingly, we
would want our acceptance of them to be governed by a principle of evidence like the
foregoing principle (E) concerning the laws of classical logic:

(E*) Where p is any law of mathematics, and e is any body of evidence, Pr(p/e) = 1.
This implies that the laws of mathematics are characterized by epistemic immunity.
When people assert (E*), it is usually because it follows from the probability
calculus in combination with the assumption that the laws of mathematics are
necessary. Here, however, I am viewing it in a quite different way—specifically, as
representing a free-standing disposition or commitment concerning the evaluation
of mathematical propositions in relation to evidence. Since Quine was a staunch
opponent of modal assumptions, it would beg the question against him to base a case
for (E*) on an assumption about necessity. There is no such problem, however, with
the idea that it serves to fix the epistemic role of mathematical concepts.
(E*) implies that empirical discoveries have no bearing on the acceptability of
mathematical theories. Given that we rely heavily on mathematics in generating
empirical predictions, it might seem puzzling that we are free to adopt such a
commitment. But our puzzlement dissipates when we recall that the applicability
of mathematics to the world depends on what is known as a representation theorem.
To secure the applicability of a mathematical theory M to an empirical domain D, it is
 CONCEPTS , TELEOLOGY, AND RATIONAL REVISION 319

necessary to prove a theorem which establishes that there is a structure-preserving

function of some sort—usually an isomorphism—linking D to preferred models of
M. Once this representation theorem has been proved, it is possible to combine M
with a theory of D to generate predictions. Suppose now that some of these predic-
tions turn out to be false. In this case, we would be compelled to reject the
representation theorem, but not M itself. This is because the theorem depends on
assumptions of two kinds—the axioms of M, and empirical hypotheses about D.
Given that there is an independent, aesthetic motivation for accepting the axioms, it
would preferable as well as possible to respond to the predictive failure by changing
the empirical hypotheses and devising a new mathematical theory that was better
suited to the resulting empirical picture of D. (Of course, if the sole reason for
introducing this new mathematical theory was to provide a model of D, we would
have no reason to hang onto it if the predictions supported by its representation
theorem also turned out to be erroneous.)
Are there other examples of immunity to revision? Perhaps. To illustrate the range
of possibilities here, I will discuss a family of examples that lies far afield from the
laws of mathematics.
Descriptional theories of meaning claim that concepts of various types are equiva-
lent with definite descriptions. If any of these theories are true, then propositions
giving expression to the posited equivalences are at least immune to evidential
revision. Thus, for example, according to the descriptional theory of kind concepts,
a kind concept K is more or less synonymous with a description that invokes the
perceptually salient characteristics that are possessed by normal members of K in the
actual world. Views of this sort met with powerful objections in the 1970s, but they
have evolved in ways that at least deprive those objections of much of their force.
(See, e.g., Jackson 2010.) To the extent that the descriptional theory can be defended,
there is room for an argument that our use of kind concepts is guided by commit-
ments concerning the allocation and evaluation of evidence that provide an a priori
guarantee that certain propositions linking kind concepts to descriptions will be
supported by evidence. Similar remarks apply to certain propositions that link proper
names to descriptions, such as the proposition that Willard Van Orman Quine is the
person x such that information about x is actually transmitted by testimony involving
the name “Willard Van Orman Quine.” Further, as is shown in a classic paper by
David Lewis (Lewis 1970), there are grounds for thinking that theoretical concepts
come to the same things as descriptions based on the Ramsey sentences of the
theories in which they occur. Insofar as this is true, it is possible to argue that a
range of propositions equating theoretical concepts with Ramsified descriptions are
guaranteed to have evidential support.
Descriptional propositions of these kinds are in effect abbreviative definitions, and
are therefore immune to evidential revision for the same reason as the proposition
that a fortnight is a period of fourteen days is immune. Do they have any claim to be
considered immune to teleological revision as well? Possibly so, for it might be that
320 PART IV: KNOWLEDGE

we would wish to think about kinds and theoretical entities as possibilities even if we
became convinced that they do not actually exist. Moreover, we might wish to
continue to attribute beliefs and other attitudes involving the relevant concepts to
actual and possible agents. We have lost interest in using the notion of phlogiston in
first order descriptions of the world, but we retain an interest in characterizing the
beliefs of the proto-chemists who accepted the theory of combustion and rusting in
which phlogiston played a role. Perhaps we also wish to reserve the right to use the
concept in characterizing the attitudes of imaginary characters.
It is arguable, then, that our two notions of immunity have spheres of application
that are both large and highly variegated. At all events, they seem robust enough to
merit philosophical attention.

