OXFORD STUDIES IN METAPHYSICS OXFORD STUDIES

l!ditoria/ Aduis01y Board IN METAPHYSICS

David Chalmers (New York University md Australasian National University)
Andrew Cortcns (Boise State University)
Tamar Szabo Gendler (Yale University) Volume 10
Sally Haslangcr (MIT)
John Hawthorne (University of Southern California)
Mark Heller (Syracuse University)
Hud Hudson (Western Washington University)
Kathrin Koslicki (University of Alberta)
Edited by
Kris McDaniel (Syracuse University)
Brian McLaughlin (Rutgers University) Karen Bennett
Trenton Merrick<; (University of Virginia) and
Kevin Mulligan (Universitc de Gcncvc) Dean W. Zimmerman
Laurie Paul (University of North Carolina-Chapel Hill)
Theodore Sider (Cornell University)
Timothy Williamson (Oxford University)

Managing Editor
Peter van Elswyk (Rutgers University)

OXFORD
UNIVERSITY PRESS
 OXFORD PREFACE
VNIYBRSITY l'RllSS
Creal Clarendon Street, Oxford, OX2 GDP.
Unhcd Kingdom
Oxfind Studies in Metaphysics is dedicated to the timely publication of new
Oxford U niversity Press is a de pan mcn r of the University of Oxford.
(t furthers the Univcrsiry's ohjecrivc of cxcd!ence in research, scholarship,
work in metaphysics, broadly construed. '[he subject is taken to include not
and education by publishing worldwide. O xford is a regisrercd trade mark of only perennially central topics (e.g. modality, onwlogy, and mereology) but
Oxfi) rd University Press in the UK and in ccnain mhcr coum rics
also metaphysical questions that emerge within other subfields (e.g. phil-
@ lhc several contrib urnr.s 20 17
osophy of mind, philosophy of science, and philosophy of religio n). Each
·rhc moral righcs or the ;tuthor;; have been asscncd volume also contains an essay by the winner of the Sanders Prize in
First Edition published in 2017 Metaphysics, an annual award described within.
Im pression: 1
K. B. & D . W. Z.
All rights reserved. No pan of this publication may be rcpro<luccd , stored in
a retrieval system, or transmiued , in any fo rm or by any tneans_. withma the !tht1ct1, NY & New Brunswick, NJ
prior permissio n in writing of Oxford Un iversity Press, or as expressly perm itted
by law, by licence or under terms agreed \\rith the appropriate rcprographics
rights organizm ion. Enquiric$ <:(inccrning reproduction outside the sc.o pc (lf the
above should be sent w the flights Dcpan mc nr, Oxford Unive rsity Press, ~H the
address above

You must nm circulate this work in any mhc r form

and you must impose this same condition on any acqu irer

Pub lished in the United States of Amerirn by Oxford UnivcrSity Press

198 Madison Avenue, New York, NY 10016, United States o f Ainerica

British Library Cacaloguing in Publication Data

Data avai lable
Library of Congress Conuol Number: 20042'\3307
ISBN 978--0- 19- 879 197-3 (hbk.)
978- 0- l'J-879198-0 (pbk_)
Printed in Great Britain by
Clays Lid, St Ivcs pie
Links co th ird party websites a rc p rovided by Oxford in good faith a nd
for in fornudo n o nly. Oxford disclaims any responsibility for d1c materials
conca.incd in <tny third party 1,vcbsitc referenced i11 thi.s work.
CONTENTS

IJg St11ulers l'riz.e in Mett1physics ix

DISPOSITIONS

Teleological Dispositions 3
Nick Kroll
2 Indirect Directness 38
Jennifer McKitrick
3 Dispositions without Teleology 47
Dt1vid Mtmley t1nd Ryt1n Wt1sserman

ANALYTICITY REVISITED

4 Devious Stipulations 63
john Horden
5 Stipulations and Requirements: Reply to Horden 74
Louis deRosset

WHAT REALITY IS LIKE

6 Colors as Primitive Dispositions 85

Ht1git Benbaji
7 Arc There Ineffable Aspects of Reality? 124
Thomas Hojiveber
8 The Metaphysics of Quantities and Their Dimensions 171
Bradfoid Skow

MODALITY AND EXISTENCE

9 Vague Existence 201

Alessandro Torza
10 Ersatz Counterparts 235
Richard Woodward
Vlll Contents

GROUNDING AND EXPLANATION

THE SANDERS PRIZE IN METAPHYSICS
11 The Principle of Sufficient Reason and Probability 261
Alexander Pruss
Sponsored by the Marc Sanders Foundation* and administered by the
l 2 Crounding Ground 279 editorial board of Oxford Studies in Mettlphysics, this annual essay competi-
Jon Erling Litlmzd tion is open to scholars who are within fifteen years of receiving a Ph.D. or
students who are currently enrolled in a graduate program. (Independent
Author lndex 317 scholars should enquire of the editors to determine eligibility.) The award is
$10,000. Winning essays will appear in Oxford Studies in Mettlphysics, so
submissions must not be under review elsewhere.
Essays should generally be no longer than 10,000 words; longer essays
may be considered, but authors must seek prior approval by providing the
editor with an abstract and word count by 1 November. To be eligible for
next year's prize, submissions must be electronically submitted by
31 January. Reforeeing will be blind; authors should omit remarks and
references that might disclose their identities. Receipt of submissions will
be acknowledged by e-mail. The winner is determined by a committee of
members of the editorial board of Oxford Studies in Metaphysics, and will be
announced in early March. At the author's request, the board will simul-
taneously consider entries in the prize competition as submissions for
Ox_/(Jrd Studies in Metaphysics, independently of the prize.
Previous winners of the Sanders Prize are:
Thomas Hohveber, "Inexpressible Properties and Propositions", Vol. 2;
Matthew McGrath, "Four-Dimensionalism and the Puzzles of Coincidence",
Vol. 3;
Cody Gilmore, "Time Travel, Coinciding Objects, and Persistence", Vol. 3;
Stephan Leuenberger, "Ceteris Absentibus Physicalism", Vol. 4;
Jeffrey Sanford Russell, "The Structure of Gunk: Adventures in the Ontology
of Space", Vol. 4;
Bradford Skow, "Extrinsic Temporal Metrics", Vol. 5;
Jason Turner, "Ontological Nihilism", Vol. 6;
Rachael Briggs and Graeme A. Forbes, "The Real Truth About the Unreal
Future", Vol. 7;
Sharnik Dasgupta, "Absolutism vs Comparativism about Quantities", Vol. 8;
Louis deRosset, "Analyticity and Ontology", Vol 9;

' The Marc Sanders Foundation is a non-profit organization dedicated to the revival of
systematic philosophy and traditional metaphysics. Information about the Foundation's
other initiatives may be found at <http://www.marcsandersfoundation.com/>.
x The Sanders Prize in Met11ph_ysics

Nicholas K. Jones, "Multiple Constitution", Vol. 9;

Nick Kroll, "Teleological Dispositions", Vol. 10;
Jon Erling Litland, "Grounding Grounding", VoL 10.
Enquiries should be addressed to Dean Zimmerman at:
dwzimmer@rci.rutgers.edu

DISPOSITIONS
 1
Teleological Dispositions
Nick Kroll

l. INTRODUCTION

Some things are disposed to break when struck. Some things are disposed to
bend when stressed. Some things arc disposed to dissolve in water. Things
have dispositions.
But dispositions need not manifest. A vase disposed to break when struck may
meet its end by melting rather than breaking. A rod disposed to bend when
stressed might succumb to rust without ever bending. And a chunk of salt
disposed to dissolve in water may be used up while making hydrochloric acid.
These are mundane observations. Reflecting on them, however, leads to
two important intuitions about dispositions. The first is that dispositions
have some kind of directedness. The second is that dispositions have some
kind of connection with conditionals.
Consider the vase that meets its end by melting rather than breaking. The
vase doesn't break. Yet it seems that in virtue of being disposed to break
when struck, the vase is in a state that is in some sense directed at a state of
<lffairs in which the vase breaks. Likewise, even though the wire never bends,
it seems that in virtue of being disposed to bend when stressed, the wire is in
a state that is in some sense directed at a state of affairs in which the wire
bends. Generalizing, the idea is that something with a disposition is in a
state that is in some sense directed at a manifestation of the disposition. Put
simply: a disposition need not manifest but it is in some sense directed at
manifesting.
I've characterized the directedness intuition as one that needs an explan-
ation. What needs to be explained is the sense in which dispositions are
directed at manifesting. Keeping this in mind, let's turn to the intuition that
dispositions have some kind of connection to conditionals.
What if the vase had been suitably struck instead of being placed in the
furnace? Presumably, it would have broke and so its disposition to break
when struck would have manifested. Likewise, what if the wire had been
4 Nick Kroll
suitably stressed before corroding? Presumably, it would have bent and so its
disposition to bend when stressed would have manifested. Generalizing, it
seems that while a disposition need not manifest, it nonetheless wouU
manifest if certain conditions were to obtain. There thus appears to be
some kind of connection between dispositions and conditionals.
Suppose the connection between dispositions and conditions is a strong
connection. In particuhr, suppose that we have an informative and coun-
terexample free conditional analysis of dispositions. Then perhaps we could
explain the directedncss intuition: the sense in which a disposition is
directed at manifesting is that the disposition would manifest if conditions
C were to obtain (where conditions Care specified by the given conditional
analysis).
Orthodoxy would have it that the correct account of dispositions rests on
a conditional analysis of dispositions and so such an explanation of the
dircctcdness intuition is the right explanation. I argue otherwise. In particu-
lar, 1argue f()r a teleological account of dispositions. According to this account
of dispositions, the connection between dispositions and conditions is
explained in terms of the directedness of dispositions and the dircctcdness
of dispositions is a teleological directedness.
We begin by undermining orthodoxy. Following Molnar (2003) and
Fara (2005), I believe the project of analyzing dispositions in terms of
conditionals is a lost cause. But the purpose of our overview of conditional
analyses is not to establish such a strong conclusion. The purpose is rather to
motivate a turn towards a teleological account of dispositions.
2. CONDITIONAL ANALYSES
2.1 The simple conditional analysis
Our starting point is the so-called simple conditional analysis of dispositions.
The basic idea behind this analysis is that every disposition has a stimulus
condition, and if that stimulus condition were to obtain, the disposition
would manifest. More explicitly, the proposal is this:
(SCA) Necessarily: xis disposed to M when C iff x would M if C were
the case. 1
For example, if (SCA) is correct, a vase is disposed to break when struck just
in case the vase would break if it were struck.
1 intuition that (SCA) is subject to counterexample by situations involving finks and masks.
See Ryle (1949), Goodman (1954), and Quine (1960).
Teleological Dhpositions 5
(SCA) is obviously compatible with the observation that a disposition
need not manifest: if the vase is never struck, its disposition to break when
struck never manifests. But (SCA) has some problems. We'll mention two
of the most notorious.
First, there is the problem offinks. The problem here is that the stimulus
condition for a disposition may also be a condition for losing or acquiring
the disposition. The classic example is C. B. Martin's electro-jink. 2 It's
connected to a wire that is not disposed to conduct electricity when wuched
by a conductor. I-!owever, the fink ensures that if the wire were to be
couched by a conducror, the wire would acquire the disposition. Thus,
while the wire is not disposed to conduct electricity when touched by a
conductor, it would conduct electricity if it were touched by a conductor.
Martin also provides ~m example going the other way. Suppose the
electro-fink has a reverse cycle. On reverse cycle, the electro-fink ensures
that any wire connected to it that is disposed to conduct electricity when
touched by a conductor loses this disposition when it is touched by a
conductor. Consider such a wire. While the wire is disposed to conduct
electricity when touched by a conductor, it would not conduct electricity if
it were touched by a conductor. On the contrary, if it were touched by a
conductor, it would lose this disposition. So, we have counterexamples to
(SCA) in both directions.
Second, there is the problem ofmt1Sks. A mask is something that prevents a
disposition from manifesting when the stimulus condition obtains, and it
docs so without taking away the disposition. 3 A standard example involves
an antidote for a poison. 4 The poison is disposed to kill when ingested. But
when ingested, the poison takes some time to do its work. During that time,
if you were to take the antidote, you would be saved. The antidote, however,
does not remove the poison's disposition. It simply prevents the poison
from doing any more damage. Put it this way: even though the poison is
killing you, it need not kill you. You could take the wtidote. So, there are
situations where the poison wouldn't kill ifit were ingested even though the
poison is disposed to kill when ingested. We have another counterexample
to (SCA). 5
2 3
See Martin (1994). Masks owe their name to Johnston (1992).
4
This example is due to Bird (1998).
5
Choi (2008) denies that (SCA) is subject to counterexample from finks and masks.
Somehow, the right-hand side of the relevant instance of (SCA) is such that the possibility
of finks and masks doesn't arise. Like many others, I'm not convinced. One reason why
I'm not convinced is that if Choi is right, it's hard to sec why so many have had the
G Nick Kroll Te!eologictd Dispositions 7

that, in this scenario, the wire is not disposed to conduct electricity when
2.2 Appealing to the exclusion of external interference touched by a conductor.
The scenario involving the electro-fink running on reverse cycle is han-
One response to situations involving finks and masks is that they involve
dled in a similar manner. In this case, the electro-fink makes the wire, when
some kind of external interferencc. 6 In the case of finks, some kind of
touched by a conductor, lose some intrinsic property, and consequently lose
external interference causes an object to acquire or lose the relevant dispos-
the disposition to conduct electricity when touched by a conductor. But
ition when the stimulus condition for the disposition obtains. In the case of
what would happen if the wire were touched by a conductor and it were to
masks, some kind of external interference prevents the manifestation of a
retain this intrinsic property? It would conduct electricity, so it seems. So,
disposition from obtaining even when disposition's stimulus condition
(LCA) correctly predicts rhat the wire is disposed to conduct electricity
obtains and the disposition remains.
when touched by a conductor.
Perhaps, then, we can avoid the problem of finks and the problem of
Masks, however, are still a problem. You ingest the poison but take the
masks by appealing to a clause which excludes external interference.
antidote. The antidote doesn't remove the poison's disposition to kill when
(ECA) Necessarily: xis disposed to M when C iff x would M if C were ingested. It just prevents the manifestation of the disposition from obtaining.
the case and nothing external were to interfere. So, the poison retains whatever intrinsic property grounds its disposition.
(LCA) thus predicts that you die. But you don't, thanks to the antidote.
(ECA) should only be seen as a first attempt. For it raises the following
There is a modification of (LCA) that avoids this counterexample.
question: nothing external to what interferes with what? More needs to be
Consider (LCA') and one of its instances.
said.
I'm going to attempt to fill in the details. We'll begin with a conditional (LCA') Necessarily: xis disposed to M when C iff x has some intrinsic
analysis of dispositions inspired by Lewis (1997). property I in virtue of which: if C were to obtain and x were to
retain I, there would be a process p such that if nothing
(LCA) Necessarily: x is disposed to M when C iff x has some intrinsic
external top were to interfere with p, x would M (as a result).
property I in virtue of which: if C were the case and x were to
retain I, x would M. (1) Necessarily: the poison is disposed to kill when ingested iff: the
poison has some intrinsic property I in virtue of which: if it were to
One might ask: why the appeal to intrinsic properties? Lewis answers, in
ingested by x and it were to retain I, there would be a process p
effect, by claiming that dispositions are intrinsic properties of their bcarers.7
such that if nothing external top were to interfere with p, the poison
This claim is controversial. Some have argued that some dispositions are
would kill x (as a result).
extrinsic properties. 8 But we need not concern ourselves with whether
dispositions arc intrinsic properties. More important for present purposes
Now take some time t after you have ingested the poison but before you
is that the appeal to intrinsic properties seems to solve the problem of finks.
have taken the antidote. The poison is causing damage to your organs at t.
Consider once again the wire connected to the electro-fink that is not
Let this process be p. 'fhe question, then, is whether p is such that if nothing
disposed to conduct electricity when touched by a conductor. !his wire
external to it were to interfere with it, the poison would kill you as a result.
doesn't have an intrinsic property I in virtue of which the wire would
It seems plausible, at least from the rather limited description of the case,
conduct electricity if it were touched by a conductor and retain I. The
that p is such a process. Shortly afrer t, you take the antidote. In doing so,
wire does have such an extrinsic property: namely, being connected to the
another process obtains, one which is external top. This process prevents x
electro-fink. And, sure enough, if the wire were touched by a conductor, the
from developing into one in which the poison kills you as a result. But if it
electro-fink would make the wire acquire such a intrinsic property. But this
weren't for this external interference, p would have developed into such a
is neither here nor there as far as (LCA) is concerned. It correctly predicts
process. Thus, ( 1) correctly predicts that the poison is disposed to kill you
when you ingest it.
(LCA') is the best I can do to spell out the "provided nothing external
6
Sec Johnston (1992). 7
See Lewis (1997, p. 155). interferes" response. But my best is not good enough: the analysis is subject
8
Sec McKitrick (2003). to counterexample.
8 Nick Kroll
Our counterexample is a variant of one due to Manley and Wasserman
(2008).'> Suppose there is a concrete block that withstands any damage done
to it by a sledgehammer unless it is hit in just the right spot with just the
right amount of force. Furthermore, it is nearly impossible to hit the block
in this spot with the right amount of frHce. But if the block is hit in the right
spot with the right amount of force, it shatters as a result. The block has an
Achilles' heel. Now while this block could shatter from being struck, it seems
that we would quite rightly think that it is not disposed to shatter when
struck.
Now suppose that, due to some fre<tk occurrence and despite no intrinsic
change to the block, conditions arc temporarily just right for the block to be
hit in the just the right spot with just the right amount of force, provided
someone is around to strike the block. For the next two seconds, if anyone
were to pick up a sledgehammer and strike the block, the block would be
struck in just the right spot with the right amount of force. But no one is
around to strike the block.
While conditions did momentarily obtain such that the block would
shatter if it were struck, these conditions did not thereby render the block
disposed to shatter when struck. We thus have a counterexample to (LCA +).
During the above described two second interval, it is not the case that the
block is disposed to shatter when struck. However, it is the case that
the block has some intrinsic property (its Achilles' heel) in virtue of which,
if the block were struck and retain this property, there would be a event
such that if this event were to con tinuc without interruption, the block would
shatter (as a result). So my best is not good cnough. 10
9
Manley and Wasserman (2008) use their example to refute conditional analyses that
reply to the problem of finks and rhe problem of masks by appealing to hyper-specific
stimulus conditions. The basic idea behind such analyses is that when we say, for example,
"the vase is disposed to break when struck," the disposition ascribed is not simply being
disposed to break when struck but being disposed to break when struck in some hyper-
specific way. (This seems to be Lewis's response to the problem of masks: see Lewis 1997,
p. 153.) I don't consider "hyper-specific" conditional analyses here, mainly because they
seem to be ad hoc responses to the problems of finks and masks. Manley and Wasserman
offer a more substantive (and conclusive) argument against hyper-specific conditional
ana~scs.
1
Contessa (2013) offers an "interference free" concli~ional analysis of disposition
ascriptions. 13ut his best is no better than mine: it is subject to counterexample as well.
Bypassing the details of his analysis, it snffices to note that his analysis predicts that ifx is
intrinsically disposed to lvf when C and y is something that interferes with x's being
intrinsically disposed to M when C, then it's not the case that x would 1Vf if C were the
case. But this prediction is incorrect. Here's why.
Suppose x is intrinsically disposed to break into pieces when struck. Further suppose
that y is something that interferes with x' s disposition in the following way. lfx were struck,
a process P of x breaking into pieces would begin as a result. However, this process P
Teleological Dispositions 9
2.3 Appealing to normal conditions
One might worry that the counterexamples we've considered so far invoke
situations that arc in some sense abnormal or atypical. Certainly, the
situation involving the block with an Achilles' heel is not a normal situation.
Likewise, it seems that wires, whether or not they are clisposccl to conduct
electricity when touched by a conductor, arc not normally connected to
some sort of electro-fink. And in the situation where you ingest the poison
but take the antidote, it would be natural to say something like "The poison
didn't kill you, but normally it would have."
Perhaps, then, we should consider a conditional analysis of dispositions
that explicitly appeals to normality.
(NCA) Necessarily: xis disposed to M when C iffin normal conditions,
x would Iv! if C were the case.
Supposing that situations involving finks, masks, and Achilles' heels arc not
normal situations, the hope is that (NCA) allows us to properly ignore these
situations.
But in what sense arc such situations abnormal? Fara (2005) points out
there need not be anything bizarre about a mask:
Dispositions of objects arc being masked all the time. I'm disposed to go to sleep
when I'm tired; but this disposition is sometimes masked by too much street noise.
Cylinders of rubber are disposed to roll when placed on an inclined plane; but this
disposition can be masked by applying a car's brakes ... [T]he masking of disposi-
tions is such a humdrum occurrence that any adequate account of [dispositions]
11
must accommodate it.
Similar remarks apply to Achilles' heels: they are so common that any
adequate account of dispositions must accommodate them. In short, then,
wouldn't culminate because some state of affuirs S involvingy would obtain that stops the
process. So, while x would have started cracking when S obtains, it wouldn't yet be broken
into pieces. But clue to S obtaining another state of aff:<irs S' would obtain. Due to S'
obtaining, another process!" ofx breaking into pieces would begin and culminate. So, ifx
were struck, it would (after all <md despite y) break into pieces.
Now, since S is a state of affairs that prevents the manifestation of x's disposition to
break into pieces when struck without taking away x's disposition, y is some kind of mask.
Furthermore, since S' is not a state of affairs that involves x being struck, x being broken
into pieces in S' is not a manifestation ofx's disposition to break into pieces when struck.
Putting this all together, we have a counterexample to Contcssa's analysis: (i) x is
intrinsically disposed to break into pieces when struck, (ii) y is something that interferes
with x being disposed to break into pieces when struck, and (iii) x would break into pieces
if it were struck.
11
Fara (2005, p. 50).
10 Nick Kroll
unless the ddender of (NCA) further specifics the sense of normality she is
appealing to, the threat of counterexample remains.
A defender of (NCA) may protest that she does not need to specify the
exact sense of normality in (NCA). She may simply rest her case on the
intuitive plausibility of the "Yeah ... but normally ... " response to situ-
ations involving finks, masks, and Achilles' heels. But this will not do.
To see why, let us begin with an observation due to Fara (2005). This is
the observation that situations that are normal with mpect to x being diJposed
to M when C are situations where x would M if C were the case. Following
Fara, we further note that it follows that if the sense of normality appealed to
in (NCA) is one that is relativized to x being disposed to M when C, the
right-hand side of (NCA) is subject to trivialization. It amounts to the trivial
claim that situations where x would M if C were the case are situations
where x would M if C were the case.
Now recall the incident involving the poison and the antidote. It may be
natural to respond with "Yeah, the poison didn't kill you, but normally it
would have." But if asked why the poison would normally would kill you, it
seems to me that one would say "Because it is disposed to kill when
ingested." But then the sense of normality appealed to is one that is
relativized to the poison being disposed to kill when ingested. If so, the
"Yeal1 ... but normally ... " response in this case actually undermines
(NCA). It suggests that the notion of normality appealed to in (NCA) is
one that is relativized to x being disposed to M when C, and thus the right-
hand side of (NCA) is subject to trivialization.
So, it would be a mistake to a defender of (NCA) to rest her case on the
"Yeah ... but normally ... " response to situations involving finds, masks,
and Achilles' heels. Furthermore, we've seen that unless the defender of
(NCA) gives us some reason to think otherwise, we have reason to believe
that the sense of normality appealed to in (NCA) is one that is relativized to
x being disposed to M when C. Consequently, we have reason to believe
that (NCA) is subject to trivialization. 12
Similar remarks apply to appeals to ideal conditions, typical conditions,
or ceteris parihus conditions. 13 Moline (1975) nicely captures the typical
attitude towards such appeals:
[T]hcy are fundamentally dodges ... They amount to muddled ways of disguising
from ourselves more or less serious ignorance of the dispositional properties of
individual things or persons and of rypcs of things or pcrsons. 14
12
Sec Hauska (2008) for further concerns about appealing to normality,
15
See Mumford (1998) and Steinberg (2009). 14 Moline (1975, pp. 244~5).
Teleologiml DiJpositions [l
I agree, ceterispcirihus. That is, unless an "escape-clause" conditional analysis
of dispositions can show that it is legitimate to appeal to its escape-clause(s),
it is fundamentally a dodge. 'fhus, the wony isn't just that such analyses are
subject to trivialization. It is also that, trivialization aside, they are
uninformative. 15
2.4 Appealing to proportionality
The conditional analyses we've considered so far face either the threat of
counterexample, the threat of trivialization, or a wony about informative-
ness. Finks, masks, and Achilles' heels provide the threat of counterexample.
Adding some qualification or "escape-clause" to avoid this threat brings with
it either the threat of trivialization or a worry about informativeness.
Perhaps, then, we should try to avoid the threat of counterexample some
other way. This is what Manley and Wasserman (2008) try to do with
(PROP).
(PROP) Necessarily: x is disposed to M when C iff x would M in a
suitable proportion of C-cases.
The basic idea behind (PROP) is that instead of looking at what would
happen at the closest world(s) where C obtains, we look at what would
happen in situations (some actual, the rest merely possible) where C obtains.
If a suitable proportion of these situations are situations where x Ms, then x
16
is disposed to M when C. The converse is alleged to hold as well.
An interesting feature of (PROP) is that situations involving finks and
masks are not ignored. To illustrate, consider a vase disposed to break when
struck and the following instance of the (PROP).
(2) Necessarily, the vase is disposed to break when struck iff the vase
would break in a suitable proportion of cases where it is struck.
15
Ir should be noted that Moline's complaint isn't avoided by inventing a new type of
conditional with a semantics that is supposed to model normal conditions, ideal condi-
tions, or ceteris paribus conditions. So, the conditional ;malyses of Maurrcau (1997),
Gundersen (2002), and Bonevac ct al. (2006) do not avoid Moline's complaint. Similar
remarks apply to an appeal to context to avoid the counterexamples (sec Fara 2005).
16 One might wonder whether (PROP) is a conditional analysis. Certainly, the right-
hand side of (PROP) is not a subjunctive or indicative conditional and no such condi-
tional is embedded in (PROP). So, how exactly is it a conditional analysis? Perhaps the
idea is that the truth or falsity of the right-hand side depends upon a bunch of
countcrfacrual facts, facts like x would M if C1 were the case, x wouldn't Iv! if (0_ were
the case, x would Iv! if C:1 were the case, and so on. In any case, Manley and W<L,se1man
take it to be a conditional analysis, and we'll follow suit.
l2 Nick Kroll
There is no shorcage of (nomologically) possible cases where the vase is
struck and its disposition is masked by something or other. There is also no
shortage of (nomologically) possible cases where the vase is struck but some
fink makes it lose the disposition to break when struck. The right-hand side
of (2) takes these cases into consideration. ff Manley and Wasserman arc
correct, such cases arc outweighed by those cases where the vase is struck and
breaks: there is a "suitable" proportion of cases where the vase is struck and
breaks. So, Manley and Wasserman's strategy for avoiding the problem of
finks and the problem of masks is not to add at some qualification to SCA so
that such situations are properly ignored. Rather, the strategy is to take such
situations into consideration but maintain that there arc enough non-finkish
and non-masking situations where the relevant disposition manifests. Simi-
lar remarks apply to problems with Achilles' heels.
I'm not convinced that this is a successfol stratet,'Y. 17 It's clear enough
what a counterexample going from left to right would look like. We would
need x to be disposed to M when C yet across the relevant region of modal
;pace C-cases are, by and large, cases where x' s disposition to NI is masked or
finked <md so x doesn't M.
My computer's CPU provides such a counterexample. The CPU is
disposed to overheat when running a large number of (tasking) processes.
That's why the computer has a hcatsink and fans. When the CPU is running
a large number of processes, the hcatsink and fans mask the CPU's dispos-
ition to overheat when running a large number of processes. Now creatures
smart enough to design such a CPU arc also smart enough to realize that its
disposition to overheat when running a large number of processes needs to
be masked/finked when the CPU is running a large number of processes.
And this isn't an accident. It holds across the relevant region of modal space
that cases where the CPU is engineered arc, by and large, cases where the
engineers realize that its disposition to overheat when running a large
number of processes needs to masked/finked in some way. So, the relevant
region of modal space is such that cases where the CPU is running a large
number of tasking processes arc predominantly cases where the CPU's
disposition to overheat when running a large number of processes is masked
or finked. So, on any reasonable understanding of"suitablc proportion," it's
not the case the CPU would overheat in a suitable proportion of cases where
it is running a large number of tasking proccsscs. 18 Yet the CPU is disposed
17
I should also say that I'm not convinced by '111 argument Manley and \'V'assermann
make against competing conditional analyses. Since this argument also applies to my
teleological account of dispositions, I leave discussion of it for the Appendix.
18 One might worry that there are at least continuum-many cases where the CPU's
disposition is neither masked nor finked, the CPU is running a large numher of processes,
Te!co!ogiml Dispositions 13
to overheat when running a large number of tasking processes. That's why
che heatsink and fans arc there. We have a counterexample to (PROP). 19
Herc is similar but more fanciful counterexample. Suppose advanced
creatures have engineered an artifact that is disposed to shatter when struck.
While the blueprints for this artifact were being drawn, the creatures realized
rhat the artifact will be disposed to shatter when struck. So, not only did
they engineer a mask, they also made sure that the mask would be applied
upon creation of the artifact. One more bit of fantasy. This artifact is so
sophisticated that it can only be engineered by creatures smart enough to
re<1lize that the artifact will be disposed to shaner when struck and also
realize that some kind of mask or fink will need to be engineered and applied
upon creation. Well, maybe lesser creatures could somehow "accidentally"
engineer the artifact but the probability is minisculc. The important point is
that across modal ;pace there arc hardly any cases where the artifact is struck
and shatters. Granted there arc cases where the mask/fink is defective, cases
where the mask/fink is not applied, and cases where lesser creatures create
the artifact. But such cases arc hardly worth noticing. What is worth
noticing is that on any reasonable understanding of "suitable proportion,"
it's false that the artifact would shatter in a suitable proportion of cases
where it is struck. Yct the artifact is disposed to shatter when struck. That's
why the mask was applied upon creation.
Some may not be convinced by these alleged counterexamples. Regarding
the alleged counterexample involving my computer's CPU, one might
object that because of the heatsink and fans, the CPU is not disposed to
overheat when running a large number of processes. This objection, how-
ever, seems (to me anyway) to confuse my CPU with my computer. The
latter is not disposed to overheat when the CPU is running a large number of
processes. The former is so disposed. That's why the hcatsink and fans arc
there. One might also wony about my claim that, across the relevant region
of modal space, cases where the CPU is engineered arc, by and large, cases
where the engineers realize that its disposition to overhear when running a
large number of processes needs to be masked/finked in some way. Maybe
and so its disposition manifests. We are, after all, taking modal space into consideration.
So, let x be the set of such cases. Surely the set of cases where the CPU's disposition is
masked or finked but the CPU is running a large number of processes has the same
cardinality as x. So, how can it be that there arc more of the latter cases than the former
cases? Good question. It is, however, a question for Manley and Wasserman to answer.
They .orfo.r some suggestions. Whatever merric they use to justify rhe appeal to propor-
tionality, 1t had better come out that my CPU wouldn't overheat in a suitable proportion
of cases where it is running a large number of (t<L,king) processes.
19
This scenario is also a counterexample to the variant of (PROP) found in
Wasserman (2011).
14 Nick Kroll
engineers (across modal space) don't have to be that smart to engineer the
CPU. I don't think so, but suppose I'm wrong. What about the counter-
example involving the super-sophisticated artifact? I find it hard to believe
that such an artifact is impossible. 20
2.5 A different direction
The problems raised fCJr the above conditional analyses do not show that
there is no satisfactory conditional analysis of dispositions. But they do
motivate a move in a different direction. That is, they at least provide
some reason for considering an account of dispositions that docs not rest
on a conditional analysis of dispositions. So, I propose that we look in a
different direction. In particular, I propose that we look towards a teleo-
logical account of dispositions.
3. A TELEOLOGICAL ACCOUNT OF DISPOSITIONS
3.1 Preliminary remarks
The move towards a teleological account of dispositions is not motivated
solely by the shortcomings of the above conditional analyses. It's also
motivated by the intuition that dispositions are, in some sense, directed at
their manifestations.
Some may claim to not have this intuition. To them, I point to Good-
man's famous characterization of dispositions in terms of "threats and
promises." 21 Goodman's metaphor captures an important intuition about
dispositions. And it seems pretty clear that this intuition is the directedness
intuition. So, if you are not sure whether you have the directedness intu-
ition, check whether you get the metaphor. If you get it, you have the
intuition.
Some may doubt that the directedness intuition should be given much
weight when evaluating an account of dispositions. Perhaps the stronger
intuition is that there is an important connection between dispositions and
conditionals, and if we could only get this connection straight, we would
20
Manley and Wasserman (2008) suggest that (PROP) may have to be revised so that
some C-cascs arc weighed more heavily than others. So, perhaps this is a way to avoid the
above counterexamples. But then, as Cross (2013) notes, it's hard to see how (PROP) is
an improvement over appeals to normal conditions, ideal conditions, or ceteris paribus
conditions.
21
Goodman (1954, p. 40).
Te!eofogic(d Dispositions 15
have a conditional analysis of dispositions that explains, in some deflationary
way, intuitions about the directedness of dispositions.
As a first step in developing a teleological account of dispositions, [ turn
this concern around and into an argument for appealing to teleology in
giving an account of dispositions. In particular, l argue that by appealing to
teleology we can actually provide a counterexample free and informative
connection between dispositions and conditionals.
3.2 The first step
Consider the following passage from Lewis (l 997).
Sometimes it takes some time for a disposition to do its work. When stimulus S
arrives and the disposition is present, some process begins ... When the process
reaches completion, then that is, or that causes, response r. But if the disposition
went away part-way through, the process would be abortcd. 22
To foreshadow what is to come, I claim that the processes Lewis describes
are teleological processes. But before saying anymore about this, we need to
slightly amend what Lewis says.
Masking cases show that even if the stimulus condition arrives and the
disposition remains, there need not be some process that begins and ends
with a manifestation of the disposition. So, taking masks into consideration,
suppose Lewis had said something slightly different. Something like this:
Sometimes it takes some time for a disposition to do its work. When the disposition
is activated, some process begins. When the process reaches completion, then that is,
or that causes, response r. But if the disposition went away part-way through, the
process would be aborted. Nonetheless, if nothing were to interfere with this process,
there would be response r.
We would then have the following activation principle in place:
(AV) If x's disposition to M when C is activated, then either x imme-
diately Ms or there is some process such that if the process were to
continue without interruption, x would M.
I claim that (AV) is counterexample free. Before offering support for this
claim, I need to clarify a distinction (AV) relies upon. This is the distinction
between a disposition being activated and the stimulus condition for the
disposition obtaining.
Certain masking cases illustrate the distinction. Distinguish between two
types of masking cases: those in which a disposition is manifesting but docs
22
Lewis (1997, p. 146).
16 Nick Kroll
not manifest because of a mask, <md those in which a mask prevents even a
partial manifestation of the disposition even though the stimulus condition
obtains. Call the former "Type-1" masking cases and the latter "Type-2"
masking cases. The case of the poison that is killing you but doesn't kill you
because of the antidote is an example of a Type-1 masking case. For an
example of a 'T'ype-2 masking case, consider a vase disposed to break when
dropped but wrapped in bubble-wrap. It's dropped and doesn't break. It
seems plausible that the bubble-wrap not only prevented a manifestation of
the disposition but also a partial manifestation of the disposition. Further-
more, it seems plausible that the bubble-wrap prevented even a partial
manifestation of the disposition because it prevented the disposition from
being activated in the first place. The disposition remained dormant even
though the stimulus condition obtained. Similar remarks apply to my
CPU's disposition to overheat when running a large number of processes.
The fans and heatsink prevent the disposition from being activated when
the CPU is running a large number of processes. Generalizing, ·rype-2
masking cases are cases where the stimulus condition obtains, the dispos-
ition remains, but the disposition is not activated.
Cases involving finks also illustrate the distinction. When does the
electro-fink running on reverse cycle make the wire lose its disposition to
conduct electricity when touched by a conductor? Well, when the wire is
touched by a conductor. But docs the fink do its work instantaneously? 23 If
so, the instant the wire is touched by a conductor, it is not disposed to
conduct electricity. If not, there is an instant or interval of time where the
wire is touched by a conductor and disposed to conduct electricity when
touched by a conductor. In the first case, the disposition is not activated
because it's no longer present. In the second case, it's not clear whether the
disposition is momentarily activated before the fink docs its work. If it isn't,
then we have a case where the stimulus condition obtains but the disposition
isn't activated. 21 Generalizing, it seems that (reverse) finks can work in one
of three ways. They can make it so that the disposition goes away the instant
the stimulus condition obtains, and so the disposition is not activated
because it's no longer there. They can make it so that the disposition goes
away when the stimulus condition obtains but before the disposition is
activated. And they can make it so that the disposition goes way after the
disposition is activated but before the disposition manifests. Instances of the
first two ways bring out the distinction between a disposition being acti-
vated and the stimulus condition of the disposition obtaining.
23
Lewis (1997) calls this "a dilemma about timing."
24
My intuitions, though not entirely clear, are that the disposition isn't activated.
J 'eleo!ogica! Dhpositions 17
It should be clear, then, that there is a distinction between a disposition
being activated and the stimulus condition of the disposition obtaining. Of
course, there is a connection between the two. For a disposition to be
activated, its stimulus condition must obtain. Likewise, though perhaps
this is obvious, the activation of the disposition requires that the disposition
is present.
Let's return to (AV). I claim that (AV) is counterexample free. Type-I
masking cases pose no threat. In such cases, the disposition is activated but
doesn't manifest because of the mask. Nonetheless, when the disposition
was activated, there was thereby a process such that were that process to
continue without interruption, there would be a manifestation of the
disposition. Type-2 masking cases pose no threat. In such cases, the dispos-
ition isn't activated. F;inks pose no threat. Either the fink makes it so that the
disposition isn't activated when the stimulus condition obtains or it makes it
so that the disposition goes away after it has been activated but before it
manifests. In the first case, there is obviously no threat to (AV). In the
second case, there is no threat to (AV) because if it weren't for the fink the
disposition would have manifested. And this is so because when the dispos-
ition was activated, there was thereby a process such that if the process were
to continue without interruption, there would be a manifestation of the
disposition. So, (AV) has no problems with the problem of finks or the
problem of masks.
What about Achilles' heels? My intuitions suggest that the concrete block
with the Achilles' heel has two dispositions. Because it is sturdy, it is
disposed to remain intact when struck. Because of its Achilles' heel, there
is a particular spots and a particular amount of force f such that the block is
disposed to shatter when struck with force J in spots. Suppose the block is
• struck with force /in spot s and its disposition to shatter when struck with
force fin spot s manifests. My intuitions suggest that even though the
block was struck, its disposition to remain intact when struck wasn't
activated. What was activated was its Achilles' heel.
But some may have the intuition that if the block is disposed to shatter
when struck with force Jin spot s, then it can't be disposed to remain
intact when struck. To them, I say consider Achilles. The greatest warrior of
the Trojan War was not disposed to fall when struck. He was disposed to
withstand harm when struck. However, because of his Achilles' heel, he was
also disposed to fall when struck in the just the right spot. When Achilles
was struck in just the right spot, his disposition to fall when struck in just the
right spot was activated and subsequently manifested. Was his disposition to
withstand harm when struck activated? My intuitions suggest that it wasn't.
IfI' m right, there is no threat to (AV). IfI' m wtong, there is a threat to (AV)
only if when Achilles was struck in just the right spot, there wasn't thereby a
18 Nick Kroll
process such that if the process were to continue without interruption,
Achilles' disposition to withstand harm when struck would manifest. My
intuitions aren't entirely clear because I don't have the intuition that
Achilles' disposition to withstand harm when struck was activated in the
first place. But it does seem to me that i/this disposition was activated, there
was such a process. And what interfered with it was t:he activation of
Achilles' Achilles' heel. So, in either case, there is no threat to (AV).
Generalizing, we can think of cases involving Achilles' heels as special
cases of masking. If the case is a Type-! masking case, then the relevant
disposition is manifesting but doesn't manifest because of the activation of an
Achilles' heel. If the case is a Type-2 masking case, then the activation of an
Achilles' heel prevents the relevant disposition from being activated in the
first place. In either case, there is no threat to (AV).
So, since I can think of no other potential threat to (AV), I conjecture that
(AV) is counterexample free. 25 However, there is still an issue that needs to
be resolved.
The issue is that (AV) is an "escape-clause" account of the relationship
between dispositions, their activation, and their manifestation. To give a few
simple examples:
• If Dis activated, then, ceteris paribus, it manifests.
• In normal/ide;tl/typical circumstances, if D activated, it manifests.
• If Dis activated, then, provided nothing interferes, it manifests.
Moline's remark about such proposals bears repeating.
[T]hey are fund;uncnt<tlly dodges ... They amount to muddled ways of disguising
from ourselves more or less serious ignorance of the dispositional properties of
individual things or persons and of types of things or persons. (Moline 1975,
pp. 244-5)
As I said above, I agree, ceteris paribus. With respect to (AV), this means that
if we don't have some reason for thinking that it is legitimate to appeal to the
notion of a process continuing without interruption, (AV) is simply a way to
disguise more or less serious ignorance of the relationship between disposi-
tions, their activation, and their manifestation.
(AV), however, can be legitimized by appealing to teleology. Consider the
following passage from Makin (2006):
25
Jenkins and Nolan (2011) argue that it is possible for there to be dispositions with
impossible manifestations. However, their alleged examples pose no threat to (AV). They
offer no example of disposition such that it's possible for the disposition to be activated
but impossible for there to be even a partial manifestation of the disposition.
Teleological Dispositions 19
[I]f it is appropriate to view a [process] teleologically, it is therefore also appropriate
to apply other notions: interference, interruption, hindrance, and a normal outcome.
It makes sense to talk of a tcleologirnl process being interrupted ... That is because a
teleological process has a privileged stage to which it runs in norm<tl conditions,
26
unless interfered with or hindered: the [end] to which it is dirccted.
Makin is offering an explanation of why Aristotle often appeals to teleo-
logical notions in discussing the nature of his distinction between potenti-
ality and actuality and, in particular, why actuality is prior to potentiality.
Putting Aristotle aside, it shouldn't be too difficult to sec how Makin's
insight is relevant to (AV).
Makin's insight is that if a process is a teleological process, then it is
legitimate to appeal to the notion of the process continuing without inter-
ruption. It's legitimate because of the following principle governing teleo-
logical processes.
(TP) If a process p is directed at end E, then: in virtue ofp being directed
at end E, if p were to continue without interruption, E would be
the case.
(AV) follows from the conjunction of (TP) and (TI).
(Tl) If x's disposition to M when C is activated, then either (a) x
immediately Ms, or (b) there is some process directed at the end
that x Ms.
Given (TI), we thus have good reason for thinking (AV) isn't fundamentally
a dodge. On the contrary, it's a consequence of (Tl), which is a substantive
thesis about the relationship between dispositions, their activation, and their
manifestation.
So I offer (Tl) to those looking for an interesting and counterexample
free connection between dispositions and conditionals. If you accept my
offer, you'll get (CDC).
(CDC) Necessarily: if xis disposed to M when C, x's disposition to M
when C is activated, and x doesn't immediately M, then there
is some process such that:
(*) ifthe process were to continue without interruption, x would M.
(Tl) may seem like a steep price to pay for an interesting and counterexam-
ple free connection between dispositions and conditionals. However, the
point remains that by appealing to teleology, we can actually provide an
interesting and counterexample free connection between dispositions and
26
Makin (2006), p. 194.
20 Nick Kroll
conditionals. Thus, those who think that the directedness of dispositions
can be explained, in some deflationary way, by some interesting connection
between dispositions and conditionals have a worry to contend with if we
can offer an account of dispositions that explains the directedness of dis-
positions and, in doing so, explains why (Tl) holds. The wony is that the
directedness of dispositions is what explains why there is some interesting
connection between dispositions and conditionals.
The stage is now set for a teleological account of dispositions.
3.3 The proposal
Following Molnar (2003), I claim that directedness is what sets dispositions
apart from non-dispositional properties. 'f'here seems to be no sense in
which a triangular object is, in virtue of being triangular, in a state directed
at the occurrence of some event. 27 On the other hand, there seems to be
some sense in which a vase disposed to break when struck is, in virtue of
being so disposed, in a state directed at the occurrence of an event in which
the vase breaks. Such intuitions provide some initial justification for the
claim that directedness is what sets dispositions apart from non-dispositional
properties. But, of course, more needs be said about exactly what type of
directedness is alleged to distinguish dispositions from non-dispositional
properties.
I say that teleological directedness is what sets dispositions apart from non-
dispositional properties. (T2) spells out the details.
(T2) Necessarily: a property Pis a disposition iff there is a condition C
and event-type M such that necessarily, Pis the property of being
in a state directed at the end that one Ms when C.
(T2. l) follows from (T2).
(T2. l) Necessarily: a property P is a disposition iff there is a condition
C and event-type M such that: necessarily, x has P iff x is in a
state directed at the end that x Ms when C.
So, what makes a disposition a disposition is that the property just is the
property of being in a state directed at a certain teleological end. Conse-
quently, to have a disposition just is to be in a state directed at a certain
teleological end.
27
lnspirc<l by Mellor (1974), some might claim that a triangular object is, in virtue of
being triangular, in a stare directed at the occurrence of an event in which its sides are
counted and the result is three. I don't have this intuition.
Tefeofogim! Di;positions 21
(T2) does not specify the relation between the property of being disposed
to M when C and the property of being in a state directed at the end that
one Ms when C. Sure enough, the property of being disposed to M when C
is a disposition. But all that follows from (T2) is that there is some C* and
M *such that the property of being disposed to M when C is the property of
being in a state directed at the encl that one M *s when C *.
I propose that the identity relation holds between the property of being
disposed to M when C and the property of being in a state directed at the
end that one Ms when C.
(T3) Necessarily: the property of being disposed to M when Cjust is the
property being in a state directed at the end that one !vfs when C.
It follows from (T3) that necessarily, xis disposed to M when C just in case
xis in a state directed at the end that.>: Ms when C.
With (T3) in hand, we turn to the question: what happens when C
obtains and x is disposed to M when C?
My answer is that what happens depends on whether x's disposition is
activated when C obtains. Ifx's disposition isn't activated, then nothing of
interest happens with respect to x being disposed to M when C. That is, the
disposition remains dormant. However, ifx's disposition is activated, some-
thing of interest does happen. In particular, in virtue ofx being disposed to
M when C, either x immediately Ms or there is a process directed at the end
that x Ms. Given (T3), we thus get (T4).
Cf4) Ifx' s disposition to M when C is activated, then in virtue ofx being
in a state directed at the end that x Ms when C, either x immediately
Ms or there is a process directed at the encl that x Ms.
So, the teleological directeclness of a disposition does some work when the
disposition is activated. It also explains why (Tl) holds.
The conjunction of (T2), (T3), and (T4) constitute my teleological
account of dispositions: (TAD), for short.
A concrete example may help clarify (TAD). Consider the property of
being disposed to dissolve when in water. (T3) tells us that this property just
is the property of being in a state directed at the encl that one dissolves when
in water. (T2), then, tells us the property of being disposed to dissolve when
in water is a disposition. (Nothing new.) But it also tells us that what makes
this property a disposition is that it is the property of being in a state directed
at the end that one dissolves when in water. Now consider a chuck of salt
disposed to dissolve when in water. Suppose the salt is placed in water and
it's disposition is activated. (T4) tells that in virtue of the salt being in a state
directed at the end that it dissolves when in water, either the salt immedi-
ately dissolves or there is a process directed at the encl that the salt dissolves.
22 Nick Kroll
Supposing the salt does not instantaneously dissolve, there is then a process
directed at the end that the salt dissolves. [f this te!os of the process were to
obtain, there would be a manifestation of the salt's disposition. If this te!os
doesn't obtain, there is no manifestation in this particular circumstance (but
if the process had continued its normal course without interruption, the telos
would have obtained).
The argument for (TAD) is that it explains the directedness of disposi-
tions, and in doing so, provides an interesting and counterexample free
connection between dispositions and conditionals. This wouldn't be much
of an argument ifa competing account of dispositions provides both a better
explanation of the directedness of dispositions and an interesting and
counterexample free connection between dispositions and conditionals.
So, we'll have to sec what the competition has to offer.
Before comparing (TAD) to its rivals, though, we should address a
concern that I'm sure has been gnawing away at some.
4. A DETOUR
4.1 Seriously?
The concern can be put like this: "You can't be serious." We can categorize
those with such a concern into three groups.
First, there are those who, setting aside the analogy, would dogmatically
agree with Francis Bacon's remark that "inquiry into fin<ll causes is sterile,
and, like a virgin consecrated to God, produces nothing." 28 Second, there
are those who think that inquiry into final causes has a place, provided that
place concerns the goals of agents or the purposes of the artifacts they design.
For them, a teleological account of dispositions is committed to projecting
mental states to properties or treating them as artifacts, and so is absurd.
Third, there are those who allow for so-called "natural teleology" but restrict
it to the function or proper function of features of organisms. For them, a
teleological account of dispositions is, at best, committed to treating dis-
positions as biological functions, and so is hardly worth considering.
The dogmatists can be ignored since nothing e<m be said to make them
change their minds. The simple response to the other two groups is that
while teleology is often tied up in talk of goals, purposes, design, function,
proper function, and sometimes talk of certain outcomes being better than
others, there is no reason to assume that teleology must be tied up in such
28
De Augmentis Scientiflrum, Bk. iii, Ch. 5, quoted in Woodfield (1976, p. 3) ;md
Hawthorne (2006, p. 268).
Te!eologicrd Dispositions 23
talk. So, don't make this assumption when considering (TAD). There is a
general notion of teleological directedness that outstrips talk of goals,
purposes, design, and function. It is this generaJ notion at play in (TAD).
Some might not be convinced by the simple response. 'They might want
some reason, independent of my argument for (TAD), to take seriously a
general notion of teleological directedness that outstrips talk of goals,
purposes, design, and function.
To this end, we take a detour from dispositions and turn to sentences in
the progressive aspect.
4.2 The progressive aspect and events in progress
Here are some examples of sentences in the progressive aspect.
(3a) Steve is driving to Boston.
(3b) A chunk of salt is dissolving.
(3c) The universe is expanding.
(3a) says that there is an event of Steve driving to Boston in progress; (3b)
says that there is an event of some salt dissolving in progress; and (3c) says
that there is an event of the universe expanding in progress. So it is in
general: a sentence in the progressive aspect says that there is an event of
some type in progress. Indeed, this is the core semantic intuition about the
progressive aspect.
The orthodox approach to capturing the core intuition is to offer a modal
analysis of the progressive. 29 The appeal to modality usually starts with the
observation that an event in progress need not culminate. 30 Steve could be
driving to Boston but be forced to turn around due to car troubles. A chunk
of salt could be dissolving but be taken out of water before it (fully)
dissolves. In such cases, there is an event in progress that doesn't culminate.
But what if Steve's drive hadn't been interrupted by car troubles? Presum-
ably, he would have driven to Boston. And what if the salt hadn't been taken
out of the water? Presumably, it would have (fully) dissolved. Generalizing,
the basic idea is that while an event in progress need not culminate, it
nonetheless would culminate if it were to continue without interruption. As
29
See Dowty (1979), Asher (1992), Landman (1992), Bonomi (1997), Portner
(1998), Hallman (2009), and Higginbotham (2009).
50
This observation is one way of putting what is called the "imperfective paradox" in
the literature on the progressive. Another way of putting the imperfective paradox is that a
past progressive docs not, in general, entail its perfective correlate. For instance, "Steve
was driving to Boston" does not entail "Steve drove to Boston."
24 Nick Kroll
far as I can tell, this "no-interruption" intuition is what spurs modal analyses
of the progressive.
Bypassing the details of a semantics for the progressive that captures the
no-interruption intuition, let's just focus on the account of events in
progress suggested by the no-interruption intuition. It's helpful here to
appeal to resultant states. So, let me say a word about resultant states, and
then offer the account of events in progress suggested by the no-interruption
intuition.
A resultant state of an event is a state of the event having occurred or
taken place. For example, suppose Mirah drew a circle. As a result of this
event taking place, a state of Mirah having drawn a circle obtains. This state
of Mirah having drawn a circle is a resultant state of the event.
If this example doesn't help, here is a heuristic that may. Take a sentence
in the simple past which describes an event and form its present perfect
correlate. Then, think of the present perfect sentence as describing the
(relevant) resultant state of the event described by the simple past sentence.
For example, in the sentences that follow think of each b-sentence
as describing the (relevant) resultant state of the event described by the
a-sentence.
(4a) Steve drove to Boston.
(4b) Steve has driven to Boston.
(Sa) A chunk of salt (fully) dissolved.
(Sb) A chunk of salt has (fully) dissolved.
(6a) The universe expanded.
(6b) The universe has expanded.
To put it yet another way. think of (4b) as saying that a state of Steve having
driven to Boston now holds, think of (Sb) as saying that a state of the chunk
of salt having (fully) dissolved now holds, and think of (6b) as saying that a
state of the universe having expanded now holds. Such states are resultant
states of particular events. 31
With resultant states suitably clarified, we can now offer the account of
events in progress suggested by the no-interruption intuition.
(NI) Necessarily, e is a cp event in progress at tiff e would bring about a
resultant of a cp event if it were to continue past t without
interruption.
31
There are analyses of the perfect that involve quantification over resultant states, and
so take the above paraphrases to be semantically significant. See Parsons ( 1990), Kratzer
(2000), and Higginbotbm (2009).
Teleologiml DiJpositions 2S
For instance, an event is an event of the universe expanding in progress (at t)
just in case the event would bring about a state of the universe having
expanded if it were to continue (past t) without interruption.
There are, however, two problems with (NI). The first is that the right-to-
left direction is subject to counterexample. The second is that the lcft-to-
right direction is subject to a worry about informativeness.
Our counterexample is a variant of a scenario due to Landman (l 992).
Suppose Mary is delusional and thinks she needs to swim to the other
side of the Atlantic to save her soul. Mary is not only delusional but also a
very bad swimmer. So, she enters the Atlantic around Boston, swims for
an hour, and then drowns. Mary was trying to swim to the other side of
the Atlantic but she wasn't actually swimming to the other side of the
Atlantic. Indeed, this case is a nice example of the difference between
trying to do something and actually doing what you arc arc trying to do.
But what if Maty's swim had continued without interruption. Well, she
wouldn't have drowned. Likewise, she wouldn't have been eaten by a
shark or saved by a fishing boat. In short, the swim would have continued
on until she miraculously reached some place on the other side. So, we
have an event that is not an event in progress of Mary swimming to the
other side of the Atlantic but one that would bring about a state of Mary
having swum to the other side of the Atlantic if it were to continue
without interruption. Thus, we have a counterexample to the right-to-left
direction of (NI).
The worry about informativeness is that unless more is said about what
counts as an interruption of a cp event in progress, we have no reason not to
think of an interruption of a cp event in progress as simply something that
prevents it from bringing about the resultant state of a cp event. We should
just as well treat an interruption of an event in progress of Steve driving to
Boston as simply something that prevents the event from bringing about a
state of Steve having driven to Boston. So, the claim that an event in
progress of Steve driving to Boston would bring about a state of Steve
having driven to Boston ifit were to continue without interruption amounts
to the uninformative claim that this event in progress would bring about
such a state unless something prevented it from doing so. In short, unless
something more is said about what counts as an interruption of a cp event in
progress, the left-to-right direction of (NI) amounts to the uninformative
claim that a cp event in progress would bring about a resultant state of a cp
event unless something prevents it from doing so.
Perhaps there is a way to modify (NI) so that it avoids the threat of
counterexample and the worry about informativeness. Suffice it to say that
as in the case of conditional analyses of dispositions, there is no modal
analysis of the progressive that is widely recognized to be informative and
26 Nick Kroll
counterexample free.: 12 So, someone who is sympathetic to a modal account
of events in progress has her work cut out for her.
There is, however, an account of events in progress that not only explains
the no-interruption intuition but also explains another intuition about
events in progress. This is the intuition that an event in progress is in some
sense directed at bringing about a certain state of affairs. Surely, an event in
progress of Steve driving to Boston is in some sense directed at bringing about
a state of affairs in which Steve has driven to Boston. Surely, an event in
progress of a chunk of salt dissolving is in some sense directed at bringing
about a state of affairs in which the salt has (fully) dissolved. Surely, an event
in progress of the universe expanding is in some sense directed at bringing
about a state of affairs in which the universe has expanded. So, it is in
general: events in progress have some kind of directedness.
I claim that the directedness of events in progress is a teleological
directedness: for an event in progress to be directed at a certain state of
affairs is for the relevant state of affairs to be a te!os of the event in progress.
More specifically, I propose the following teleological account of events in
progress.
(EIP) Necessarily, e is a <p event in progress at t iff e is, at t, directed at
the end that it cause the resultant state of a <p event to obtain at
some t'>t.
So, for example, an event is an event in progress of Steve driving to Boston
(at t) iff it is an event that is (at t) directed at the end that it cause a state of
Steve having driven to Boston to obtain (at some later t'). 33
(EIP) explains the directedness intuition about events in progress. It also
explains the no-interruption intuition. Recall Makin's insight.
32
See Szah6 (2004), Szab6 (2008), and Kroll (2015) for arguments against modal
an~lJ5es of the progressive. . . . primes. So, it's not possible for there to be a resultant state of an event in which Frank
· · (EIP) should not be understood as a complicated way of saymg that an event m
progress is directed at the end that it culminate. To see why, let e be a sufficiently
extended event in the progress of the universe expanding. For each expansion of the
universe during this time, there is a corresponding resultant scare of the universe having
expanded that is brought about bye. Bute doesn't culminate each time it brings about a
state of the universe having expanded. Indeed, as an event in progress of universe
expanding, e is not associated with any kind of culmination. What (EIP) basically tells
us is that even if e brings about a state of the universe having expanded at some moment t,
it is still, at t, directed at bringing about a later state of the universe having expanded. So,
while e is not directed at any kind of culmination, it is still, at each moment, directed at
bringing about further expansion of the universe. In short, some events in progress arc
(teleologically) directed at culminating, others aren't, but all are (teleologically) directed at
bringing about a kiter resultant state.
Teleological Dispositions 27
(TP) If a process pis directed at end E, then: in vhtue ofp being directed
at end E, if p were to continue without interruption, E would be
the case.
(EIP) treats events in progress as teleological processes. Thus, (E 1) follows
from (EII').
(El) If e is a <p event in progress at t, then e would bring about a
resultant state of a (p event if it were to continue past t without
interruption. 31
So, (EIP) explains the intuition that spurs modal accounts of events in
progress. Furthermore, the informativeness wony is no longer a worry. (El)
is a consequence of (EIP), which is a substantive thesis about events in
progress.
Here, then, is an argument, independent of my argument for (TAD), to
take seriously a general notion of teleological directedness that outstrips talk
of goals, purposes, design, and function. (EII') is an account of events in
progress that explains the directedness intuition and explains the intuition
that motivates its rivals (i.e. modal accounts). So, we have good reason to
take (EIP) seriously. The notion of teleological directedness in (EIP) is a
general notion that outstrips talk of goals, purposes, design, and function.
For example, bringing about a state of the universe having expanded is not a
goal, purpose, or function of an event in the progress of the universe
expanding, and I see no reason to think such an event in progress was
designed to bring about such a state.
54
(El) is simply the left-to-right direction of (NI). So, I am committed to this
direction of (NI) being counterexample free. The only alleged counterexample I aware
of is due to Szab6 (2008). Szab6 asks us to consider a young boy, Frank, who starts
enumerating prime numbers in sequence: two, three, five, seven, eleven, thirteen, and so
on. According to Szab6, uttering "Frank is enumerating the primes" is an accurate
description of what Frank is doing. But it's not possible for Frank to enumerate the
enumerates the primes. Thus, if "Frank is enumerating the primes" is an accurate
description of what Frank is doing, we would appear to have a counterexan1ple to (El).
I'm not convinced. Suppose Frank's mother utters "Frank is enumerating the primes"
w describe what Frank is doing and you overhear the utterance. Suppose you ask: "All of
the primes or some of the primes?" If the mother were to respond with "All of the primes,"
she would be saying something false. On the other hand, if the mother were to respond
with "Some of the primes," she would by saying something true. It seems, then, that
taking the mother to be saying something true when she utters "Frank is enumerating the
primes" rests on taking her to be communicating the proposition that Frank was
enumerating some of the primes. If this is right (and I think it is), then the scenario is
not a counterexample to (El). To be such a scenario Frank would have to be enumerating
all of the primes, which he obviously is not doing. Rather, the scenario is one where Frank
is enumerating some of the primes, and he obviously can enumerate some of the primes.
28 Nick Kroll
So, since we have reason to take (EIP) seriously, we have reason to take
seriously a general notion of teleological dircctcdness that outstrips talk of
goals, purposes, design, and function.
I have not provided a detailed defense of (EIP) over its rivals. 35 But the
purpose of our detour isn't to provide a detailed defense of (EIP). It's to
provide some independent reason to take seriously a general notion of
teleological directedncss. Let us, then, return to dispositions by comparing
(TAD) to its rivals.
4.3 Against the competition
The competition falls into two camps. The first camp consists of accounts of
dispositions that rest upon a conditional analysis of dispositions. The second
camp consists of accounts of dispositions that appeal to a different type of
directedness. 36
We have yet to see a satisfactory (i.e. counterexample free and inform-
ative) conditional analysis of dispositions. This is a problem for the first
camp. It also makes comparing (TAD) to the competition from the first
camp rather difficult. Furthermore, there doesn't seem to be much motiv-
ation for continuing the search for a satisfactory conditional analysis if we
can offer an account of dispositions that explains both the directedness of
dispositions and provides a connection between dispositions and condition-
als. (TAD) is such an account of dispositions.
Those sympathetic to conditional analyses might respond by claiming
that we want a deflationary account of dispositions, and we get what we
want only if there is a counterexample free and informative conditional
analysis of dispositions. They might remind me of something else Goodman
said-namely, that it would be ideal to explain what dispositions are
without "any reference to occult powcrs." 37 I have two things to say in
response.
35
Sec Kroll (2015) for such a defonse of (EIP).
36
There is a possible third camp. Following Fara (2005), one might reject conditional
analyses but offer a "habitual" account of dispositions in which being disposed to M when
C just is having an intrinsic property in virtue of which one Ms when C. I do not consider
this possible third camp to provide t1ctut1! competition to (TAD) because the counterexam-
ples I offored to (PROP) serve as counterexamples to a habitual account of dispositions: my
CPU docs not have an intrinsic property in virtue of which it overheats when running a
large number of processes, yet it is disposed to overheat when running a large number of
processes (that's why the he'dtsink and fans arc there). If you are skeptical of my counter-
examples to (PROP), see Wasserman (2011) for reasons to be skeptical of the prospects ofa
habitual account of dispositions.
37
Goodman ( l 954, p. 40).
Teleological Dispositions 29
First, a conditiomtl analysis of dispositions does not tell us what disposi-
38
tions are. All it provides is a schematic biconditonal whose instances arc
alleged to be truths. Recall (SCA).
(SCA) Necessarily: xis disposed to M when C iffx would M if C were
the case.
As Manley (2012) notes, you could accept (SCA) and hold a reductive
functionalist account of dispositions in which the property of being disposed
to M when Cjust is the second-order property of having a non-dispositional
property pin virtue of which one would M if C were the case. Or you could
hold a similar but non-reductive functionalist account of the dispositions.
Or you could accept (SCA) yet hold that the disposition to M when Cjust is
the property of being such that one would M if C were the case. Of course,
it would be a mistake to accept (SCA). But the point is that if you are
worried about occult powers, a conditional analysis of dispositions by itself
isn't enough to relieve your worries. You need an account of dispositions
that does not require occult powers.
Second, we don't want a deflationary account of dispositions ifa deflationary
account of dispositions is one that rejects the possibility of a dis1x)sition being
fondamental property (i.e. a property that "carves the world at its joints").
Dispositional essentialists argue that any account of dispositions that rejects the
possibility offundamental dispositions is badly mistaken. 39 According to such
theorists, our best science tells that at least some fundamental properties of the
actual world are dispositions. Just consider spin, charge, and mass. Our best
science tells us that these properties are fundamental properties. And surely
spin, charge, and mass are dispositions. Witness Ellis and Lierse (1994):
With few exceptions, the most fundamental properties that we know about arc all
dispositional ... Therefore, we must either suppose that these basic properties arc not
truly fundamental ... or else we must concede that categorical realism is f:.Jsc. 40
We'll understand categorical realism is the view that no disposition is a
fundamental property (i.e. no property that carves the actual world by its
joints is a disposition). Ellis and Lierse' s challenge to the categorical realist is
straightfotward. Either show that charge, for instance, is not a disposition or
show that it's not a fundamental property. Good luck with either disjunct.
I will not take a stand on whether or not luck is on the categorical realist's side.
Maybe our best science will change. Or maybe there is some interpretation of
58
Manley (2012) stresses the import<ulCe ofrhis point, and, I should say, inspired this
first response.
19
· Sec Ellis (20Cll), Molnar (2003), and Bird (2007).
40
Ellis and Licrsc (1994, p. 32).
 Nick Kroll
our current best science in which charge is not a fundamental property or at least
not a disposition.
I will, however, take a stand on the possibility of there being something
like charge that is a disposition m1d a fondamental property. Let's stipulate
cl-charge is the property the fundamental particles of the actual world would
have if what the dispositional essentialist says about charge is correct. In
other words, if the dispositional essentialist is right about charge, then
d-charge is charge. Could there be a world w in which the fundamental
entities of w have cl-charge as a fundamental property? The dispositional
essentialist says the actual world is such a world. I say that such a world is
possible and so it's possible for there to be a disposition that is a fundamental
property.
Let's now put aside whether (TAD) undermines the motivation for
analyzing dispositions in terms of conditionals. We'll just suppose that we
have a counterexample free and informative conditional analysis. I argue
that even with such an analysis, there is reason to favor (TAD), and the
reason comes from the possibility of a basic or fundamental disposition.
For the sake of concreteness, suppose (PROP) is a counterexample free
and informative analysis of dispositions. Now suppose that x is disposed to
/VJ when C. Then, (7) follows from (PROP).
(7) x would /VJ in a suitable proportion of C-cases.
Further suppose that the property of being disposed to /VJ when C is a basic
or firndamental property. Presumedly, then, there is something about the
disposition itself that explains why (7) is the case. What could it be?
One option is that the property of being disposed to /VJ when C just is
the property of being such that one would /VJ in a suitable proportion of
C3 -cases. 41 Taking this approach, we would be committed to primitive
counterfactual facts, and so would be committed to denying the plausible
principle that what something would do depends on how it is. But maybe
that principle needs to be rejected once we consider the possibility of
fimdarnental dispositions.
Another option is to add (PROP) to (TAD) but deny that the property of
being disposed to /VJ when C just is the property of being such that one
would /VJ in a suitable proportion of C-cases. Taking this approach, we
would be committed to primitive teleological facts that explain certain
counterfactual facts. In other words, the position would be that (7) is the
case in virtue of (8) being the case.
(8) x is in a state directed at the encl that x /VJ s when C.
41
See Manley (2012).
Teleological Dispositions 31
So, by appealing to (TAD), we don't have to reject the principle that what
something would do depends on how it is. Thus, there is reason to favor the
appeal to (TAD).
Generalizing, I make the following objection to the first cmnp. We want
an account of dispositions that does not rule out the possibility of a
disposition being a firndamental property. The first camp promises to
provide a counterexample free and informative conditional analysis of
dispositions. Suppose they provide such an analysis. There is still reason
to favor (TAD) so long as (TAD) is compatible with the m1alysis. The reason
is that (TAD) provides an account of dispositions that allows us to retain the
principle that what something would do depends on how it is even in the
case where that thing has a fundamental disposition. 42
I conclude, then, that there isn't much of a challenge from the first camp.
First, they have yet to provide a satisfactory conditional analysis from which
they can issue a challenge. Second, even if they can provide such an analysis,
there is a challenge only if they can show that the analysis is incompatible
with (TAD) or show that it is impossible for a disposition to be a foncla-
mental property.
Let's turn to the second camp. The competition from the second camp
appeals to a different type of clirectedness in providing an account of disposi-
tions. Whereas I claim that the clirectedness of dispositions is teleological, the
second camp claims that the clirecteclness of dispositions has something to do
wid1 intentionality. For lack of a better name, we'll call members of the
second camp "intentionality-based" accounts of dispositions. 43
Intentionality-based accounts of dispositions are inspired by some parallels
between intentional mental states m1cl dispositions. Suppose Johnny believes
that Santa brings him presents, lives at the North Pole, has red cheeks, etc.
There is a certain sense in which these beliefs ofJohnny's are directed at Santa.
Since Santa doesn't exist, there is also a certain sense in which these beliefs are
directed at something that doesn't exist. And so it is for intentional mental
states in general: there is a certain sense in which they are directed at
'
12
It should be noted that accepting (TAD) does not imply accepting the possibility of
fundamental dispositions. Suppose we were to add ('T"5) to (TAD):
(T5) Necessarily: the property of being in a state directed at the encl that one Ms when
C just is the second-order property of having a non-dispositional property I in
virtue of which: if C were to obtain, either x would immediately M or there
would be a process directed at the end that xM s.
(TAD) would now entail that it is impossible for a disposition to be a foncLuncntal
property. Of course, (T3) would be subject to counterexample, and (T4) would be subject
to overdetermination worries. But that seems to go with the territory.
43
See Place (2005), Martin and Heil (1998), and Molnar (2003).
32 Nick Kroll
something, and they can be directed at something that doesn't exist. 44 But
dispositions have these two features as well: there is a certain sense in which a
disposition is directed at its manifestation, and, since a disposition need not
mm1ifest, there is a certain sense in which a disposition can be directed at
something that docs not exist. So, since some dispositions are not mental
properties, this suggests that intentionality might not be the m<lfk of the
mental. It also suggests that maybe we should try to offer an account of
dispositions in which the clirectedncss of dispositions is explained in terms of
the directedness of intentionality. In other words, perhaps we should offer an
intentionality-based account of dispositions.
There is an obvious objection to this line of thought. The sense in which
an intentional mental state is directed at something is that the mental state is
about or represents something. Johnny's belief that Santa brings him
presents is directed at Santa in the sense that his belief is about or represents
Santa. But dispositions are neither about nor represent their manifestations.
Thus, the directedncss of dispositions cannot be explained in terms of the
directedness of intentionaliry. 45
If it weren't for (TAD), l don't think this objection would be decisive.
A defender of an intentionality-based account of dispositions could respond
by saying that the numerous failed attempts to provide a satisfactory
conditional an<llysis of dispositions provide some motivation for looking
in a different direction. Perhaps the parallel between the directedness of
intentional mental states and the directedness of dispositions isn't perfect.
But the parallel doesn't have to be perfect for us to grasp a more general
notion of intentionality, one in which there is both mental intentionality
and physical intentionality.
(TAD) undermines this response. It provides an explanation of the
directedness of dispositions, and it is not motivated by appealing to any
parallel between intentional mental states and dispositions. As illustrated by
the above detour, part of the motivation for (TAD) is that there is some
parallel between events in progress and dispositions. But this seems to be
correct. Contrast the following:
(i) the sense in which an event in progress of Steve driving to Boston is
directed at its culmination.
(ii) the sense in which a vase's disposition to break when struck is directed
at its manifestation.
(iii) the sense in which Johnny's belief that Santa has red checks is directed
at Santa.
4
" This observation is famously due to Brentano ( 187 4).
15
' Bird (2007) argues that other alleged parallels between intentional mental states and
the directedncss of dispositions are weak at best.
Teleological Dispositions 33
It strikes me that (i) and (ii) are very similar but (iii) is very different. So, I think
it is safe to appeal to a parallel between events in progress and dispositions as
fXU't of the motivation for (TAD). In any case, ('TAD) offers an explanation of
the directedness of dispositions that does not require postulating a more
general notion of intentionality. Thus, the burden is on the second camp
to show that this explanation is inadequate but an explanation that rests on
(ii) and (iii) being similar in some way is adequate. I'm skepticaJ.
The second camp might respond by claiming that my teleological notion
of directedness falls under the general notion of intentionality. If this can be
shown, then I join the second camp.
4.4 Taking stock
Let's take stock. The first camp needs to provide a satisfactory conditional
analysis that is incompatible with (TAD) or they need to provide a satisfac-
tory conditional analysis and show that it is impossible for a disposition to be
fundamental property. The second camp needs to provide a convincing
argument that a teleological explanation of the directedness of dispositions is
inadequate but an explanation that appeals to similarities between inten-
tional mental states and dispositions is adequate. I've argued that neither
camp has provided what it needs to provide.
I conclude we have good reason to favor (TAD) over the competition.
5. CONCLUDING REMARKS
I've argued that whenever something has a disposition, something is in
a state with a telos. Along the way, I've also sketched an argument for
the claim that whenever there is an event in progress, there is an event
with a telos. Absent a strong argument that nothing has any disposition at any
time and that nothing is ever happening at any time, I take it that there is
thus good reason to think that Bacon was terribly mistaken about teleology.
Inquiry into final causes is far from sterile. On the contrary, it reveals the
. ..
nature o f cl 1spos1t1ons an cl events Ill
. progress. 0 r so I' ve argue cl .46
Nick Kroll
Franklin & Marshall College
46
If we conjoin (TAD) with a dispositional essentialist account of the laws of narure,
we also have an argument that teleology reveals something about the laws of nature.
According to a dispositional essentialist account of the laws of nature, the laws, or at least
some of the laws, arc grounded in those dispositions that arc fundamental properties. (See
Ellis 2001 and Bird 2007.) Combining this accoum of laws with (TAD), it follows that at
34 Nick Kroll
ACKNOWLEDGMENTS
I would like to thank George Bealer, Keith DeRose, Tamar Gendler, and Raul
Saucedo for discussion and comments on earlier versions of this paper. My deepest
thanks to Zoltan Szabo.
APPENDIX
Manley and Wasserman (2008) argue that a problem with competing conditional
analyses is that they cannot account for the fact that dispositions come in degrees.
Simplifying somewhat, their argument runs as follows. Take the adjectives "fragile,"
"sturdy," and "soluble." These adjectives denote dispositions. And these adjectives
are clearly gradablc adjectives. Witness: "This vase is more fragile than that one,"
"The concrete block is very sturdy," "How soluble is salt?" So, since gradablc
adjectives denote gradable properties, it follows that some dispositions are gradable
properties. Thus, some dispositions come in degrees. The problem, then, for
conditional analyses like (SCA) wd (LCA) is that while dispositions come in degrees,
conditionals do not. (PROP), however, doesn't have this problem because propor-
tions come in degrees.
One problem with this argument is that on what is perhaps the standard
semantics for gradablc adjectives-such adjectives do not denote properties of
individuals. Rather, they denote measure functions: functions from individuals to
degrees on a scale. (Sec Kennedy 2007.) For example, "tall" is taken to denote a
function from individuals to degrees on a scale of height and "cold" is taken to
denote a function from individuals to degrees on a scale of temperature. Comparative
morphemes, then, are taken to establish an ordering relation between degrees on the
relevant scale. Bypassing the compositional details, the upshot is that "xis taller than
y" is true just in case x's degree of height is greater than y's degree of height. A covert
morpheme is postulated for the positive form of a gradable adjective (occurrences in
clauses without any overt degree morphology like "x is G" where "G" is a gradable
adjective). Semantically, this covert morpheme takes a measure function ;rnd returns
a context sensitive function from individuals to truth values. Relative to a context,
this function takes an individual and returns the value True just in case the the value
of the measure function applied ro the individual is a degree on the relevant scale that
"stands out" in the context. Bypassing the compositional details, the upshot is that
"x is tall" is true relative to a context c just in case x' s degree of height stands out in c.
Let's go back to the gradablc adjective "fragile." Under the above semantics for
gradable adjectives, "fragile" does not denote a disposition. It denotes a measure
function from individuals to degrees on a scale. But what is this scale measuring?
least some of the laws of nature arc grounded in teleological properties. That is, it follows
that at least some of the laws are teleological laws.
Teleological Dispositions 35
;\ssuming that fragility has something to do with breaking and striking, it is probably
further safe to assume that the scale is measuring the degree to which something is
disposed to break when struck. Suppose the degrees on this scale arc dispositions. At
the bottom of the scale arc things that arc disposed to withstm1d any dttmage when
struck. Going up, we find things that arc disposed to crctck ti little when struck. Going
further up, we finds things that arc disposed to shr1tter when struck. So, x is more
fragile than y just in case x's position on this scale is higher than /s position on this
scale. Likewise, "x is fragile" is true relative to a context c just in case x' s position on
this scale stands out in c. The defender of (SCA) can now claim that the degrees of
the scale (i.e. the dispositions) correspond to certain countcrfactual properties.
Anything at the bottom of the scale is such that it would withstand any d~unage if
struck. Going further up, we find things that would crack a little if struck. Further up
yet, we find things that would shatter if struck. Generalizing, the degrees on the sea.le
can be mapped onto certain counterfactual properties, and so (SCA) has no problems
with the fact that "fragile" is a gradablc adjective.
On the other hand, it could be that the degrees on the scale are dispositions, but
rowarcls the top of the scale arc things that are disposed to break when struck ever so
lightly. Going clown are things disposed to break when struck not ever so lightly but
with a modemte ttmozmt offarce. Going further down are things disposed to break
when struck not with a moderate amount of force but with a great dMl offorce. At the
bottom, we find the unbreakable: things that are disposed to withstand ;my striking.
So, x is more fragile than y just in case x position on this other scale is higher than y' s
position on this other scale. Likewise, "xis fragile" is true relative to a context c just in
case x's position on this other scale stands out inc. And, once again, the defender of
(SCA) can now claim that these degrees on this other scale correspond to certain
counterfactual properties.
One can devise other scales, but so long as the degrees on this scale are disposi-
tions, the defender of (SCA) will not have a problem. Similar remarks apply to any
account of dispositions and any gradable adjective that is usually taken by philo-
sophers to denote a disposition. So, Manley and Wasseman's objection has no force
against (TAD).
REFERENCES
Asher, N. (1992). "A Default, Truth Conditional Semantics for the Progressive."
Linguistics cmd Phdosophy 15(5), 463-508.
Bird, A. (1998). "Dispositions and Antidotes." Philosophical Quarterly 48, 227-34.
Bird, A. (2007). Nature's Metaphysics: Lflws and Properties. New York: Oxford
University Press.
Bonevac, D., J. Dever, and D. Sosa (2006). "The Conditional Fallacy." The
Philosophicctl Review 115(3), 273-316.
Bonomi, A. (1997). "The Progressive and the Structure of Events." journal of
Semmztics 14(2), 173-205.
Brentano, F. (1874). Psychology fiwn an Empiricttl Strmdpoint. London: Routledge.
 Nick Kroll Teleological IJiJpositions 37

Choi, S. (2008). "Dispositional Properties and Counterfactual Conditionals." Mind Maurrcau, M. ( 1997). "fainthearted Conditionals." The journal ofPhilosophy 94(4),
117(468), 795-841. 187-211.
Contessa, G. (2013). "Dispositions and Interferences." Philosophical Studies 165(2), Mellor, D. H. (1974). "In Defense of Dispositions." Phi!osophim! Review 83(2),
401-19. 157~8l.