VII. Analyticity
Quine’s critique of immunity to revision was developed in concert with a systematic
attack on analyticity. Now that the former notion has been somewhat rehabilitated, it
is natural to inquire whether it might be used to illuminate the latter notion, perhaps
by adopting the theory that consists of the following two claims: First, a statement is
analytic just in case it expresses a proposition p such that (i) p is immune to revision,
and (ii) p carries no non-logical existential commitments. And second, given that
there are three types of immunity to revision (epistemic immunity, teleological
immunity, and the combination of the two), analyticity comes in three different
flavors.
I put this theory forward in a tentative spirit, for it is beyond the scope of the
present chapter to determine how well it squares with our pre-theoretical, intuitive
ascriptions of analyticity. It is clear at the outset, however, that the theory has certain
virtues. For instance, since it is empirically assessable whether a proposition would be
rejected in various evidential circumstances, the theory implies that ascriptions of
analyticity have testable implications. This is a big plus in the present dialectical
context, for as we saw, it was one of Quine’s main complaints about traditional
conceptions of analyticity that they are empirically empty. I also note that Quine
himself held that analyticity squares up with unrevisability pretty well. Thus, as we
noticed earlier, he says that an analytic statement is one that “holds come what may.”

References
Bennett, J. (2003). A Philosophical Guide to Conditionals. Oxford: Clarendon Press.
Boghossian, P. (1997). “Analyticity.” In B. Hale and C. Wright (eds.), A Companion to the
Philosophy of Language. Oxford: Blackwell Publishers Ltd.
Carey, S. (2009). The Origin of Concepts. Oxford: Oxford University Press.
Gibbins, P. (1987). Particles and Paradoxes. Cambridge: Cambridge University Press.
 CONCEPTS , TELEOLOGY, AND RATIONAL REVISION 321

Grice, H. P. and Strawson, P. F. (1957). “In Defence of A Dogma,” The Philosophical Review 65:
141–58.
Gupta, A. (2008). “Definitions,” The Stanford Encyclopedia of Philosophy. <http://plato.stan
ford.edu/>.
Harris, J. H. (1982). “What’s so Logical about the ‘Logical’ Axioms?” Studia Logica 41: 159–71.
Hill, C. S. (2002). Thought and World. Cambridge: Cambridge University Press.
Horwich, P. (1998). Truth, 2nd edition. Oxford: Oxford University Press.
Horwich, P. (1999). Meaning. Oxford: Oxford University Press.
Jackson, F. (2010). Language, Names, and Information. Oxford: Wiley-Blackwell.
Lewis, D. K. (1970). “How to Define Theoretical Terms.” Journal of Philosophy 67: 427–46.
Quine, W. V. O. (1976). “Carnap and Logical Truth.” In Quine, The Ways of Paradox and
Other Essays, 2nd edition. Cambridge, MA: Harvard University Press.
Quine, W. V. O. (1980). “Two Dogmas of Empiricism.” In Quine, From a Logical Point of
View, 2nd edition. Cambridge, MA: Harvard University Press.
Quine, W. V. O. (1986). Philosophy of Logic, 2nd edition. Cambridge, MA: Harvard University
Press.
Schechter, J. (forthcoming). “Juxtaposition: A New Way to Combine Logics,” Review of
Symbolic Logic.
Wittgenstein, L. (2001). The Philosophical Investigations, 3rd edition. Oxford: Blackwell.
Index