Cross, T. (2013). "Recent Work on Dispositions." Anr1lysis 72(1), 115-24. Moline, J. ( 1975). "Provided Nothing External Interferes." Mind 84( 1), 244-54.
Dowty, D. ( l 979). Word lvfeaning rmd Montague G'rmnmar: The Semantics of Verbs Molnar, G. (2003). Powas: A Study in Dispositions. Oxford: Oxford University Press.
and Times in Generative Senumtin ttnd in J'vfont11gues PTQ. Dordrccht, Nether- Mumfrml, S. (1998). Dispositions. New York: Oxford University Press.
lands: Springer. Parsons, T. (1990). Events in the Semantics of English. Cambridge, MA: MIT Press.
Ellis, B. (200 I). Scimtific Essentitt!ism. Cambridge: Cun bridge U nivcrsity Press. Place, U. T. (2005). "Intentionality as the Mark of the Dispositional." Dit1!ectim
Ellis, B. and C. Licrse (1994). "Dispositional Essentialism." Australasian }ournrt! of 50(2), 91-120.
Philosophy 72(1), 27-45. Portner, P. (1998). '"fhc Progressive in Modal Semantics." Lang11ttge 74(4), 760--87.
Fara, M. (2005). "Dispositions and Habitmls." Noiis 39(1), 43--82. Quine, W. (1960). Word and Obiect. Cambridge, MA: MIT Press.
Goodman, N. (1954). 1-~zct, Fiction,' and Forecflst. Cambridge, MA: Harvard Ryle, G. ( 1949). The Concept ofMind. Chicago, IL: University of Chicago Press.
University Press. Steinberg, J. R. (2009). "Dispositions and Subjunctives." PhilosophiCt1! Studies
Gundersen, L (2002). "In Defence of the Conditional Account of Dispositions." 148(3), 323-41.
Synthese 130(3), 389-411. Szabo, Z. G. (2004). "On the Progressive and the Perfective." No1is 38(1), 29-59.
Hallman, P. (2009). "Proportions in Time: Interactions of Qu;mtification and Szabo, Z. G. (2008). "Things in Progress." Philosophical Perspectives 22(1), 499-525.
Aspect." Natural Language Semantics 17, 29-61. Wasserman, R. (20 l I). "Dispositions and Generics." Philosophical Perspectiues 25(1),
Hauska, J. (2008). "Dispositions and Nom1al Conditions." Phdosophica! Studies 139(2), 425-53.
219-32. Woodfield, A. (I 976). Teleology. Cambridge: Cambridge University Press.
Hawthorne,]. (2006). Metr1physiCt1l Essays. Oxford: Oxford University Press.
Higginbotham, J. (2009). Tense, Aspect, and !ndexietility. Oxford: Oxford University
Press.
Jenkins, C. and D. Nolan (2011). "Disposition Impossible." Noiis 46(4), 732-53.
Johnston, M. (1992). "How to Speak of the Colors." Philosophical Studies 68(3),
221-63.
Kennedy, C. (2007). "Vagueness and Grammar: The Sem;mtics of Relative and
Absolute Gradablc Adjectives." Linguistics and Philosophy 30(1), 1-45.
Kratzer, A. (2000). "Building Statives." In L. Conathan (ed.), Proceedings of the 26th
Annual Meeting of the Berkeley Linguistic Society. Berkeley: Berkeley University
Press.
Kroll, N. (2015). "Progressive Teleology." PhilosophiCtzl Studies 172(11), 2931-54.
Landman, F. (1992). "The Progressive." Natuml LmzguageSemantiu 1(1), 1-32.
Lewis, D. (1997). "Finkish Dispositions." Philosophicrtl Quarterly 47(187), 143-58.
McKitrick, J. (2003). "A Case for Extrinsic Dispositions." Austra&sicm Journal of
Philosophy 81 (2), 155-74.
Makin, S. (2006). Aristotle: Metr1physics Book 0. Oxford: Clarendon Press.
Manley, D. (2012). "Dispositionality: Beyond the Biconditionals." Australasifln
}oun1t1l of Philosophy 90(2), 321-34.
Manley, D. and R. Wasserman (2008). "On Linking Dispositions and Conditionals."
Mind 117(465), 59.
Martin, C. B. ( l 994). "Dispositions and Conditionals." Philosophical Quarterly
44(174), 1-8.
Martin, C. B. and J. Heil ( 1998). "Rules and Powers." Nozis 32( 12), 283-312.

2 but I will not rehearse them here. Rather, I want to focus on Kroll's
Indirect Directness
Jennifer McKitrick
In "Teleological Dispositions," Nick Kroll appeals to teleology to account
for the way that dispositions seem to be directed toward their merely
possible manifestations. :He argues that his teleological account of disposi-
tions (TAD) does a better job of making sense of this directedness than rival
approaches that appeal to conditional statements or physical intentionality.
In this short critique, I argue that, without satisfactory clarification of a
number of issues, TAD does not adequately account for the directedness of
dispositions. I focus on two aspects of TAD: the Activation Principle, and
the proposed necessary and sufficient conditions for being a dispositional
property.
It is common in the dispositions literature to say that a disposition has a
trigger, also known as a stimulus, stimulus condition, or circumstance of
manifestation. 1 For example, a stimulus condition for fragility is said to be
'being struck.' Insofar as a counterfactual conditional statement is true of an
object in virtue of having a certain disposition, the antecedent of that
conditional characterizes the stimulus condition for that disposition. For
example, if the conditional "if it were struck, it would break" is true of the
fragile glass in virtue of its being fragile, then "being struck" characterizes
the stimulus for its fragility. As Kroll points out, some philosophers analyze
"the glass is fragile" in terms of a conditional such as "if the glass were struc~,
it would break." Kroll effectively characterizes many of the reasons condi-
tional analyses have been on the defensive in recent years. Many philo-
sophers have recognized numerous scenarios in which the truth values of the
disposition ascription and that of counterfactual diverge. 2 To take a simple
1 Some dispute the claim that dispositions have stimuli. See V~tter, Barbara. "Disposi;ions
without Conditions." Mind 123 (2014): 129-56 and Mumford, Stephen and Ram L. AnJum,
Getting Ciwses ftom Powm. Oxford: Oxford University Pr=. \2011): 37.. . 1md Philosophy of'Mind. Berlin: Walter de Gruytcr, 2009.
2 3
Numerous objectors to conditional analyses of d1spos1t1on ascnpt1ons mcludc Vetter
(op. cit.) as well as Smith, A. D. "Dispositional Properties." Mind 86 (1977): 439-45;
Martin, C. B. "Dispositions and Conditionals," Philosophic11l Qwtrterly 44 (1994): 1-8;
Indirect Directness 39
example, the counterfactual "if it were struck, it would break" is not true of
the fragile glass when it is wrapped in bubble wrap. At this point, of course,
there arc many moves available to defenders of the conditional approach,'1
alternative to conditional accounts of dispositions-his teleologirnl account.
I begin with a key principle of Kroll's account, the Activation Principle:
"(AV) If x's disposition to M when C is activated, then either x
immediately M~ or there is some process such that if the process
were to continue without interruption, x would M."
On the face of it, this principle seems subject to the same kinds of
counterexamples as conditional analyses. If the glass's disposition to break
when struck is activated by striking while it is wrapped in bubble-wrap, the
glass does not immediately break, and quite possibly, there is no process
which is such that, if it were to continue without interruption, the glass
would break. However, Kroll claims that the activation principle is coun-
terexample free. How could this be? Kroll claims that his account avoids
such counterexamples by making a distinction between a disposition being
"activated" and its stimulus condition obtaining. Accordingly, a disposed
object can be subject to the stimulus condition for its disposition without
that disposition being activated. In other words, the disposition can be
"stimulated" without being "activated." Kroll gives the following examples
to illustrate stimulation without activation: a vase wrapped in bubble wrap is
dropped, and is consequently subject to the stimulus condition for "the
disposition to break when dropped," but the disposition is not activated; a
computer's disposition to overheat is subject to the stimulus condition-
running a large number of processes-but the computer's disposition to
overheat is not activated due to the computer's cooling mechanisms.
The distinction between "a disposed object being in stimulating circum-
stances" on the one hand, and "an object having its disposition activated" on
the other, makes some intuitive sense. But I suspect that part of this intuitive
Bird, Alexander. "Dispositions and Antidotes," The Philosophical Quarterly 48 (1998):
227-34; Molnar, George. "Arc Dispositions Reducible?" Philosophical Quarterly 49
(I 999): 1-17, 145-67; Clarke, Rudolph. "Intrinsic Finks," The Philosophical Quarterly
58 (2008): 512-18; Everett, Anthony. "Intrinsic Finks, Masks, and Mimics," Erkennmis
71 (2009): 191-203; Schrenk, Markus. "Hie Rhodos, Hie Salta: From Reductionist
Semantics to a Realist Ontology of Forceful Dispositions," in Damschen, Gregor, Robert
Schnepf; and Karsten R. Stlibcr, eels., Debating Dispositions: Issues in Metaphysics, Epistemolog;•
Sec Choi, Sungho. "The Simple vs. Reformed Conditional Analysis of Dispositions."
Synthese 148 (2006): 369-79; Gundersen, L. (2002). In Defence of the Conditional
Account of Dispositions. Synthese, 130, 389-411.
40 Jennifer McKitrick
appeal draws on the plausible distinction between the stimulus condition
partially obtaining and the stimulus condition folly obtaining. Arguably,
the stimulus condition for the disposition to break when struck does not
frtlly obtain if something is dropped while wrapped in bubble wrap. If the
distinction between partial and complete stimulus conditions is not the kind
of distinction Kroll has in mind, then it must be that every aspect of the
stimulus condition could folly obtain, and yet the activation of the dispos-
ition does not occur. What, then, is this activation? What kind of thing is it?
What are the grounds or truth-makers for "x's disposition to lvf when C is
activated"?
Clearly, an activation is supposed to be something that happens at a
time-prior to, or simultaneous with the disposition's manifestation, as
(AV) suggests. I assume that an activation happens in a place as well,
somewhere in the vicinity of the disposed object. Since an activation has a
duration and a location, it seems like an event. But if the activation is an
event, it is clearly supposed to be an event that is distinct from the stimulus
event. So, consider a case where an unprotected glass is struck and it breaks.
If the striking started a process that continued without interruption, when
and where did the activation happen? One possibility is that it happened at
the same time and place as the striking. If that is so, we need some criteria
for event-identity beyond spatio-temporal location to differentiate the
stimulus event from the activation. Perhaps these two supposedly different
events involve instantiation of different properties. But what properties does
an activation event have? Another possibility is that the activation happens
after the stimulus occurs. So, the striking happens, and then the disposition
to break when struck is activated, initiating a process which continues
uninterrupted until the glass breaks. And if it just so happens that the
process never starts, that's no problem for the account, because we can
just say that, while stimulated, the disposition was never activated. And
what's the reason for thinking the disposition was not activated? Perhaps we
should think that the activation did not occur because neither the manifest-
ation, nor a process leading to the manifestation, occurred. It's not clear if
there could be any independent empirical evidence that an activation did or
did not occur.
What if the manifestation happened immediately upon the occurrence of
the stimulus? Suppose the glass shatters instantaneously upon being struck.
Then the activation must be simultaneous with both the striking and the
breaking. If the activation is a third event happening at the same place and
time, again it is hard to see any independent empirical evidence for its
occurrence. Either an activation is a mysterious and ad hoc third event
(in addition to the stimulus and the manifestation) or it is indistinguishable
from the initial stage of the process leading to the m<mifcstation, or it is
Indirect Directness 41
indistinguishable from the manifestation itself. lf the activation is construed
as either the manifestation, or as the initial stage of the process leading to the
manifestation, then the account becomes circular and trivial. It would
essentially say:
Ifx's disposition to 1'>1 when C is such that x immediately A1s or some process biding
to x M-ing commences, then either x immediately Ms or there is some process such
that, if the process were to continue without interruption, x would M.
None of these options seem very attractive, so I assume Kroll will want to say
something else about the nature of activation, or the truth-makers of
activation claims. In doing so, perhaps he could address another concern
about his stimulus/activation distinction. Clearly, on Kroll's view, the
stimulus is not sufficient for the activation of the disposition. Nor is the
stimulus sufficient for the manifestation, nor for any process that would lead
to the manifestation. However, the activation is sufficient for the manifest-
ation, or it is sufficient for the commencement of a process that would lead
to the manifestation. Consequently, it seems like activation is where the
action is at. So, then what is the role of the stimulus? ls it possible to activate
a disposition without the stimulus occurring? If so, then the stimulating
circumstance C seems irrelevant. Something that has "a disposition to Min
C" may or may not Min C, but it must M (or commence a processes leading
to M-ing) when it is activated. Perhaps "the disposition to Min C" should
be called "the disposition to M when activated." Cases in which fragile
glasses stay intact when strucl< would be irrelevant, since striking is merely a
stimulus, and the thing that matters for manifesting is activation. This
makes the questions about the nature of activation more pressing, because
it seems like the only rationale for positing the idea that an activation
occurred (or did not occur) is the occurrence (or nonoccurrence) of a
manifestation (or a process leading to a manifestation). Such a rationale
for making claims about when activations occur mal<es the Activation
Principle counterexample-free by fiat.
An alternative to making the stimulus incidental to the manifestation
process is to say that activation does not occur unless the stimulus occurs-
that the stimulus is necessary for the activation. But what sort of necessity
could this be? Perhaps the activation is grounded in the stimulus? But seeing
as most accounts of grounding consider grounds to be sufficient for the
grounded, 4 this would have the consequence that a stimulus is sufficient
for an activation, contraty to Kroll's account. Then perhaps the activation
4
Mett1physical Grounding: Underst11ndi11g the Structure of Reality, ed. F. Correia and
B. Schnieder. Cambridge: Cambridge University Press, 2012.
42 Jennifer McKitrick
causally depends on the stimulus? In other words, perhaps the stimulus
causes the activation. If this is right, it raises the question:
(1) Why docs a stimulus sometimes cause an activation, and some-
times not?
Maybe sometimes a stimulus has all that it takes to cause an activation, and
other times the stimulus lacks something. Note the similarities to the earlier
discussion about partial and complete stimulus conditions. Also, note the
similarity between question (1) and a question implicitly considered earlier
with respect to conditional analyses:
(2) Why does a stimulus sometimes cause a manifestation, and some-
times not?
Kroll might answer question (2) by saying that, sometimes, stimulus e<mses
an activation, and sometimes it doesn't. But if there's no answer to question
(I), then this answer to question (2) is unsatisfying. And answering question
(I) would seem to require getting specific, or appealing to ideal conditions,
or normal conditions, or any of the other moves that defenders of condi-
tional analyses have tried-moves which Kroll criticizes. Another way to put
the point is as follows. According to a simple account of the manifestation
process, when a disposition is stimulated, it manifests. We noted a prob-
lematic mismatch between stimulated dispositions and manifesting disposi-
tions. Kroll' s alternative offers a perfect match between activated
dispositions and manifesting dispositions. 5 Yet the account entails an unex-
plained mismatch between stimulated dispositions and activated disposi-
tions. So, the introduction of "activation" adds another element or step to
the manifestation process, thereby relocating, but not solving, the problem
with the simpler account.
Kroll goes on to develop his Teleological Account of Dispositions (TAD)
beyond the Activation Principle. Insofar as dispositions are directed at their
manifestations, a teleological directedness seems like a plausible way to go.
However, the details of the account warrant clarification. One of the key
tenets of TAD states necessary and sufficient conditions for a property to be
a disposition:
"(T2) Necessarily: a property Pis a disposition iff there is a condition C
and event-type M such that necessarily, Pis the property of being
in a state directed at the end that one Ms when C."
5
I am simplifying slightly. Kroll's perfect match is between activated dispositions and
manifestations (or interruptible processes leading to manifestations).
Indirect Directness 43
One implication of this analysis is that the existence of the stimulating
condition is necessary for a property to be a disposition. It is implausible to
think that this means that a particular occurrence of the stimulus is necessary
for the disposition to exist. So, it must mean that the stimulus-type is
necessaty for the disposition to exist. I am not sure what the existence
conditions for stimulus-types are. Perhaps, in order for a type of event to
exist in a world, an instance of that type must occur in that world. That
would have the implication that, say, in a world where immersion in water
has never occurred, there is no such thing as water-solubility. Also, in the
actual world, there would no dispositions to manifest in merely possible
kinds of circumstances.
This implication aside, my first question about (T2) is, what does Kroll
mean by "a property of being in a state"? To answer that, one would have to
say what a state is. Perhaps a state is a state of affairs, which some philo-
sophers construe as a particular instantiating a property. 6 Then "a property
of being in a state" would be "a property of being a particular instantiating a
property," and the account does not say what this further property is. If a
disposition is a property of being a particular with a certain property, this
suggests that a disposition is a second-order property-a property that a
thing has in virtue of having some other property. This further suggests that
a disposition must have some sort of basis, or grounds. This claim has been
disputed by a number of dispositions theorists.7
Whatever a state is, according to (T2), some states are directed at an end.
So then a disposition is a property of being in a state, and this state is
directed at an end. According to Kroll, it is the state that is directed at the end,
not the disposition. So, despite being promised an account of a disposition's
directedness, what's directed is not the disposition, but a state. Is the dispos-
ition indirectly directed at the end in virtue of being a property of a state that
is directed at an end? Then dispositions are at best indirectly directed, so
to speak. Perhaps there's nothing problematic here, but featuring "states" in
this analysis adds apparently unnecessary complications, since the motivation
for doing so is unclear.
My second line of questions about (T2) are about the end at which the
state is directed-"that one Ms when C." What does it mean ro be directed
at /Vf-ing when C? Does it me<ll1 that, when C happens, one is directed at
6
See, for example, Armstrong, D. M. A World ofStates ofAffilirs. Cambridge: Cambtidgc
University Press, 1997.
7
See, for example, McKitrick, Jennifer. "The Bare Metaphysical Possibility of Bare
Dispositions." Philosophy and Phenomenological Research 96 (2003): 349-69; Mumford,
Stephen. "The Ungrounded Argument." Synthese 149 (2006): 471-89; Bird, Alexander.
'The Regress of Pure Powers?" The Philmophical Quarterly 57 (2007): 513-34.
44 Jennifer McKitricl<
M-ing? Or, docs it mean that one is directed at C happening so that one can
M? Or does it mean that one is directed at both C happening and Nf-ing?
For example, if the end of"the disposition to break when struck" is that one
breaks when struck, does that mean that one aims at getting struck <md
consequently breaking? If so, then fragile things are, in part, directed at
getting struck. It is implausible to think that disposed objects are in a state
such that they are directed towards triggering their dispositions, even in part.
Perhaps, instead, one aims at breaking only when one is struck. This suggests
that when one isn't struck, one isn't aiming at breaking. This would have
the consequence that when dispositions aren't in the stimulating circum-
stances, they are not directed at their manifestations. If so, then this analysis
does not account for the directedness of dispositions when stimulating
circumstances do not obtain.
Furthermore, when Kroll writes that the state is directed at the end that
"one Ms when C," what is the referent of "one"? Anything? So, perhaps the
state is directed at an existential fact that something Ms when C? We get
some clarification in (T2. l), which is said to follow from (T2):
"(T2. l): Necessarily: a property Pis a disposition iff there is a condition
C :md event-type M such that: necessarily, x has P iff xis in a
state directed at the end that x M~ when C."
So, tbe object that has the disposition is the "one" that Ms, if the end is
realized. Since (T2) does not specif}' what thing Ms in the end, it is not clear
how (T2. l) is supposed to follow from (T2). At any rate, it is questionable
whether we should accept (T2. l ), for it entails that the locus of manifest-
ation is always the disposed object. Consequently, it rules out the possibility
that a thing can have a disposition for something else to M. But examples of
such dispositions are common: being lethal, poisonous, soporific, attractive,
or provocative, for example. Perhaps Kroll would want to say that, in such
cases the manifestation is causing death, causing poisoning, causing sleep, etc.
and these are things that the disposed object does when its disposition is
activated. But is "causing death" an event-type? It sounds like a causal
process, and various kinds of causal processes can be causings of death.
Furthermore, "M-ing when C" is supposed to be "the end." If the end is
"causing death," the end at which the state is directed is itself a causal
process, and not the end of that causal process. But if, instead, the mani-
festation is the end of that causal process-sleeping, dying, being angry,
etc.-then the particular that is "M-ing" is not the particular that had the
disposition in question, and (T2. l) should be rejected.
Even if these questions about (T2. l) have satisfacto1y answers, there are
further questions to consider about ends. Kroll approvingly quotes Makin:
"a teleological process has a privileged stage to which it runs in normal
indirect Directness 45
conditions." lf the end is truly a privileged state, then not just any Af-ing will
do--only NI-ing when C. So the characterization of the stimulating circum-
stances matters. And M-ing is said to be one event-type. Typing events has
its own challenges: docs shattering into thousands of shards, chipping,
cracking, and splitting in two all count as instances of the event-type
"breaking?" Regardless, a surprising consequence of this account is that it
rules out multi-track dispositions. Multi-track dispositions manifest via
different types of evencs in different types of circumstances. 8 For example,
courage can manifest by rushing into a burning building, or by standing up
for an unpopular political position. One rmmifestation of electrical charge is
amaction (in the stimulating circumstances of being in proximity to certain
kinds of particles) and another manifestation of electrical charge is repulsion
(in the stimulating circumstance of being in proximity to other kinds of
particles). Kroll could say that attraction and repulsion are really the same
type of event, but then the criteria for event-typing looks suspiciously ad hoc.
Kroll could side with Alexander Bird and say that charge is merely a cluster
of different dispositions with different manifestations. 9 And Kroll could say
that, while there are many different dispositions to do many different
kinds of dangerous or frightening things, there is no such thing as courage.
Such costs follow from defining dispositions as exclusively "single-track."
Furthermore, recall that (T2. l) says that when x has a dispositional
property to Min C, " ... x is in a state directed at the end that x Ms when
C." So, consider a puddle of water. It has many dispositions: to freeze
when cold, to evaporate when hot, to dissolve salt when salt is immersed
in it, and many others. So, according to Kroll's account, the puddle is in a
state directed at the end that it freezes when cold, and it is in a state
directed at the end that it evaporates when hot, and it is in a state directed
at the end that it dissolves salt when salt is immersed in it. If each
particular <stimulus-type, manifestation-type> pair corresponds to a dif-
ferent disposition, and we differentiate event-types in a relatively fine-
grained way, this list is innumerably long. So, how many states is the
puddle in? Is it just one state that is directed at all of these different ends?
Then the puddle would be in one state simultaneously directed at innu-
merably many ends, most of which could not be jointly realized. If there
8
See Ryle, Gilbert. 7/;e Concept of Mind. New York: Barnes and Noble, 1949, for an
inrroduction to the multi-rrack/singlc-trackdistincrion. Sec Vetter, Barbara. Potentiality: hwn
Dispositions to Modality. Oxford: Oxford University Press, 2014, pp. 36-46, for arguments
that all dispositions are massively multi-track.
9
s
Bird, Alexander. Nrtt11re Met11physics: Laws and Properties. Oxford: Oxford Univer-
sity Press, 2007, pp. 21-4. Vetter points out that this conflicts with our best scientific
understanding of such properties (Potentiality: From Dispositions to Modality).
46 Jennifer lvhKitrick

is a different state for each stimulus-manifestation pair, the puddle would

simultaneously be in innumerably many states, and simultaneously dir-
ected at innumerably many ends, most of which could not be jointly
realized. And the simple puddle would be no anomaly, in terms of its 3
massively-multi-directedncss.
Summing up, the main reasons why 'fAD does not adequately account
for a disposition's directcdness are the following. first, TAD depends on the
Dispositions without Teleology
idea that dispositions lead to their manifestations when they are activated,
but it is unclear what it means to say a disposition is activated as opposed David Manley and Ryan Wasserman
to being stimulated. Second, TAD docs not attribute directedness to the
disposition itself. Third, TAD does not account for directedness when the
stimulating circumstances do not obtain. fourth, if TAD does give us
In "Teleological Dispositions," Nick Kroll oilers a novel theory of disposi-
directedness, it gives too much, for it seems to entail that everything is always
tions in terms of primitive directed states. Kroll is clear that his notion
directed in innumerably many different directions.
of directedness "outstrips talk of goals, purposes, design, and function" (23),
Jennifer McKitrick
and that it commits him to "primitive teleological facts" (30). This notion
University of Nebraska, Lincoln
may strike some as outdated and unscientific, but Kroll argues that it has an
important theoretical role to play. In particular, he holds that a primitive
notion of dircctedncss can provide a theory of dispositions, an explanation
of the link between dispositions and conditionals, and an account of the
progressive aspect in English. In this paper, we raise some worries for each of
these claims.

l. DIRECTEDN ESS AND DISPOSITIONS

Kroll's first and most important claim is that a primitive notion of directed-
ness can provide a plausible theory of dispositions. He summarizes this
theory as follows:
(T3) Necessarily: the property of being disposed to M when C just is the
property being in a state directed at the end that one Ms when C.
According to this view, something is disposed to dissolve in water, for
example, just in case it is in a state that is directed at the end that it dissolves
when placed in water. On Kroll's view, directed states of this sort are not
reducible to non-teleological facts. 1

1
Note that one could accept (T3) without accepting this last claim. For example, one
could reduce facts about dispositions w facts about directedness, and then reduce facts
about clirectcclness to facts about singular causation, or primitive laws, or counterfactuals.
(Sec the following section on Directeclncss, Dispositions, and Conditionals for more on
this last suggestion.) Kroll rejects these reductions on the grounds that "it's possible for
there to be a disposition that is a fundamental property" (30). But he offers no argument
48 Drwid Manley and Ryan \Y/asserman
We have two main concerns for this account.
First, (T3) is only a plausible account of dispositions if the proffered
exp!mwns is less mysterious than the explm1andum. We are not sure this is
the case. Suppose, tt)f example, that Adam is disposed to order an extra shot
of espresso in the morning whenever he's been up late the night before
(unless it's the weekend or a holiday). Docs this mean that Adam alwavs
walks around "in a state directed at his ordering an extra shot of espresso In
the morning when he's been up late, provided that it's not a weekend or a
holiday"? We have no idea how to understand this question-not, that is,
unless we understand it as a roundabout way of asking whether Adam
has the relevant disposition. To us, this suggests that clirecteclness is more
naturally understood in terms of dispositions, rather than the other way
around.
Our second worry for (T3) is that it is unclear whether it can account
for three interrelated features of our talk about dispositions. First, our talk
about dispositions can take the comparative form. We can say that a
substance is volatile, for example, but we can also say that one substance
is more volatile than another. Second, our talk about dispositions is
gradable. We can say that an object is fragile, for example, but we can
also say that it is highly fragile, or that it is somewhat fragile. Third, our talk
about dispositions is sensitive to context. The standards for counting as
"irascible" in a context, for example, depend on who else is relevant in that
context, and how short their fuses are. These three features of dispositions
are closely related. In fact, if we follow the standard approach to gradable
adjectives, the second and third features can be analyzed in terms of the
first. We begin, for example, with the relation of being more Jmgile than.
This relation provides an ordering of objects along a scale. We can then
say that an object is fragile just in case it meets a contextually-determined
cutoff point on that scale, and that it is highly fragile (for example) just in
case it easily meets that cutoff point. In this way, we can provide a unified
account of the comparativiry, gradability, and context-sensitivity of our
disposition talk. 2
However, this approach is in tension with many theories of dispositions.
Consider, for example, the simple conditional analysis of dispositions:
for the claim that fundamental dispositions arc possible, and we confess to lacking any
direct intuitions about whether it's true. As a result, we think this question should be left
as spoils to the victor.
2
Elsewhere we have argued that things are a little more complicated, because there is
more than one dimension to the context-dependence of gradable adjectives (see Manley
and Wasserman 2008: 78-9).
Dispositions without Teleology 49
(SCA) Necessarily, N is disposed to M when C iff N would M if C:0
According to (SCA), an object is disposed to break when dropped, for
ex<tmple, just in case it would break if it were dropped. One problem for
this idea is that that it does not allow for comparisons. Two objects might be
such that they would both break if dropped, but it's not as if one of these
counterfactual facts is somehow "more of a fact" than the other. Since
conditionals do not come in degrees, (SCA) does not establish a scale.
And, since it does not establish a scale, it docs not allow for the standard
approach to gradability and context-sensitivity. 4
Kroll's account seems to face the same problem. As he elucidates his view,
it appears that a state either has a certain te!os or it does not-there is no such
thing as having this te!os to a greater degree than some other object does.
However, if directedness does not come in degrees, it is unclear how (T3)
could establish a scale. And without a scale, there is no way to apply the
standard treatment of gradabiliry and context-sensitivity.
Kroll anticipates this objection, and offers a response in the appendix to
his paper. 5 There, he focuses on the case of fi·agility, which (he says) has
something to do with striking and breaking. Since both of these things
admit of degrees, Kroll says that we can establish an appropriate ordering for
fragility. More specifically, we can establish a ranking of more specific
dispositions-there is, for example, the disposition to break when struck
very hard, the disposition to break when strnck with at least moderate force,
the disposition to break when struck ever so lightly, and so on (where each
of these more specific dispositions can be analyzed in accordance with (T3)).
He then says that one object is more fi'agi!e than another just in case the first
3
Co.nditional "analyses" arc often formulated as biconditionals, sometimes prefixed by
a necessity operator. However, biconclitionals and modal claims only report patterns across
modal space, whereas an analysis should arguably explain such patterns. For more on this
point, see below. Sec also Wasserman (forthcoming a and forthcoming b) and Manley
(2012).
4
Manley and Wasserman (2007, 2008).
5
We have some concerns about the way Kroll characterizes the standard account. He
wdtes. that "on what is perhaps the sta~dard semantics for gradablc adjectives, such
adjectives do not denote properties ot individuals. Rather, they denote measure
fimctlons .... A covert morpheme ... takes a measure function and returns a context
sen.sitive function from individuals to truth values" (34). This strikes us as misleading.
It _is true that some semanticists-:--like Christopher Kennedy (2007)-take gradable
ad;ectwes to express measure functions rather than properties (i.e. fi.mctions from indi-
viduals to truth values). But everyone including Kennedy agrees that the verb phrase as a
whole-111clud1ng any covert clements-expresses a property relative to a context. For
example, in an utterance of "Amal is tall," the function of the verb phrase, saturated by
''.'hatever contextual contribution, is to attribute a property to Amal. The same thing goes
for dispositional predicates.
50 David Manley and Ryan Wmserman
object's more specific disposition ranks higher than the second's. For
example, if one object is disposed to break when struck with at least
moderate force and the other is not, then the first is more fragile than the
second. With this kind of comparison in place, we can go on to give the
standard analysis of gradahility and context-sensitivity for "fragile."
Unfortunately, there are two interrelated problems with this approach. 6
The first is that the dispositions on Kroll's scales will themselves come in
degrees. For example, two objects might both be disposed to break when
lightly struck, and yet one of those objects might he more so disposed. After
all, there are lots of different angles at which something might be lightly
struck, a lot of different places where it might be struck. and lots of different
environmental factors that might be in place when it is lightly struck. 7
Intuitively, all of these things (and more) matter when it comes to making
comparisons of fragility. Suppose, for example, that one object is disposed to
break when lightly struck at any angle, and that a second is disposed to break
when lightly struck at many (but not all) angles. All else being equal, this
would mean that the first object is more fragile than the second. What this
goes to show is that Kroll will have to get much more specific about the
dispositions that appear on his scales-indeed, he will have to get maximtl!ly
specific, in the sense that the "stimulus condition" for each disposition will
have to specify not just the exact amount of force that is applied, but also the
exact angle at which it is applied, the exact place it is applied, and so on.
This leads directly to the second problem: Achilles' heels. Let's suppose
that we specify a particular angle, place, etc. in which a force is to be applied.
We then allow the specific amount of force to vary in order to establish a scale
of dispositions-there is the disposition to break when struck with a force of
1.425 millinewtons at angle a, in place p, etc., the disposition to break when
struck with a force of 2.386 millinewtons at angle a, in place p, etc. and so
on. Now imagine an unusual block-shaped object. If you hit the block at
almost any angle, in almost any place, with almost any force, it will not break.
In fact, it will not even be scratched. However, if you hit it at exactly angle a,
in exactly place p, with any force whatsoever, it will shatter. This is the block's
Achilles' heel. Here is the problem: given this block's unusual nature, its
disposition to break when struck will rank at (or near) the top of the relevant
scale-the scale in which we hold fixed place p, angle a, etc. and allow the
amount of force to vary. So, if an object's fragility is determined by where its
disposition appears on this particular scale, our unusual block will turn out to
be more fragile than a delicate crystal vase. But it clearly is not.
6 ncss and Progressives.
These problems are raised, in a more general form, in Manley and Wasserman
(2008).
7 and Manley and Wasserman (2008).
Including the presence or absence of various finks and masks.
Dispositions without Teleology 51
This kind of case shows that we cannot analyze fragility as meeting a
specifi'c threshold on a specifi'c scale of specific dispositions. Rather, in order to
understand fragility-in all of its fonns--we must somehow take ({ff of the
different stimulus conditions from all of the more specific dispositions into
account. Kroll does not offer any suggestions about how this is to be done,
so his reply is, to that extent, incomplete. 8
2. DIRECTEDNESS, DISPOSITIONS,
AND CONDITIONALS
One task for a theory of dispositions is to explain the evident link between
dispositions and conditionals. In most cases, a device is disposed to circulate
air when turned on just in case it would circulate air when turned on. So too,
a person is di;,posed to smoke when nervous just in case he would smoke if
nervous. There are, of course, exceptions to these rules, 9 but an adequate
theory of dispositions should explain why there is an exception-admitting
rule at all-that is, it should explain the link between dispositions and
conditionals.
According to Kroll, his account of dispositions suggests the following
"interesting and counterexan1ple free connection between dispositions and
conditionals:"
(CDC) Necessarily: if xis disposed to M when C, x's disposition to M
when C is activated, and x doesn't immediately M, then there is
some process such that: if the process were to continue without
interruption, x would M.
Here, the notion of continuing without interruption is to be understood in
terms of directedness (22): to say that x would M if the relevant process were
to continue without interruption is to say that the process is directed at x's
M-ing. So, for example, if something is disposed to kill when ingested, and
this disposition is activated (by, for example, someone's ingesting it), then it
will either immediately kill the person, or there will be a process in place that
is directed at killing that person. More generally, an object with a disposition
to M when C will M when C if the disposition is activated and there is no
interruption. According to Kroll, this is the link between dispositions and
conditionals.
8
We suggest an alternative approach to these issues in the final section on Dirccted-
'> For a discussion of these exceptions, see Martin (1981), Lewis (1997), Fara (2005),
52 D(fvic! Manley m1c! Ry(fn Wr1ssem1tm
We have three general worries for this suggestion.
First, we worry about Kroll's notion of teleological processes. However,
we will put off discussion of this point until the fi.nal section of this paper,
Directcdncss and Progressives.
Second, we worry about Kroll's notion of "activation." In the example
just given, a poison's disposition to kill when ingested is activated when
the relevant stimulus condition obtains-Le. when sorncone ingests the
poison. However, Kroll tells us that these two things can come apart. For
example, he tells us that a vase's disposition to break when dropped is not
activated when it is dropped while in protective packing. In this case, he
says, the disposition is "stimulated" without being "activated." But, given
this distinction, we arc unsure whether we have an adequate grasp of the
latter concept.
Herc is one way of putting pressure on this idea. Let's suppose that there are
fundamental probabilistic dispositions. More specifically, let's suppose that
there arc A-particles and B-fields, and that A-particles arc probabilistically
disposed to decay when they enter B-fields. Most of the time, when an
A-particle enters a B-field, it decays instantaneously. However, somctimcs-
by chance-an A-particle enters a B-field without decaying. (This is not a
case in which a decay-process begins and is interrupted-it is simply a case
in which no decay-process begins.) Now, suppose that an A-particle has
just entered a B-field and that-by chance-it does not decay. Was this
particle's disposition to decay activated? To the extent that we understand
this question, we think that the answer is left open. On the one hand, it
seems like indeterminism might enter in at the activation stage--i.c. it
might be indeterminate whether the disposition will be activated when the
stimulus condition obtains. But it also seems as if indeterminism could
enter in at the manifestation stage-i.c. the disposition might always be
activated when the stimulus condition obtains, but it might be indeter-
minate whether the activated disposition manifests. If this is indeed
possible, we have a counterexample to (CDC)-the particle's disposition
to decay when entering a B-field is activated, the particle doesn't imme-
diately decay, and yet there is no process such that: if the process were to
continue without interruption, the particle would decay.
Here is a second, related example. 10 Suppose that magical spells involve
direct causation at a distance, so that there is no process connecting the
casting of a spell to its effect. Suppose further that the Froschkonig spell is
disposed, when cast, to turn the prince into a frog at midnight. Suppose,
finally, that Merlin casts the Frosch!?onig spell at noon and the prince turns
10 This case is inspired by Schaffer (2000).
Dispositions without Teleology 53
into a frog at midnight. To the extent that we understand mlk of "activa-
tion," this strikes us as a clear case of a disposition being activated. After all,
we arc talking about the disposition to JrogiJY the prince when mst, the spell has
heen cast, and the prince does tum into (f frog. It may be possible for stimulus
conditions <rnd activation conditions to come apart, but we do not see how
this is possible in this mse. And, given this, we have yet another counter-
example to (CDC)-the spell's disposition is activated (at noon), the prince
does not instantaneously turn into a frog, and yet there is no process such
that: if the process were to continue without interruption, the prince would
turn into a frog.
Of course, Kroll might claim that these kinds of cases are all impossible,
but this strikes us as a cost. 11
Our third and final worry about (CDC) is that we think it understates the
connection between dispositions and conditionals. This is because we think
the following principle is correct:
(PROP) Necessarily, N is disposed to M when C iff N would M in a
suitable proportion of Ceases.
We have provided a detailed explanation of this principle elsewhere. 12
Here we will limit ourselves to one quick example involving two fragile
objects. Holding fixed the actual laws of nature, consider all of the possible
situations in which an object could be dropped. If one object is such that,
for a gre(fter proportion of those situations, it would break if it were in
them, then it is more disposed to break when dropped than the other. And
if it would break in a suitable proportion of dropping-cases, then it is disposed
to break when dropped simpliciter (where what counts as a "suitable propor-
tion" is determined by context). Finally, if the relevant object would break
in an especially high proportion of dropping-cases, it will count as "highly"
disposed to break when dropped. In this way, PROP explains the link between
dispositions and conditionals, while also accounting for the comparability,
gradability, and context-sensitivity of our disposition talk. u
11
One could rule out our kind of case by treating "x's disposition to M when C is
activ;ued" as synonymous with "either x immediately M,, or there is some process such
that: 1f the process were to continue without interruption, x would 1W." However, this
would render (CDC) an uninformative truism about a term of art .
12
Sec Manley and Wasserman (2007, 2008, 2011).
u Much more could be said about PROP, but we will stress just one point. PROP is
nothing more than a modal claim--it says that there is a necessaty connection between
dispositions and conditionals, but it docs not say why this connection holds. For this
reason, we do not take PROP to be a theory of what dispositions arc. Of course, one could
~rrengthe1: the principle by replacing "iff' with an "iff and because," or convening PROP
lllto a chum about what it is for N to be disposed to M in C. (Sec Manley 2012.) An
<1itcrnat1vc approach would be to take certain dispositional facts as basic, and to use those
54 David Manley and Ryan Wasserman
Kroll, however, raises the following objection to PROP:
My computer's ... CPU is disposed to overheat when running a large number of
(tasking) processes. That's why the computer has a heat sink and fans. When the
CPU is running a large number of processes, the heat sink and fans mt1sk the CPU's
disposition to overheat when running a large number of processes. Now creatures
smart enough to design such a CPU arc also smart enough to realize that its
disposition to overheat when running a large number of processes needs to be
masked/finked ... And this isn't an accident. It holds across the relevant region of
mod1tl ;pt1ce that cases where the CPU is engineered arc, by and large, cases where
the engineers realize that its disposition to overheat when running a large number
of processes needs to be masked/finked in sornc way ... So, on any reasonable
understanding of "suitable proportions," it's not the case the CPU would overheat
in a suitable proportion of cases where it is running a large number of tasking
processes. (22)
The idea is that CPUs like this only lack masks in a small minority of
nomological possibilities-in a sense of"minority" that incorporates object-
ive probability. 14 So there is not a "suitable proportion" of cases in which
they overheat.
There are two problems with this argument. The first involves the jump
from "small minority" to "not a suitable proportion;" the second involves
the gap between Kroll's notion of a "case" and our notion of a "C-case".
Here is the first problem. As we have argued elsewhere, it is sometimes
sufficient for a disposition to manifest in a pretty small proportion of
stimulus cases. 15 (Consider, for example, the disposition of a disease to
spread upon contact.) So it is not enough to argue that cases with the
relevant kind of CPU are, "by and large," cases where the engineers mask
the dispositions. What counts as a "suitable proportion" is a comext-
dcpcndent matter, after all, and ofi:en involves a comparison with other
salient objects. In the context of comparing various kinds of CPU, to say
that one's own is disposed to overheat may require only that it is more
facts to explain the truth of certain conditionals (rather than the other way around). Yet
another option would be to take certain teleological facts as b<L,ic, and to use those facts to
explain both the truth of ccnain conditionals and the presence of various dispositions. For
this reason, we take PROP to be compatible-at least in principle-with a teleological
approach to dispositions.
14 There arc infinitely many cases of both kinds, but presumably the idea is that in
imposing a measure on the space of cases, we should take objective probability into
account. W c arc happy to grant all of this, even though we' re not exactly sure how to cash
out the relevant notion of probability if the laws arc deterministic (so that it can't be
understood directly in terms of objective chance), or how to compare the probabilities of
various sets of initial conditions compatible with the laws.
15 M;mley and Wasserman (2008: 75; 2011: 1203).
Dispositions without Teleology 55
disposed to overheat than some other salient CPUs. (And clearly this dispos-
ition comes in degrees: the new Intel Core M, for example, is less disposed
to overheat than its predecessors.) There is no fixed minimum for what
proportion of cases is required to count as "suitablc." 16
But this docs not get to the heart of the issue. Indeed, Kroll could avoid
this objection by recasting his argument in terms of comparatives. Suppose
CPU-I is highly sophisticated-say, a quantum processor-while CPU-2 is
extremely basic. And suppose that intuitively they are equally disposed to
overheat--f-or every folly specific situation in which one would overheat, the
other would as well. Still, let's suppose that across nomological possibilities,
processors like CPU-1 are more likely to have their disposition masked.
(Engineers capable of creating a processor like CPU-1 will be especially
motivated to protect it with masks.) So, probabilistically speaking, there are
more possibilities in which CPU-1 would overheat than in which CPU-2
would overheat. But they are equally disposed to overheat-a violation of
our account of the comparative.
This leads us to our second response. In the description of the revised
objection we wrote:
(1) 1~or evety fully specific situation in which one would overheat, the other
would as well.
But we also wrote:
(2) There are more possibilities in which CPU-1 would overheat than in
which CPU-2 would overheat.
These claims are consistent only because we mean different things by
"situation" and "possibility." To illustrate, suppose we ask what Donald
would do in a very specific situation-say, meeting President Obama by
chance in a cafe for example. We can also ask what Ted would do in that very
same situation. Even a highly specific situation involves more than one
(centered) possibility--Donald's meeting Obama in just that way, Ted's
meeting Obama in just that way, etc. These possibilities are comprised of
the same situation saturated by different individuals.
16
Kroll may have in mind a context in which one is explaining the presence of the heat
sink and fan. This adds a layer of complication, because it is very natural to interpret an
Utterance like "There's a fan because the CPU is disposed to overheat" as eliding the word
"othe1wise": it is otherwise disposed to overheat. (In another context it's perfoctly fine to
say that the CPU is not disposed to overhcat--bea111se of the heat sink and fan.) Arguably
the function of "otherwise" in such contexts is to hold fixed the absence of the explicitly
mentioned mask. (Compare: "Why did I buy insurance? Well, I'd go broke if there were a
fire." Here the lack of insurance is held fixed in evaluating the counterfactual.)
56 Drwid Manley and Ryan Wassermrm
17
The point is that our C-cascs arc situations, not possibilitics. Think of
each C-case as corresponding to an open sentence describing all the specific
factors that are causally relevant to the disposition. For example, comparing
the fragility of two objects, we evaluate each one with respect to the truth of
a huge list of counterfactuals of the form "N would break if it were dropped
from l meter onto a surface with a Shore hardness of 90A, through air
with a density of l .2 kg/m 5 ... etc." And we ask in what proportion of these
situations each object would break. It will make no difference to this
comparison whether, for some of those situations, one object is more likely
to be in that situation than the other. Thus a very fancy vase and a very plain
vase might be such that they would break in all the same situations, even
though the fancy vase is less likely to be in many of the situations that induce
breakage. Even if this were true across nomologically possible worlds, it
would make no difference to their relative fragility on our view-and this
18
seems like the right result. "19
3. DIRECTEDNESS AND PROGRESSIVES
At this point, we have addressed Kroll's first two reasons for believing in
primitive dircctcdncss: the claim that it provides a successful account of
dispositions and the claim that it explains the link between dispositions and
conditionals. We now turn to his third and final reason: the claim that
primitive directedncss can provide an account of the progressive aspect in
English.
A sentence in the progressive says that a certain type of event is in
progress. For example, "Steve is driving to Boston" says that a drive to Boston
by Steve is in progress. The challenge is to say what it is for a given type of
event to be "in progress," and to do this without presupposing that this
17 See Manley <UJd Wasserman (2008: 74-5): "We can introduce the term 'stimulus
condition case' or 'C-case' for every precise combination of values for heights, Shore
measurements, densities of the medium, and so on ... "
18 It might make a difference if: though rhey would otherwise break in all the same
situations, one object would break in a more probable situation, while the other would
break in a less probable situation. When imposing a measure on situations, we may take
into account rhe objective probability of rhe situations, independently of the objects that
sarnrate them. (For example, the situation of being struck by an iron hammer may be more
probable rhan being struck by a fi·ancium hammer-even across nomologically possible
worlds.)
I'! The comparative version of this argument does seem ro cause trouble for Barbara
Vetter's version of PROP, since she explicitly treats cases as "triples ofa world, a rime, and
an object" (2012: §2; 2014: §2.3).
Dispositions without Teleology 57
event will culminate (since Steve could be driving to Boston, for example,
even if he will never reach his destination).
Kroll suggests the following:
(EIP) Necessarily, e is a cp event in progress at tiff e is, at t, directed at
the end that it cause the resultant state of a (p event to obtain
at some I > t.
So, for example, what makes it the case that there is a "drive to Boston by
Steve" in progress is that that there is an event going on that is directed at
b1inging it about that Steve has driven to Boston. Crucially, such an event
could be going on, even if this telos is never achieved-this, Kroll claims, is
what makes teleological talk ideal for understanding the progressive aspect.
We have three main worries for this claim.
First, we worry that Kroll's approach requires too much of events in
progress. Of course we agree with Kroll that there is some sense in which
Steve's drive is directed toward his having driven to Boston. Similarly, if
Sanjay is building a house, then there is some sense in which his activity is
directed toward his having built a house. rfowever, once we get beyond
these kinds of examples, teleological talk seems far less natural. Suppose, for
example, that Sela is laughing, or standing still, or shaking uncontrollably.
All of these statements arc in the progressive aspect, but none of them need
have anything to do with teleology. There is no sense--or, at least, no
intuitive sense-in which Sela's laughter is "directed at the end that it
cause Sela to have laughed." Since (EIP) requires this to be the case, that
principle strikes us as implausible.
Our second worry is that Kroll's approach seems to require too little of
events in progress. Suppose Bill kills Bob with his gun. While he is firing
the gun, Bill is killing Bob; but while he is planning the murder and
buying the gun, he is not yet killing Bob. But why not? On Kroll's view, it
must be that no event has begun that has Bill's killing Bob as its end. But
that seems false, as far as we can understand it. It seems that Bill's
planning is directed at this end in exactly the same sense in which his firing
the gun is. But Kroll will need to draw a sharp distinction between these
events in order to get the data right. This strikes us as a difficult position
to maintain.
Our third and final worry concerns the relation between progressives,
outcomes, and dircctedncss. To begin, consider the following case, which
Kroll discusses elsewhere:
Suppose Mary needs to cross a minefield. Besides the first few feet, rhe minefield is
densely filled. So, while there is some probability that she would cross it if she tried,
the probability is minuscule. Also, the mines are spread out fairly evenly: besides the
58 David Manley and Ryan YVasserman
first few feet, for each step she could take forward, there is roughly the same, and very
high, probability that she would step on a mine. Mary, however, is unaware that the
field is a minefield. All she knows is that she needs to make it to the other side. So,
she steps out into the minefield, showing absolutely no caution. She takes a few steps
and then, not surprisingly, steps on a mine. Let t be some time after Mary first steps
out into the minefield but before she steps on the mine that kills her. (6a) is true at t.
But what about (6b)?
(6a). Mary is walking in the minefield.
(6b). Mary is crossing the minefield.
It seems to me (and many others) that (6b) is false at t. Mary was trying to cross
the minefield but wasn't crossing it. (2015: 2934)
We agree with Kroll's judgment in this case: (6b) is false in the case where
Mary will soon step on a landmine. In order to generate this result, Kroll will
have to deny that there is (at that time) an event that is directed at bringing it
about that Mary crossed the minefield. So far, so good. But now what about
the case in which Mary does-by a string of lucky steps-happen to cross
the minefield? In that case, Kroll says that (6b) is--at some point-true
(2015: 2935, fn 9). Indeed, even if Mary had said after the first few steps,
'Tm crossing the field!", we would have to admit in retrospect that she was
speaking the truth. But, in this case, Kroll will have to say that there is an
event that is directed at bringing it about that Ma1y crossed the minefield. In
other words, whether or not there is a directed event of the relevant kind will
depend upon the occurrence (or non-occurrence) of certain events in the
future. 20 This strikes us as a strange thing for the teleologist to say--one
would think that whether or not a primitive teleological fact obtains at t
would. be independent of what goes on after that time.
In fact, this final worry seems to bring out a tension between Kroll's
theo1y of dispositions and. his account of the progressive. After all, the fact
that d.irected.ness is outcome-independent is precisely what was supposed to
make it suitable for analyzing dispositions in the first place: whether or not
an object is disposed to break when dropped-or whether it is in a state
directed at breaking when dropped-does not turn on whether it will
actually be dropped or broken. But, in that case, directedness seems ill--
suited for analyzing the progressive since some progressive statements differ
from disposition ascriptions in exactly this respect.
David Manley
University of Michigan, Arm Arbor Ryan
Wasserman Western Washington University
Dispositions without 1{,feo!ogy 59
REFERENCES
Fara, Michael. 2005. "Dispositions and Habituals." Notis 39(1): 43-82.
Kennedy, Christopher. 2007. "Vagueness and Grammar: The Semantics of Relative
and Absolute Gradable Adjectives." Linguistics and Phiu1sophy 30(1): 1-45.
Kroll, Nick. 2015. "Progressive Teleology." Philosophiml Studies 172( I I): 2931--54.
Kroll, Nick. 2016. 'Teleological Dispositions" fin this volume, 3--37).
Lewis, David. 1997. "Finkish Dispositions." The Philosophiml Quarter&' 47(187):
143-58.
Manley, David. 20 l 2. "Dispositionality: Beyond the Biconditionals." Austrrdmimt
journtI! of Phi/osopl~y 90(2): 321-.34.
Manley, David and Wasserman, Ryan. 2007. "A Gradable Approach to Dispositions."
The Philosophical Quarterly 57(226): 68-75.
Manley, David and Wasserman, Ryan. 2008. "On Linking Dispositions and
Conditionals." Mind 117(465): 59-84.
Manley, David and Wasserman, Ryan. 20 l l. "Dispositions, Conditionals, and
Counterexamples." Mind 120(480): 1191-227.
Martin, C. B. 1994. "Dispositions and Conditionals." IlJe Philosophical Quarterly44
(174): 1-8.
Schaffer, Jonathan. 2000. "Overlappings: Probability-Raising without Causation."
Austmlmian journal of Philosopl~y 78( l): 40-6.
Vetter, Barbara. 2012. "On Linking Dispositions and Which Condition<1ls?" Mind
120(480): 1173-89.
Vetter, Barbara. 2014. "Dispositions without Conditionals." Mind 123(489):
129-56.
Wasserman, Ryan. Forthcoming a. "Theories of Persistence." Philosophical Studies.
Wasserman, Ryan. Forthcoming b. "Vagueness ;md the Laws of Metaphysics."
Philosophy and Phenomenological Research.

20
Suppose Bill is firing his gun at Bob and intending to kill him. ls he killing Bob? If
Bill happens to miss every rime-however unluckily-the answer is "no." If he hits
Bob---however luckily-and causes Bob to die, the answer is "yes."
 4
Devious Stipulations
john Horden

1. INTRODUCTION

Traditionally, analytic truths have been thought, in some good sense, to

require nothing of the world. And it is a corollary of this thought that
combining a theory with its analytic consequences never produces a theory
that requires more of the world than the original. Rudolf Carnap (l 950)
employed this idea in an attempt to reconcile his empiricism with a liberal
acceptance of abstract entities; claiming that the existence of numbers,
properties, etc. can be trivially deduced by the rules governing our expres-
sions for such things, once those expressions and rules are introduced into
our language. In a highly influential response, Willard Van Orman Quine
(l 951) argued that the analytic/synthetic distinction Carnap relied on for
that result is untenable. Thereafter many so-called analytic philosophers,
reputedly a majority for a while, followed Quine in disavowing analyticity
altogether.
However, since then, analyticity has slowly crept back into philosophical
respectability. Nowadays, it seems, analytic philosophers are mostly inclined
to accept both that there are analytic truths (see Bourget and Chalmers,
2014), and that such interpreted sentences are characteristically metaphys-
ically undemanding, at least in the sense of being neutral with respect to our
position in logical space. Rather than rejecting analyticity outright, most
would simply deny that there are any metaphysically interesting analytic
truths. Meanwhile, going against the grain somewhat, there have in recent
years been several notable attempts to answer ontological questions through
conceptual analysis (see Hale and Wright, 200 l; Schiffer, 2003; Thomasson,
2007, 2015; Hirsch, 2011; HofWeber and Velleman, 2011; Steinberg,
2013). And, by and large, all sides agree that if the existence of certain things
analytically follows from sentences we already accept, then explicitly acknow-
ledging the existence of those things cannot sensibly be regarded as an extra
theoretical cost.
64 john Harden
In his 'Analyticity and Ontology', Louis deRosset challenges this consensus.
rie tries to refl1te two principles (2015: 131, 139):
DAO If P analytically entails the existence of certain things, then a
theory that contains P but does not claim that those things exist
is no more ontologically parsimonious than a theory that also
claims that they exist.
GAO If P analytically entails Q, then (P /\ Q) requires nothing more of
the world than docs P.
(DAO is what deRosset calls 'the doctrine of analyticity in ontology' and is
his main target. GAO is a generalization thereof.)
He glosses these principles as follows: 'a sentence </> is analytic iff it is
entailed by true sentences i/1 1 , ih, ... such that failure to accept any if1n
constitutes some measure of linguistic incompetence' (2015: 133); 'a sen-
tence r/> analytiad!y entails a sentence if; iff the material conditional (r/> =? if;)
is analytic' (2015: 133); 'the parsimony of a theory is given by what the
truth of the theory requires of the world with respect to what there is' (2015:
138). So the more ontologically parsimonious a theory is, the less it requires
of the world with respect to what there is.
DeRosset does not define what it is for a sentence or theo1y to require
something of the world, but makes clear that 'the world' here rigidly desig-
nates the actual world:
The argument [against GAO] here relics on the assumption that t1ctu11l!y: </J requires
no less of the world-the 11ctut1l world, that is-than docs</). ( ... W]hat's required
of the actual world for actur1l!y: grass is green to be true is just for the actual world
to meet whatever requirements there arc for grass is green to be true. (2015: 143;
sec also n. 20)
Perhaps we should accept something along the following lines: a sentence</>
requires of the world that R iff (i) 'R' does not logically follow from
sentences that are true entirely because of their meanings, and (ii) whichever
world is actual, if q) is true, this is partly because R. And then: a sentence </>
requires nothing more of the world than a sentence if1 iff, however 'R' is
replaced, if r/> requires of the world that R, then if; requires of the world
that R. Thus 'Actually grass is green' requires nothing more or less of the
world than 'Grass is green'.
I won't rely on the correctness of the above definitions exactly as stated; just
as dcRosset, so he tells us (2015: 135), docs not rely on his 'rough and ready'
definition of analyticity. lfowcver, I shall make two assumptions. First:
(schematically) if a sentence requires of the world that R, then, whichever
world is actual, that sentence is true only if R. Second: if a sentence is true
Deuiotts Stipu!tttions 65
entirely because of its meaning, then it requires nothing of the world. (Notice
that I say 'entirely because' here, not 'only because'. This is to allow for
overdetermination. Plausibly, 'Everything is sclf~identical' is true partly
because everything is self-identical (see Quine, 1954: 113; Boghossian,
1996: 364). Nonetheless, that sentence is fully determined to be true by its
meaning; so in tl1c relevant sense, it requires nothing of the world (see Russell,
2008: § 1.2).)
Now, to sec the prima j{zcie appeal of DAO and GAO, consider:
(M 1) Someone is married.
(M2) Someone is married to someone else.
(M 1) explicitly affirms the existence of one person only, whereas (M2)
explicitly affirms the existence of two. Nonetheless it seems clear chat the
conjunction of (MI) and (M2) requires nothing more of the world (and in
particular, is no less ontologically parsimonious) than (M 1) alone, because
(Ml) analytically entails (M2). And likewise, we might think, for more
metaphysically interesting analytic entailments, if such there be.
However, dcRosset denies that DAO and GAO hold in full generality.
He suggests that in the metaphysically interesting cases just alluded to, these
principles are liable to break down. So even if, to take his leading example,
'There are particles arranged tablewise in location L' analytically entails
'There is a table in L', it remains plausible that the conjunccion of these
sentences is less ontologically parsimonious than the former sentence alone.
(Here he targets Amie Thomasson in particular. For the relevant sense of
'arranged tablcwise' and the like, sec her 2007: 16-17.)
Thus while unQuineanly granting the tenability of the analytic/synthetic
distinction, deRossct disputes its metaphysical significance. To this end, he
purports to produce counterexamples to DAO and GAO by means of
linguistic stipulations. He offers two main candidate counterexamples just
to GAO (though he admits that the first of these is not decisive) and one
specifically to DAO. I aim to show where these arguments go wrong.
2. 'VERDANTLY*'
Here is dcRossct's first candidate counterexample to GAO. (I bypass his initial,
illustratively unsuccessful 'verdantly' stipulation.) Suppose we introduce a new
sentential operator 'verdantly*' by stipulating the following rules of inference:
P ~Verdantly* P
Verdantly* P ~Actually grass is green
Verdantly* P ~Possibly P
66 john Horden
'Actually' here has its familiar interpretation from two-dimensional seman-
tics: r Actually (/>' is true at a world iff </> is true at the actual world. Hence
what deRosset calls Steuenson ~- constraint on linguistic stipulation (a gener-
alized version of the constraint proposed by Stevenson, l 961) initially
appears to be satisfied:
[AJ linguistic stipulation succeeds only if there is a consistent way to assign truth
conditions to sentences containing the introduced term that makes the content of
the stipulation true. (20 l 5: l 41)
To see this, suppose grass were not green. Then by the above introduction
rule, 'Verdantly* grass is not green' would be true, and so by the first
elimination rule, 'Actually grass is green' would be true. But even if grass
were not green, grass would still be green at the actual world, so no incon-
sistency results.
If his 'verdantly*' stipulation succeeds, deRosset tells us, we have a
counterexample to GAO. For then 'Snow is white' analytically entails
'Actually grass is green'. But 'Snow is white and actually grass is green'
requires more of the world than 'Snow is white'. For the latter only requires
that snow be white, whereas the former also requires that grass be green.
However, deRosset admits some uncertainty as to whether this stipula-
tion succeeds (2015: l43-4). For perhaps when applying Stevenson's
constraint we are not entitled to take for granted tl posteriori necessities
such as the fact that grass is actually green. Perhaps we should also consider
how to evaluate sentences containing 'verdantly*' on the assumption that
grass is not actually green. But then consider what truth value should be
assigned to
Verdantly* grass is either green or not green.
If it is true, the first elimination rule for 'verdantly*' is invalid, because
'Actually grass is green' is by hypothesis false. If, on the other hand, the
evaluated sentence is false, the introduction rule is invalid, because 'Grass
is either green or not green' is (we may assume) logically true. So the
'verdantly*' stipulation fails to meet Stevenson's constraint on this more
stringent construal. DeRosset ultimately leaves open how stringently
Stevenson's constraint should be construed, and so admits that his 'ver-
dantly*' stipulation does not provide a decisive counterexample to GAO.
Fair enough; though whichever way we construe Stevenson's constraint,
I wonder how useful it will be for adjudicating on attempted stipulations in
cases where the bounds of conceptual possibility are in dispute. Should we
consider how to evaluate sentences containing 'number' or 'proposition' on
the assumption that there are only finitely many things, for example? If so,
then the broadly Fregean and Carnapian views of abstracta targeted by
Deuious Stipu!tztions 67
deRosset are in trouble; but any such objection to those views-which, to be
fair, deRosset does not himself raise-seems question-begging.
In any case, there is another reason to think that the 'verdantly*' stipu-
lation fails. For 'verdantly*', as defined, is by reasonable standards dearly
and fatally nonconservative. N uel Belnap (l 962) recommended a conserva-
tiveness constraint on stipulation in response to Arthur Prior's (1960)
parody of implicit definition, wherein 'tonk' is defined by the rules:
P~ P tonk Q
P tonk Q ~ Q
Thus any sentence analytically entails any other; hence every sentence is
analytic. Obviously the stipulation fails; the only question is how.
J. T. Stevenson (1961) proposed one constraint in response: every truth-
functional connective must have a consistent truth table (notice that this is
significantly weaker than the constrnint deRosset names after Stevenson, on
either construal of the latter). Belnap proposed another constraint: the
introduction of any new vocabulary must yield a conservative extension of
the language. That is to say, the rules stipulated to govern new vocabulary
cannot allow the derivation of any sentence of the old language (i.e. any
sentence without the new vocabulary) that was not already derivable. 1 In
footnotes, deRosset describes this conservativeness constraint, and observes
that it is met by his subsequent stipulations-presumably regarding this as a
point in their favour-but conspicuously does not claim the same for his
'verdantly*' stipulation. And it is easy to see why: 'Actually grass is green' is a
synthetic sentence of English before this stipulation, but would become
derivable from any sentence whatsoever (and hence analytic) were the
stipulation to succeed. Thus Belnap's constraint is manifestly violated.
Admittedly, many regard Belnap's constraint as too strong (see Read,
1988: §9.3; 2000: 125-7; Peacocke, 1993: §§3-4; 2004: 18-21; Prawitz,
1994: 374; Shapiro, 1998: §3). For example, it seems legitimate to add
second-order quantifiers or a truth predicate to a first-order system of
arithmetic, though doing so enables the derivation of the original system's
Godel sentence. Also, abstraction principles such as Hume's Principle
The number of Fs = the number of Gs iff there are just as many Fs as Gs
if successfully stipulated, allow us to derive results about the (infinite) size of
the domain of first-order quantification that were previously statable but
perhaps unprovable. Hence Bob Hale and Crispin Wright endorse a weaker
1
This principle of conservativeness, or 'noncreativity', was previously endorsed by
others; sec e.g. Frcge, 1914: 208. For a historical discussion, see Urbaniak and l·famari,
2012.
68 John Horden
version of conservativeness (cf. Field, J 980: ch. 1; Wright, 1997: §9; 1999:
§2. 5; Schiffer, 200.'3: §2.2), also mentioned by dcRossct (2015: 14 l, n. 19),
with the following proviso attached:
Tt is our view that a stipulation may have consequences which can be expressed in the
;mtcccdent l;mguagc, and to which there need have been no previous commitment,
without compromise of its legitimacy pro1tided rhe truth of these consequences makes no
dern;rnds on the previously recognized ontology. (Hale and Wright, 2000: .302, n. 32)
However, the truth of 'Actually grass is green' clearly makes demands on the
previously recognized ontology. l;or it demands that grass be green. So the
'verdantly" stipulation fails to meet even this weaker version of conservative-
ness. So if either Belnap's or Hale and Wright's version of conservativeness is
correct, this stipulation cannot provide a genuine counterexample to GAO.
Moreover, regardless of the specific constraints on stipulation, deRossct's
'verdantly*' example faces a ruinous dilemma. Pirst horn: 'Actually grass is
green' retains its prior meaning. Then the nonanalyticity of this sentence,
given its empirical status, is far more certain than the success of the
stipulation. Given our understanding of analyticity, regrettably imprecise
as that may be, it is absurd to think that we could make this sentence
analytic without changing its meaning. So we should conclude, in Moorean
fashion, that the stipulation somehow fails. If there were no known con-
straint that prohibited this stipulation, we would have to posit one. That,
mutatis mutrmdis, is the lesson of 'tonk'. Second horn: 'Actually grass is
green' loses its prior meaning. Then we don't know what, if anything, this
sentence means after the stipulation: its subsequent meaning seems to be
radically underdetermined. So we have no reason to accept that its conjunc-
tion with 'Snow is white' requires more of the world than 'Snow is white'
alone. Either way, we do not get a genuine counterexample to GAO.
3. 'GRASSGREEN'
Let us now examine deRosset's second candidate counterexample to GAO:
Suppose we stipulate that 'grassgrcen' is to be a predicate that expresses the property being
green if, as a matter of fact, grass is green, and not being green otherwise. (2015: 144)
This stipulation, deRosset tells us, meets Stevenson's constraint even on its
more stringent construal, as well as meeting other proposed constraints
including conservativeness, generality, and harmony.2
2
The latter two constr-aints arc proposed by Hale and Wright (2000)---in addition to
their version of conservativeness--and are clcscribed by deRosset thus: 'Genemlity: the
Devious Stipulations 69
Now, he continues, we have a more decisive counterexample co GAO. Por
if this stipulation succeeds, then 'Grass is grassgrcen' is thereby guaranteed co
be actually true, no matter what colour grass actually has. 3 So 'Snow is white'
analytically entails 'Grass is grassgreen'. But 'Snow is white and grass is
grassgreen' requires more of the world than 'Snow is white'. Por since grass
is actually green, 'grassgrccn', according to the stipulation, expresses the property
of being green. So whereas 'Snow is white' only requires that snow be white,
'Snow is white and grass is grassgrcen' also requires that grass be green.
Clearly something fishy is going on here. If 'Grass is grassgrccn' is
guaranteed to be actually true, no matter what colour grass actually has,
then it cannot require of the actual world that grass be green. Even if we
grant that the stipulation succeeds, it seems sensible to examine the seman-
tics of the novel term 'grassgreen' a little more closely before we embrace
inconsistency. So what is its stipulated meaning? That, I think, depends on
whether the stipulation is conditionti! or unconditional. This yields another
dilemma.
Pirst horn: 'is grassgrecn' is conditionally stipulated to mc<m is green if grass
is actually green, and is not green otherwise. Then the stipulation does not by
itself settle the meaning of this predicate. The inscription-type 'Grass is
grassgrcen' is thereby guaranteed to express some truth at the actual world
in our extended language, but which truth it expresses depends on the actual
colour of grass. In fact, since grass is actually green, the resulting interpreted
sentence is exactly synonymous with 'Grass is green'. So we have no reason to
think that it is analytically entailed by 'Snow is white'. Second horn: 'is
grassgreen' is unconditionally stipulated to mean is green iflgmss is tictually
green. 4 Then 'Grass is grassgreen' does not require of the actual world that
grass be green. Nor does it require anything else of the world. So its conjunc-
tion with 'Snow is white' requires nothing more of the world than 'Snow is
white' alone. Either way, we do not get a genuine counterexample to GAO.
DeRosset concedes that it is unclear what, exactly, we should take 'grass-
green' to mean, but insists that this does not matter for his purposes, so long
as his stipulation succeeds (2015: 148). On the contrary, his argument here
apparently rests on an equivocation between two intensionally equivalent
stipulation should enable the interpretation of a wide enough range of relevant sentences';
'Harmony: the introduction and elimination rules should not allow us to infor more (or
problematically less) than our warrant for the premises allows us to infer' (2015: 141, n. 19).
:i At least, 'Grass, if it exists, is grassgreen' is guaranteed to be actually true; like
deRosset, I set aside this nuance. Also I grant that analytic truths can be contingent. As
deRosset explains in his paper, he regards that assumption as inessential to his arguments
anyway.
4
Or, more longwindcdly: is such that either (11) it is green and actually gn1Ss is green or (b)
it is not green and it is not the case that actually gmss is green.
70 john Horden
interpretations of the introduced predicate, either of which is consistent
with the explicit content of the stipulation. (Likewise for deRosset's third
candidate counterexample, as we shall shortly see.) On one interpretation,
'Grass is grassgreen' requires something of the world, but is not guaranteed
by its meaning to be actually true. On another interpretation, 'Grass is
grassgrecn' is guaranteed by its meaning to be actually true, but requires
nothing of the world. And yet, on an intensional individuation of properties,
'grassgrcen' expresses the same property on either interpretation: the prop-
erty of being green.
As it happens, deRosset anticipates the second interpretation, though he
relegates this point to a brief footnote:
It is plausible, perhaps, to think that the mnh of ['Grass is grassgreen'] requires that
grass be green iff grass is actually green; this is the view most naturally suggested by
the content of the 'grassgrecn' stipulation. But this requirement clearly goes beyond
the requirement for the truth of ['Snow is white'], as evidenced by the fact that the
requirement is not satisfied in circumstances in which grass is purple but snow is still
white. (2015: 161-2, n. 47)
This response is clearly inadequate, however. For deRosset, recall, takes the
expression 'requires nothing more of the world', as it appears in GAO, to be
equivalent to 'requires nothing more of the actual world'. And though
'Grass is green iff grass is actually green' plausibly requires of nonactual
worlds that grass be green, it requires nothing of the actual world (see Evans,
1979: §4; Davies and Humberstone, 1980: §2). So if 'Grass is grassgreen'
merely abbreviates that biconditional, then its conjunction with 'Snow is
white' requires nothing more of the world than 'Snow is white' alone. It is
irrelevant that at some other possible world the biconditional is false while
'Snow is white' is true.
4. 'PRIMAN BEINGS'
Now for deRosset's third candidate counterexample to GAO, which he also
intends to serve as a counterexample to DAO:
Suppose we stipulate that 'priman being' is to be a predicate that expresses the
property being a human being if, as a matter of fact, there arc more than seven billion
human beings, and being a prime number otherwise. (2015: 146)
The stipulation here closely parallels that of deRosset's previous example.
And as with his previous example, he assures us that this stipulation meets all
the constraints mentioned so far.
If this stipulation succeeds, he continues, then the material conditional
'If there are more than 7 billion prime numbers, then there are more than
Devious Stipulations 71
7 billion priman beings' is thereby guaranteed to be actually true, no matter
how many humans there actually are, hence it is analytic. So 'There are more
than 7 billion prime numbers' analytically entails 'There arc more than
7 billion priman beings'. But 'There are more than 7 billion prime numbers
and there are more than 7 billion priman beings' requires more of the world
(and in particular, is less ontologically parsimonious) than 'There are more
than 7 billion prime numbers'. F;or since there arc actually more than
7 billion humans, 'priman being', according to the stipulation, expresses
the property of being human. So whereas 'There are more than 7 billion
prime numbers' does not require the existence of any humans, 'There are
more than 7 billion prime numbers and there are more than 7 billion
priman beings' requires the existence of more than 7 billion of them.
The analogy with deRosset's 'grassgreen' example should be clear; hence
I shall respond analogously. If, given the existence of more than 7 billion
prime numbers, 'There are more than 7 billion priman beings' is guaranteed
to be actually true, no matter how many humans there actually are, then it
cannot require of the actual world that there be more than 7 billion humans.
Even if we grant that the stipulation succeeds, we should examine the
semantics of the novel term 'priman being' a little more closely before we
embrace inconsistency. So what is its stipulated meaning? Again, that
depends on whether the stipulation is conditional or unconditional. This
yields a dilemma.
First horn: 'is a priman being' is conditionally stipulated to mean is
human if there are actually more than 7 billion humans, and is a prime
number otherwise. Then the stipulation does not by itself settle the meaning
of this predicate. Provided that there are more than 7 billion prime num-
bers, the inscription-type 'There are more than 7 billion priman beings' is
guaranteed by the stipulation to express some truth at the actual world
in our extended language, but which truth it expresses depends on how
many humans there actually are. In fact, since there are actually more than
7 billion humans, the resulting interpreted sentence is exactly synonymous
with 'There are more than 7 billion humans'. So we have no reason to think
that it is analytically entailed by 'There are more than 7 billion prime
numbers'. Second horn: 'is a priman being' is unconditionally stipulated
to mean is human if there are actually more than 7 billion humans, and a
prime number otherwise. 5 Then 'There are more than 7 billion priman
beings' requires nothing more of the actual world than whatever is required
for there to be more than 7 billion prime numbers. So its conjunction with
5
Or, more longwindedly: is such that either (a) it is human and actually there are more
than 7 billion humans or (b) it is a prime number and it is not the case that actually there are
more than 7 billion humttns.
72 john Horden
'T'here are more than 7 billion prime numbers' requires nothing more of
the world, and is no less ontologically parsimonious, than 'There are more
than 7 billion prime numbers' alone. Either way, we do not get a genuine
counterex<unple to either GAO or DAO.
5. CONCLUSION
So, as we have seen, none of deRosset's stipulations yields a genuine
counterexample to either of the principles he targets. Despite his efforts to
discredit them, the doctrine of analyticity in ontology and its generalization
remam. as p laust"bl e as ever. 6
john Horden
National Autonomous University of Mexico
REFERENCES
Belnap, N. [)., 1%2, 'Tonk, Plonk and Plink', Analysis, 22(6): 130-4.
Boghossian, P. A., 19%, 'An;1lyticity Reconsidered', Notis, 30(3): 360--91.
Bourget, D. and D. J. Chalmers, 2014, 'What Do Philosophers Believe?', Philosophiml
Studies, 170(3): 465-500.
Carnap, R., 1950, 'Empiricism, Semantics, and Ontology', Revue lntemrttionale de
Philosophic, 4: 20-40; reprinted in his 1956, Meaning ttnd Necessity: A Study in
Semantics and Modt1! Logic, 2nd cdn, Chicago, IL: University of Chicago Press.
Davies, M. and L. Humbcrstonc, 1980, Two Notions of Necessity', Philosophiml
Studies, 38(!); 1-30.
deRosset, L., 2015, 'Analyticity and Ontology', in K. Bennett and D. W. Zimmerman,
eds, Oxford Studies in Metaphysics, vol. 9, Oxford: Oxford University Press.
Evans, G., 1979, 'Reference and Contingency', The Monist, 62(2): 161-89.
Field, H. H., 1980, Science \Vithout Numbers: A Defence ofNomimtlism, Princeton,
NJ: Princeton University Press.
Frege, G., 1914, 'Logic in Mathematics', trans. P. Long and R. White, in
H. Hermes, F. Kambartel, and F. Kaulbach, eds, 1979, Gottlob Frcge: Posthumous
Writings, Oxford: Blackwell.
Hale, B. and C. Wright, 2000, 'Implicit Definition and the A Priori', in
P. Boghossian and C. Peacocke, eds, New Esst1ys on the A Priori, Oxford; Oxford
University Press; reprinted in Hale and Wright, 2001.
6
Thanks to Philipp Blum, Aurelien Darbellay, Miguel Hocltje, Dan Lopez de Sa,
Giovanni Merlo, Sven Rosenkranz and Louis deRosset for comments ;md cliscnssion.
My research was supported by projects CSD2009-0056, FFl2012-35026, and FFI2015-
66372-P, Gobierno de Espana, and the LOGOS group, grant 2014SGR-8l, Generalitat
de Catalunya.
Devious Stipu!tttions 73
Hale, B. and C. Wright, 200!, The Ret1sonj· Proper Study: limzys towards a
Neo-Fregean Philosophy ofMathmMtics, Oxford: Oxford University Press.
Hirsch, E., 20 l l, Qumuijii.'I" Vmiance rmd RMlism: Esst1ys in !vfettlonto!ogy, Oxford:
Oxford University Press.
Hofweber, T. and J. D. Vclleman, 20 l l, 'How to Endure', Philosophicrd Qzuzrterly,
61 (242); 37-57.
Pcacockc, C., 1993, 'Proof and Truth', in J. lhldanc and C. Wright, eds, Reality,
Representt1tio11, and Projection, Oxford: Oxfrird University Press.
Peacocke, C., 2004, l/Je Realm of Ret1son, Oxford: Oxford UniversiLy Press.
Prawitz, D., 1994, review of M. Dummett, The Logiml Br/Sis of Metrtphysics, /\1ind,
103(411): 373-6.
Prior, A. N., 1%0, 'The Runabout Inference-Ticket', Arwlysis, 21(2): 38-9.
Quine, W. V., 1951, "fwo Dogmas of Empiricism', Philosophicrd Review, 60(1):
20--43.
Quine, W. V., 1954, 'Carnap and Logirnl Truth', in his 1976, TJ1e Wt1ys of Pamdox
and Other ESSttys, 2nd edn, Cambridge, MA: Harvard University Press.
Read, S., 1988, Relevant Logic: A Philosophicttl Er:mnination of inference, Oxford:
Blackwell.
Read, S., 2000, 'Harmony and Autonomy in Classical Logic',jourwtl ofPhilosophicrd
Logic, 29(2): 123-54.
Russell, G., 2008, Truth in Virtue of Met1ning: A Defence of the Analytic/Synthetic
Distinction, Oxford: Oxford University Press.
Schiffer, S., 2003, The Things We Mem1, Oxford: Oxford University Press.
Shapiro, S., 1998, 'Induction and Indefinite Extensibility; The Godel Sentence is
True, but Did Someone Change the Subject?', Mind, 107(427): 597-624.
Steinberg, A., 2013, 'Plconastic Possible Worlds', Philosophiml Studies, 164(3); 767-89.
Stevenson, J. T., 1961, 'Roundabout the Runabout Inference-Ticket', Analysis,
21 (6): 124-8.
Thomasson, A. L., 2007, Ordinmy Objects, Oxford: Oxford University Press.
Thomasson, A. L., 2015, Ontology Made Et1sy, Oxford: Oxford University Press.
Urbaniak, R. and K. S. Hamari, 2012, 'Busting a Myth about Ldniewski and
Definitions', Histoiy and Philosophy of Logic, 33(2): 159-89.
Wright, C., 1997, 'On the Philosophical Significance of Frege's Theorem', in
R. Heck, ed., Language, Thought, and Logic: Esst1ys in Honour ofMic/;t1el Dummett,
Oxford: Oxford University Press; reprinted in Hale and Wright, 200 l.
Wright, C., 1999, 'Is Hume's Principle Analytic?', Notre Dmne journal of FonntJ!
Logic, 40(1); <5--30; reprinted in Hale and Wright, 2001.