AA (argument from analogy) Basbaum, A.I. 159, 182

and type–materialism 110–13 Bechtel, William 144
abbreviative concepts 302, 306–11 behaviourism
acquaintance theory and Quine 85, 87
and experiential awareness 9–10, 207–8 beliefs
and perceptual awareness 219 and lottery propositions 253–72
action, law of 63 and perceptual experiences 218
Adair, E.R. 183 and process reliabilism 250–1
adverbialism see also empirical evaluation of beliefs
and experiential awareness 207, 208 Belnap, N. 55, 62
Aizawa, K. 144 Bennett, D.J. 213, 218
analyicity 320 best explanation principle
Quine’s critique of 297, 298–303 and Cartesian skepticism 16–8
Angle, Steve 91 and Hawthorne 257–9, 261, 265
animals and representationalism 10, 218–9, 257–9,
and the multiple realization 261, 26
argument 141, 142 and type–materialism 100, 102, 108, 138–41,
and perceptual awareness 157–8 145–6
Anton–Erxleben, K. 189 Bigelow, J. 197
appeal to other species argument Blakeslee, S. 184
and type–materialism 114, 115 Block, N. 107, 114, 139, 145, 197, 218, 232
appearance properties Boghossian, P. 281, 299
and visual experience 220–3, 228–9, 232–5 Boyd, R. 103
and visual qualia 211, 213–14 brain processes/states
appearance/reality distinction and identity theory 138, 146, 147
and qualitative awareness 12, 148, 157, 175, and the mind–body problem 118–31
199–201, 205–6, 210, 290–1 and the multiple realization argument 142–4
a priori modalities 275–81 and qualia 192–3
argument from analogy (AA) and qualitative states 149–51
and type–materialism 110–13 and token–materialism 98, 102
Armstrong, D.M. 105–6 and type–materialism 5–7, 8, 99, 102, 103–4,
Arnauld, A. 291 107–10
assertion argument Brentano, Franz 58
and Hawthorne’s lottery propositions 257–8, Brindley, G.S. 99
259–63 Broackes, J. 218
attention Brueckner, A.L. 241
and the peripheral state theory of pain 182, Burge, T. 80, 191, 234
183, 189–90 Bushnell, M.C. 182, 183, 184, 188
attention mechanisms Byrne, A. 12, 202
and visual qualia 202, 209
awareness 9–13, 106, 150, 156 Camp, J.L. 55
experiential character of qualitative Campbell, J. 87
awareness 190–3 Carey, S. 192, 305
visual 197–217 Carnap, R. 6
of pain, 156–60, 166, 174–5 Carrasco, M. 189–90, 227
see also perceptual awareness; Cartesian arguments
representationalism modal arguments 6, 7, 14, 117–35, 148, 288–92
and type–materialism 103–5, 107
Balog, K. 136, 149 Cartesian intuitions
Barsalou, L.W. 192 and conceivability 131
324 INDEX
Cartesian intuitions (cont.)
and imagination 124–5
Cartesian skepticism 13, 16–9
and process reliabilism 239–52
causal argument
for type–materialism 140–1
causal chain problem
of perceptual concepts 77, 78–80, 88
central state identity theory 5–8, 97–152,
177–8, 190, 193, 194
certainty–based practical reasoning
and Hawthorne’s lottery
propositions 266, 267
Chalmers, D. 117
Chisholm, R.M. 6, 220
Christensen, D. 297
Churchland, P.M. 111
Churchland, P.S. 115
circularity argument
and process reliabilism 245, 250–1
classical logic
disjunction 312–17
empirical evaluation of 298, 300–3, 306
Cleve, J. Van 197
Coghill, R.C. 180
cognition
cognitive abilities and concepts 91
cognitive interests and concepts 303–6
core cognition 192
dual system theories of 14
cognitive psychology
and perceptual awareness of pain 159
cognitive role of concepts 5
pain 161–71, 175, 297–321
cognitive science
and perceptual concepts 87
and representationalism 9
Cohen, S. 239
color qualia 200, 205–6
and conceptual dualism 149
and the knowledge argument 148
and the multiple realization argument 141
and visual experience 221–2
conceiving 274–5
conceivability
and Cartesian modal arguments 126–34,
288–92
constrained conceiving 275, 276, 277–8
the faculty of conception 126–31
and modal knowledge 14, 273–296
concepts 15–16, 66–94
abbreviative 302, 306–11
and cognitive interests 303–6
cognitive role of 5
and deflationism 66–74
determinacy of 4–5, 23–31, 69–70
epistemic relations of 66
logical 304–5, 312–7
mathematical 90–92, 93, 273, 280, 286
meaning of 1, 2
and modal knowledge 274, 275, 276,
279–80, 286
natural relations of 76–8
and perceptual awareness of pain 157–8
and reference 66–7, 69–93
representational relations of 66
and thought 52
see also logical concepts; mathematical concepts;
perceptual concepts; semantic concepts
conceptual awareness
and visual qualia 209–10
conceptual dualism 6–7, 8, 11, 149, 150, 177–8,
191–2, 193
conceptual necessity 284, 285–6
conceptual possibility 282–3
congruent sentences
and the determinacy of meaning 26
constancy thesis 225
constancy transformations
and visual experience 224–8, 235
constrained conceiving 275, 276, 277–8
constructive idealism 6
continuity argument
and visual experience 227–8
Cooper, L. 121
Copi, I.M. 112–13
core cognition 192
correlation laws
and arguments for type–materialism 139–40,
145–7
correlation thesis
and type–materialism 98–102, 104, 110, 114–15
correspondence 1, 2
see also semantic correspondence
counterfactual conditionals 15
and metaphysical necessity 292–6
counterfactual reasoning
and nomological modalities 287
Craig, A.D. 160, 183, 184
Cummins, R. 78
David, M. 51
Davidson, D. 98
De Armey, M. 111
De Morgan’s laws
and disjunction 304
deductive arguments
and Hawthorne’s lottery propositions 270–1
defeaters
and possession conditions for the concept of
pain 163–4
deflationism 2–4, 15–7, 51–65, 66–78, 93
arguments for 69–72
definition and versions of 66–7
deflationary theories of truth 2–4, 51, 53–7
and minimalism 2–4, 67, 68, 69–70