5 requirement is that that vety individual have that vety property.
Stipulations and Requirements:
Reply to Horden
Louis deRosset
In (deRosset, 2015) (henceforth, AO), I argued that there are counterexam-
ples to each of the following two principles:
(DAO) If P analytically entails the existence of certain things, then a
theoty that contains P but does not claim that those things exist
is no more ontologically parsimonious than a theo1y that also
claims that they exist.
(GAO) If P analytically entails Q, then (PAQ) requires nothing more
of the world than does P.
John Borden has offered interesting criticisms of that argument. In this
short note, I reply to the criticisms and briefly indicate what I take their
lesson to be.
The putative counterexamples to GAO and DAO are of essentially the
same sort. For the sake of simplicity, then, I will focus on a putative
counterexample to GAO. In AO I asked readers to suppose that we had
stipulated that 'grassgreen' is to be a predicate that expresses the property
being green if, as a matter of fact, grass is green, and not being green
otherwise. To simplify the discussion, let's dispense with suppositions:
I do hereby so stipulate. I trust you know enough to recognize that grass is
grassgreen. Also, as you may know, the flesh of kiwi fruit, clover, and most
freeway signs in the U.S. are grassgreen. Neither roses nor violets are
grassgreen.
My argument that this stipulation yields a counterexample to GAO relied
on three premises.
A My stipulation is successful, where "success" in this sense entails that,
as a result of the stipulation, its content is true.
B "Disquoting" the content of a successful stipulation yields an analytic
truth.
Stipulations r1nd Requirements: Reply to Horden 75
C Requirements for the truth of a sentence of the form '7 is grassgreen'
are given by which individual (if any) is the referent of T and which
property is expressed by "grassgreen" in a straightfo1ward way: the
Given (A), we may conclude that, if grass is, as a matter of fact, green, then
"grassgreen" expresses beinggreen, and, otherwise, "grassgreen" expresses not
being green. Application of (B) then yields the analyticity of
(l) if, as a matter of fact, grass is green, then something is grassgreen
iff it is green; and, otherwise, something is grassgreen iff it is not
green.
It is then very simple to show that
(2) snow is white
<Ulalytically entails
(3) grass is grassgreen.
Application of (C), together with the fact that grass is, as a matter of fact,
green, yields the conclusion that (3) requires for its truth that grass have the
property being green. Since that requirement clearly goes beyond require-
ments for the truth of (2), we have a counterexample to GAO.
In bis comment on AO, Hordcn offers two reasons for thinking this
argument unsound. First, he contends that its conclusion is inconsistent.
Second, he contends that my argument faces a dilemma, depending on
which of two hypotheses about the meaning of (3) turns out to be true. I will
discuss each contention in turn.
Horden's charge of inconsistency is made on p. 69: "If'grass is grassgreen'
is guaranteed to be actually true, no matter what colour grass actually has,
then it cannot require of the actual world that grass be green. Even if we
gr<mt that the stipulation succeeds, it seems sensible to examine the seman-
tics of the novel term 'grassgreen' a little more closely before we embrace
inconsistency" (emphasis original). It is a little difncult to sec what is
supposed to be inconsistent here. The guarantee is epistemic, while require-
ments for the truth of a claim concern the metaphysical or semantic
question of what the world has to be like for the claim to be true. Given
(C), any residual air of inconsistency dissolves once we realize that the
success of the stipulation entails that the truth of (3) requires one thing if,
as a matter of fact, grass is not green and another thing if, as a matter of fact,
grass is green.
Horden does not explicitly say why an epistemic guarantee of the truth of a
sentence is supposed to be inconsistent with there being substantial
7G Louis deRosset
requirements for its truth. But he provides a clue when he sug_e;csts a schematic
biconditional governing requirements: "a sentence (/> requires of the world
that Riff (i) 'R' does not logically follow from sentences that are true entirely
because of rheir meanings, and (ii) whichever world is actual, if(/> is true, this
is partly because R" (p. 64). 1 Horden only tentatively advances this bicondi-
tional, but says that his argument assumes the following consequence of
its left-to-right direction (p. G4):
Ll if a sentence requires of the world that R, then, whichever world is
acttul, that sentence is true only if R.
With (LI) in hand, it's pretty clear why Horden thinks the argument against
GAO entangles us in inconsistency. Consider the relevant instance of (LI):
L2 If (3) requires of the world that grass is green, then, whichever world
is actual, (3) is true only if grass is green.
If the argument from (A)-(C) is sound, then, because grass is in fact green,
(3) in fact requires of the world that grass is green. l·fowevcr, if grass turns
out as a matter of fact to be purple-if, that is, a world in which grass is
purple turns out to be actual--then the soundness of the argument implies
that (3) docs not require that grass be green. This would contravene (L2),
which entails that (3)'s requiring of the world that grass be green be
insensitive to how things turn out to be as a matter of fact. So, the
contention that the argument involves us in an inconsistency can be made
good if we assume (Ll).
But (Ll) faces counterexamples. l have just flipped a coin. l hereby
stipulate that "TrueThat" is a syntactically atomic sentence that requires
for its truth that grass be green if the coin hndcd heads, and requires for its
truth that grass not be green otherwise. The stipulation passes all of the
standard tests for successful stipulation. In particular, it is conservative and
harmonious, and passes what in A 0 (pp. 141-4) l called Stevenson's con-
straint. I have gone on to use "TrucThat" in conversation with my col-
leagues, friends, and family. As you will see, I will even be using it in this
paper. Given its success, the content of my "TrueThat" stipulation is true.
So, (L l) has a false instance:
L3 If "TrucThat" requires of the world that grass is green, then,
whichever world is actual, "TrueThat" is true only if grass is green.
It turns out that the coin landed heads. This provides enough information
for you to recognize that TrucThat iff grass is green. If, however, a world in
1
I have benefited here from personal correspondence with Horden.
Stipu!r1tions and Requirements: Reply to Horr/en 77
which the coin landed tails is actual, then TrueThat iff grass is not green. In
general, we appear to be able, under the right circumstances, to successfully
stipulate the requirements for the truth of new sentences. What's more, we
have at our disposal devices for conclitionalizing on how things arc as a matter
of fact in the content of such stipulations. So, (Ll) appears to implausibly
restrict our stipulativc powers. It thus provides no reason to think that the
argument from (A)-(C) against GAO entangles us in any inconsistency.
Horden's second objection to the argument from (A)--(C) is that it faces a
dilemma,_ depcn_ding on what we take (3) to mean. In AO (p. 148), I argued
that the force ot the putative counterexamples was independent of which of
several plausible hypotheses about the meaning of, e.g. (3) turned out to be
correct, and otherwise avoided explicit discussion about the meanings of my
stipulated terms. A word of explanation for why we might want to avoid
claims about what (3) means may, however, be appropriate. The issue is that
there arc a dizzying array of different semantic values that might be assigned
to (3). Assume for illustration that the occurcncc of "grass" in (3) is a
singular term, so that (3) has the form "a is F," where a is a term and F is
a prcdicate. 2 Then it is plausible to claim that the sentence expresses a
Russellian proposition represented by (grass, being green). Certainly, a
proponent of DAO is in no position to deny that sentences like (3) express
such Russellian propositions, given how "Russcllian proposition" and
related vocabulary are used in the philosophical community. But such a
proponent is also in no position to deny that (3) has other semantic values,
including a truth value, a carnapian intension, a character (Kaplan, 1989),
or perhaps even a primary intension (Chalmers, 2002). Which, if any, of
these semantic values should we hold gives the meaning of (3)? lt seemed to
me that nothing interesting hung on which of many plausible answers we
might give to this question.
Horden disagrees. Here is his dilemma. If (3) means
(4) grass is green
then, Horden contends, (B) is false: though the stipulation succeeds and the
truth of (3) requires that grass have the property being green, (3) is not
analytic. Alternatively, if (3) means
(5) grass is green iff grass is actually green
then, Horden contends, (C) is false: the requirements for the truth of (3) arc
trivially satisfied and so cannot include that grass have the property being
2
lf.(3) lacks such a st.ructurc, then it will express some slightly more complicated
Russellian smgular proposmon, bur the point in the main text would be unaffected.
78 /,ouis deRosset
green. More generally, Horden's argument here might be taken to suggest
that there is no way of fixing a meaning for any sentence on which both (B)
and the analogue of (C) are true. That is, there is no way of fixing a meaning
for a sentence so that, like (4), it requires for its truth that grass be green,
but, like (5), it is analytic.
Let's consider each horn of the dilemma in turn. On the first horn, the
defense of GAO would require that (3) be synthetic, even though it has what
in AO (p. l 58) I called the trappings ofant1/yticity: the conventions governing
the use of (3) guarantee that it is entailed by certain sentences such that
failure to accept any of these sentences constitutes some measure of linguis-
tic incompetence. For instance, one of the conventions governing the use of
(3) is given by the content of my "grassgreen" stipulation. From those
conventions, instances of certain disquotation principles, and a little logic,
it's easy to work out that (3) is true.
This certainly makes it look as if (3) is analytic, and it seems to me that
there is little reason to deny that it really is analytic, short of antecedent
commitment to GAO. Thus, I would argue that the most plausible verdict
on this horn of the dilemma is that my "grassgreen" stipulation has fixed the
meaning of (3) in such a way as to make both (B) and (C) true.
Suppose, however, that I'm completely wrong about that. Suppose, that
is, that appearances in this case really are misleading in the way that
Horden contends: (3) has the trappings of analyticity but is not analytic.
Horden's contention here is an instance of a more general strategy for
defending DAO and GAO from the putative counterexamples. This
defense denies that the possession of the trappings of analyticity is enough
to guarantee analyticity. But the defense thereby gives up an important
methodological advantage on which the philosophical applications of
DAO rely. If a sentence can have the trappings of analyticity without
being analytic, then we can't rely on our appreciation of the trappings of
analyticity in cases like
(6) If particles Pi, ... ,p,, are arranged table-wise in L, then there is a
table in L
to warrant the application of DAO. Thus, a radical omologist who agrees
that particles Pi, ... ,p,, are arranged table-wise in L but denies that there
is a table there can admit that (6) has the trappings of analyticity while
reasonably denying that it is analytic. In AO (p. 159), I concluded on this
3 conversation with my colleagues, friends, and family. You know enough
basis that, on this response, "DAO is saved, but it's rendered toothless."
J Please see AO §4.2 for more extended discussion.
Stipulations and Requirements: Reply to Horden 79
A defender of DAO and GAO might reply by distinguishing mere uninter-
preted strings from meaningfol expressions of an interpreted hnguage. 1 The
idea is that (3) fails to be analytic because "grassgreen" is a mere uninterpreted
string: the content of my stipulation gives us reason to think that that string
will be true when it acquires a meaning, but no reason to think that any
meaningfol interpreted sentence is true. lt is only meaningfol, interpreted
sentences that are analytic or synthetic, so (3) is not even the right sort of thing
to be analytic. This reply runs afoul of the fact that "grassgreen" is already a
term being used meaningfully (if only by me) to characterize things. In the
second paragraph of this paper, for instance, I used it three times. On each
occasion, I asserted something true. Perhaps you count yourself as falling short
of ideal competence with the expression, but you know enough to recognize
the truth of (3). So, (3) is already a meaningfol sentence of an interpreted
language, and the content of my stipulation provides an epistemic guarantee
of its truth. It won't, then, fail to be analytic for want of semantic significance.
Consider now the second horn of Hord.en's dilemma, on which we
assume that (3) means the same thing as
(5) grass is green iff grass is actually green.
The defender of GAO holds that (5) requires nothing of the world. If this
claim is granted, then the semantic equivalence of (3) and (5) yields the
falsity of (C): like (5), (3) requires nothing of the world.
Horden is correct to think that (C) entails that (3) and (5) are semantically
discernible. In particular, the biconditional (5) won't bear an interpretation
on which its truth requires that grass have the property being green. So, if (3)
and (5) were semantically indiscernible, then (C) would fail because (3), like
(5), would also fail to require that grass have the property being green. I would
think, however, that the claim that (3) and (5) are semantically indiscernible is
independently implausible. (3) has semantic values that (5) lacks, including
the Russetlian proposition represented by (grass, being green). So, Horden is
correct to contend that (C) is, plausibly, false if (3) means the same thing as
(5); but it is not plausible to hold that (3) means the same thing as (5).
In any case, we can tweak the argument a little to avoid this horn of the
dilemma entirely, using a stipulation much like my "TrueThat" stipulation.
I hereby stipulate that "GreenThat" is a syntactically atomic sentence that
requires for its truth that grass be green if, as a matter of fact, grass is green;
and that grass not be green otherwise. I have gone on to use "GreenThat" in
about grass to recognize that GreenThat iff grass is green, so you could use it
4
Herc again I benefit from correspondence with Horden.
80 Louis deRosset Stipulations cmd Requirements: Reply to Horden 81
too if you were so inclined. My stipulation passes all of the standard tests for
successful stipulation. In p;uticular, it is conservative and harmonious, and REFERENCES
passes Stevenson's constraint. The stipulation takes the analogue of the second
horn of Hordcn's dilemma off the table. In fact, it takes the denial of the Chalmers, David]. "On Sense and Intension." In James E. Tomberlin, ed., Philosophicrtl
analogue of (C) off the table if the success of my stipulation is granted. If the l'enpectives: rangur1germdMind, volume 16, pp. 135-82. Oxford: Blackwell, 2002.
"GreenThat" stipulation is successful, then its content is true, so it requires for dcRosset, Louis. "An:tlyticity and Ontology." In K. Bennett and D. Zimmerman,
its truth that grass be green. For this reason, insofar as one is willing to eds., Oxford Studies in Metaphysics, volume 9, pp. l29-70. Oxford: Oxford
countenance the idea that linguistic conventions or stipulations fix meanings University Press, 2015.
fi)r sentences, my "GrecnThat" stipulation illustrates a way to fix a meaning for Kaplan, David. "Demonstratives." In Joseph Almog, John Perry, and Howard
a sentence on which both (B) and the analogue of (C) appear to be true. Wettstein, eds., Themes From Kaplan, pp. 481--563. New York: Oxford University
Press, 1989.
I have argued that neither the objection based on (Ll) nor the dilemma
ultimately threatens the cogency of the putative counterexamples to DAO
and GAO. But we shouldn't let these difficulties for the objections obscure
the more general lessons of Borden's discussion. Hordcn is admirably
explicit about the assumption (Ll) governing requirements for the truth
of a sentence on which his first objection depends. Though, as [ have
argued, (Ll) itself is implausibly restrictive, I believe that this is a fruitful
line of inquiry for the analyticity theorist to pursue. (Ll) is part of Hordcn' s
proposal for a partial theory of requirements. I think (Ll) itself is false, but
there arc other aspects of that theory that arc quite attractive. For instance,
on Hordcn's tentative proposal, the requirements for the truth of a sentence
are canonically given by a specification of the form
(7) Sentence S requires that qJ.
Further, the requirements for the truth of a given sentence are conceived by
Horden as playing an explanatory role: they specify the conditions in virtue
of which S is true (if it is). These seem like fruitful starting points for a
theory of requirements.
To my mind, the most promising line of defense for GAO and DAO will
follow Horden's lead: first, articulate, motivate, and defend a (perhaps
partial) theory of requirements; then, show that, on that theory, the pro-
posed counterexamples fail. Moreover, the discussion ofHorden's dilemma
has outlined the strategy that the defense must follow. The theory of
requirements on which the defense relics must be supported well enough
to motivate the rejection of the othe1wise plausible claim that such stipula-
tions as the ones introducing "grassgrecn" and "GreenThat" are successful. 5
Louis deRosset
University of Vermont

5
Thanks arc due to Mark Moyer and Kate N olfi for comments on earlier drafts of this
paper. Special thanks arc due to John Horden for hclpfol correspondence. Thanks also ro
Amanda Lowe and Caley Millcn-Pigliucci fr)r help in preparing the text.
 6
Colors as Primitive Dispositions
Hagit Benbaji

This paper offers an account of colors that combines prnmuv1sm with

dispositionalism. "Primitivist dispositionalism," as I shall call it, may sound
like an oxymoron. Primitivism takes redness to be a sui generis simple property
of an object, wholly borne on its surface, and fully revealed to us in visual
experience. 1 Dispositionalism takes redness to be the disposition of an object
2
to look red to normal perceivers under normal perceptual conditions. But it
is puzzling how a sui generis property, wholly borne-like shape-on the
surface ofa physical object, is essentially connected to experiences, and, from
the other direction, puzzling how a disposition to cause experiences is entirely
revealed in these experiences.
Primitivist dispositionalism is not only hardly intelligible, but also diffi-
cult to offer a motivation for. Perhaps primitivism is the view of colors that
best captures the phenomenology of color experience 3-colors do indeed
appear to be simple properties wholly borne on the surface of objects-but
there arc no such properties; only in Eden, it seems, was the apple "glori-
ously, perfectly, and primitively red" (Chalmers 2006, 49). By contrast,
while no one can deny that objects <li'e disposed to look colored, disposi-
tionalism has been criticized for misrepresenting the phenomenology of
color experience. 4 However we might want to analyze or unpack the concept

1
Although primitivism emerged relatively recently, there are now many diHcrent
variants: McGinn's impressionism (1996); Campbell's simple view (1993); Johnston's
hylomorphism (unpublished manuscript); Yablo's "naive objcctivism" (1995); Wcst-
phal's phenomcnalism (1987, 2005); Broackcs's "ways of changing light" ( 1997, 2007);
Gert's "unmysterious" primitivism (2008); Kalderon's pluralism (2008, 2013); Allen's
selccrionism (2007, 2009); Warkins's "a posteriori" primitivism (2010).
2
Dispositionalism has put forward by Locke. Contemporary advocates include Ryle
1949, Pcacocke 1984, McDowell 1985, Evans 1980, McGinn 1983, Wiggins 1987,
Johnston 1992. All page rderences to articles reprinted in Byrne and Hilbert 1997 refer to
that edition.
1
Sec Chalmers 2006, Johnston manuscript, McGinn 1996.
,; McGinn 1996, 541. See also Johnston 1992, 140-2.
8G Hagit Benbaji
of color, experientially speaking, the critics insist, colors simply do not look like
dispositions. Indeed, colors seem to be stubbornly non-dispositional. Combine
this phenomenological charge-colors don't look like dispositions--with the
metaphysical extravagance of primitive properties, and primitivist disposition-
alism would seem to be a non-starter.
This paper has three goals. hrst, to show how primitive dispositions are
possible. Second, to claim that colors are primitive dispositions. Third, to
argue that if the primitive property that is directly revealed to us in experience
is none other than the apple's disposition to look red, then objects do have
these Edcnic colors <1frcr all.
The key to achieving all three goals is the notion of an "appearance
property," 5 such as my aunt's looking young, her youthful appearance.
When I look at my aunt, I sec that she looks young, not that she is
young. That is, I see an appearance property: the property of looking
youthful. On my conception, appearance properties are to be understood,
not in terms of mental effects they generate in an observer-sensations,
qualia, and the like-but as genuine properties of physical objects. This
paper will argue that to see a color is to see an appearance property, just as to
see my aunt's youthful appearance is to see an appearance property.
I begin by articulating the basic insights worth saving from disposition-
alism and primitivism, as well as the tension raised by their conjunction
(section 1). In the main body of the paper, I then outline a model for
appearance properties that establishes the possibility of primitive disposi-
tions (section 2); spell out what colors have to be, namely, primitive
dispositions, for this model to be applicable to them; and address the
metaphysical implications of primitivist dispositionalism (section 3).
I close by replying to the phenomenological objection, showing that colors
do not look non-dispositional (section 4), and furthermore, colors do indeed
look like dispositions (section 5). Thus my argument may also shed light on
our inclination "to feel the pull of Revelation," 6 Russell's famous claim that
"I know the colour perfectly and completely when I see it" (1912, 47).
5 Suggested (for different purposel) in Pettit 2003, Noe 2004, Phillips 2006.
6Johnston 1992, 139. Several characterizations of this thesis have been offered.
Campbell takes it as saying that experience provides us with "knowledge, for each
particular color, of which particular property it is-the qualitative characters of the colors"
(2005, 4); according to Johnston, Revelation is the view that, "Canaty yellow is counted
as havinu just those intrinsic and essential features which are evident in an experience of
canary y~llow" (1992, 139); Byrne and Hilbert (2007) argue that Revelation implies that
r1ny necessary truth P about the nature of colors is given in experience, so that if it seems to
us (under normal viewing conditions) that P, then Pis true (Infallibility), and if Pis true,
then it seems to us that Pis true (Sclf~lntimation).
Colors as Primitive Dispositions 87
The resulting account of color is dispositional, in that there is nothing to
being red beyond looking reel. Nevertheless, by giving due credit to the
phenomenology of color experience that makes primitivism appealing, it
offers a way for us to remain on Earth, yet foe! like we're in Eden.
1. MOTIVATIONS AND PUZZLES
Primitivism claims that colors are more than dispositions. Regardless of what
that extra something consists in, this claim Rady contradicts rmy version of
color-dispositionalism. Likewise, a reductive dispositional account is straight-
fo1wardly incompatible with primitivism, since it purports to identify recl-
type experiences without invoking the concept of redness. Thus, primitivist
dispositionalism can accommodate only non-reductive clispositionalism,
which denies that experiences of red can be identified without invoking the
concept of redness, and some inspirations of primitivism. Specifically, I extract
each view's core motivation: from dispositionalism, the objective-subjective
duality of colors (section 1.1), and from primitivism, the claim that colors
are experienced directly (section 1.2). These motivations are, I will show,
compatible.
To say that these motivations arc compatible is not to eliminate the initial
conflict their conjunction may generate. Clarifying the tenets of disposi-
tionalism and primitivism does, in fact, bring to the fore a certain tension
between them (section 1.3).
1.1 The motivation for dispositionalism
Colors, like shapes, are properties of material objects, with all that this
familiar belief implies; for instance, that colors do not disappear the minute
we stop looking at them, when it gets dark, or when lighting conditions
change, and that they leave room for error and illusions. In this sense, colors
are objective.
In another sense, colors arc, unlike shapes, subjective. We understand
what it is for something to be a triangle if, say, we take it to be a closed
three-sided figure, that is, by having a non-perceptual description of it. By
contrast, a foll understanding of what it is for something to be red requires
an understanding of what it is for that object to look red (McDowell
1985, 203).
Setting aside the rubric of "dispositions," which may create the impres-
sion that the underlying insight is excessively complicated, all that is implied
by the dispositional account is that for an object to be red, say, is for it to
88 Hagit Benbaji
look red. The connection between colors and experiences is vindicated by
commonplace facts such as the following:
1. All we can do, <ltld all we need do, in reply to the question, "What docs
'red' mean?", is point to the apple and say "Look!", ·whereas shapes can be
taught by touch or geometry.
2. It seems to be the case that the congenitally blind cannot understand the
concept of redness, though they can understand shape concepts.
Why can't the congenitally blind understand what redness is? It does not
suffice to reply that redness implies looking red, for it is also true that
squareness implies looking square, and those with congenital blindness can
understand what squareness is. To explain this phenomenon we have to
acknowledge that what it is for something to be red is simply for it to be such
as to look red. This-namely, that being colored is solely a matter of
appearances, of how things look to us, rather than a matter of having
some underlying geometric structure, or non-perceptual, hidden (physical,
chemical) structure-is the grain of truth common to all versions of the
dispositional account. Hence so many believe that "nothing can be seen as a
colour without being seen as essentially connected with vision" (Boghossian
and Velleman 1989, 96).
This is what I call the 'objective-subjective duality' of colors, which is the
core claim of dispositionalism. What it amounts to is that colors, inasmuch
as they are visible properties of objects, are objective, while inasmuch as
they are essentially connected to experience, they are subjective. Although
colors are, in a sense, mind-dependent, they are, as Allen aptly puts it, "still
properties that exist without-in the spatial sense of being outside of--the
mind" (2007, 142).
1.2 The motivation for primitivism
On the face of it, primitivism is focused on what colors are, that is, on the
nature of colors, only insofar as that nature is derived from how colors arc
perceived-what it's like to see redness. Indeed, primitivism seems to follow
from Revelation: the nature of colors is wholly revealed in experience, but
since it is not revealed to us in experience that colors have microstructures, or
(allegedly) that they are dispositions, colors must be primitive, viz. sui generis
properties.7
Granting that Revelation implies primitivism, why should this mean that
primitivism is true? Does the phenomenology of color experience really
7
This argument is discussed in Byrne and Hilbert 2007.
Colors as Primitiue Dispositions 89
make any difference to our philosophical analysis of color? 8 For it might be
argued, so what if dispositions or physical properties are not fully revealed to
us in experience, or cannot be seen at all? If an otherwise-well-supported
view, whatever it is, docs not mesh well with color experience-so be it.
A better motivation for primitivism than the radical requirement that the
'nature of colors be fully revealed in experience is the more modest require-
ment that at least part of the nature of colors be revealed in experience.
Although weaker than Revelation, "Manifestation," as Kalderon calls it, is a
strong claim. Manifestation claims that when we perceive redness, we ipso
facto perceive at least part of its nature; we perceive what red is like (e.g. that
it is reddish and not at all bluish). Manifestation claims that colors arc what
they appear to be.
To say that redness is what it appears to be is to say that redness determines
the phenomenal character of our experience of it-what it's like to see redness.
Thus, Manifestation implies "Inheritance" (Kaldcron 2007), 9 the view that
"the qualitative character of a colour-experience is inherited from the quali-
tat.ive property of the colour" (Campbell 1993, 189). 10
8
Some have taken phenomenological considerations seriously enough to withdraw
their support for dispositionalism and instead endorse primitivism. Johnston, unpub-
lished manuscript; McGinn 19%. Other philosophers are not bothered by the charge that
their view violates our commonsense experience of color; sec Johnston 1992, Boghossian
and Velleman 1989, Chalmers 2006, Misccvic 2007.
9
As characterized by Johnston (1992) and Kaldcron (2007), my conrention that
Manifestation implies Inheritance is, perhaps, controversial. Johnston argues that color
dispositions satisfy Manifestation in virtue of the fact that experiences of colors are the
manifestations, i.e. the triggering, of the dispositions. "About any disposition ofobjects to
produce a given experience, it is plausible to hold that if one has an experience of the kind
in question and takes that experience to be a manifestation of the disposition in question,
one thereby knows the complete intrinsic nature of the disposition" Oohnston 1992,
167). On Johnston's view, Manifestation is satisfied by any account of visual experience
one accepts, whether construed in terms of mere sensations, mental representations, or
primitive relations of subjects to objects. However, I am arguing that without a proper
understanding of visual experiences, Manifestation cannot be true. If experiences arc mere
sensations without intentionality, a disposition to cause red sensations is manifested only
in the way a disposition to cause pain or nausea is manifested Oohnston's example). This
raises the concern that experiences of redness cannot be distinguished from bodily
sensations Oohnston 1998).
10
Inheritance alone is satisfied by reductive reprcsentationalism that accepts that the
phenomenal character of the experience of something's looking red is simply a matter of
its representing an object as red, representation being a matter of the perceiver's causal or
teleological connection to red objects. A physical property might determine the phenom-
enology of color experience in this sense, without that property's manifesting its nature in
experience. Tye and Dretske argue that physical redness determines the phenomenal
character of_ experience, and thereby eliminates qualitative properties not only from the
world, but from the mind as well. Thus, according to Tye and Drctskc, Inheritance does
not imply Manifestation.
90 Hagit Benbaji Colm:r as P1imitive Dispositions 91

Since the claim that Manifestation implies [nheritance is the main recap, the epistemic insight is that if colors are not directly perceived, the
rationale for primitivism, let me accept the following constraint on colors: empirical world itself becomes invisible, unknowable, or even unjustified.
MI (Manifestation cum Inheritance): a color is a property of an object,
such that by being manifested in experience determines the phenomenal
1.3 The tension
character of that experience.
The central claim of this paper is that colors are pnmmve dispositions.
MI seeks to uphold both the irreducibility of colors and the idea that they
Primitive dispositions are irreducible dispositions that satisfy MI. Let me
arc directly perceived. First, like Revelation, MI implies primitivism. On the
explain why the core motivation for primitivism, namely MI, seems to
one hand, Manifestation implies that colors cannot be solely a matter of
conflict with the core motivation for dispositionalism, namely, the subject-
physical structure, since experience does not reveal to us any physical aspects
ive aspect of color, whereby colors are essentially connected to experience.
of colors. On the other, Inheritance implies that colors cannot be reducible
Both primitivists and dispositionalists accept the following conditional: if
dispositions, since reductive dispositional accounts identify color experi-
colors are essentially connected to experiences of them, colors cannot
ences independently of any putative color property of things in the world,
explain these experiences. The concern that underlies this conditional is
and thus no such property can determine the phenomenology of color
that if we explain what it is for something to be red by saying that it looks
experience. Hence, colors are sui generis. Second, MI explicates what it
red, "we cannot explain why something looks reel by saying that it is red, for
means to see a property of an object directly. Hence, I take primitive colors
this involves an explanatory circle" (McGinn 1983, 15).
to be properties of objects that satisfy MI.
Dispositionalism accepts the antecedent, that is, the claim that colors are
Thus far, it might seem that the basic reasoning behind the "so what?"
essentially connected to experiences, hence, the consequence: it deems
retort to Revelation has not been undermined, but merely reformulated: "so
colors to be explanatorily redundant. The implication is that colors can
what if the nature of colors is not manifested to us in experience? The nature
neither satisfy MI, nor be directly perceived.
of water is not manifested in experience, yet this is no reason to claim that
water is more than H 2 0." The nature of color might not, it could be argued, 1. Inheritance is an explanatory thesis. It claims that the phenomenal
be manifest in perception at all, because it is the upshot of either a scientific character of a color experience "depends on and derives fi'om the qualita-
discovery, or philosophical analysis (McGinn 1983, 135; Cohen 20 IOa) . 11 tive character of the presented color" (Kalderon 2007, 3). To render
Nevertheless, the "so what?" retort is not apt in the case of colors. We colors explanato1y powerless is to deny Inheritance.
might not care if the nature of water is not manifest to us in experience, but 2. Since Manifestation implies Inheritance, it is false as well.
if colors are not manifest to us in experience, no property is. Primitivism is
3. Since MI gives content to the idea of direct perception, colors are not
driven by the insight that were colors not directly revealed to us in experi-
directly perceived. Indeed, we see solubility by seeing a manifestation of
ence, we would be unable to gain any direct knowledge of the mincl-
solubility-the sugar's dissolving, just as we see water by seeing its liquid-
inclepenclen t world (Campbell 2002), and worse, we could not even take
ity, transparency, etc. If Manifestation is false, the manifestations of color
our experiences to supply reasons for beliefs about the world (Brewer 1999).
dispositions do not reveal the nature of the color any more than the visible
The underlying idea here is that perception of color is not merely a form
properties of water reveal its nature. Thus, we do not sec colors directly.
of "Morse code transmission" Qohnston 1992, 166), "a reliable effect of
properties whose nature remains unknown" (Kalcleron 2007, 587), as The tension between MI and the subjectivity of colors seems to be
perception of water might be. grounded in the idea of a property that explains the very experiences
I cannot even begin to fully justify these epistemic claims here. My point which constitute it, not in any particular construal of these experiences,
is to emphasize that primitivism is rooted in an epistemological or cognitive i.e. reductive or non-reductive.
outlook, rather than solely in the ideal of phenomenological adequacy. To The dispositionalist's modus ponem is the primitivist's modus to/lens. Since
the thrust of primitivism-Inheritance, Manifestation, and the very idea of
direct experience-is lost if we reject the explanatory force of colors, primi-
11
Some argue that we can see the wrongness of an act, though it might turn out that tivism concludes that colors are not essentially connected to experiences. This
this wrongness is entirely a matter of some complex consequentialist feature. explains why it is so important for primitivism to reject the consequence
92 Hagit Benbaji
of the conditional. If the way things look depends on their colors as much
as it depends on their shapes, colors are prior to, and independent of, our
experiences of them.
2. APPEARANCE PROPERTIES AS
DISPOSITIONS TO APPEAR F
To explain the possibility of primitive dispositions, let me introduce the
notion of an rtppemmzce property (section 2. l) and claim that to see a
property as an ttppearance property is to see it as a disposition of an object
that is inherently connected to visual experience (section 2.2).
2.1 Appearance properties
I look at my colleague, who is only thirty, and it occurs to me that he looks
old. He does not appear to be old; he appears to me to look old-he has the
timeworn look older people have. Being aware of his actual age, the property
I notice when I see my colleague is not that of "being old," but that of
"looking old."
Once we become aware of the phenomenon of attending to the appear-
ances of things, it seems ubiquitous. As Phillips (2006) remarks, we spend
much time glancing at our appearance in car and shop windows, and each
time we look quite different. My face looks more animated, more natural,
from this angle, or in daylight as opposed to fluorescent light. We note how
much she looks like her sister, how vodka looks like water, or how a certain
face looks familiar. In all these cases, we intuitively take appearances, the
way things look, to be bona fide properties of the things in question.
"Looking like water" is taken to be a property of vodka, just as "looking
old" is a property I ascribe to my colleague. As O'Shaughnessy puts it, "It
was the look of Helen's face, rather than its chemical or electrical or
pheronomic properties, that caused such a horror" (2000, 570).
This talk of appearance properties raises an immediate objection, namely,
that appearnnce properties are not genuine properties at all, but mere
illusions. Upon seeing a stick in water, I do not sec a new property of the
stick-"looking bent"-but rather, an illusion: a straight stick that appears
bent. Similarly, my seeing my colleague's aged appearance is not seeing a
property, but being deluded as to his real age. More generally, seeing an
appearance property is not, on this view, genuine perception of a property.
Of course, I need not be misled by the illusory experience, and I may be
sophisticated enough not to believe that the stick is bent. But it is
Colon flS l'rimitiue Dispositions 93
nonetheless an illusion. The objection, then, is that any statement asserting
the occurrence of an illusory experience (IE), such as
(IE) It merely seems to me that the stick is bent
can be systematically converted into a statement asserting the occurrence of
a veridical experience (VE) of an appearance property:
(VE) I sec that the stick is bent-looking.
Yet clearly, this linguistic maneuver cannot endow the stick with a genuine
property-so goes the objection.
Hence the challenge confronting one who seeks-as I do--to uphold the
existence of appearance properties is to distinguish them from illusions. 'This
challenge can be met by showing that we can distinguish between real and
illusory appearance properties. Let us deem an object's "normal" appearance
to be the way the object looks under normal perceptual conditions. 12 In the
case of the stick, since the normal appearance of the shape always manifests
the stick's real shape, the only way the appearance can be misleading is if the
stick's normal appearance changes due to its being viewed under unusual
conditions, e.g. submerged in water. This is indeed a case of illusion,
inasmuch as the water alters the stick's normal appearance so that it appears
bent. On the other hand, we might be misled about someone's age even
without any change in his normal appC11rance, as in the case of my colleague.
Purely observational properties, such as shapes, cannot be separated from
their appearances, that is, from the way they look, whereas properties that
are not purely observational are separable from their appearances. Inasmuch
as the connection between being old <md appearing old is not as dose as that
between being straight and appearing straight, the property of being old is
less observational than that of being straight. It is observational, for there is
indeed an appearance associated with it, but this appearance can be instan-
tiated even in cases where the instantiating individual is not old. In such a
case, I see the look associated with oldness, without seeing oldness; the look
has a life of its own, as it were.
What makes the submerged stick a case of illusion is the fact that water
alters the stick's normal appearance. By contrast, my colleague's normal
appearance hrls not changed flt all, and thus, though we cannot see that he's
not as old as he looks, making this a case in which we might be misled by an
appear;mce, we do nevertheless see something else-namely, the look of
agedness, which is a genuine appearance property.
12
I am assuming that there is no non-circular description of the best perceptual
conditions for seeing colors and shapes.
94 Hr1git Benbrtji
This reply establishes a distinction between an illusion and perception of
an appearance property, thereby making itself vulnerable to the deflationist
concern about hypostatization of"looks." After all, there is a simpler way to
distinguish the appearance of agedness from the bent-stick illusion: when
I see that my colleague looks old, I do not sec any special kind of property,
any putative "look;" rather, I see wrinkles, a slight shuffle, and other visible
properties on which "looking old" supervenes. There is no need to reify
appearances in order to distinguish the bent-stick illusion from my old-
looking colleague.
The bold reply to the deflationist concern is that it involves phenomeno-
logical falsification. When I see that my colleague looks old, I do not merely
see lines or brown marks, but wrinkles and age spots. What is it to see the
lines as wrinkles or the brown marks as age spots? It is to see them as
expressions of oldness, signs of age. And this is precisely to see my col-
league's aged "look."
The modest reply points out that even if talk of "looks" is redundant in
the case of oldness, appearance properties cannot be thoroughly eliminated.
Consider first the famous example of aspect seeing. We can shifr at will from
seeing the duck-rabbit as a cluck (Figure 6.1 ), to seeing it as a rabbit, and we
cannot specify the content of the experience of that shift without ascribing
to the picture a certain "look." We see the picture first as rabbit-shaped and
then as duck-shaped, so we see that something in the figure changes. But we
do not see the figure as changing its shape; rather, part of the experience is
that the figure remains unchanged when we change perspective. What
changes is the way it looks, the picture's appearance: from one point of
view it looks like a cluck, from another it looks like a rabbit.
The language of appearance properties is, then, indispensable to a plaus-
ible account of ambiguous figures.
Aspect shifting is exemplified (in a way) in the ordinary phenomenon of
constanc')'. I look at a round coin, and then view it from another angle. The
coin looks round when it is in front of me, or when I look at it from above,
Figure 6.1 A famous example of aspect seeing
Colors t!S P1imitioe Dispositions 95
but when I see it from an oblique angle, it looks elliptical. In contrast to the
case of the stick in water, in which the real shape cannot be seen, in this case
I sec all along that the shape of the coin has not changed, that is, I sec that
the coin remains round through all the changes in viewing perspective.
Indeed, it is an integral part of the capacity to perceive shapes that we can
keep track of this property through movements in egocentric space. Note:
we do not merely believe the coin to be round through all the changes in the
observer's position, but rather, we see its shape as constant through these
changes. Yet in gazing at the coin, something does seem to change; what is
it? The only property that can be experienced as changing here, I submit, is
the look of the shape.
Let me briefly consider deflationist strategies that try to explain away the
reification of looks, and why these strategies fail.
(a) Denial of constancy: we do not really experience the coin as having
one constant shape, but as having changing shapes. Reply: to reinforce the
claim that the coin's shape is not perceived as changing, Siewert points to
cases in which things clearly appear to change shape. "Gaze at the shapes in a
lava lamp. Watch as a balloon is inflated. See someone's mouth break into a
smile" (2006, I 0). By contrast, the coin does not appear to change its shape.
Phenomenologically speaking, there is a difference between experiencing
something's changing shape, and experiencing a constant shape the appear-
ance of which fluctuates.
(b) Denial of changing appearances: we do not see that the coin usually
looks elliptical when viewed from an angle; we cannot see that the coin
remains round all along, and at the same time experience it as elliptical.
Rather, we can only experience it as elliptical from a "detached" point of
view. u Alternatively, it might be conceded that there is an elliptical aspect
to the situation, that is, the projection of the coin against a hypothetical
screen, though it takes some practice to be aware of this elliptical aspect.
After all, we are "not inclined, the tiniest bit to take the tilted penny to be
elliptical" (Smith 2002, 182). Reply: notice that even those who deny that
the coin usually looks elliptical do not do so unreservedly-they ofter excuses
(conflicting appearances imply different points of views), and alternatives
(an elliptical aspect to the situation). Bur the excuses are unnecessary, and
the alternatives are not phenomenologically apt. On the view I am arguing for,
there are no conflicting experiences-we do not experience the coin as round
and as elliptical at the same time; rather, we experience the coin as round and
as elliptical-looking at the same time, hence there is no need to invoke a shift
in perspective. The suggested elliptical aspect of the situation is not a
13
Kelly 2004, Smith 2002.
96 Hagit Benbaji
satisfactory alternative, because phenomenologically, it is not merely that the
14
situation has an elliptical aspect, but that it has an elliptical appeanmce. As to
the epistemic argument that we are not inclined, even "the tiniest bit," to
believe that the coin is elliptical, this is fully in line with my view: this
disinclination is rooted in the fact that we experience the coin as looking
15
elliptical, not as being elliptical.
(c) The changing property is internal: it is a property of our experience
(sense-data theorists), or a sensory property of regions of the visual field
(Peacock 2008). Reply: even apart from the contentious appeal to sense
data, this answer does not accord with the phenomenology of experience.
It is the coin's properties that we experience as changing, not sense data or
any other objects that might be distinct from us, yet internal, or mind-
dcpendent.
(d) The changing property is the coin's location relative to the perceiver:
the round coin looks round and tilted away from me. Reply: I do not deny
that egocentric locations are represented in experience; I deny that egocen-
tric location exhausts the content of the visual experience of the coin as
viewed from that location. It is true that in one case the coin appears to be
facing straight ahead of me, whereas in the other, it appears to be tilted away
from me. But something else differentiates the content of the two experi-
ences: in the one case, experience presents the coin as round, in the other,
experience presents it as looking elliptical. Many figures are not impacted in
this way by movement in egocentric space. Thus, a tilted "A" changes place
relative to the perceiver, but as opposed to a tilted square or a tilted circle,
still looks like an A (Macpherson 2006, 91) (Figure 6.2).
By contrast, viewed from an oblique angle, the coin's shape does have a
different appearance than when viewed frontally. So a rotated circle looks
like an ellipse, and a tilted square looks like a diamond, but a tilted A still
looks like an A.
14 Someone congenitally blind can know that the projection of the coin against a
hypothetical screen is elliptical. However, she would presumably not know what it is for
the coin to be experienced as both round and as having an elliptical appearance.
15 A "parsimonious view of looks" has recently been suggested by Martin (20 I 0).
This view reduces appearance properties to objects' visible properties: colors, shapes,
sizes, etc. On this view, since the coin docs not change shape, it also doesn't change in
appearance. The parsimonious view is thus an example of strategy (b): \~hat chan~es is
how the coin strikes us, that is, our psychological state, and the assemlnlity of d1ffe:·ent
claims about the coin that express those psychological states, e.g. the assertibiliry ol the
claim that the coin looks elliptical when tilted (Martin 2010, 221--2). [cannot do justice
to Martin's rich semantic proposal here, but the general arguments against strategy
(b) apply to his view as well.
Colors tis Primitive Dispositions 97
Figure 6.2 Tilted 'A'
(e) The changing property is the disposition "to generate in us an instance
of the type of perceptual state we undergo when perceiving an ellipse straight
on" (Cohen 20 lOb, 17). This property is the input of "the computation
whose output is a representation with the content that there is a distal
instance of round" (17). What seems to change is not the distal shape, i.e.
roundness, but the net perceptual signal transduced (to which distal shape
and visual angle are among the contributing factors). Reply: even apart from
its controversial assumption that "causally intermediate states in the sub-
personal visual system" (17) have representational content, the reply falsifies
the phenomenology of the experiences in question. As I already emphasized
(in (c) above), the experience presents aspects of the coin as changing, not our
sensations or perceptual signals at sub-personal levels.
(f) The changing property is "situation-dependent" (Schellenberg
2008). Reply: indeed, the coin is elliptical only in a given situation, and it
is round independently of the situation. Yet in the situation where the coin
is tilted, the coin is both elliptical and round. But admitting that one
property is situation-dependent and the other situation-independent docs
nothing to resolve the puzzle of contradictory properties. Adding "in the
given situation" doesn't make it okay to say that the coin is elliptical.
(g) The changing property is "apparent or perspectival shape" (Noe
2004): we experience a change in apparent shape, together with a constant
(plain old) shape; we experience the circulariry of the coin "in its merely
[apparent] elliptical shape" (167). Reply: the question is what the property
of being elliptical in "apparent" shape is. If it is simply being elliptical from
here (164), the puzzle remains, because the coin also looks round from here.
So it has to be "the shape that would be exactly hidden by an elliptical patch
placed in a plane perpendicular to my current line of sight" (Siewert 2006, 7).
But that shape is none other than elliptical! Siewert suggests that talk of
an "apparent" shape is redundant; all we need is the following counterfactual:
"ifyou were to block the appearance of the plate [coin], you would need to
put differently shaped patches in the way, given the change in how the plate
[coin] appears to you as it turns" (2006, 8). Ifl understand Siewert's construal
of Noe's suggestion correctly, it implies that the appearance of the coin
98 Hagit Benbaji
changes-it changes in a way that is captured by differently-shaped patches.
'fhat is to say, what is changing is the coin's elliptical look.
Once we concede the existence of appearance properties, the answer to
the puzzle of contradictory properties is straightfotward: we experience the
coin as round, and as elliptical-looking. Let me emphasize that my purpose
here is not to provide a full or even a partial account of either constancy or
aspect shift. Nor do I utilize the phenomena to argue for the perceptual or
epistemic primacy of"looks" (as do Noe 2004 and Schellenberg 2008). My
appeal to constancy is made solely to lend credence to the possibility of seeing
somethings "look"-·an aged look, an elliptical look, and so on. The capacity to
perceive the same shape through changes in its appearance presupposes the
distinction, within visual experience, between visible properties themsclves-
squareness, roundness, etc.-and their more fine-grained appearances, for
example, the way round coins look from an oblique angle. Constancy assists in
demonstrating that visible properties such as shape and size can be associated
with finer-grained appearances, which are sensitive to changes in spatial
location. My conclusion is that to solve the puzzle of contradictory properties
we must recognize in the existence of appearance properties that explain our
experience of constancy and change, and concede that we see these fincr-
grained properties as appearance properties.
2.2 Appearance properties are dispositions to appear F
So far, I've shown that some properties are appearance properties, and that
we see them as such. I now want to argue that an appearance property is a
disposition.
Reflection on experiencing appearance properties brings to the fore two of
their most distinctive characteristics.
( 1) Objectivity. The ellipse-like look seems to be a property of the coin, not a
mental experience of some kind. We do not experience such "looks" as
properties of our experiences (Noe 2004, 85). Similarly, my colleague's
aged look is not experienced as a mental entity, but as an objective
property inhering in my colleague. Neither the coin's elliptical look,
nor my colleague's aged look, disappears when I avert my gaze. The
elliptical look is the same for any perceiver viewing the coin from that
specific angle (Schellenberg 2008, 11). I may fail to see the coin's elliptical
look (because it is dark, or because I'm viewing the coin from here,
16
as opposed to there), but it can be seen when conditions are right.
16 In the other direction, something can appear to have a certain look, though it docs
not in fact have that look, so there is a gap between seeming to have a certain appearance
Colors as Primitive DiJpositiom 99
As Austin puts it: "I am not disclosing a fact about myself, but about
petrol, when I say that petrol looks like water" (1962, 43).
(2) Subjectivity. Even so, the experience of seeing looks, that is, things'
appearances, is not just like the experience of seeing shapes, say. The
second distinctive characteristic of experiencing an appearance
property-the coin's elliptical look, say-is that the property is experi-
enced only as a look, as a property whose nature is to appear a certain
way. This is the import of constancy: we experience the changing
appearances as appearances (looking round, looking elliptical). \Y/e
don't experience the coin as being elliptical, but as looking elliptical. 17
To recapitulate, the two essential features of appearance properties are
objectivity (appearances arc-and are experienced by us as-genuine prop-
erties of physical objects) and connection to experience (an appearance
property is-and is experienced by us as-a way of appearing). But for a
property to have these two features amounts to its being, and being experi-
enced as, a response-disposition. As I've already emphasized (in section l. l),
all that is implied by the dispositional account is that redness is a property of
objects that is nonetheless essentially connected to visual experience. Hence,
to experience the coin as looking elliptical is to experience it as looking a
certain way (viz. elliptical) under certain conditions, or in other words, to
experience it as being disposed to look elliptical. 18 I suggest, then, that an
elliptical appearance is a disposition, of an object, to appear elliptical, and it
is experienced as such.
Yet the identification of an elliptical appeamnce with a disposition to !Mk
elliptical might seem confusing. If; as I suggest, the look is the appearance,
how can it be a disposition to have an appearance? The appearance seems to
be the manifestation of the disposition; if not, i.e. if it is the disposition
itself, what, exactly, is the manifestation of this disposition? Furthermore, if
something is a disposition, then there is a property that is exemplified when
and having that appearance (she doesn't really look old, she just hasn't slept for three
nights in a row).
17
Notice that this claim, viz. that "looks" arc properties whose nature is simply to be
manifested visually in some specific way whenever the conditions are right, describes how
we conceive "looks," insofar as it describes how we experience "looks." It is the special
nature of being a property whose essence is just its appearance that we cannot conceive F to
be a "look" without seeing it as such.
18
Although Schellenberg 2008 argues that these relational and perspcctival properties
are--crucially---not to be analyzed in terms of how things look. Her reason for this denial
seems to be that it is "e<L>y to imagine" (72) the illusion of an intrinsic property
(cllipticalncss) when what is perceived is a situation-dependent property (looking
elliptical). But the fact that we might confuse :m appearance property with an intrinsic
one does not show that in ordinary cases we do not perceive the former as a "look."
100 Hrtgit Benbrtji
the characteristic manifestation of the disposition occurs. That property
cannot be the very disposition itself, for were that the case, it would be a
disposition that is always triggered. So on this reasoning, it isn't the dispos-
ition itself that appears to us, but rather, its characteristic manifestation.
This objection is the key to understanding the relevance of the
appearance-property model for the idea of experiencing a disposition dir-
ectly, and thus, to understanding the possibility of primitive dispositions.
The objector rightly assumes that when the characteristic manifestation of a
disposition occurs, say, sugar is put in water and dissolves, then the property
that is exemplified is "dissolving in water," that is, the manifestation of
solubility. But a visual disposition is unlike solubility, in that it is necessarily
manifested when its mf!rzijestations occur.
To clarify this point, we need to disambiguate the term "manifestation."
A property, the table's squareness, say, is manifested to me in visual
experience, meaning, it is presented or revealed to me in visual experience.
[n the more technical sense, the term "manifestation" is used to refer to the
triggering or exercise of the disposition (e.g. dissolving in water is a mani-
festation of solubility). 19
Since the manifestations-in the "trigge1ing" sense--of a visual disposition
are the experiences of tl1at visual disposition, the disposition manifests-in the
"presenting" sense-itself when its manifestations (in the triggering sense)
occur. The disposition to look elliptical can be an appearance property of say,
a coin, without always being triggered, because the disposition to appear
elliptical manifests itself (in borh senses of "manifestation") only under
specific conditions (when tilted, for example).
So what happens when I see a coin from an oblique angle twice within,
say, an hour? I have two different experiences of the disposition to look
elliptical. But I don't see two different manifestations (in the triggering
sense) of the disposition in question (the disposition to look elliptical) as
I would vis-a-vis the disposition to be breakable, were a window to be
cracked twice during that time, while I'm looking at it. Rather, in having
different (token) experiences, I see the srtme efliptiml f!ppertrf!rzce, i.e. a
property of the coin. The same property is revealed to me twice.
We now have a model for what it is for an object to have an appearance
property-a look-as, for example, when we see the coin's elliptical look.
a coin's disposition to look elliptical = a coin's elliptical appearance
manifestations of the disposition to look elliptical = experiences of the
coin as looking elliptical
19
Both the term "appearance" and the term "look" arc also ambiguous-they can
denote a property of ;m object, as well as a property of a subject (that is, an experience).
Colors tis Primitive Dfrpositions 101
According to this model, the disposition is manifested (in the sense of
"triggered") by causing experiences of itself, and these experiences manifest
to us (in the sense of "reveal to us") that disposition, namely, the coin's
elliptical look. The disposition is simply the coin '.r elliptical f!ppeamnce,
which lies wholly on its surface.
Let me emphasize how far this model is from the traditional view, on
which the relevant dispositions arc dispositions to affect experience. On
the traditional view, there is no genuine appearance property on the
object's surface-the appearance property is reduced to the shape of the
coin and internal experiences of the perceiver. On the model proposed
here, there is, literally, an elliptical appearance, a dispositional property,
that is manifested (i.e. presented) on the surface of the coin when I look at
it. Looking at the coin, I have an experience of it, which is a manifestation
(i.e. a trigger) of the disposition; but the experience does not somehow
come between me and the coin. Rather, the experience reveals to me the
coin's elliptical appearance.
On my model, an ellipse's disposition to look elliptical is not identical
to a (tilted) coin's disposition to look elliptical. The manifestations of the
former disposition arc experiences of the ellipse; they present an ellipse to
us. The manifestations of the latter disposition are experiences of an
elliptical look; they present the disposition itself. This might seem to be
an unwelcome conclusion. Under some circumstances one might be
unable to tell the difference between the tilted coin and a fronutl ellipse,
and the explanation for this inability is that the experience of the coin
from an angle and the experience of a frontal ellipse are introspectively
indistinguishable.
However, as I've already argued, the intuition regarding the phenomenon
of constancy is that the tilted coin does not look like an ellipse viewed from
the front, because we also experience the coin as constant in shape, namely,
as round. Since we see the elliptical appearance of the tilted coin together
with its roundness, the experience of the tilted coin and that of a frontal
ellipse are introspectively distinguishf!b!e. The rationale for introducing
"looks" or appearance properties to begin with is to explain differences in
our experiences of the coin in terms of properties whose co-instantiation does
not constitute a contradiction (being round, looking elliptical), rather than
in terms of the paradoxical co-instantiation of incompatible properties
(being round and being elliptical).
Granted, experiences of the coin from an angle and of an ellipse viewed
frontally may be introspectively indistinguishable, to some observers, under
some circumstances. This is perfectly compatible with the model of appear-
ance properties proposed here. After all, the experience of the coin from an
angle presents the coin as looking elliptical; it is no surprise that one might
 Hagit Benbaji
102
confuse it with an ellipse. But this happens when we <lo not experience the
tilted coin as also being round: on this scenario, we arc under the illusion
that it is an ellipse. Ordinarily, however, observers familiar with the phe-
nomenon of constancy have enough cues to see the coin not as an ellipse,
but as looking elliptical.
The model of appearance properties enables us to understand how a
primitive disposition is possible. The coin's elliptical look is none other
than a primitive disposition of the coin. Let me now articulate how the model
eliminates the puzzling features of primitive dispositions.
( 1) Inheritance. The coin's appearance can determine the phenomenal
character of our experience in the explanatorily ambitious way required
by Inheritance: we explain our experience of the coin in terms of its
appearance-a property of the coin's surface. Indeed, Inheritance-that
is, explaining the phenomenology of constancy and change in terms of
appearance properties of objects-is, in fact, the main thrust of the
model. Similarly, the phenomenal difference between an ellipse viewed
from the front and a tilted coin is explained by the differing properties
of those objects, namely, ellipticalness and having an elliptical appear-
ance, respectively.
(2) Manifestation. The nature of the property is experientially transparent-
we perceive the tilted coin as it is, namely, as looking elliptical, not as
being elliptical.
(3) Direct experience. The coin's elliptical appearance is directly m<mifested
to me in experience; it is not that I see it by seeing its manifestations. On
this model, experience reveals the coin's elliptical appearance to me
20
directly, just as it directly reveals the coin's roundness.
The model for appearance properties is thus able to accommodate the
notion of a disposition that satisfies Ml, that is, the notion of a primitive
disposition.
20 Thus, one might argue that I can see that a piece of fruit is such that were I to pick it
up, it would be squished by my hand; or I see that it is such that pushing it will cause it to
roll. Its tendency to be squished or to roll is visible (to a suitably sophisticated perceiver).
This notion of a visible disposition is weaker than the one I am proposing, since it implies
that I can see the coin's tendency to look elliptical when tilted even when it is not tilted,
but simply in front of me. This notion is close to Noe's idea that we sec patterns or
regularities in the way things look. But on the model developed in this paper, by contrast,
when the coin is tilted, I can see directly its elliptical appearance-not merely see tts
potentiality or regularity.
Colors as Primitive DiJpositions 103
3. WHAT COLORS ARE: APPLYING THE MODEL
I have argued that experiencing appearances-an elliptical look, a youthful
look, and the like-provides a model for the experience of visual disposi-
tions, dispositions to appear a certain way. The dispositional account
identifies redness with the disposition to look red, so it is natural to apply
the model suggested in the previous section to colors. In what follows,
I apply the model to colors (section 3.1), and explain the metaphysical
implications of ~loing so, thereby distinguishing the appearance-property-
based account of colors from dispositionalism and primitivism (section 3.2).
3.1 Applying the model
'fhe core contention of the dispositional account is that what it is for the
apple to be red is for it to look red to normal observers under appropriate
conditions for seeing color (I will omit the latter qualifying clauses in the
following equations, and often, thereafter):
(Dl) redness =the disposition of an object to look red
The model proposed in the last section explicates the dispositional nature of
a tilted coin's elliptical appearance. The relevance of the model is premised
on the idea that "looks" or appearances are properties of physical objects--
the elliptical look is a property of the coin, the youthful look, a property of
my aunt, and so on. Accordingly, the dispositional account has to pick out
the object's appearance, or some feature thereof. When I point to the apple's
redness, according to the appearance-property model, I point to a distinctive
feature, a distinctive look it has, just as I do when I point to the coin's
elliptical appearance when viewed from an angle:
(D2) the disposition of an object to look red = a reel look
This equation states that redness is an appearance property. In being such as
to look red, the apple has an appearance property, borne on its surface,
namely, redness.
(D3) redness = a red look
D3, then, follows from Dl+D2.
As explained above, on this model, the disposition is manifested (that is,
triggered) by causing experiences that reveal the disposition to us: the coin's
elliptical appearance causes experiences of its looking elliptical, not of its
being elliptical. Similarly, the apple's reel appearance-its redness-causes
experiences of itself; namely, of redness.
104 Hdgit Benbaji
(D4) redness = the disposition to cause experiences of redness
On this model, the experiences referred to on the right-hand side of (D4) arc
experiences of the property referred to on the left-hand side of the equation.
We cannot identify the experiences without specifying that they are experi-
ences of redness, as the non-reductive account of dispositions asserts.
One might object that (D4) is thus circular. The problem with circularity
is that it deprives the account of "any interest" Qohnston, manuscript, 18),
or renders it vacuous, inasmuch as it does not differentiate colors from
shapes-after all, square things arc also disposed to look square when
conditions arc right for seeing squares (Broackcs l 997, 204).
The model is indeed, circular-none of the equations purport to be an
analysis or definition of redness, or color in general. They are not intended
to be informative to someone who docs not understand what redness is.
However, the model is not vacuous, since these equations offer an iriform-
ative account of redness. They do not merely affirm that redness is being
disposed to look red; this is true of squareness as well: squareness is being
disposed to look square. They also claim that redness is nothing over and
above a reel look But as I showed in section l. l, squareness, unlike redness,
is more than a look. The equations are thus informative-they tell us what it
is for rm object to be red, whereas affirming that squareness is being disposed
to look square does not tell us what it is for something to be square. The
appearance-property model apprises us of the nature of colors, namely, that
colors, unlike shapes, are looks-appearance properties-or to use the
traditional characterization, dispositions to cause experiences.
In order to know which property redness is, as opposed to, say, yellow-
ness, or blueness, one has to look at a red object, under the appropriate
circumstances (lighting, etc.). Accordingly, the experience of seeing redness
can be characterized by pointing, under the right conditions, to something
red:
(D5) a disposition to cause experiences of redness = a disposition to
look like that (pointing to, say, a red apple)
The demonstrative "that" signifies the way the object looks, presents itself to
us, appears-namely, red. 21 While (D4) tells us what the nature of colors is,
21
Watkins (2002, 80-6) suggests the demonstrative formula as a solution to the
circularity problem. I am presently working on a critique of that solution. Shoemaker
provides a similar reply: "Particular ways of looking can be picked out demonstratively-
' looking way Wis looking that way'-or by descriptions like 'the way chartreuse things
look to observers with thus and such perceptual systems in thus and such conditions'"
(2006, 466). Shoemaker, however, denies that these appearance properties are properties
of external objects.
Colon as Primitive Dispositions 105
(D5) tells what redness is. Both are circular (the demonstrative presupposes
the concept of red), but not vacuous.
It might be objected that the model leaves no room for illusions. The
experience of a white object might, under abnormal conditions, be intro-
spect~~ely indistinguishable from the experience of a red apple under normal
cond1t1ons. If the same experience can occur when the object docs not have a
red appearance, then it still seems that the experience somehow "intrudes,"
as it were, between the perceiver and the color. Just as on adverbial or sense-
data theories, we come to know about objects in the external world by
means. of knowin~ something internal, so too, according to this line of
reasonmg, the theSIS that the experience reveals an alleged red "look" is no
more than wordplay.
Two replies to this objection suggest themselves, in accordance with the
two chief conceptions of perception, the representational and the rcla-
tional. 22 Representationalists maintain that the phenomenal character of
our experiet~ce of an object is determined by an object's properties as
represented 111 experience. The experience of an object's looking red is a
matter o'. its r~presenting that object as having the property of redness,
representmg bemg a matter of, for example, the experience's being reliably
caused by red objects in the environment. Such representation can when
conditions are abnormal, be caused by a white object, so this a~count
implies that there is a common factor to veridical experience and illusion.
Yet the experience of an object's looking reel is a direct representation of the
object's color: it satisfies "Transparency," the near-universal observation that
in perception our attention is directed outward to the world, not to any
inward sensation.
On the relational view, experience is a primitive cognitive relation between
an o~ject a'.1d a perc~iver, not merely a representation of it. According to the
~elation~ view, mamfest pr~perties of things themselves figure in experience,
not their mental surrogates Oohnston, manuscript), be the latter sensations
or representa:!~ns. The experience does not "come between the perceiver
and the col~r II~ .a'.1y sense at all; the experience is simply the presentation
of the color itself. 1he color constitutes the experience. Hence, if the color is
not instantiated, the relational experience docs not hold. Since an illusion of
a pro.Per~ i~ not a manifestation of an actual property of the object in
quesuon, It 1s not phenomenally identical to a veridical experience, which
presents the color to us. We cannot introspectively distinguish an illusion of
redness from a veridical experience of redness, but since only the latter
22
There is also a third conception of illusion that should be mentioned in this context.
Johnston (2004) combines a relational view of experience with postulation of a common
factor between veridical experience and illusory experience.
106 Hagit Benbaji
presents to us the object's red look, the illusoty and veridical experiences are
not phenomenologically identical.
The disjunctive conception of experience that is committed to the
possibility of introspective indistinguishability without phenomenological
identity is controversial. Although I cannot defend this possibility here, let
me point out that it is in line with the core motivation for primitivism, that
is, the epistemic significance of direct perception, the premise that in
manifesting a property to us, experience provides us with knowledge of
which property is experienced. If the same experience sometimes-namely,
when it is illusory-fails to provide such knowledge, it is hard to understand
how perception can have this epistemic significance.
In any case, I take it as a strength of the model proposed here that it leaves
room for both views of experience. Neither the physicalist nor the classic
dispositional account can make sense of the strong notion of perceptual
presentation, as opposed to mere representation of a property. Appearance
properties, by contrast, arc the natural candidates, not merely for represen-
tation in experience, but for literally constituting experience.
The picture that emerges is as follows:
redness = tl disposition to look red = a red look = a disposition to cause
experiences of redness = a disposition to look like thtit
There are, then, red looks, green looks, yellow looks, and so on. They ate
mind-independent dispositions of objects, and what they cause are experi-
ences the phenomenology of which is determined by the property in question
(redness, greenness, yellowness). But of course something can look red (green,
yellow, etc.) to me when it isn't, if the conditions are abnormal.
3.2 Metaphysical implications: between
dispositionalism and primitivism
The central import of the proposed model is that introducing appearance
properties, that is, "looks," allows us to invoke primitive dispositions. Let me
elaborate on the metaphysical implications of a primitivist-dispositional
account of color by distinguishing it from classic dispositional accounts,
on the one hand, and from prirnitivist accounts, on the other hand.
Recall that the underlying conditional, common to both primitivists and
dispositionalists, claims that if colors arc essentially connected to experiences
of them, they are explanatorily redundant. The possibility of primitive
dispositions proves the falsity of this conditional. It shows that the condi-
tional is tempting only given that the dependence between colors and
experiences of colors goes in one direction. The model of appearance
Colors as Primitiue Di.1positions 107
properties displays a double-dependency between appearance properties
and experiences. Appearance properties explain why, in gazing at the coin,
we experience it as changing in some way: it is because something is,
indeed, changing-the coin's look. Nevertheless, the coin's look is, and is
experienced as, a property whose nature is to present itself, that is, as an
appearance property.
When the model is applied to colors, the experience of redness is
dependent on and derived from redness, as Inheritance requires, and
hence, not musally prior to redness. On the other hand, redness is not
conceptually prior to the experience of redness, because we cannot under-
stand what it is for an object to be red except in terms of its looking red.
Since the priority and dependency of colors and experiences have different
senses (namely, a causal sense, and a conceptual sense), there is no need for
concern about explanatory circularity.
Why, then, do dispositionalism and primitivism claim that if colors are
essentially connected to experiences they cannot explain our experiences?
Because this conditional is, I believe, ontologically biased.
Dispositionalism accepts the consequence of the conditional, in order to
downgrade the status of colors: a property that lacks explanatory power does
not seem to be a genuine property. The upshot is that even a non-reductive
account of dispositions does not undermine a "dualistic metaphysical
scheme" according to which all properties arc "either physical or mental,
or some combination of the two" (McGinn 1996: 548).
Primitivism rejects the consequence of the conditional, in order to
upgrade the status of colors: our experiences do not constitute the nature
of colors any more than they constitute the nature of shapes. The upshot is
that colors are more than dispositions; redness is taken to be the categorical
ground for objects' dispositions to look red.
On primitivist dispositionalism redness is manifested in the experience of
redness, just as the elliptical appearance of the tilted coin is manifested in
visually experiencing the tilted coin. Thus, primitivist dispositionalism
denies ontological dualism. It acknowledges, with primitivism, "an extra
layer for colors .... To the old question, 'Are colors mental or physical,
subjective or objective?', we must answer, 'Neither: they constitute a third
category, just as real as, but distinct from, mental and physical properties'"
(McGinn 1996, 548).
Nevertheless, in contrast to primitivism, primitive dispositionalism main-
tains that the primitive property that is manifested on the surface of the
apple is the apple's red look, its red appearance. Although the apple's red
appearance is neither mental nor physical, or even a combination of physical
and mental properties, it does not constitute a mysterious third category: all it
t<tkes for an object to be red is for it to look red.
108 nigit Benbaji
Acknowledging such sui generis properties might provoke the charge that
prirnitivist dispositionalism collapses into either dispositionalism or primi-
rivism. If colors are irreducible properties that explain our experiences of
them, then there are two possible causes of our experiences: irreducible
properties (primitive or dispositional), and their rnicrophysical bases. Primi-
tivist dispositionalism is either incompatible with physicalism, in insisting
that irreducible properties arc genuinely causal, or philosophically superflu-
ous, in deeming them causally inefficacious. lfprimitivist dispositionalism is
incompatible with physicalism, then it is no less extravagant than primitiv-
ism. If primitivist dispositionalism concedes that colors are causally ineffi-
cacious with respect to our visual experiences, then colors are explanatorily
redundant, as is the case on classic dispositionalism.
This dilemma is familiar from discussions of the over-determination of
the mental, as are the proposed ways to resolve it. One suggested way out is
to deny that the scientific explanation of color experience in terms of
wavelengths excludes, overdetermines, or renders superfluous the simple
explanation that the apple looks red because it is red. The scientific and the
commonsense explanations have distinct explananda. Colors explain what
the apple looks like, whereas invoking wavelength explains the visual-
neurological processes that underlie color experienccs.2·" A non-reductive
account of colors, primitive or dispositional, is, then, committed to explana-
tory pluralism.
The dilemma is false, then, because there is nothing mysterious in the
idea of irreducible colors that explain our experiences of them. What makes
primitivism mysterious is not the commitment to irreducible colors, but the
claim that these irreducible colors are something over and above disposi-
tions. If colors are more than dispositions, it is incumbent on primitivism to
explain "in what a grasp of the supposed residue would consist" (Evans
1980, 273). Primitivists concede that there is no adequate non-sensory
description of colors, as colors are primitive. So what is it for the apple to
be red, that is, red-as-we-see-it, in the dark? In the case of shapes, it will be
recalled, we can understand what it is for the apple to be round because we
have a geometric account of roundness. But in what sense is an apple
primitively red in the dark? "Berkeley's 'master argument"' (1713/34, §§
1.22-4) shows that if we ny to imagine redness as persisting "out there," e.g.
in the dark, exactly as we see it, but without any connection to experience,
21
The two explanations belong to difterent "explanatory spaces" (Campbell, 1993,
182), in the same way that reasons constitute a genuine explanatory space with respect to
actions, a space that is independent of the neurological explanations of complex physical
movements. For alternative accounts of explanatory pluralism, sec Yablo ( 1992), Child
(1994), Bennett (2003), Benbaji (2010).
Colon as Primitive Dispositions 109
we do not succeed in doing anydiing "but perceiving a color unperceived by
anyone else" (Evans 1980, 272-4; for critical discussion, see Allen 2007).
Primitivists may, switching on the light and pointing to the apple, insist that
for an object to remain reel in the dark is for it to remain like thrtt. But as
Evans notes, "switching on the light merely tests for the dispositional
property; what could show whether or not objects did in fact retain these
other colour properties [i.e. the primitive ones, H.B. I in the dark?" ( l 980,
273). The purported distinguishing feature that renders colors more than
dispositions is mysterious-what could the "more" possibly be? Primitivist
dispositionalism dissolves the mystery: the sui generis properties borne on
objects' surfaces are the objects' dispositions to look colorecl. 24
4. DO COLORS LOOK NON-DISPOSITIONAL?
So far, I have sought to render intelligible the idea of primitive dispositions,
utilizing the idea of an appearance property, e.g. someone's looking old.
f;urthermore, I proposed that an object's redness should be thought ofas the
property of presenting a certain appearance. What remains to be established
is whether colors are, in fact, such appearances. For a critic might well accept
my proposed analysis of seeing an appearance property, and with it, the
possibility of primitive dispositions, but for phenomenological reasons, such
as the objection that colors do not look like dispositions, reject the claim
that colors are, in fact, such properties.
This phenomenological objection is crucial, because it purports to
undermine application of the appearance-property model to colors. If
colors are dispositions, and look non-dispositional, experiencing them
involves some error. The property revealed to us in experience cannot be
redness, namely, the disposition we attribute to objects. Redness is not
what it appears to be. Hence, MI is false: an experience of redness can be
neither the manifestation of redness, nor derived from redness. To estab-
lish that colors are primitive dispositions, we need to refute the objection
that colors look non-dispositional.
The objection that we see colors as non-dispositional rests on two claims:
the phenomenological observation that colors, like shapes, are experienced
as simple properties of objects, and the metaphysical claim that dispositions
are complex in nature, being comprised of relations between subjects and
objects. Together, the two assumptions imply that we see colors as non-
clispositional. Let me present and respond to a few variants of this objection:
2'i I develop this argument against primitivism in Why Color Primitivism (2016).
110 Hagit Benbrtji
that we see colors as monadic (section 4.1), constant (section 4.2), and
categorical (section 4.3). I will argue that primitive dispositions are indeed
monadic and constant, as they look, but that they neither are categorical,
nor look categorical.
4.1 Monadic (i.e. non-relational) properties
The first variant of this objection assumes that a disposition is a relational
property, indeed, that the very idea of a "look," and the invocation of the "to
whom?" and "under what conditions?" provisos, suggest relativization to
perceivers and perceptual conditions. An appearance property-the vivid
redness of this tulip, say-would presumably not be the same for someone
with binocular vision as it would for someone with monocular vision; the
way things look to a hawk surely differs from the way they appear to us-
and so on. Hence, redness is a relational property. The experience of a red
object, however, is the experience of a non-relational, monadic property of
the object's surface. Colors look non-dispositional, since they look non-
relational. 25
On primitivist dispositionalism, colors are not relations. "The apple is
red" is not short for "The apple is red for S in C," where "5" refers to a
species of perceiver, and "C" specifies a set of visual conditions. Rather, it
provides determinate answers to the questions "to whom?" and "under what
conditions?": to humans with normal vision, in daylight, in which case one
can ascertain the colors of things by looking at them. 26 Neither someone
with monocular vision, nor a hawk, can see the apple as red.
But human beings could disappear from the scene, and our notion of
normal vision could also change; would this apple continue to be red, if
we-humans with normal vision-were not around? More generally, can
objects lose their color merely by changing their relations with other objects?
It has been suggested that this question-"suppose that x is red; as we
modify things other than x, and thereby modify the relations x bears to other
things, will x (necessarily) continue to be red?" (Cohen 2009, 9)-is our
pre-theoretical test for relational properties, and that dispositionalism is
compelled to answer in the negative (9-12). By contrast, I argue that the
apple would retain that look (pointing to the apple) even were we to
disappear, because for the apple to be red is for it to look red to us, with
our perceptual apparatus, under the actual conditions. The experience
cannot be identified other than by actually pointing to the apple, and we
25
Johnston 1992, 142; McGinn 1996, 302.
26
This characterization of perceivers and perceptual conditions is based on that
proposed in McDowell 2004 and 2011.
Colors r1s Primitive Dispositions 111
can only point to the color of the apple by using our actual apparatus, under
the actual conditions. Hence, a primitive disposition is characterized
"rigidly" (McDowell 2004, 399).
Undeniably, it is only contingent that we have a visual system that
provides for such uniformity and richness in our normal experience of
colors. As McDowell says, "The contingent uniformity in chromatic visual
sensations enters into the very constitution of the properties themselves.
There is nothing to those properties ... except being such as to look colored
to subjects like us, subjects who come within the scope of a concept of
normality in color vision only because there is that match in the chromatic
looks that things present to them" (McDowell 2011, IO). But we are lucky
enough to enjoy this uniformity in chromatic visual experiences, and given
that we are so lucky, our perceptual apparatus reveals to us that the apple
really is red, that is, it would continue to be red even had we never been
around. I conclude that primitivist dispositionalism can be accused of
anthropocentrism (Wiggins 1976, 107), but not relationalism: the apple
has, over and above its physical properties, a red look that is not a relation to
perceivers.
However, for those who are convinced that the phenomenon of percep-
tual variation is widespread, at least with respect to some shades of colors,
let me point to an intriguing non-relational explanation of the phenom-
enon, compatible with the account proposed in this paper. Kalderon
develops a theory of color pluralism, according to which objects have
multiple colors. Colors are not relations between object, perceiver, and
perceptual circumstances; rather, the relations between these three factors
determine the perceptual availability of colors. "According to the color
pluralist, then, the relativist conflates the conditions for the perception
of a color with the perceived color" (2007, 577). Although I cannot go into
color pluralism here, it is consistent with primitivist dispositionalism (as
it is with primitivism).
Even though appearance properties are not relations to perceivers, it
might be claimed that they are relations to the environment (Noe 2004,
149). Boghossian and V elleman argue that when one turns on the light in a
dark room, "the colours of surrounding objects look as if they have been
revealed, not as if they have been activated." This implies that colors look
non-dispositional, for otherwise, "they would seem to come on when illu-
minated, just as a lamp comes on when its switch is flipped. Turning on the
light would seem, simultaneously, like turning on the colors; or perhaps it
would seem like waking up the colors, just as it is seen to startle the cat"
(1989, 85).
The underlying assumption of this objection is the already familiar idea
that we see the disposition by seeing its manifestations occurring: we see
112 Hagit Benhrlji
solubility by seeing sugar dissoluing. I have <tlready shown that this assump-
tion is false with regard to appearance properties, since in the case of
appearance properties, we do not see the manifestations-namely, the
cxperiences--at all, so there is no reason to expect to see any occun·ences.
Redness, that is, the disposition to look red, manifests, that is, presents itself
as a sustained property of the apple, not as an event that transpires when
I "illuminate" the apple.
4.2 Constant properties
Color constancy is, like shape constancy, a familiar, commonplace phenom-
enon. The white wall looks gray in the shade, but this is not an illusion of
grayness; I can sec that the wall does not change its color, as the coin does
not change its shape when tilted. There seems to be a perfect parallel
between constancy of shapes and constancy of colors. This parallel invites
the distinction between (i) experiencing an object as red, and (ii) experien-
cing an object as having a red look. If there is such a distinction, as suggested
by constancy, the model proffered here is actually giving an account of an
object's looking red, not its redness (just as the model is an account of the
coin's elliptical appearance, not of its shape). Redness would have to be
more than its various "looks," that is, it would have to be, not a primitive
disposition, but primitive. Otherwise, if redness itself is nothing but an
appearance, it cannot be the constant color that we see along with its various
appearances.
To explain this charge, consider a specific shade of red, red-215, say. This
shade now, at dusk, seems dark and shadowy, whereas earlier in the day it
was bright and bold. Yet we experience red-215 as one constant color
that unites and explains its various appearances, just as the roundness of
the coin is the property that unites and explains its various appearances
(round from above, elliptical from an angle). The concern, presumably, is
that on the dispositional account, red-215 is nothing beyond what it looks
like under various circumstances, that is, it is merely a series of various more
finely differentiated determinate appearances (Allen 2009; Kalderon, manu-
script). 27 The objection contends that the appearances are in constant flux,
and the putative disposition is nothing but a set of changing appearances.
However, having established that the disposition is a primitive property
on the surface of objects, primitivist clispositionalism need not deny that the
specific shade of color, red-215, is a constant property that unites and
27 To favor one very specific appearance, in a certain kind of lighting (say, the kind of
bright daylight present here right now, at 2.45 p.m.), is completely arbitrary; red-215 has
no such defining and privileged look. But see Allen (201 O).
Colon tis Prhnitiue Dispositions l 1:3
explains its various appearances. First, it has to be shown that there is a fact
of the matter as to whether a given shade is recl-215. And we can indeed do
so via a simple model. Following a suggestion by Broackes (2007, 59), we
can group together as red-215 things that stand in a certain phenomenal
equivalence relation, such as is indistinguishable in co!mfrom, to the master
red-215 sample. Thus, I can check whether my bright red copy of the
Ii·r1ctatus is recl-215 by holding it next to the sample to see if it is or is not
distinguishable from it under various illumination conditions. Now if the
upper half of the book cover is under illumination 11, and the lower half
illumination I2, what unites these two appearances? The common factor is
that both are the same shade of red, namely, the shade that is indistinguish-
able in color fiwn the master red-215 sample (viewed under the same
conditions). This is why I expect the following counterfactuals to be true:
were the upper half of the cover viewed under 12, it would look exactly the
way the lower half looks now; were the lower half viewed under I l, it would
look exactly the way the upper half looks now. To experience the color of
the cover as a constant color, over and above its various appearances, is to
expect it to vary in appearance in "red-215 appropriate" ways, to use Noe's
expression (2004, 143), that is, in the phenomenally salient ways recl-215
changes under different lighting conditions. To account for constancy, there
is no need to assume a conception of color on which it is something over and
above an appearance property.
To sum up, I have argued that from the correct observation that colors are
experienced as non-relational and constant, it does not follow that they look
non-dispositional, because primitive dispositions are non-relational and
constant.
4.3 Categorical properties
Primitivists also argue that colors look categorical. This claim is essential to
the regress argument against the non-reductive account of dispositions.
According to the objection from infinite regress, the problem with the
non-reductive dispositional account is that one "cannot see what colour an
object has. For he cannot see that particular colour of an object except by
seeing the particular way the object tends to appear; and he cannot see the
way it tends to appear except by seeing the way it tends to appear as tending
to appear; and so on, r1d infinitum" (Boghossian and Velleman 1989, 90).
This regress renders the content of the experience of redness completely
vacuous (89). But, I would reply, why not stop the regress by invoking the
relevant color by name-to appear reel is to appear disposed to appear recl-
and put an encl to the matter? Boghossian and Velleman concede that we do
this for squareness ("objects cm appear disposed to look square just by
114 Hagit Benhaji
looking square") but only because things look square "intrinsically and
categorically" (89). 'fhe problem for the dispositional account is that
when something looks red it looks "non-dispositionrd(y red" (89).
Bur if this is the objection, then it just restates the core objection addressed
in this section, namely, the claim that colors look non-dispositional. 28 "fhe
regress argument does not provide any reason for claiming that colors look
categorical-it presupposes that they look non-dispositional, that is, categor-
ical. Furthermore, "categorically" docs not add anything to "intrinsically"
(understood to mean non-relationally). Once we acknowledge a primitive
non-relational dispositional property on the surface of objects, which we can
point to, the regress can be preempted.
I conclude that there is no reason to think that colors look non-
dispositional. This conclusion also serves another purpose: it constitutes a
refutation of primitivism. I have already argued (in section 3.2) that primi-
tivism has not articulated any coherent sense in which colors are more than
dispositions. Having refuted the phenomenological charge, I can argue that
there is no plausible motivation for making colors more than dispositions:
there is no way to explain why more is needed. Primitive dispositions can
provide for robust possibilities (e.g. the possibility that the apple will remain
red even when we arc no longer around); non-relational properties; and
constancy. 29 This covers the entire list of (metaphysical) reasons usually
offered for the claim that colors are categorical properties; if there is no
(phenomenological) reason to claim that colors look non-dispositional, there
is no reason to take colors to be categorical.
5. DO COLORS LOOK LIKE DISPOSITIONS?
Ironically, the response to the objection that colors look non-dispositional
reinforces the phenomenological objection that they nonetheless do not
look like dispositions. I have claimed that from the fact that colors, like
shapes, are experienced as simple, constant, and monadic properties of
objects, it does not follow that they are experienced as categorical, i.e.
non-dispositional. However, if colors are experienced as simple, constant,
and monadic, they seem to be on a par, phenomenologically speaking, with
shapes. It does not appear to us that colors are dispositions, for they are not
28
For a different reply to Boghossian and V cllcman, sec Byrne and Hilbert (2011).
29
Robust possibilities and causal explanation Qohnston 1992, Campbell 1993); con-
stancy (Allen 2009, Kalderon, manuscript); non-relational properties (McGinn 1996);
direct intervention (Campbell 2006); categorical base Uohnston 1992, 147, Campbell
1993).
Colon as Primitive Dispositions I 15
presented in experience in a manner that differs in any way from the manner
in which shapes are presented, and surely, it does not appear to us that
shapes are dispositions.
While the claim that colors look non-dispositional implies that the per-
ception of colors involves an error-it deceives us as to their nature (colors are
dispositions and present themselves as non-dispositional), the present objec-
tion is more modest. It merely raises the possibility that perception is neutral
with regard to the question of whether colors arc dispositions: we cannot
tell, simply by looking at them, whether colors are dispositions. Colors do not
look non-dispositional, but they do not look like dispositions either.
The claim that "colors do not look like dispositions" tends to elicit a range
of strong reactions, from "why would anyone think that colors look like
dispositions?" 30 to "how would they have to look to look like disposi-
tions?"31 The more cautious and perhaps dominant response is to claim
that the experience of perceiving color is silent about the question of
whether colors are dispositions-that both dispositionality and non-
dispositionality are not visible properties at all. 32
Granting-for the sake of argument-that colors are not generally per-
ceived as dispositions-why should this undermine the dispositional
account of colors? If a critic grants the previous point, that colors do not
appear positively non-dispositional, it's hard to see how the failure to appear
dispositional should be taken as evidence for anything. Colors are presented
to us in experience, as primitivism claims, but not as dispositions. Assuming
that primitivist dispositionalism is not committed to Revelation, colors may
well he dispositions though they do not look like dispositions, just as water
can be H 2 0 even though it doesn't look like H 2 0.
However, the core claim of the model for appearance properties adduced
the fact that from an angle, we experience the coin as looking elliptical, not
as being elliptical. The coin looks round and elliptical-looking. The model
proffered for appearance properties implies, then, that an appearance
property-an elliptical look, a youthful look, and the like-fr a visih!e
property. Appearance properties appear to us as they are, namely, as ways
of looking. If we experience the coin as elliptical-looking, then perception is
not neutral with respect to appearance properties, as it is regarding H 2 0. If
an advocate of primitivist dispositionalism claims that colors are appearance
properties, but do not look like appearance properties, she cannot simply
invoke the neutrality of perception. The reply to the retort "so what if colors
30
McGinn 1983, Johnston 1992, Dancy 1986, Boghossian and Velleman 1989.
31
Johnston 1992, 141. For a dismissive reading of the problem, see McDowell 1985, 112.
2
'' Byrne 200 l, Levin 2000.
116 Hagit Benhaji
don't look like dispositions?" is that if they do not, they might not be
appearance properties!
Let me close by pointing out some phenomenological data that supports
the experiential disparity of colors and shapes, and thus reinforces the claim
that colors are appearance properties (section 5.1 ). I then consider the
implications of this disparity for Revelation (section 5.2).
5.1 Seeing colors as appearance properties
I contend that the disparity between the phenomenology of colors and that
of shapes is attested to by various experiences, and indeed falsifies the alleged
parity between colors and shapes. Let me list the perceptual clues that, taken
together, indicate that shapes are not presented in experience as dispositions,
whereas colors are.
(l) First and foremost, when I see something as square, I see it as having
four equal sides and four right angles. When I see something as red, I see it
as having tht1t shade (pointing). What does it mean that the only way to
capture the content of experience is by pointing to the color? It means,
precisely, that redness is manifested in experience by looking red, that's all.
The fact that the geometric description of shapes is given to us in visual
experience, whereas no description of color is given to us in visual experience,
attests to the disparate natures of colors and shapes. The conceptual differ-
ence between colors and shapes expresses itself in the experience of these
properties: colors are experienced as purely visual, shapes are not. 33
(2) Consider the difference between Molyneux's question and its coun-
terpart vis-a-vis colors. Molyneux asks whether someone born blind, who
later gains vision, would be capable of distinguishing between a square and a
circle solely by sight. 34 Answering in the affirmative makes sense, I would
JJ It might he objected that it is just a matter ofluck that when we see a square we sec a
figure with four equal sides, and there could, conceivably, be creatures who arc able to tell
squares from pemagons and circles just by looking, but cannot sc>e that there rs a four-
sided object out there (such a suggestion is put forward in Kulvicki 2007). I agree, of
course, that a child might have a rudimcntaty capacity to track some visible pattern of
squareness, that is, she might be able to pick out squares quite successfully from a group of
circles and assorted rectangles. But this is not to say that she secs the pattern as squareness.
Suppose we ask her, pointing to a rectangle, why it is not a square. She r'.uy not reply,
"because it docs not have four equal sides and four right angles." But if she docs not
exhibit any grasp of the difference between the sides of rect~ngles _and those of the squares,
it is hard to ascribe to her anything more than sheer practical abrhty. What d1st1ngmshcs
the capacity to sec something as square, as opposed to the more primitive ability to track a
visible pattern of squareness, is an inkling, a modicum of comprehension that the shape
has a certain geometric structure.
34 The original question is about a rnbe and a sphere.
Colors as Primitive Dispositions 117
argue, precisely because there is a perspective-independent description in
light of which the shape's appearance is intelligible. Given the absence of
such a description in the case of color, the parallel question-whether
someone born blind who later acquires the ability to see would be able to
distinguish red from blue on seeing them for the first time--cannot be
answered in the affirmative. Now an affirmative answer to Molyneux's
original question means that the content of the visual experience of square-
ness is the same as the content of the tactile experience of squareness: the
subject visually identifies the shape as that given to him in tactile perception
of square things (as having four equal sides and so on). 'T"hat the experience
of shapes is cross-modal in this way illustrates that we experience shapes as
more than "looks." That the experience of colors is modality-specific indi-
cates that colors are not more than "looks."
(3) A related indication of the cross-modality of shape-experience
resides in the fact that to experience a shape, as well as its changing
appearances, one must move oneself or the shape, and these movements
are correlated to movements of tactile perception. Noe illustrates this
correlation between visual and tactile experience of shapes with the
following example: "if something looks square, then one would need
to move one's head in characteristic ways to look at each of the corners.
One would have to move one's hands the same way (at the appropriate
level of abstraction) to feel each corner" (Noe 2004, 102). By contrast,
movements of the observer and object will have no effect on the appear-
ance of color. The correlation between movements of the perceiver or
perceived object in the context of visual perception of shapes, and hand
movements in the context of tactile perception, and the absence of such
a correlation in the context of visual perception of colors, is a feature of
the capacity to perceive shapes and colors. That is, it is inherent in our
capacity to see shapes that they are accessible through touch. It is
inherent in our capacity to see colors that they are inaccessible through
other modalities.
(4) When I see the coin from an oblique angle, l have no choice but to
describe it as elliptical-looking, because it is not elliptical (see section 2.1).
By contrast, l do not have to describe different experiences of the same color
in terms of how the object in question looks. I can appeal to variations in
illumination: the white wall is shaded here, and fully illuminated there. 1 Glll
avoid the claim that the white wall looks gray in the shaded area by
mentioning darkness or shadows. This isn't to deny that it is also true that
the wall does look gray in the shaded area, But in this case, we have an
alternative description, which we lack in the case of shape. The fact that we
can experience the variation as a change in illumination shows not merely
that we take both the wall's various appearances, and the constant color, as
118 Hag# Benhaji
metaphysically on a par, but that we experience the constant color and its
various appearances as "looks."
(5) To see something as having a certain property is a capacity entailing
expectations with regard to the success and failure of our ascription of that
property. The capacity to see something as having a certain shape entails
different expectations than does the capacity to see something as having a
certain color. For example, the capacity to see a figure as square entails the
expectation that in cases of uncertainty, where we cannot tell, just by
looking, whether the figure is a square or merely a rectangle, there is an
effective method for resolving this uncertainty, namely, measuring the
figure. By contrast, the capacity to see the shirt as green entails the expect-
ation that in cases of uncertainty, where we cannot tell, just by looking,
whether the color is green or teal, all we can do is improve the perceptual
conditions: look at it in daylight, check it under non-fluorescent illumin-
ation, etc. To verify our perception of colors we only need to improve the
conditions under which they are viewed; to verify our perception of shapes,
we need to appeal to methods beyond how they look.
One might object that these expectations set very high standards for
being able to see that something has a shape or color. It seems implausible
to argue that we don't have the capacity to see squares unless we under-
stand how to use protractors, for example, or that we can't see that an
object is red, unless we know (or at least "expect") that fluorescent lighting
can have strange effects. What about people who lived before fluorescent
lighting was invented?
Of course people who lived before fluorescent lighting was invented
could see colors, but this is, in part, because they were "sophisticated"
enough to recognize the distinction between normal and abnormal percep-
tual conditions, or what amounts to the same thing, the distinction between
an object's color and an illusion of that color. 35 The capacity to see colors
presupposes possible failures of the exercise of that capacity, which occur
when there is a breakdown of normal conditions. Similarly, to have the
capacity to see something as square, not merely to have a rudimentary ability
to track some visible pattern of squareness, requires a grasp of what is
involved in correcting misleading appearances beyond the way it looks.
(1)-(5) are differences between what it is to see redness and what it is
to see squareness. Although appearances are simple, constant, and monadic
properties of objects, as shapes are, the experience of seeing colors is not
the same as that of seeing shapes. Upon reflection, the parity thesis falsifies
35 See Sellars (1956, 48), the Myth of Jones.
Colon· as Primitive Dispositions 119
the phenomenology of experience: colors, unlike shapes, do look like
dispositions, and more specifically, appearance properties.
5.2 Revelation?
The conclusion that colors look like dispositions might seem to be an
affirmation of Revelation. If colors are primitive dispositions, and their
being primitive dispositions is given to us in experience, has not the whole
nature of redness been revealed to us?
Revelation raises two sorts of issues in my view. First, one might accept
the model proposed here: in particular, one might accept that there are
appearance properties; that we perceive them as appearance properties; and
that these appearance properties arc dispositions. Nevertheless, one might
argue that it does not follow from these claims that we sec appearance
properties r1s dispositions.
Recall, however, what it is to see an appearance property. To see the coin
as looking elliptical is to experience it as a property of an object that is such as
to look a certain way (elliptical) when viewed from this angle (when tilted).
This exhausts the notion of seeing a property as a visual disposition: to see it
as a property of objects that is such as to look a certain way under certain
conditions. In other words, there is no gap between seeing something as
looking F and seeing it as being disposed to look F 36
The second concern regarding Revelation is the thought that science
could discover the physical basis for dispositions, and since a disposition
is, by dint of its dispositional nature, constituted by a categorical physical
base, part ofits nature remains hidden. If having a physical base is part of the
nature of primitive dispositions, then Revelation must be qualified.
However, proponents of primitivist dispositionalism might concede that
primitive dispositions are necessariry grounded in physical categorical bases,
and yet refuse to take this fact to be part of the nature of colors. 37 In this
case, Revelation might indeed be true.
So although it's not obvious whether Revelation is true, I suggest a way in
which it might be. I have argued for both dispositionalism and the irredu-
cibility of colors to any physical properties of objects; and also that colors
ic> For those unconvinced by this reply, let me qualify it as follows: acknowledging that
there are appearance properties, that they are perceived as such and thar they tire
dispositions, is all I need to establish the possibility of primitive dispositions and their
manifestation in experience.
37
For a conception of Revelation as knowledge of things, in the spirit of Russell's
notion of acquaintance, see Campbell (2006). Gert (2008, 142) suggests a qualification of
Revelation that seems to be warranted on my account as well.
120 Htigit Benbaji
appear to be dispositions; from which it follows that Revelation can be true.
Rather than simply assuming its truth, however, this paper has endeavored
to argue fur it.
This vindication of Revelation completes my project of showing how
it is possible to fulfil the promise of primitivist dispositionalism. Recall
the dream of Eden, where the apple was wholly revealed in experience as
"gloriously, perfectly, and primitively red" (Chalmers 2006, 49). [ have
shown that colors can, indeed, be gloriously and perfectly revealed in
experience--as dispositions, or as I called them, "looks" of physical objects.
Htigit Benbaji
Ben Gurion Univer:rity
REFERENCES
Allen, K. 2007. "The Mind-Independence of Colour," European journal of Philoso-
phy 15(2): 137-58.
Allen, K. 2009. "Being Coloured and Looking Coloured," Cmuidian journal of
Philosophy 39(4): 647-70.
Allen, K. 2010. "Locating The Unique Hues." Rivistrt di Estetica 43(43): 13-28.
Austin, J. L. 1962. Sense and Sensibilirt (Oxford: Oxford University Press).
Bcnbaji, H. 2010. "Token Monism, Event Dualism, and Ovcrdetermination,"
Cmaditm journal of Philosophy 40(1): 20.
Bcnbaji, H. 2016. Why Color Primitivism AustmLlsirm joumal ofPhilosophy, Vol. 94.2.
Bennett, M. R. 2003. Philosophical Fo1mck1tions ofNeuroscience (Oxford: Blackwell).
Berkeley, G. 1948-57. "Three Dialogues between Hylas and Philonous." IY1e Works
of Georg,e Berkeley, Bishop of Cloyne, ed. A. A. Luce and T. E. Jessop (London:
Thomas Nelson and Sons), 9 vols.
Boghossian, P. and Vclleman, ]. D. 1989. "Color as a Secondary Quality," Mind
98(389): 81-103. Reprinted in Byrne and Hilbert 1997: 81-104.
Brewer, B. 1999. Perception and Retlson (Oxford: Clarendon Press).
Broackes, J. 1997. "The Autonomy of Colour," in Byrne and Hilbert 1997,
191-225.
Broackes, J. 2007. "Colour, World and Archimedean Metaphysics: Stroud and the
Quest for Reality," Erkenntnis 66(1): 27-71.
Byrne, A. 2001. "Do Colours Look Like Dispositions? Reply to Langsam and
Others," Philosophical Quarterly 51 (203): 238-45.
Byrne, A. and Hilbert, D. R. (eds.) 1997. Re11dings on Co/01; Vol. I: The Philosophy of
Color (Cambridge, MA: MIT Press).
Byrne, A. and Hilbert, D.R. 2007. "Color Primitivism." Erkemztnis 66 (1/2): 73-105.
Byrne, A. and Hilbert, D.R. 2011. "Are Colors Secondary Qualities?" in L. Nolan
(ed.), Primmy and Secondmy Qualities (Oxford: Oxford University Press).
Campbell,]. 1993. 'The Simple View of Colour," in Byrne and Hilbert 1997, 177-90.
Colors as Pri miti11e Dispositions l21
Campbell, J. 2002. Reference and Consciousness (Oxford: Clarendon Press).
Campbell, J. 2005. "Transparency vs. Revelation in Color Perception," Philosophiml
Topics 33(1): 105~15.
Campbell, J. 2006. "Manipulating Colour: Pounding an Almond," in Gendler and
Hawthorne 2006, 31-48.
Chalmers, D. 2006. "Perception and the Pall from Eden," in Gendler and
Hawthorne 2006, 49-125.
Child, W. 1994. Causali~y, Interpretation, rmd the Mind (Oxford: Oxford University
Press).
Cohen, J. 2009. The Red and the Real: An Essay on Color Onto!Dgy (Oxford: Oxford
University Press).
Cohen, J. 2010a. "Color Rclationalism and Color Phenomcnoloi,>y," in B. Nanay
(ed.), Perceiving the World (New York: Oxford University Press), 13-32.
Cohen, ]. 20l0b. "Perception and Computation," Philosophical !rsues 20(1):
96-124.
Dancy, .J. 1986. "Two Conceptions of Moral Realism," Proceedings ofthe Aristotelian
Society, supp. vol. 60: 167-87. Reprinted in J. Rachels (ed.), Moral The111y
(Oxford: Oxford University Press, l 997), 227-44.
Evans, G. 1980. "Things Without the Mind," in Z. van Straaten (ed.), Philosophical
Subjects: Essays Presented to P. F. Strawson (Oxford: Oxford University Press),
76-116. Reprinted in Evans Collected Papen (Oxford: Clarendon, 1985), 249-90.
Gendler, T. and Hawthorne, J. (eds.) 2006. Perceptutt! Experience (Oxford: Oxford
University Press).
Gert, J. 2008. "What Colors Could Not Be: An Argument for Color Primitivism,"
journal of Philosophy 105(3): 128-57.
Johnston, M. l 992. "How to Speak of the Colors," Philosophical Studies 68(.3):
221-63. Reprinted in Byrne and Hilbert 1997, 137-72.
Johnston, M. l 998. "Is the External World Invisible?" Philosophical Issues 7: 185-98.
Johnston, M. 2004. "The Obscure Object of Hallucination," Philosophical Studies
120(1-3): 113-83.
Johnston, M. The Manifest, unpublished manuscript.
Kalderon, M. E. 2007. "Color Pluralism," Philosophical Review 116(4): 563·-60!.
Kalderon, M. E. 2008. "Metamerism, Constancy, and Knowing Which," Mind
117(468): 935-71.
Kalderon, M. E. 'Experiential Pluralism and the Power of Perception'. In Themes
fi'om Travis. (Oxford: Oxford University Press, forthcoming).
Kalderon, M. E. "Color and the Problem of Perceptual Presence," unpublished
manuscript.
Kelly, S. D. 2004. "Seeing Things in Mcrlcau-Ponty," in T. Carman and
M. B. N. Hansen (eds.), The Cambridge Companion to Merleau-Ponty (Cambridge:
Cambridge University Press).
Kulvicki, ). 2007. "What is What It's Like? Introducing Perceptual Modes of
Presentation," Synthese 156(2): 205-29.
Levin, J. 2000. "Dispositional Theories of Color and the Claims of Common Sense,"
Philosophical Studies 100(2): 151-74.
122 Hagit Benbrtji Colon as Primitiue Diipositions 123