problems with 72–78
and reference 66, 67, 68, 71–4, 76
and substitutionalism 51–68, 67–8, 76
and truth 15–7, 67–8, 70–1
denotation 1
and the concept of pain 128
and deflationism 66, 67, 68
and semantic relations 57
Descartes, R. 275, 282, 288, 291
Cartesian arguments 103–5, 107, 148, 288–92
Cartesian intuitions 124–5, 131
Cartesian skepticism and process
reliabilism 13, 16–19, 239–52
evil demon hypothesis 17–18, 244, 288
First Meditation 241, 242
descriptional theories of meaning 318–20
desire, laws of 63
Devitt, M. 32
disembodiment, intuitions about 121, 123–6,
129–30
disjunction
and cognitive interests 304–5
and logical concepts 312–17
disquotation schemata/sentences 33, 34, 35–7, 43
diversity argument
and deflationism 71
Dobelle, W.H. 99
doxastic theory
of experiential awareness 207, 208–10
Dreier, J. 91, 297
Dretske, F. 12, 79, 80, 179, 202
dual system theories of cognition 14
dualism 5–6
and type–materialism 138–9
see also conceptual dualism; mind–body
problem; property dualism
Edgington, D. 287
Einstein, A. 140
emotional qualia 197, 199
emotions
and identity theory 137
empirical concepts
and deflationism 72
empirical developments and disconfirmation
and abbreviative concepts 311
empirical evaluation of beliefs 15–6, 297–320
abbreviative concepts 302, 306–11
concepts and cognitive interests 303–6
and epistemic immunity 317–19, 320
logical concepts 298, 306, 307, 312–17
and Quinean skepticism 297, 298–303, 320
and telelogical revision 317, 319–20
empirical relation theory of reference 4, 34,
35–6, 36–7, 43
empirical relations
and semantic relations 64
INDEX 325
empiricism
and conceptual dualism 191–2
epistemic capacities
and concepts 73
epistemic immunity to revision 317–19, 320
epistemic justification 17
internalist and externalist 16–19
and process reliabilism 13–14, 239–40, 242,
243, 244, 247, 250–1
epistemic possibility argument
and Hawthorne’s lottery propositions 257,
258, 263–4
epistemic probability
and abbreviative concepts 307
and logical concepts 313
epistemology of modal knowledge 14–15
equipotentiality argument
and type–materialism 114–15
Euclidean geometry 306, 311, 317, 318
Evans, G. 38
evidence argument
and process reliabilism 241–2, 245–9, 251
evil demon hypothesis (Descartes) 17–18,
244, 288
Existential Introduction, rule of 57
existential quantification 57
experience
representationalist theory of 8, 9–13
type–materialist theory of 5–8
experiential awareness
representationalist theories of 201–10
explanationism 16–18
explanatory power argument
for type–materialism 139–40
expression
and semantic relations 57
externalist concept of evidence
and reliabilism 248
externalist justification of beliefs 16–19,
239–52
faculty of conception 126–31
Feigl, H. 136
Field, H. 32, 40
Finke, R.A. 121–2
Fodor, J.A. 58, 78, 79, 107, 114
folk psychology
and pain 160, 172, 184, 185
and substitutionalism 62–5
and visual experience 230–1
and visual qualia 201, 205, 206, 210–11
Fox, I. 32, 117, 297
Frege, G. 52, 61, 171
Freud, S.
The Interpretation of Dreams 5
Friedman, M. 32
Fuller, S. 189
326 INDEX

Fumerton, R. 155 iconic representation

functionalism and visual qualia 203–4
and type–materialism 103, 142, 145, 146, 147 idealism
constructive and non–constructive 6
Gardner, E.P. 159 identity
gavagai problem (Quine) argument for the necessity of 130
and the determinacy of meaning 23–31, 77, and the determinacy of meaning 26–9,
78, 83–7, 87 30–1, 86
Gelman, S. 281 and type–materialism 101, 131–4
Gentzen, G. 56 identity theory 8, 11, 136–52
Gibbins, P. 315 and laws of nature 140
Gibson, J.J. 286 and the multiple realization argument 141–5
Gillett, C. 144 and property dualism 7, 131, 138,
Gobell, J. 227 146, 147
God and qualia 136–7, 140–1, 179
and type–materialism 101 and representationalism 146, 147
Gödel, K. 277 and the "same insides" principle 144–5
Goldman, A. 239 and type identity 136
Goodman, N. 6 see also type–materialism
Gozzano, S. 136 imagination
Grahek, N. 143 and the mind–body problem 120–6
grain argument 148, 204 and modal knowledge 274
grammar, rules of logical 274 indexicals
Grice, H.P. 299 and reference 46–7
Gricean principle and Hawthorne’s lottery individuation
propositions 254, 262–3, 269, 270 and the multiple realization argument 144
and subjunctive conditionals 294–5 inference, role of concepts in 5, 302, 304,
grounding theory of truth 62 306, 312–7
Grover, D. 55 information
Gupta, A. 32, 51, 53, 62, 66, 218, 297, 307 and the causal chain problem 78–9
and reference 65
Harman, G. 7–8, 100, 179, 193, 232 information processing
Harris, J.H. 301 and representationalism 9
Hart, W.D. 117 informational uptake, law of 62, 63–4
Hawthorne, J. 14 intermundane reference 44–6
lottery puzzle and the nature of belief 253–72 internalist concept of evidence
Hefer, D. 273 and reliabilism 248, 249
Heider, Fritz 142 internalist justification of beliefs 16–18
heuristics introspection
and dual system theories of cognition 14 and the mind–body problem 5–6, 7
Hill, C.S. 3, 8, 15, 48, 66, 67, 68, 136, 139, 149, and pain 188
157, 178, 179, 184, 210, 211, 219, 308 and representationalism 9–10
Consciousness 13, 136, 179, 197, 206, 291 intuition
Sensations 7, 128, 129, 136, 243 and conceivability 131
Thought and World 51, 52, 59, 67, 68 and the faculty of conception 129–30, 134
Hobbes, T. 136 and imagination 120–6
Hofbauer, R.K. 180, 188 modal intuitions 134, 277
holism and reference 77, 78 and representationalism 10
Holmes, G. 99
Hopper, Edward Jackson, F. 111, 131
Sun in an Empty Room 221 Jessell, T.M. 159, 182
Horwich, P. 2–3, 4, 53, 66, 67, 299, 308
Huemer, M. 232 Kahneman, D. 14, 143, 174
human faces Kandel, E.R. 159
and visual experience 231–2 Kant, I. 15, 275, 291
Hume, D. 191–2, 275 Kantianism and representationalism 10
Hyslop, A. 111 Keil, F. 281
 INDEX 327

Kim, J. 115, 197–8, 200–1 Lycan, W.G. 97, 100, 179

knowledge 13–19
and dual system theories of cognition 14
and lottery propositions 253–72
and truth 14, 15–16
see also modal knowledge
knowledge account of assertion 260–2
knowledge argument
and type–materialism 6, 131, 148
Kriegel, U. 197
Kripke, S. 7, 62, 91, 92, 103
and Cartesian modal arguments 289, 291
Kripkean paradox 78
Kripkean skepticism 69, 70
and metaphysical modalities 281–3, 285
and the mind–body problem 117–20, 121, 125
Naming and Necessity 104
and a priori modalities 276
laws of nature
and identity propositions 140
and substitutionalism 52
laws of reference 37–9, 40, 41–2, 49–50,
63–5, 75–6
and the similarity hypothesis 45
Leeds, S. 32
Leibniz, G. 101
Leibniz’s Law 31, 86
Lewis, D. 40, 107, 295, 319
Liu, T. 189
Loar, B. 136, 149, 177, 178
Lockean appearance properties
and visual qualia 212–13
logic, laws of
and modal knowledge 273, 275, 276–7, 278,
279–80, 286
and Quine’s revisability thesis 299–303,
304–5, 312–7
logical concepts
empirical evaluation of 298, 306, 307, 312–17
logical positivism 275
lottery propositions (Hawthorne) 253–72
assertion argument 257–8, 259–63
and deductive arguments 270–1
epistemic possibility argument 257, 258,
263–4
and the Gricean principle 254, 262–3,
269, 270
multi–premise closure (MPC) 256–7,
269–9, 271
and paradigm cases of knowledge 270–1
and parity reasoning 254–7, 269
practical reasoning argument 257, 258–9,
264–9
simplification–permitting solution 267–8
Löwenheim–Skolem Theorem 78
Lucretius 136
Lyons, J. 155
McGee, V. 31, 297
Mach, E. 226
McLaughlin, B. 31, 136, 139, 178, 218
Malcolm, N. 111
Marcus, R. 48
Marr, D. 233
Martin, J.H. 159
mathematical concepts 88–92, 93, 273, 280, 286
abbreviated 311
and cognitive interests 306
Euclidean geometry 306, 311, 317, 318
immunity to revision 317–19
and Quine’s revisability thesis 299–300,
301, 302–3
Meagher, M.W. 182
meaning 1–5
determinacy of 4–5, 23–31, 69–70, 77–87,
88–93
see also concepts; reference; substitutionalism
Melzack, R. 181, 182, 184
memory
and numerical concepts 91, 92
metaphysical modalities 281–5, 287–8
metaphysical necessity 15, 282, 283–5, 285–6
and subjunctive conditionals 292–6
metaphysical possibility 282
Millikan, R. 215, 297
Minar, Edward 241
mind–body problem 5–8, 117–34
and the faculty of conception 126–31
and imagination–based intuitions of
separability 120–6
Kripke’s line of thought 117–20, 121, 288–92
and modal knowledge 274
and representationalism 9, 10–11, 13
and visual qualia 210–11
minimalism
and deflationism 67, 68, 69–70
misrepresentation
and the peripheral state theory of
pain 189–90
Mladejovsky, M.G. 99
modal intuition 277
modal knowledge 273–96
Cartesian modal arguments 117–35, 148,
288–92
empirical revision of modal beliefs 303
epistemology of 14–15, 273–96
and imagination 274
metaphysical modalities 281–5, 287–8
and modal intuitions 134
nomological modalities 285–8
a priori modalities 275–81
see also conceivability; necessity; possibility
328 INDEX