McDowell, J. 1985. "Values and Secondary Qualities," in T. Honderich (ed.), Wiggins, D. 1976. "'fruth, Invention, and the Meaning of Lifr," reprinted in his
Morality and Objectiuity (Boston: Routledge & Kegan Paul), l l 0-29. Needs, Values, Truth: Essays in the Phi/{)sophy ofVrilue (3rd edn) (Oxford: Clarendon,
McDowell, J. 2004. "Reality and Colours: Comment on Stroud," Philosophy and 1998), 87-138.
Phenomenological Research 68(2), 395-400. Yablo, S. 1992. 'Mental Causation'. Philosophical Reuiew 101(2): 245-80.
McDowell, J. 2011. "Colors as Secondary Qualities," in J. Bridges, N. Kolodny and Wiggins, D. l 987. "A Sensible Subjectivism?" reprinted in his NeerLr, Vrtlues, huth:
W. Wong (eds.), The Possibility of Philosophical Understanding: Reflections on the Essays in the Philosopl~y of Vrtlue (3rd cdn.) (Oxford: Clarendon Press, 1998),
Thought o/Bttny Stroud (Oxford: Oxford University Press). 185-214.
McGinn, C. 1983. The Subjectiue View (Oxford: Oxford University Press).
McGinn, C. 1996. "Another Look at Colors,".foimutl ofPhilosophy 93(11): 537-53.
Macpherson, F. 2006. "Ambiguous Figures and the Content of Experience," Nazis
40(1): 82-117.
Martin, M. G. F. 2010. "What's in a Look?" in B. Nanay (ed.), Perceiving the World
(New York: Oxford University Press), 160-225.
Misccvic, N. 2007. "Is Color-Dispositionalisrn Nasty and Unccologirnl?" Erkenntnis
66(1-2): 203-31.
Noc, A. 2004. Action in Perception (Cunbridge, MA: MIT Press).
()'Shaughnessy, B. (2000). Consciousness and the World (Oxford: Oxford University
Prc->s).
Peacocke, C. 1984. "Color Concepts and Color Experience," Synthese 58(3): 365-82.
Peacocke, C. 2008. "Sensational Properties: Theses to Accept and Theses to Reject."
Re1me lnterm1tiont1!e de l'hilosophie 62: 7-24.
Pettit, P. 2003. "Looks as Powers," Philosophical Issues 13: 221-52.
Phillips, I. 2006. "Perception and Context,'' paper delivered at the NPAPC, War-
wick, July 2006.
Russell, B. 1912. The Problems of Philosopf~y (London: Oxford University Press).
Ryle, G. 1949. The Concept of Mind (London: Penguin).
Schellenberg, S. 2008. "The Simation-Dependency of Perception," journal of Phil-
osophy I 05(2): 55-84.
Sellars, W. 1956. "Empiricism and the Philosophy of Mind," in H. Feig! and
M. Scriven, eds., Minnesota Studies in the Philosophy of Science, volume 1,
The Foundations of Science and the Concepts of Psychology rmd Psychoanalysis
(Minneapolis: University of Minnesota Press), 253-329.
Shoemaker, S. 2006. "On the Ways Things Appear," in Gendler and Hawthorne
2006, 461-80.
Siewert, C. 2006. "Is the Appearance of Shape Protean?" Psyche 12(3): 1-16.
Smith, A. D. 2002. lfJe Problem ofPerception (Cambridge, MA: Harvard University
Press).
Watkins, M. 2002. Rediscovering Colors: A Study in Pollyrmnrt Rerdism (Dordrecht:
Kluwer).
Watkins, M. 2010. "A Posteriori Primitivism." Philosophic,tl Studies 150(1 ): 123-37.
Westphal, J. 1987. Colour: Some Philosophictil Problems from Wittgenstein (Oxford:
Blackwell).
Westphal, J. 2005. "Conflicting Appearances, Necessity and the Irreducibility of
Propositions About Colours," Proceedings of the Aristotelian Society 105(2):
219-35.