model theory and defense of type–materialism 5–6, 103,

and reference 78, 90
Moore’s Paradox 261
multi–premise closure (MPC)
and Hawthorne’s lottery propositions 256–7,
269–9, 271
multiple realization argument (MRA) 7,
107–10, 141–5
Mundale, J. 144
Nagel, T. 118
“What Is It Like to Be a Bat?” 120–4
natural kind properties
and visual experience 219, 231–2
natural properties
and semantic correspondence 75–6
natural relations
and conceptual reference 76–8, 87, 93
naturalism
and reference 58, 82–3
naturalistic constraints
on theoretical concepts 76–8, 89–90
necessity
and metaphysical modalities 281–2
and modal knowledge 14–15, 273, 275–7,
278, 279–80
see also metaphysical necessity
Nelson, R.J. 97
neurological kinds 97
neurological qualities
and type–materialism 97, 106, 109
neuroscience
and the concept of pain 133, 159, 167,
171, 191, 180–90
and the multiple realization argument 143–4
and qualia 149–50
and sensory concepts 127–9
and type–materialism 114–15
Newton, I. 140
Nisbett, R.E. 113
nomological modalities 285–8
nomological possibility 277
numerical concepts 88–92, 93
objectual quantifiers 57
Occam’s Razor 100, 102, 138, 141
occclusion argument
and visual experience 227
optimal perceptual circumstances 82
pain 11, 155–76, 177–94
appearance/reality distinction for 156–7,
175, 290–1
and Cartesian modal arguments 288–9,
289–92
cognitive role of the concept of 161–71, 175
and damage 160, 172–5, 181, 184–8, 193
104–5, 106–7, 110
experiential awareness of 157–161, 191
and identity statements 131–4
and identity theory 136, 137, 146, 147–8, 149
injured soldiers and athletes 155, 156–7, 165,
171, 181
and the mind–body problem 118, 119, 120,
122–3, 124
and the multiple realization
argument 141–2, 143
perceptual awareness of 150, 157–61, 175,
178, 192
peripheral state theory of 180–93
phantom limb pain 150–2, 156, 165, 171,
181, 184
and qualia 150, 179–80, 193, 234
and representationalism 10–11, 12
semantic representations of 167–8, 169, 170–1
as a sensory concept 127–8
thermal grill illusion 181, 184
truth conditions for pain statements 168–71
Palmer, S.R.
Vision Science 213, 224, 225, 227
Papineau, D. 136, 149, 178
paradigm cases of knowledge
and Hawthorne’s lottery propositions 270–1
paradoxes
and conceptual reference 76–8
and deflationism 74–8
of pain 155–76
semantic 47–8, 61–2
parity reasoning
and Hawthorne’s lottery
propositions 254–7, 269
Pautz, A. 183, 234
Peacocke, C. 161
Peano arithmetic 89, 92, 93
Penfield, W. 99
perception
and content 218–36
justification of perceptual beliefs 16–19
and type–materialism 6–7
perceptual awareness
of pain 150, 157–61, 175, 178, 192
and qualia 150, 178
and representationalism 9, 10, 11
and visual experience 220–3
and visual qualia 207
perceptual concepts 78–88
and the causal chain problem 77, 78–80, 88
and cognitive interests 304
and the new Quinean argument 83–7
and the problem of error 77, 78, 80–3, 88
perceptual experience 218–20
and identity theory 137
see also visual experience
 INDEX 329

perceptual imagination and visual experience 13, 234

and the mind–body problem 121–6 property dualism 7, 131, 138, 146, 147
perceptual qualia 197–8, 199 and visual qualia 210
perceptual relativity propositional attitudes 50
and visual experience 223–9 and perceptual experiences 218
peripheral states role of concepts in 5
and pain 180–93 propositional truth 2–3, 51, 54
and qualia 178 prosentences 55
perspective prothoughts 55
and perceptual awareness 9 Putnam, H. 7, 32, 38, 78, 98, 297
phenomenal concepts 6–7, 8, 11, 149, 150, and Cartesian modal arguments 289
177–8, 191–2, 193 and Cartesian skepticism 240–1
phenomenal kinds 97–8, 114 first John Locke lecture 41
phenomenal qualities and metaphysical modalities 281–2, 285
and type–materialism 97–8, 106, 109, 110, and the multiple realization argument 141–5
113, 115 Pylyshyn, Z. 121
phenomenology
and visual experience 229–30 qualia 97–116, 117–35, 136–52, 155–76,
physical concepts 177–96, 197–217, 218–29, 232–5
and deflationism 71–2 and brain states 149–51, 192–3
physicalism experiential character of qualitative
and qualia 201–2, 214 awareness 190–3
and semantic relations 58 and identity theory 97–116, 117–35, 136–52,
and type–materialism 5–6, 8, 114 177–80, 190–3
Place, U.T. 136, 178 and pain 150, 179–83, 193, 234
Plato 277 properties of 199–200
Polger, T. 136 and property dualism 146
Porro, C.A. 180 and quantitative similarity/difference 199
Posner, M.I. 190 and representationalism 10, 11–13, 155–76,
possession conditions 177–96, 197–217, 218–29, 232–5
for the concept of pain 161–6, 167, 170, 171–2 and type–materialism/central state identity
possibility 273, 274–5, 275–6, 277, 280 theory 97–116, 117–35, 136–52
and Cartesian modal arguments 288–92 see also visual qualia
conceptual 282–3 quantification
metaphysical 282 and the determinacy of meaning 24–6
nomological 277 see also substitutional quantification
practical reasoning quantum logic 298, 300, 301–2
and dual system theories of cognition 14 disjunction 314–17
and Hawthorne’s lottery propositions 257,
258–9, 264–9 Quine, W.V.O. 2, 4, 5, 15, 139, 318
prediction and behaviorism 85, 87
and Quine’s revisability thesis 301–3 critique of analyticity 297–320
Price, D.D. 159, 181, 182, 188 on abbreviative concepts 309–10
Prinz, J.J. 192, 230, 233 and the concept of truth 54
probability–based practical reasoning gavagai example 23–31, 77, 78, 83–8
and Hawthorne’s lottery Quinean skepticism about reference 69–70
propositions 266–7, 268 revisability thesis 15–16, 297–320
problem of error 77, 80–3, 88 “Two Doctrines of Empiricism” 297,
process reliabilism 298–303
and Cartesian skepticism 13, 16–19, 239–52
and the circularity argument 245, 250–1 Rainville, P. 180, 182
and the evidence argument 241–2, Ramachandran, V.S. 184
245–9, 251 Ramsey, F.P. 249
objection to 242–3 Rasmassen, T. 99
theory of 239–40 real world hypothesis (RWH)
Profitt, D. 174 process reliabilism and Cartesian
projection hypothesis skepticism 240–2, 244, 245
330 INDEX

reduction Scheffler, I. 23
and substitutionalism 51–2, 58–62 Schroeder, T. 155
redundancy theory scope
of reference 33–4, 34–5, 36–7, 43 of perceptual awareness 9, 219
of truth 32 Sellars, W. 148, 204
See also deflationism; substitutionalism semantic concepts
reference 1, 2, 3, 32–50 and deflationism 68–9, 70–1
and deflationism 66, 67, 68, 71–4, 76 rudimentary theory of reference for 32–50
determinacy of 4–5, 69–70, 72–93 and substitutionalism 51–2, 57–65
and disquotation schemata/sentences 33, 34, truth–related 1, 2
35–7, 43 semantic correspondence 51, 57–8, 59–60, 63
intermundane 44–6 and deflationism 68
laws of 37–9, 40, 41–2, 45, 49–50, 62–5, 75–7 linkage with natural properties 75–6
and perceptual concepts 78–94 semantic paradoxes 47, 61–2
a posteriori concept of 39, 40, 41–2, 46, 47 semantic substitutionalism 51–65, 67–9
a priori concept of 39, 40–1, 42, 46, 47 Senglish (imaginary language) 37–48
referential ambiguity 48 and the similarity hypothesis 42–8
and semantic relations 51, 57, 58–9, 63, 64–5 Senor, T. 155
and the similarity hypothesis 42–8 sensitivity
and substitutional quantification 63–5, 76 and perceptual concepts 88
reliability condition sensory concepts 126–134
and type–materialism 112–13 sensory qualia 197, 199
reliable indication, laws of 62 sentences
reliable inference, laws of 62–3 and prosentences 55
representational qualities separability
of conscious experience 97, 98 and the faculty of conception 126–31
representationalism 8, 9–13, 155–76, 177–96, imagination–based intuitions of 120–6
197–217, 218–235 Shaffer, J. 105, 106
and identity theory 146, 147, 190–3 Shephard, R. 121
and perceptual experience 218–19 Sidelle, A. 283
and qualia 10, 11–13, 178–9, 179–80, 183, Siegel, Susanna 218
190–3, 193–4, 197–217 similarity hypothesis 42–8
use of the term 202, 208–9 Simmel, M.–A. 142
and visual awareness 197, 201–10, 215 Simmons, K. 51
resolution simplicity argument
and perceptual awareness 9, 219 for type–materialism 137–9, 141
response expansion simplification–permitting solution
and pain 183 and Hawthorne’s lottery propositions 267–8
revisability 15–6, 297–320 skepticism
epistemic immunity to 317–19 and deflationism 69–70
of logical concepts 314–17 process reliabilism and Cartesian
Quine on 299–303 skepticism 13, 16–19, 239–52
telelogical immunity to 317, 319–20 Quine’s critique of analyticity 297,
revision theory of truth 62 298–303, 320
Rhudy, J.L. 182 see also Cartesian skepticism
risk aggregation Smart, J.J.C. 100, 137–8, 178
and Hawthorne’s lottery Soames, S. 32
propositions 256, 269 Stalnaker, R. 139
robust concepts Stevens, S.S. 183
semantic concepts as 4 Strawson, P.F. 25, 111, 299
Russell, Bertrand 6, 58 subjective probability
Russellian theories of awareness 9–10, and reliabilism 249
207–8, 219 subjunctive theory
and metaphysical necessity 292–6
“same insides” principle substitutional quantification 55–7, 62, 63,
and identity theory 144–5 64–5, 67–9
Schechter, J. 13–14, 253, 297, 302 and the concept of pain 170

and deflationism 67, 68–9
and reference 76
substitutionalism 51, 57–65, 67–70
and deflationism 67–8, 69–70, 76
and the laws of folk psychology 62–5
and reduction 51–2, 58–62
and semantic relations 51, 57–65
and truth 51, 53–7, 61–2
sufficient reason, principle of 100
Swampman 214–16
sympathetic imagination
and the mind–body problem 121, 126
System 1 heuristics 143
Tarski, A. 36, 40, 53, 90
telelogical immunity to revision 317, 319–20
Thagaard, P. 115
thoughts
and concepts 89–90
and meaning 1–2
and prothoughts 55
and semantic relations 60–1
use of the term “thought” 1–2, 52
Thouless properties 233–5
Timmermann, L. 180
token identity 136
token–materialism 98, 101–2
defining 117
transparency
and representationalism 9–10, 13
and type–materialism 7–9, 13
triangulation
and perceptual concepts 79–80
truth 1
concept of 15–16, 303
and deflationism 67–8, 68–9, 70–1
and knowledge 14, 15
Liar paradox 48
propositional 2–3, 51, 54
redundancy theory of 32, 36
revision theory of 62
substitutional theory of 51, 53–7, 61–2
of thoughts 2
truth–related semantic concepts 1, 2, 4
truth value gaps 25–6
see also propositional truth
truth conditions
for abbreviative concepts 308
for pain statements 168–71
Turatto, M. 189
Turing machine analogy 91, 92
Tversky, A. 174
Tye, M. 12, 179, 202, 215, 232
type identity 136–7
INDEX 331
type–materialism 5–8, 97–116, 136–152
causal argument for 140–1
conditional justification of 100–2
and the correlation thesis 98–102
defense of 6–7, 97–116
defining 117
explanatory power argument for 100–2,
139–40
and neurological qualities 97, 106, 109
objections to 7–8, 147–9
and phenomenal qualities 97–8, 106, 109,
110, 113, 115
a priori objections to 103–13
simplicity argument for 137–9, 141
theory of qualia 177–8, 190, 193
and token–materialism 98, 101–2, 117
see also identity theory; mind–body problem
universal quantification 57
visual experience 99, 218–36
appearance properties of 220–3, 229, 232–5
and folk psychology 230–1
and human faces 231–2
and perceptual relativity 223–9
and Thouless properties 233–5
visual qualia 178, 187–217, 220–3, 232–5
appearance properties of 211, 213–14
appearance/reality distinction in 199, 200,
201, 206, 210
color qualia 141, 148, 149
experiential awareness of 203–6
phenomenological and epistemic senses of
“looks” 198–9
and Swampman 214–16
visually guided action
and the multiple realization argument 142
Wall, P.D. 181, 182, 184
Warren, L. 155
Warren, W. 218
Westendorf, D. 97
White, S. 131
Williamson, T.
Knowledge and its Limits 260–2
Wilson, M. 66
Wittgenstein, L. 115
Philosophical Investigations 303
Wright, C. 241
Yablo, S. 117
zombihood, intuitions about the possibility
of 124–6, 129–30