7 closer look at a hidden assumption that was relied upon until then, but
Are I'here Ineffable Aspects of Reality?
Thom11s Hofweber
l. INTRODUCTION
Should we think that some aspects of reality arc simply beyond creatures like
us, in the sense that we are in principle incapable of representing them in
thought or language? Or should we think that beings with a mind and
language like ours arc able to represent every truth and every fact? In other
words, should we think that some truths are ineffable for us: beyond what
we can think or say? Whatever the answer is, it likely has substantial
consequences. If it is no, i.e. no truth is ineffable for us, then this might
shed light on what reality is like, or what our minds are like, or why the two
match up so well. If the answer is yes, some truths are ineffable for us, then
this might affect our attempts to understand all of reality. In particular, it
might affect the project of metaphysics and its ambition to understand all of
reality in its grander features. If we had reason to think that only a limited
range of facts can be represented by creatures like us then this might give us
reason to think that metaphysics in its ambitious form is beyond what we
can hope to cany out successfully. In this paper I will argue that the question
whether there are any ineffable truths or facts is an important, although
somewhat neglected, question whose answer has significant consequences,
and I will make a proposal about what the answer is, on what this answer
depends, and what follows from it.
The paper has four parts: first, I will clarify what is at issue and make the
notion of the ineffable more precise in several ways. Second, I will argue that
there are ineffable truths using several different arguments. These arguments
will rely on a certain hidden assumption which is almost universally made
implicitly and accepted by most when made explicit, but which I will
critically investigate later in the paper, in part four. A third part will attempt
to answer a puzzle about the ineffable connected to the relationship between
effable and ineffable truths and why the ineffable seems to be more hidden
from us than would be suggested merely by the fact that it is ineffable.
Are There lne.ffrtb!e Aspects of Reri!ity? 125
The solution to this puzzle will make clear how the ineffable is significant.
I will argue that it has important consequences for metaphysics in particular,
which suggest modesty instead of ambition. In the fourth part I will have a
which might well be mistaken. [ will ;trgue that we have good, but not
conclusive, reason to think that this assumption is indeed false, and if so
then everything changes. On the natural way in which this assumption is
false we get no ineffable truths, no modesty in metaphysics, but a form of
idealism instead. I will try to make clear that the resulting form of ide;tlism
is coherenr, significant, and quite possibly true. Whether the crucial
assumption is, in the end, true I won't be able to settle here, but we can
see that there will be important consequences either way. I will pick sides
at the end. But before we can get to all this we need to get clearer about
what is at issue.
2. WHAT IS THE QUESTION?
The ineffable naturally appears as a possibility when we think about the
relationship between what reality is like, on the one hand, and what we can
truly say, on the other. The relationship between these two leads to one
unproblematic (for present purposes) area of overlap, and to two more
mysterious outlying areas (see Figure 7.1).
The area of overlap is a true description of reality: we can truly say
something and reality is like that. Although much can and has been said
about how this is to be understood in more detail, I will leave it untouched
here, since my concern is with the two more problematic cases: first whether
Correct description
The ineffable The non-descriptive
What reality is like What we can say truly
Figure 7.1 The location of the ineffa.blc
126 Thomas Hofweber
what we can truly say goes beyond what reality is like, and second whether
what reality is like goes beyond what we can truly say, which is our main
topic. The first possibility might seem incoherent. How could what we can
truly say go beyond what reality is like? If we said it truly then how could
reality not be like what we said? Those who think that this option is
coherent generally maintain that it only seems incoherent to us because
we mistake it with something else: either that reality is different from how
we say it is (and thus what we say should be false) or else that we say
something truly about something other than reality (which isn't an option,
since reality is all-inclusive). Instead, they hold, this option is coherent, since
we can say something truly that isn't descriptive at all, neither of reality nor
anything else. Some parts of speech aim to describe, while other parts aim to
do something else, for example express an attitude of the speaker. Truth
applies to both, and thus we can say something truly that goes beyond what
reality is like. It is true, but doesn't aim to describe reality, and thus reality
isn't required to be as described for it to be true. Exprcssivism about
normative discourse combined with minimalism about truth is a paradig-
nntic instance of this approach. The question, of course, still remains
whether it indeed is coherent, but since we will not focus on this outside
area in our Venn diagram, we don't have to settle this here. This part of our
diagram will only have a minor role in what is to come.
My main concern is the other outside area: parts of what reality is like that
outrun what we can truly say. This is the ineffable, that which we can't say.
Herc there should at first be no question about its coherence, but there is a
real question about whether there is anything which is ineffable. Is there
anything that reality is like that goes beyond what we can say, and thus say
truly? If so, how much of it is there? Is it merely a little sliver at the edge of
the overlap, maybe something related to the paradoxes, or to consciousness,
or is it a vast area, maybe most of what reality is like? What would follow for
inquiry in general and philosophy in particular if there were a large area of
the ineffable? To make progress on these questions we will first need to
clarify the relevant notion of the ineffable, and how this problem is different
from a number of other problems in the neighborhood. These problems arc
also real and interesting problems, but not the ones I am trying to make
progress on here. My discussion here will focus on a notion of the ineffable
that is most promising for it being metaphysically significant, in that it
captures the sense in which it just might be that minds and languages like
ours arc not good enough, in principle, to represent some aspect of reality,
and therefore are not good enough for carrying out an ambitious project of
metaphysics. Whether or not we arc limited in this way is what I hope to
find out. And to do this we should put aside some issues that I will not tty to
resolve and focus in on the relevant ones instead.
Are There lnejfi1ble Aspects of Reality? 127
Ineffable feelings. First, there is a common use of'\vords are not enough" to
articulate the limits of language, but these are not the limits I am concerned
with here. Take, for example "I can't tell you how happy I am to see you!"
"Words arc not enough to say how glad [am to see you!" It would be beside the
point to answer "Are you very, vety happy, or even happier than that?" It is not
that the first speaker has a degree of happiness such that no words can pick out
that degree of happiness. After all,"[ am maximally happy" would certainly be
good enough to do that. Where words give out is not in describing the degree
of happiness, but rather in giving the hearer a sense of what it feels like to be
that happy. Words are not enough in getting the hearer to feel the way the
speaker feels, or at least give them a sense of what such a feeling is like. But they
are enough for describing how happy someone is: very, very happy. That words
are not enough to transmit feelings in this sense is notable, but not a limitation
oflanguagc in capturing what reality is like. Words might also not be enough
to get you on the last Hight to Raleigh, in the sense that no matter what words
you utter, you won't get on that flight. This limitation oflanguage is not one in
its descriptive power, but in its limited effects to produce feelings or get an
airline seat, a limitation I can happily accept and which isn't my concern here.
What I am concerned with here is whether there are any facts, any truths, or
any true propositions, such that we cannot, in principle, state or represent these
facts, truths, or true propositions in our language.
Ineffable objects. The notion of the ineffable is often tied to objects, and as
such it is seen as problematic and paradoxical. An incfE.1.ble object is usually
understood in one of two ways. It either is one that we can't talk about at all,
or it is one about which nothing can be truly said. An example of the former
is sometimes taken to be God when God replies to Moses' question about
what his name is with "I am who I ~tm," and leaves it more or less at that. 1
One possible lesson of that is that God cm't be named, although this seems
somewhat incoherent, since I just named God with "Goel." It is God, after
all, who is supposed to be unnameable. Another lesson might be that God
shouldn't be named, which wouldn't make God ineffable, of course, just
normatively out of the naming game. It wouldn't be a limitation on our
representational capacities, just on how we should employ them.
On the other conception of ineffable objects, as ones about which
nothing can be truly said, it also is generally taken to lead to a paradoxical
2
conclusion. After all, can't we at least truly say about ineffable objects that
nothing can be truly said of them? In this sense ineffable objects can be tied
1
Exodus 3: 13-15.
2
For a discussion of the ineffable in that sense, and a form of an embrace of the
apparent paradoxes, sec Priest (2002).
128 Ihomas Hojiveber
to our main concern: ineffable truths or facts. If nothing can be truly said
about an object o then any fact involving o should be ineffable. But the
latter, ineffable facts, are not paradoxical. It is not required that notl:ing can
be truly said about ineffable facts, only that ineffable facts can't be effed, that
is, one can't utter a sentence such that this sentence expresses, states, or
represents, that fact. I can't state the fact in question, but I might well be
able to say true things about the fact, including that I can't state it. Ineffable
facts or truths are not paradoxical, although ineffable objects, that is, objects
about which nothing can be truly said, do seem to be paradoxical. An
ineffable fact is simply a case where what is true outruns what we can
truly say. It is not a paradox, but whether this is ever so is our concern here.
A gap between language and thought? A third topic connected to the
ineffable is the question whether there are certain facts or propositions that
one can think, but one can't say. That is, are there certain contents that our
thoughts can have, but there is no utterance of a sentence that has that same
content? Some think that there are. One candidate for this are thoughts that
involve phenomenal concepts. Maybe such concepts allow us to think
thoughts that we can't put in language. Another, more traditional, example
is a version of mysticism. According to it we can attain insights by various
means like fasting or mediation, but we can't communicate them to others
after we achieve them. These insights are not supposed to be feelings, but
instead have propositional content. They are thoughts with contents that
can be true or false. However, due to the nature of these contents they
cannot be put into language. Although the mystic can think a thought wit~
that content, they can't communicate it with language. You have to medi-
tate/fast/etc. to gain that insight. Whether either one of these cases obtains is
controversial, of course, but this controversy does not matter now. I am not
concerned with whether there are some limitations of language that are not
limitations of thought. Instead I am concerned with whether there are
certain facts or truths that are simply beyond us in either way, be it thought
or language. I want to find out whether there are truths that we cannot
represent at all, be it in language or be it in thought. Thus from now on
I will take the ineffable to be that which we can neither think nor say.
Whether there is a gap between language and thought thus won't matter for
what is to come, interesting as the question is otherwise.
Conceptual representation vs. other representation. Our issue here is not
whether we can represent everything in some way or other, but rather
whether for every fact or truth we can have a conceptual representation of
that fact or truth. It might well be that something ineffable is going on right
over there, and I could pull out my camera and take a picture of it, and thus
represent it in some way. The issue is not whether I can always do that, but
Are There Ineffable AJpects of Reality? 129
rather whether there is a conceptual representation of every fact or trmh.
Conceptual representations arc paradigmatically the kind of representation
we have in thought or language. The ineffable concerns the limits of
conceptual representation, not the limit of representation more generally.
Fine vs. coarse contents. If the proposition that I am hungry now is different
from the proposition that TH is hungry at t then this truth likely is ineffi1ble for
everyone but me right now. You would only be able to express it if you were me
and even then only at that particular time. This would m<1ke these truths
ineffable, not because minds like ours can't represent them, but because of who
and when you need to be to represent it. If contents are that fine-grained then it
is trivial, but insignificant, that there are many ineffoble truths. To get a more
interesting question we should see whether there are still ineffable truths even if
we consider contents more coarse-grained where perspectival elements like
who, when, and where, you are do not matter. We will thus from now on
assume that contents are coarse-grained enough so that perspectival elements
don't matter, or alternatively, we consider something ineffable if it can't be
represented no matter who, where, and when, you are.
To focus on coarse-grained contents in the following is not to take sides in
the debate whether contents are best taken to be fine-grained or coarse-grained.
It is rather to take sides on the question what the proper notion of the ineffable
is for which we should find out whether there are any ineffable aspects of
reality. If we use a notion of the ineffable tied to fine-grained contents the
answer is clearly that tl1ere are ineffable truths, but that answer doesn't have any
interesting consequences, it is simply guaranteed by how fine-grained contents
arc. 'T'he interesting question remains whether there are ineffable truths when
considering a notion of the ineffable tied to coarse-grain contents. That
question is not trivial, and has the potential to lead to substantial consequences.
We will thus consider contents to be coarse-grained in the following.
De facto ineffable vs. completely ineffable, We need to be clearer on how
the "can" in "can't be thought or said" should be understood when we
consider the ineffable. It is uncontroversial that there is a sense of"can" such
that there are some facts or propositions that we can neither think nor say.
But whether we are also limited in a more permissive sense of "can" is
controversial and a harder and more interesting question. To illustrate the
difference, take the complete sand-metric of planet earth: the precise dis-
tance that every grain of sand on earth presently has to every other one.
Since there are about l 0 24 or so grains of sand on earth this is an incredibly
complex fact. No human being will ever be able to say or think that content.
But this is merely due to a limitation of resources, in particular time. Since
we have a short lifespan we will run out of time before we will be done to
think or say that truth. It is, as we can call it, de facto ineffable. We can
130 Thomas Hofiueber
represent every part of it, in that for each pair of grains of sand we can say
what their distance is. 'fhe whole, ineffable fact is just a conjunction of
many effable facts. This fact is beyond us in the sense that it is too long and
complex, but it is not beyond us in the sense that our representational
capacities <lfe not suitable to represent it in principle. If we had more time
we could do it. We can thus distinguish the de facto ineffable, which is what
we can't think or say, from the completely ineffable, which is what we can't
think or say even with unlimited resources like time and memory. Some
contents we can't think since we are limited on a certain scale. The question
remains whether there are any that we can't think in principle, even if we
allow ourselves unlimited resources on this scale. How to make this precise
is, of course, not completely clear, but the examples of limited time and
memory are clear ones that give us a limitation on a certain scale, and there
might be other similar ones. The real question for us here is whether we are
also limited in other ways, in ways that overcoming limitations of time and
memory won't help. Are there contents that we are simply incapable of
thinking or saying, in principle, in that a mind like ours just can't represent
them conceptually? Are there facts that are simply beyond creatures like us,
even with unlimited time to say or think them? This is the question I hope
to make progress on. Thus from here on, when I wonder whether there is
anything ineffable I will thus ask whether anything is completely ineffable in
the above sense. The completely ineffable is the notion of the ineffable of
interest here. As will become clear below, this won't settle what should
count as something that we can in principle do. Here that notion can be
made precise in various ways, leading to various more precise notions of the
ineffable. All of them are equally good notions, but we need to focus on the
one that leads to the most interesting and most significant question about
whether or not there are ineffable facts. We will revisit this issue below.
Ineffable for whom? The issue I hope to make progress on is not one about
what can be represented in language in principle, but instead is what we
human beings can represent. It is about whether the world outruns our
representational capacities, the ones we can employ. We won't be concerned
here with whether there could be a language suitable for other creatures that
captures everything, or whether other creatures could think everything. Our
topic is whether we can capture everything. Naturally, we should be con-
cerned with whether anything is in principle ineffable for us. And this
question is not about what language in general can represent, or what we
could do if we were gods, with completely different minds, but what we, the
kinds of creatures we in fact are, can do.
Aspects. I stated the main question as whether there are ineffable aspects of
reality. This should not be taken as indicating that the issue is whether
Are There lne.ffizble Aspects of Reality? 1:31
reality <L~ a whole has some ineffable feature. I could have also asked whether
there are certain parts of reality that are ineffable, but this might have
suggested that some spatio temporal part is full of ineffable facts. My concern
is whether any truths or facts are ineffable. If the answer is "yes" then
presumably it won't just be a single instance. Any one ineffable fact will
be closely tied to others, and such ineffable facts considered as groups of
connected facts can be seen as giving us an ineffable aspect of reality.
Nothing should hang on that terminology, though.
Ineffable vs. unknowable vs. incomprehensible. Finally, we need to distin-
guish the ineft1.ble from the incomprehensible or unknowable. It is unknow-
able, I tal<e it, whether the number of grains of sand on planet earth on
February 18, 1923, was odd or even. But it is not ineffable. I can think the
thought that it was odd on that day, and the thought that it was even. Anything
that is ineft1.ble is unknowable, given standard assumptions about knowledge
involving at least a representation of what is supposed to be known, but not the
other way round. Similarly, some things might be beyond what we can
understand or comprehend, but they are not thereby ineft1.ble. It might be
incomprehensible why there is anything at all, but it isn't ineffable that there is
anything at all. The ineffable is simply concerned with what we can represent.
It is not an epistemic notion, but one about our representational capacities. 3
Our question thus is this: are there any facts, truths, or ttue propositions, such
that we cannot, in principle, represent them in either thought or language, even
given arbitraiy resources like time and memory, and even when we individuate
facts and propositions coarsely enough to leave aside perspectival limitations?
Since this is a yes-no question there are two possible answers. We need to find out
which one is the right one, and what follows from it. 4 If the ai1swer is "no" then all
aspects of reality are effi.ble for us, and the following effability thesis is true:
3
Others arc concerned with the ineffable in these other senses. Sec, for example,
Moore (2003a, 2003b). Colin McGinn proposed that the reason why we make no
progress in philosophy is tied to our cognitive limitations, but it is not clear whether
his position is best understood as being tied to a limit of what we can represent, or instead
a limit of what we can understand. See McGinn (1989) or (1993).
4
It could be that although one answer is true in letter, the other is true in spirit. Maybe
there are some isolated facts tied to the paradoxes which can't be represented in any
thought or language, but we can represent the rest. One unsuccessful way to argue for this
is to consider the fact that for some objccto, nothing about o is ever represented. That fact
about o can't be represented without failing to obtain. And there certainly can be some
objects o about which no fact is ever represented. But this doesn't show that this fact can't
be represented, only that when it is represented then it won't be a fact any more. The
content that nothing about o is ever represented is perfectly representable by us, even if we
never do represent it. We can represent it, and in those counterfactual circumstances it is a
false proposition, while in actuality it is a true proposition. The limits of what can be
represented are nor that easily drawn. Thanks here to A. W. Moore.
132 Thomas Ho/weber
( 1) The elfability thesis. Everything is effablc.
If the answer is "yes" then some aspects of reality are ineffable for us, and so
the ineffability thesis is true:
(2) The ineffability thesis. Something is ineffable.
As we saw above, what is at issue is whether any fact, proposmon, or
content, is ineffable, or whether all of them are effablc. And that is to ask
whether it is true that for every proposition p, there is a speech act we can
perform, or thought we can think, that hasp as its content. A proposition is
cffable in speech, we can say, just in case some utterance that we can make
has that proposition as its content. And for that to be true there has to be
some sentence and some context in which we might utter the sentence, such
that this utterance of that sentence in that context has the proposition pas its
content. A proposition is cffable in thought, correspondingly, if we can have
some propositional attitude, a judgment or a belie( that has that proposition
as its content. In either case, effability in thought or language, we would
need some representation of the proposition, either mental or linguistic, that
has the proposition as its content, in the particular context it is employed.
Whether this is so for all propositions is what we need to find out. I will
argue that we should think that the ineffability thesis is true, at least granting
widely shared assumptions.
3. IN SUPPORT OF THE INEFFABLE
Leaving aside those who hold certain unusual views, to which we will get
later, everyone should believe that there are ineffable facts in our sense. But
few live up to what follows from that, in particular for metaphysics. In this
section we will see a number of good arguments for there being ineffable
facts, and we will look at what we can say about what such facts or truths are
like. It might be tempting to say that there can be no good argument given
by us for there being an ineffable fact, since such a fact could not be specified
by us, and thus no example of such a fact could ever be given. Although it is
true that we cannot give an example of such a fact, in that we can never
truly say that the fact that ... is ineffable, there are nonetheless a number of
powerful arguments that there are such facts. We can argue for there being
certain things without giving examples of them, but instead by general
arguments that make it reasonable to accept that there are such things
nonetheless. The most important arguments for this seem to me to be the
following.
Are There Ineffable Aspects of Reality? 133
3.1 Built-in cognitive limitations
The thoughts we can think must fir into our minds. And our minds think a
certain way; they have a certain cognitive setup. So any thought we can
think must have a content that a mind like ours can represent. But our
minds didn't develop with reality as a whole as their representational goal.
They developed to deal with situations that creatures like us have to deal
with to make ic: midsize objects that are reasonably stable and have stable
properties, some of which need to be eaten, some of which need to be
avoided, and so on. "T'he question is why we should think that a mind that
developed to deal with problems in this limited situation and under those
selection pressures should be good enough to represent everything there is to
represent about all of reality. We know that not all of reality is like the world
of stable midsize objects. The very small is very difterent than that, for
example. We should thus expect that a mind that has developed like ours
won't be suitable for all of reality. Our biological setup imposes a constraint
on how we must think. 'T'his constraint arose in response to selection pressures
that came from a special kind of an environment. We can expect minds like
ours to be good enough to deal with the situation they evolved to deal with,
but not good enough to deal with any situation whatsoever. Thus we should
expect that our mind has a hardwired constraint on what it can represent. 5
3.2 The argument from analogy
Although we can't give an example of a fact ineffable for us, we can give
examples of facts that are incff:'lble for other, simpler creatures. Take a
honeybee, which can represent various things about its environment like
where the nectar is, but it is in principle incapable of representing that there
is an economic crisis in Greece. Its mind is just not suitable to represent such
facts, even though it can represent other facts. Thar there is an economic
crisis in Greece is a fact ineffable for the honeybee, but not for us. But we
can imagine that there are other creatures that relate to us like we relate to
the honeybee. We can imagine that there are vastly superior aliens or gods,
say, who look down at us like we look down at the honeybee. And
analogously, they would say that we humans can't possibly represent that
p, while they clearly c,'ln. They would be able to give examples of facts that
are ineffable for us, but not for them. Whether or not there really are such
5
Sec Fodor (1983), 119ff; Nagel (1986), 90ffi and Chomsky (1975). Chomsky's
views on this matter arc more carefully discussed in Collins (2002), which comains many
references to particular passages of Chomsky's work.
134 Thomas Hofiueber
<tlicns or gods doesn't matter for this argument. The point simply is that this
analogous reasoning makes clear that we should expect there to be such facts
that could be mentioned by the aliens or gods as ineffable for us, but not for
them. The facts are there, whether the aliens or gods are there or not. 6
It might well be that the honeybee's representations are not conceptual
representations and thus might not have propositional contents at all, but
merely indicate something about the world. They might carry information,
but not have propositional content.7 But this doesn't really change the
situation. Just as we can point to information that the honeybee can't
carry with its representational capacities, aliens might point to propositional
contents that we can't represent conceptually, besides higher forms of
representation that they have in addition. In the end we should think of
what can be represented by us as being somewhere on a scale, with the
honeybee on one side of us, and other creatures, real or imagined, on t~1e
other. And what those further over on the scale can represent is ineffable for
us, and thus we should think that some aspects of reality are ineffable for us.
3.3 Cardinality arguments
Our language is built up from finitely many basic vocabulary items with
finitely many ways to combine them to give us a sentence, which has to be of
finite length. Thus overall we have countably many sentences available to
represent reality. Similar considerations at the level of concepts support that
we can have at most countably many types of thoughts. But there are
uncountably many facts or propositions to be represented. Here is a simple
argument: for every real number r there is the fact that r is a real number. For
different real numbers these facts are different, and since there arc uncount-
ably many real numbers there are uncountably many facts to be represented.
Although such cardinality arguments are very powerful, the simple out-
line given above is a bit too quick. Although it is true that there arc only
countably many sentences in our languages, this does not guarantee that we
can only represent countably many facts. We can use the same sentence to
represent different facts on different uses of this sentence, as with sentences
that contain indexicals or demonstratives like 'Tl! have another one of
those." But while this is correct, it is not clear how this would help even
with the simple argument using the real numbers. Demonstratives and
context sensitivity don't seem to help much in referring to real numbers,
and that was only the most simple and straightforward cardinality argument.
More generally, we can argue that whatever the cardinality of the set of
effable facts might be, we can take some set S of larger cardinality, and then
consider for every a which is a member of S, the fact that ti is a member of
6 7
See Nagel (1986), 95f Sec Dretskc (1981 ).
Are There lnejj{1ble AJpects o_f'Rmlity? 135
S. There arc just as many such facts as there arc members ofS, and thus most
of these facts must be ineffable. Overall then we should thus expect that the
cardinality of facts we can represent is smaller than the cardinality of all facts.
3.4 Explaining effability
If nothing is ineffable then the effability thesis is true and thus what facts
obtain and which facts we can represent are exactly the same. But those are
two very different things which would then exactly coincide. What we can
represent is one thing, what reality is like is at first a quite different thing. If
they coincide then we should ask for an explanation of why these two things
coincide. It is conceivable that they do coincide. Maybe our representational
capacities have reached the limit of what can be represented. Maybe we just
made it to the top, while other creatures, including our ancestors, were still
on the way up, unlikely as all this may be. Maybe the world is simple enough
so that we can represent all of it. Maybe we got lucky and are able to
represent every fact. The effability thesis could be true, by accident, but does
it have to be true? If the effability thesis is true, then we should ask for an
explanation of why it is true. And if no such explanation is forthcoming we
should expect that it isn't true. We shouldn't expect that two different
things coincide, and without an explanation of the cffability thesis we
should expect it to be false. But what could explain that what reality is
like and what we can represent about it coincide?
The most natural way one might try to explain why what reality is like
and what we can represent coincide is via a connection of what reality is
like and our representational capacities. One route for such a connection is
of limited use: what reality is like affects what we can represent. This route
can explain why our representations are sometimes accurate, but not why
they exhaust all of reality. That reality affects and forms our representational
capacities makes plausible that we sometimes represent correctly, but it
doesn't explain why we can represent all of reality. The other route is
more promising here: what reality is like is affected by our representational
capacities, in particular, what there is to represent about reality is due to us
and our minds. This is a version of idealism, and it is in a sense the natural
companion of the effability thesis. Reality is guaranteed to be eftable by us in
its entirety, since we, in particular our representational capacities, are
responsible for what there is and what it is like. No wonder our minds are
good enough for all of reality, since reality somehow comes from our minds.
Idealism could in principle explain why the effability thesis holds, but we
have good reason to think that this form of idealism is false. That is, we have
good reason to think that mtlity does not depend on us in the alluded to
sense: what there is and what it is like does not depend on us in a natural
sense of dependence. There would have been electrons and they would be
136 JJ;omas Hojiveber
like what they are in fact like even if we wouldn't have existed, and so electrons
and what they are like don't depend on us. Furthermore, there was a time
before there were human beings when reality was otherwise pretty much as it is
now. So, in a natural sense of dependence, reality doesn't depend on us
globally. These are simple and possibly naive arguments, but ifidealism should
explain why the effability thesis holds it will need to be spelled out in a way that
makes sense ofa dependence of reality on us that can support and explain the
effability thesis. There certainly are options on the table. One could try to
analyze the content of statements of dependence in a way that makes them
acceptable to idealism. Or one could develop the idealism in a way that places
us in some sense outside of time, and connect time and the temporal aspect of
reality to us as well. One version of this is well known (Kant 1781), but it is not
clear whether it is coherent, what a coherent formulation would look like, and
whether it is compatible with other things we take ourselves to know to be true.
An idealist explanation is in principle possible, but ones along the lines
outlined above seem to have little going for them.
'T'hc idealist strategy to explain the effability thesis outlined above in effect
connects two versions of idealism. Of those two one is reasonably taken to
be false, and the other is closely tied to our main topic. To introduce some
terminology, let us call ontological idealism the view that what there is
depends on minds, in particular our minds, in a sense to be made more
precise. Let us call conceptual idealism the view that what in principle can be
truly said or thought about reality, what the range of the conceptual or
propositional is that can be employed in principle to apply to reality,
depends on us, in a sense to be spelled out. Conceptual idealism is in essence
the view that the effability thesis holds not by mere accident, but for a reason
tied to us. Conceptual idealism combines the effability thesis with a certain
explanation of why it holds. Ontological idealism could support conceptual
idealism. If what there is depends on us then one way this could be would tie
what there is and what it is like to our representational capacities, in a way
that would guarantee the truth of the effability thesis. But ontological
idealism is false, or so we have good reason to think. The question remains
whether conceptual idealism is nonetheless true, for other reasons. Could
conceptual idealism be true even though ontological idealism is false? For
that to be so the effability thesis has to be true, and it has to be true for a
certain reason, not just by accident. So far we have seen no reason why that
should be so; to the contraty, we have seen that there is little hope to explain
why the effability thesis might hold. However, we will revisit this connec-
tion below, in section 5.8, where this possibility is seriously explored. But
without such reasons being on the horizon so far, we should side with the
ineffability thesis and accept that reality outruns what we can represent
about it.
Are There Ineffable AJpects of Reality? 137
3.5 The sources of ineffability
We have seen the outlines of four strong arguments for there being ineffable
facts. Three direct ones, and one indirect one, via there being no good
explanation why the effitbility thesis should be true. The smart money is
thus on there being ineffable facts, even though we cannot give an example.
Before we can see what follows from this we should think a bit about what
such facts might be like, and how they might be similar or different from
cffi1blc facts in various ways. The easiest way to approach this question is to
think about how our representations of facts might be limited. And the easiest
way to do that is to think about the paradigmatic way in which we represent
the world: with a subject-predicate sentence, representing an object having a
property. How might such a way to represent the world be limited? There are
three ways in which we might be limited with such representations: (I) we
might not be able to represent a certain object; (2) we might not be able to
represent a certain property; or (3) the strucUll'C of a subject-predicate
sentence might not be suitable to represent a certain truth or fact. Let's look
at them in turn to sec which ones are the likely sources of our limitation.
Missing objects. We might be unable to talk or think about some objects.
To illustrate, let's consider one way this could be, namely that singular
thought and singular reference require a causal connection between us and
the object we think or talk about. This gives rise to two possible limitations:
objects that arc not causally efficacious at all, and objects that are causally
efficacious in general, but that arc not causally related to us. Focusing on the
second case first, we can note that we are not causally connected to all objects
in the universe, for example not those that are outside of our light cone.
Under those conditions we would thus be unable to have singular thoughts
about everything outside of our light cone. This would give us lots of
ineffable truths: all those that involve the objects we can't have singular
thoughts about. Thus assuming, again only for the moment, that if causal
connection is required for singular representation then all these truths are
examples of ineffable truths.
The notion of the ineffi1ble on which this is true is not the notion we
should be concerned with. It concerns merely the de facto ineffable, not the
completely ineffable. All that is missing in this case is our causal contact to
the object. This is something we could have if only we were closer to the
object, close enough to have it inside of our light cone so that a causal
connection could obtain. Given where we are and where the object is (and
the assumed requirement on singular thought) we can't think singular
thoughts about that object. But we could think these thoughts if we were
closer. So, in a sense the ineffabiliry of truths involving such objects is due to
138 Thomas Hofweber
our placement in the world, how we are causally isolated from them, but not
clue to our representational capacities not being good enough. If we were
closer then our capacities would be all that is needed.
The issue is what we should keep fixed and what we should allow to vary
when we ask whether we ctm think a thought. We want to keep fixed our
basic cognitive setup, but we should not fix the place in the world where we
happen to be located. This is part of what we need to get clear on when we
try to determine what the proper notion of the ineffable is that we should be
concerned with. The ineftable is that which we can't say or think, but "can"
can be understood in many ways and on each prccisification of "can" do we
get a more precise and specific notion of the ineffable. On a notion where we
fix our place in the world we might get ineffable truths about far away
objects, but that is not the notion of the ineffable that is tied to the worries
about our minds not being good enough to carry out ambitious metaphys-
ics. The more interesting and more significant notion is thus the one where
we do not fix our place in the world.
One way to illustrate the relevant notion of a truth being beyond us in
principle is the incommunictibility test: some other creature who can repre-
sent that truth couldn't communicate it to us in principle, no matter how
hard they tried. If we were to encounter some creature who can represent
that truth, and who has mastered our language and thus can communicate
with us in general, then this creature would nonetheless be unable to
communicate this truth to us. We are limited in this case that we can't
represent this truth no matter what help we might get. Even someone who
can represent this truth couldn't help us to do better.
Consider for our case of missing objects someone who can represent a
content involving an object that is beyond us. That person could commu-
nicate this content to us, and thereby allow us to think about that object.
Suppose, for purposes of illustration again, a highly advanced alien creature
is talking to us and it can represent everything there is to represent, and in
particular it c.o1.n refer to all objects. The alien could then help us to the
contents that we couldn't get without it since these contents involve objects
that are otherwise beyond our referential reach. The alien could just tell us
what its name for that object is, and we can then refer to the object in
question with that name, just as we in general exploit the referential success
of other humans when we refer to objects we learn about from them. The
referential connection can be mediated via others, be they humans, aliens, or
gods. That is how we manage to refer to Socrates, afrer all. Thus even going
back to our illustration of the need for a causal connection in our unaided
situation for reference to an object, this is not an in principle limitation. Gods
or aliens that are not limited this way could allow us to piggyback on their
success, not just with faraway objects, but even with objects that are causally
Are There Ineffable Asperts of Reality? 139
inert. Reference to an object is thus not a limitation that applies to us in
principle, unless of course some objects are in principle beyond what can be
referred to, even by God, but we have no reason to think that this is the case.
Even if aliens or gods could help us to access all objects, they might not be
able to help us with other cases of a source of the ineffable. There might be
some truths where the alien would have to tell us that although it can think
and say them, it rnn't comrnunirnte them to us, since creatures like us just
can't grasp them. 8 If there is ineffability of this later kind it would have to go
beyond merely inaccessible objects. And whether some part of reality is like
that, not just in fact out of our reach, but in principle beyond creatures
like us to grasp, that is the interesting question that might have significant
consequences for metaphysics.
The interesting notion of the ineffable is thus the one that goes beyond a
limitation to think about or refer to particular objects. We could make the
notion of the ineffable more precise along the lines where a limitation to
refer to objects would make a fact involving that object ineffable, and we
could make it precise where it wouldn't. Both are perfectly good notions of
the ineffable, but the more interesting one is the latter, which passes the
incommunicability test. To make this explicit, we should take the ineffable
to be the object-permitting ineffable: that which is completely ineffable even
if we allow ourselves access to all objects. Our arguments above for there
being ineffable facts did not on the face of it rely on just a limitation to refer
to objects, and thus these arguments should still be compelling even with
the notion of the ineffable as being object-permitting. What makes ineffable
facts beyond us is thus not simply objects which are referentially inaccess-
ible. The real source of ineffability lies somewhere else.
Missing properties. The same issue now arises with properties. What if
there are some properties that we, somehow, can't represent? And in the case
of properties we might have to distinguish two ways in which we might fail
to represent them. First, it might be that the situation with properties is very
much like the one with objects. Properties, one could argue, are simply
things or entities, just like regular objects. We might be unable to pick out
that entity with a term that denotes it. But here the aliens might again be
able to help us out. They might give us a name to use that refers to
the property that otherwise was beyond our referential abilities. Second,
we might be able to fail to represent that property in a simple subject-
predicate sentence "a is F." Maybe we can't have a predicate "is F' without
outside help, and this might be a source of ineffability for us. Herc, too, the
8
A version of this scenario is described in the novel The Ophiuchi Hotline by John
Varley (1977). Thanks to David Baker for this reference.
140 lhomm flofiueber
issue arises if the aliens could do it for us. Could they just let us use one of
their predicates, say "is wallercancsfSa?" This might be enough for repre-
senting even if it isn't enough for understanding what we thereby represent.
And there is an issue whether or not the two cases are really different. After
all, a is F just in case a has Fness. If the aliens give us a name for Fness then
maybe this is all we need to represent that tl is F, and thus the case of
properties really reduces to the case of objects.
All this might go too far, but it can be taken even further. Facts, truths, or
propositions, too, can be seen as entities. And if we can refer to any objects
or properties, why not to any fact or proposition? Any proposition p is
equivalent to the proposition that p is true. And we can eff the truth
predicate which is all that we need besides the name for the proposition to
state something equivalent to p. 9 Can we really be satisfied with a sense of
the effable where we eff every truth t since aliens or gods could give us a
name for t, and we can eff that t has the property of being true? This is
dearly unsatisfactory in some ways, and might appear to be a too shallow
victo1y for the dfability thesis. But what precisely is unsatisfacto1y about it?
One concern is about what counts as the same fact or proposition. The
equivalence of p and it's true that p holds in the sense that it is necessary that
if one is true then so is the other. A more fine-grained notion of equivalence
might separate these two propositions or contents, and effing one of them
might then not be enough to eff the other. On the other hand, a too fine-
grained notion of equivalence trivializes the issue in favor of the ineffability
thesis, as we saw above in our brief discussion of pcrspcctival contents.
It is preferable to take a thco1y-ncutral and coarse-grained notion of
contents or propositions and then to avoid trivializing the issue in other
ways. And here there is a good middle ground. On the one hand we didn't
want to accept the ineffability thesis simply on the grounds that some
objects arc too distant from us in spacetime for us to refer to them. This
is a restriction that is not one "in principle" in a sense analogous to the
restrictions of time and memory not being restrictions in principle. But
allowing objects to be free threatens to trivialize the issue, since properties
and, in particular, propositions or contents themselves can be seen as
objects, too. We can reach a middle ground by allowing us objects to be
free, but explicitly excluding objects or entities that are content-like objects,
for example propositions or facts. This leaves us with a substantial question
to consider. We can remain neutral, as should become clear shortly, on
9
Whether we indeed have a truth predicate that could be used to apply to any
proposition whatsoever is controversial (see Field (1994)), but for the moment we can
leave that issue aside.
Are There Ineffable Aspects rf'Rea!ity? 141
whether we also have to exclude properties from the objects we get for free. to
The reason for that is that the real worry about ineffability co1~1es not from
there being ineffable properties, but from somewhere else.
Missing structure. We represent the world paradigmatically with represen-
tations that have subject-predicate structure. But why should we think that
representations that have this struct11rc arc enough to represent everything
there is to represent? We can represent truths with other structures as well, but
that only pushes the issue towards all those structures together. We have good
reason t~ think that there is more to reality than objects having properties.
M:my ot our own representations of reality are not in the subject-predicate
form. Consider, for example, explanatory relationships: p because q. 'T'hat is
not a simple subject-predicate sentence, but a complex sentence with a
semen ti al connective "because." Since the negation of such a sentence is
also not in subject-predicate form, some such representations are true. And
thus some sentences of this kind truly describe reality, unless all non-subjcct-
prcdicate sentences are systematically non-descriptive, analogous to an
exprcssivist treatment of normative discourse. But this last qualification is
too far-fetched to be a real limitation. We have lots of good reasons to think
that sentences that are not in subject-predicate form are not systematically
different from those that are in their attempts to describe reality. Since it isn't
clear how the same fact could be represented with a simple subject-predicate
sentence it seems that some facts about reality require a representation with a
difterent structure than that of a subject-predicate reprcscntation. 11 Of
course, in this case we do have the resources to represent these facts. W c do
have a sentential connective that allows us to represent explanatory relation-
ships. But the worry is that if subject-predicate representations in general are
not enough, and sometimes extra resources arc needed, then why think that
we have all these extra resources available to us? Even if we could access all
objects and all properties, the worry would remain whether we can represent
all the facts, since apparently some of the facts we can represent require a
representation with a strucn1re different from a subject-predicate one. And
io A further worry about properties trivializing the debate is this. If p is the Gh'C then
any object has the property of being such that pis the case. Thus if all properties arc free
then t~i~ connection g'.vcs us at least a truth conditional equivalent content for any
proposmon p. Whether it gives us the same content can be further debated, of course. Not
to trivialize the issue in this way we would again need to restrict the properties that come
for free, for example to ones that distinguish berween objects in the same world, or some
other way.
ii I am no:' leaving asid~ the issue discussed above about trivializing this by taking
every proposit1011 p to be equivalent top being true, which is in subject-predicate form, or
by taking p to be equivalent to the universe having the property of things are being
such that p.
142 Thomas Ho/weber
once we recognize that we can, wonder why we should think that our minds
have access to all the structures required to represent all facts. Why think that
the kinds of representations that are available in human languages and to
human minds are good enough to represent all facts and truths? We can say
that a fact is structumlly inejfi1ble if the source of its ineffability is the required
structure of a representation of it, i.e. a structure we do not have access to. 12
Structural ineffability gets at the heart of the matter. Any limitation of this
kind would pass the incommunicability test: if there are facts that cannot be
represented by representations available to the human mind then <1liens who
do not have this limitation could not communicate them to us. These facts
would be in principle beyond us, not just due to our location, our access to
objects, but clue to the nature of our minds. These facts would be ineffable for
us in the sense that matters.
The arguments for there being ineffable £-acts discussed above support there
being ineffable facts in this sense: structurally ineffable facts. Just as the mind
of simpler creatures doesn't have the basic representational resources to
represent all facts, so we should think by analogy and due to our built-in
cognitive setup, that some facts are beyond what creatures like us can represent
in principle. To represent such facts would require access to ways of repre-
senting the world which are beyond those available to our minds. This is what
the arguments for the ineffable point to, ;md this is what has significant
consequences for metaphysics. It is now time to see what follows from it.
4. THE SUB-ALGEBRA HYPOTHESIS
AND ITS CONSEQUENCES
4.1 The hiddenness of the ineffable
Suppose then that we should accept that some aspects of reality are ineffable for
us, in the sense that the human mind is not in principle capable of representing
them in thought or language. What should we conclude from this? Even though
I suspect many would accept that we arc limited in what we can represent, few
draw much of a conclusion from this. But it could be taken to lead to a largely
negative and skeptical conclusion about inquiry in general and our attempts to
find out what reality is like. I believe that there are indeed significant conclusions
to be drawn, but they are neither that broad nor that skeptical. But to sec what
we should properly conclude from the ineffable we need to get clear on one
farther aspect of the ineffable that is puzzling. This is the problem why the
12
For an argument that structural inefE1bility is impossible, sec Filcheva (2015).
Are There Ineffable Aspects ofRetdity? 143
ineffable is so well hidden from us. Any fact that is ineffable would make it
hidden from us in the sense that we can't represent it. But if there arc indeed
incf£1ble facts then they are more hidden from us than simply being unrepre-
sentable, as I hope to make clear shortly. This cxtra-hiddcnness is puzzling, and
points to an account of tl1e relationship between effablc and ineffable facts,
which in turn is significant for what we should conclude from there being
ineffi1blc facts for the limits of human inquiry. To illustrate all this, let us first see
how the ineffi1blc is more hidden than it needs to be.
A first thing to note is that we never seem to perceive something ineffable.
This is not a triviality. We can well wonder why it is never the case that we
perceive something that we simply cannot describe or represent in thought. In
such a situation we would be perceptually connected to a fact which we
cannot represent in thought or language, and realize that our representational
capacities give out here. Why docs this never happen? It is not inconceivable
that one day we open a door, look inside, and recognize that what we see is
simply beyond what we can describe in words. All we can say is just "wow!"
We sec something that we can't represent conceptually, while at the same
time realizing that what we sec is simply beyond us conceptually. It is hard to
imagine what that would be like in more detail, in part because it never
happens, in part because we might have to rely on our concepts in imagin-
ation. The question is why this never happens. One answer, of course, would
be that the effability thesis is tmc and nothing is ineffitblc, but there arc also
other possible explanations. One of them could come from the philosophy of
perception. It might be that everything we perceive has to be conceptualized,
<md what we perceive has to be tied to the conceptual content of the
perception. Since conceptual content, involving our concepts, can be repre-
sented by us, it is no wonder that what we perceive can be represented by us. 13
But even if we could explain why we never perceive the ineffable without
relying on the cffi1bility thesis, the question remains why we nonetheless
never encounter the incf£1.blc in other ways. It might be that what we
perceive has to be conceptualized, and thus is cffablc, for reasons having
to do with how perception works. But this doesn't answer the question why
we nonetheless apparently never encounter the ineffable in other ways. Even
if we never perceive it directly, we might realize that there is something
ineffable right here, behind this door. In such case we could realize that what
we can perceive of the situation is cffable, but there is more to it than what
we can perceive, in the sense of represent in perception. This would be a
scenario where we encounter the ineffable, and we recognize that what we
encounter is ineffable. This could not just happen in unusual situations, but
u One way this might go is, of course, in Kant (1781).
144 Thomas Ho/weber
it could conceivably be pan of regular scientific theorizing. We might
recognize regularly during inquiry that now we are approaching the limits
of what we can represent, and that the answer to our problem lies beyond it.
But why does this never seem to happen, given all the reasons we have seen
for there being ineffable facts? Why is the ineffable so well hidden from us
and apparently so irrelevant for our attempts to understand various parts of
the world in inquiry?· ['he ineffable seems to fall out of the picture and for all
intents and purposes every fact we encounter is effable. This naturally gives
rise to the impression that the effability thesis should be true, even though
we have aood reason to think it is false. To properly appreciate the signifi-
o . b
cance of the ineffable we need to get clear on why the effable facts seem to e
all the ones we ever encounter.
4.2 The sub-algebra hypothesis
If we accept the arguments for there being ineffable facts then we need to
understand how they might be related to the effable ones, in particular why
the ineffable facts seem to be as irrelevant to ordinary inquiry as they seem to
be, and why they arc so well-hidden from us. What could explain that the
possibly vast range of ineffable facts is systematically hidden from us in a
stronger way than is suggested by their being merely ineff.'lble? There is a
way to understand this which seems to me to be the best way to make sense
of it, and it is best spelled out with a mathematical an;1logy.
Consider a simple mathematical structure, say the integers with addition,
multiplication, and subtraction:
... -3,-2,-1,0, 1,2,3, ... (+, x,-)
The integers are closed under these three arithmetical operations. The sum,
product, or subtraction of any two integers is always another integer.
The integers with these operations thus form an algebraic structure. This
structure corresponds to a language suitable to describe it: it has a name for
each integer, and function symbols for each operation, and additional basic
logical vocabulary. 1;, This we could call the language of the (particular
instance of) the structure.
Now consider the world, so to speak, from the point of view of an integer,
thinking about the integers with that structure. The integers, I imagine for_
the sake of the example, can capture the world in terms of the language of
their structure. And from their point of view it will seem perfectly natural,
14 The additional vocabulary would in this case just include "=" and Boolean con-
nectives. I don't allow variables or quantifiers here for the sake of the example.
Are There lne.ffi1b!e Aspects of Reality? 145
even obvious, that the integers is all there is. After all, the sum of any two of
them is <1lways <mother integer, and so is the product or subtraction.
Furthermore, any question the integers can ask about the world in their
language will have an answer that can be stated in their language. What's
7 x 8? It's another integer. We can say that in their situation they enjoy
question-answer completeness: if you can state the question then you can state
the answer. From the point of view of the integers, they naturally take it that
they can capture all of reality there is to capture. If we were integers with
those representational resources it would seem compelling to us that the
integers is all there is.
All this would be so even if numerical reality is much richer than that,
with the integers being embedded in different, larger structures, for example
the rational numbers with the above operations as well as also division:
l I
... - 3 ... - 2 ... - !... 2... 0... 2... l ... 2 ... 3 ... (+. x.. -.. -;-)
,
Here there are infinitely many other numbers between any two integers,
getting arbitrarily close to them. The rational numbers, just like the integers,
are closed under addition, subtraction and multiplication. In addition the
ration<U numbers are closed under division (leaving out 0, as usual), whereas
the integers are not. Although the integers are embedded in the rational
numbers, the other rational numbers arc completely hidden from them.
Since the integers form a sub-structure, or sub-algebra, of the rational num-
bers with addition, multiplication, and subtraction, these other rational
numbers can never be reached from the integers with those operations: they
always lead back to the integers. But if the integers just had access to division,
say, then they could get out of their sub-algebra and reach the rest of the
rational numbers. And if we would add just one more rational number, say!,
to them then they could use that to reach lots of other rational numbers: I !,
2 ! and so on. But from the point of view of the integers, they are all there is.
And this will seem clear and compelling to them, since after all, given their
resources, what else could there be? Sums and products of integers are always
integers. This is so even though the integers are surrounded by and thinly
spread among things that arc not integers. These other aspects of numerical
reality are completely hidden from them.
This could be our situation. The parts of reality that we can represent
might form a sub-algebra of all of reality. That is to say, it might be that we
can represent certain objects, events, and propositions, and certain relations
or operations on them such that whenever we apply the operations to things
we can represent we get something that we can represent as well. Whenever
we can represent an event, say, then we can represent the cause of that event.
146 Thomas Hojiueher
Whenever we can represent a fact we can represent the explanation of that
fact, and so on. Our representational system might be a closed structure
analogous to an algebraic structure. And that structure might be a sub-
structure or sub-algebra of all of reality. This hypothesis we can call the suh-
algehm hypothesis: the hypothesis that what we can represent forms a proper
sub-structure of all there is to represent. It is based on the analogy to an
algebraic structure and a sub-algebra of it. If it were correct then we would
be much like the integers. We can never represent aspects of the much
richer reality we arc part of, but it would seem to us that the parts of reality
we can represent are all there is. After all, all operations and relations that we
can represent and that hold among things we can represent lead to things we
can represent. So, from our point of view, what else could there be?
Everything else would be systematically hidden from us. Similarly, we
should expect that we have question-answer completeness, and so for any
question we can ask we can state the answer (whether or not we can know
that this is the correct answer). If this is our situation then it would appear to
us that eve1ything is effable, even though what we can eff is surrounded by,
and possibly thinly spread among, the ineffable. We can never get there
from our point of view, and it will be completely hidden from us.
In considering this hypothesis we can note right away that it can't be quite
our situation. First of all, we don't enjoy question-answer completeness in
the way the integers would. If the effability thesis is true then, of course, we
can state any answer to any question since we can state everything. But if the
ineffability thesis is true then we Gm ask questions where we know we can't
state the answer, for example "What are all the ineffable truths?" Still, we
might enjoy large-scale question-answer completeness, which is question-
answer completeness leaving aside questions that deal with the ineffable
and related questions. In general, and for almost all cases, it might be that
if we can state the question, we can state the answer. And just this seems to be
the case. When we ask what caused something, or what explains something,
then we can in general at least state the answer, even if we don't know that it
is the right answer. This fact we can understand on the sub-algebra hypoth-
esis even if the ineffability thesis is true: causal and explanatoty relations are
ones under which our structure is closed. We can represent the explanations
of what we can represent, and we can represent the causes of what we can
represent. And again, all this could be true even though we are surrounded by
the ineffable. And just like the very same arithmetical operations of multi-
plication and addition can be seen as applying to the integers as well as the
rational numbers, so causal and explanatory relations might hold among the
facts or events we can represent as well as those that are ineffable for us. The
ineffable aspects of reality might be very much like the ones we can eff, just
outside of our sub-structure, or they might be completely different.
Are There !neffeh!e Aspects of Reality? 147
The sub-algebra hypothesis would explain why the ineffable is systemat-
ically hidden from us even though it is possibly a vast part of reality. Since it
is outside of our sub-algebra we won't encounter it via causal or explanatory
relationships. We can expect our sub-algebra to be closed under causal and
explanatory connections. And we can expect our sub-algebra to be properly
integrated with those things that causally effect us, for example in percep-
tion. Such a sub-algebra would be a very stable resting point for a represen-
tational system. There is little need to develop it further, even if it only
captures a small part of what there is to represent. On the other hand, if a
representational system docs not form a dosed structure, or at least some-
thing reasonably close to it, then we would expect it to develop further if it
develops at all. But once it is reasonably closed it rests at a stable place. The
sub-algebra hypothesis explains why the ineffable is systematically hidden
from us even if the ineffability thesis is true. It accommodates what needs to
be accommodated, and we have good reason to think it holds. The question
remains, however, what follows from it?
4.3 Ineffability and modesty
The sub-algebra hypothesis makes clear in what sense the ineffable matters,
and in what sense it doesn't matter. It doesn't matter locally, i.e. for
particular questions of fact that aren't concerned with reality as a whole. If
I ask an ordinary question like why is there a sandwich on the table, who ate
my apple, or why is the sky blue, then I should expect that I can state the
answer. Causal and explanato1y relationships are part of my representational
system, and thus are what my algebra is closed under. For ordinary local
questions the ineffable will fall out of the picture. 15
But not so for global questions about all of reality. Here, too, our
reasonable question-answer completeness will likely allow us to state the
answer if we can state the question. But the ineffable and how it is hidden
from us will ofren mislead us into accepting the wrong answer. This is
worrisome for questions about materialism and naturalism in particular.
From our point of view it can seem perfectly compelling that evetything
is material and all there is fits into the natural world. So the questions "Is
everything material?" and "Is the natural world all there is?" are questions we
can state, and whose answers (yes or no) we can state as well. But our
representational limitations might lead us to accept the wrong answers. If
15
Quantified claims, when relevant to local issues, should be taken to be restricted to
the locally relevant domain, and thus they won't range over all of reality, in contrast
to those that arc explicitly intended to be unrestricted, like the ones that arc intended to
make claims about all of reality.
148 Thomas Ho/weber
the sub-algebra hypothesis is correct then we might be in the materialist/
naturalist sub-algebra in a largely non-naturalist world. If this is our situ-
ation we would naturally, but in this case incorrectly, hold that materialism
or naturalism is true. Just as the integers would find it compelling to think
that all there is are integers, even when there are infinitely many other
numbers before we get to the next integer. We might similarly find natur-
alism compelling even though the non-natural is infinitely close to us, and
all around us, but systematically hidden from us. The ineffable is locally
irrelevant, but globally central. And it is, in particular, central for the large-
scale metaphysical questions about all of reality.
The sub-algebra hypothesis is not a skeptical hypothesis, in the sense that
it is not a scenario that we can't rule out to obtain, and which invalidates our
entitlement to our ordinary beliefs. On the contrary, it is a scenario that we
have reason to believe obtains, but on this scenario our ordinary beliefs are
taken to be true. The parts of reality that we can represent we do represent
correctly, or so we can grant here. But when we want to make claims about
all of reality then we reach a limit. The sub-algebra hypothesis does not
warrant a rejection of ttying to answer large-scale metaphysical questions
about reality, but it does warrant a form of modesty. We must recognize that
our situation is one where these questions are to be approached with a sense
that whatever answer seems compelling to us might simply reflect our own
limitations, but not how reality actually is. The ineffability thesis combined
with the sub-algebra hypothesis in particular suggests modesty about global
metaphysical questions. Modesty is not agnosticism or quictism, but it is a
step in that general direction. 16 It does not justify the abandonment of
grand metaphysical theorizing, but it does justify that such theorizing is
different in its epistemic status from other parts of inquiry. How different it
is will depend on how strong the reasons are that we have for the incft-ibility
thesis and the sub-algebra hypothesis. It will be a difference in degree, and to
what degree is not clear so far. Modesty for grand metaphysics follows, to
what degree is left open.
The argument for modesty given here is importantly different from the
argument that we should be modest about judging how many people are in
this room, since after all there might be lots of invisible and otherwise
undetectable people all around us. We have no reason to think that there
are invisible people around us; that is just a hypothesis we might not be able
to rule out. But we do have reason to think that there arc ineffable facts, and
that these facts are systematically hidden from us. These reasons are not
16
Modesty thus contrasts with the positions taken by Gideon Rosen (1994) and Sven
Rosenkranz (2007).
Are There lneffiible 1bpects of Rert!ity? 149
conclusive, of course, but they arc good reasons nonetheless. Thus reflecting
on our own situation should lead us to conclude that there is a reasonable
expectation that we fall short of complete effability. We have reason to
believe that what we can represent is less than what is the case. We also have
reason to believe that the ineffable is systematically hidden from us and that
the sub-algebra hypothesis is correct. Thus we can expect to be misled in our
judgments about global features of reality, and so modesty in grand meta-
physics is advisable, and the ambitions of metaphysics need to be toned
down. At the same time we should expect the ineft-ible to be locally
irrelevant. The consequence of all this is modesty for metaphysics, but it
is insignificant for most of the rest.
This would be a natural place to end die chapter, but I am afraid it is only
half of the story. Although most philosophers hold views that should make
all the above arguments compelling to them, there is an assumption in the
background of our discussion so far that is worth making explicit and that
might well be false. I have to confess that I have argued elsewhere that it is
17
false. If it is indeed false then everything changes for our discussion, or so
I hope to make clear in the following part of the paper. If the assumption is
false then it naturally leads to a defense of the cffability thesis and its natural
companion: idealism. To make this assumption explicit and to argue that
everything depends on it will hopefully justify going on for a bit longer.
5. INTERNALISM, EFFABILITY, AND IDEALISM
Our starting point was to wonder about the relationship between two
different things: which facts obtain on the one hand, and which facts we
can represent in thought or language on the other. The general relationship
between these two was illustrated with the Venn diagram in section 2. Here
the idea was that on the one side is reality, and on the other side arc we, with
our attempts to represent reality. The notion of reality can be made more
precise, as usual, in two ways: the totality of what there is, or the totality of
what is the case. On the latter conception it is simply the totality of all the
facts that obtain. Reality, understood as all that is the case, is taken to be
simply there, waiting for us to represent some of it in thought or language.
The totality of facts forms an independent domain: a domain of all truths,
facts, or true propositions. Reality, on this broadly propositional conception
of it, just as reality on the broadly ontological conception as the totality of
what there is, is simply there, independently of us, with no special place for
17
See Hofaebcr (2006) (2009), and, in particular (2016b).
150 lhomas Hofiueber
us in what it in general is like. But this general picture of the relationship
between the totality of facts and us might be false. Although this picture is
widely accepted, we in fact have good reason to think that it is mistaken. In
the final parts of this paper we will look more at this standard picture of the
propositional, what its alternative is, and how the issue about which one is
right relates to the question about the ineffable. But first, let's get clearer on
the standard picture of the propositional and its alternative.
5.1 Talk about propositions: that-clauses
When we talk about propositions or facts we do so most directly with a that-
clause, for example in the ascription of content to an utterance or
representation:
(3) A said that p.
The most common way to understand such sentences is to take them to
involve two semantically singular terms, that is two phrases that aim to pick
out, refer to, or denote some thing or entity. 18 In the case of (3), one term that
stands for a person: "A"; and another term that stands for a proposition: "that
p". Although the term that picks out the proposition is different in kind from
the term that picks out the person both in effect do the same thing, standing
for some thing or entity. And on the common way of fleshing out this picture,
reality contains not just a domain of persons, it also contains a domain of
propositions. When A says that p then some relation, the saying relation,
holds between two things: a sayer and a proposition. On this picture it is
natural to think of the domain of propositions in analogy with the domain of
persons. It is something that is part of reality, something that is simply there,
waiting to be picked out with our that-clauses. 19
But there is an alternative way oflooking at this. It is partly motivated by the
fact that that-clauses are first and foremost clauses. As (complement) clauses
that-clauses are of the same general category as "where I hoped it was" or
"whether she ate it," and the like. Such clauses are not naturally taken to be
terms that pick out things or entities. They seem to have a different semantic
fonction. It is tempting to hold that that-clauses in examples like (3) specify
18 Although it can be argued that we should recognize a difference between referring,
denoting, and picking out, I will gloss over such differences in the following, since it won't
matter which one of them that-clauses do, only whether they do any of them.
19 A rare exception of this picture of propositions to be there, but not be language
independent is Stephen Schiffer's (2003) view. Almost all other authors who believe in a
domain of propositions take them to be there independently of us. Whether Schiffer's view
makes a difference to our debate here is discussed among other issues in Hof\vebcr (2016a),
where I hold that on a natuml reading it does not, although an alternative might be possible.
Are There lneffr1b!e Aspt'Cts of Reality? 151
what was said, but don't refer to some object which was said. They do not refer
to a content, but say what the content is. On such a non-referential picture of
the function of that-clauses, "said" does not express a relation between two
things, one denoted by "A" and the other denoted by "that p." Instead, "said"
predicates of one thing, A, that it said that p. Complement clauses on such a
picture have a quite different function than names for objects.
Which one of these two positions is correct should be seen at least as an
open question in semantics, one that is widely debated in the present
20
literature. Whichever side comes out on top in this debate, it will have
significant consequences for our discussion of the ineffable, or so 1 will argue
now. These two pictures of the semantic function of that-clauses corres-
pond, on their natural development, to two different pictures of the prop-
ositional, and correspondingly two different answers to the question about
what is effable. To see that we first need to say more about how the second,
non-referential, picture of that-clauses is naturally developed in more detail.
The non-referential view of that-clauses might intuitively be very plaus-
ible, but it faces a serious obstacle when it comes to making sense of quite
obviously valid inferences like the inference from (3) to
(4) A said something.
How can a believer in that-clauses being non-referring or non-denoting
expressions account for tl1e validity of this inference? If A said something
then it seems that there must be some thing or entity to which A be<U"S the
saying relation. Whatever the believer in non-referential and non-denoting
clause wants to say here, it has to make sense of such inferences. But as it turns
out, similar inferences are also possible in cases tl1at have nothing to do with
clauses and also involve apparently non-referential complements. For example,
(5) I need an assistant. 21
implies that
(6) I need something.
even when I don't need any p<U"ticular person to be my assistant, just some
assistant or other. Similarly,
20
Sec Schiffor (1987) and (2003), Bach (1997), Molnrnmn (2003b), Rosefeldt
(2008), HofWeber (2007) and (2016b) and many others. There is a further issue here
whether that-clauses and more generally proposition terms like "the proposition that p"
have w be e1~l;er a'.l refe'.·enual or all non-referential, or whether a mixed view might be
true mstead. I his issue 1s discussed in more detail in Chapter 8 of Hofweber (2016b).
I will leave these mixed positions aside here.
21
Sec Moltmann (2003a).
152 l71orruts Hofiueber
(7) My daughter wants a unicorn.
(which is true) implies that
(8) She wants something.
Even though she doesn't want a particular unicorn, she would be happy
with any one, and even though there arc no unicorns at all. What all this
suggests is that there is more going on with quantifiers than the simple
objection to the non-referential picture of that-clauses suggests.
5.2 Talk about propositions: quantifiers
Some story about quantification must make sense of all this, and my favorite
one is the following. 22 Quantifiers are semantically underspecified. They are
polysemous expressions that have at least two different readings. These read-
ings arise from two different functions they have in ordina1y communication.
One is to make a claim about the domain of objects, or entities. On this use of
quantifiers we say that among all the things there are, whichever they might
be, one of them has a certain feature (in the case of a use of the particular
quantifier). This we will call the domain conditions, or external, reading, since
it makes a claim about the domain of entities, which is something language
external. On the other reading they are supposed to inferentially relate to other
expressions in ones own language. This we will call the inferential, or internal,
reading. On this reading the quantifier has a certain inferential role. In the case
of the particular quantifier, it simply is the inferential role
(9) F(t)/Hsomething)
whereby "t" could be any expression of a limited range of syntactic categor-
23
ies, be it a singular term, or a clause, or some other complement. On the
inferential reading of "something" the quantifier has this inferential role and
thus inferentially relates some sentences to others internal to the language.
To illustrate the difference, consider an example that uses the universal
quantifier "everything":
(10) Everything exists.
22 See Hofweber (2000) and (2005), and, in particular, Chapter 3 of Hofweber
(2016b).
2:l Which precise range of cases belong to this group is not completely clear. Quan-
tifiers in natural language do not interact with just any syntactic category. For example, we
can't quantify into determiner position, and we will simply accept this limitation here
without hoping to explain it.
Are There Ineffable Aspects of Reality? 153
On the polysemy account outlined above this sentence has two readings. On
one it should have a ring of truth, on the other it is quite clearly false. The
true reading is the external one, where you are saying that all the things in
the domain have in common that they all exist. And that, leaving some
philosophical views to the contra1y aside, seems true. But on the inferential
reading it is quite clearly false. On this reading the sentence has the
inferential role
(11) F(everything)/Ht)
where "t" can be any instance in my language. So (lO) implies Santa exists,
the Easter Bunny exists, etc. which are false. Since it implies false things it
must be false itself. So, on the inferential reading, (IO) is false.
To say that quantifiers have an inferential reading is not to accept an
inferentialist semantics in general. An inferential role can be the result of a
contribution to the truth conditions. The question remains so far what
contribution to the truth conditions gives qu<mtificrs their inferential roles
on the internal reading. One thing is clear: the truth conditions on the
inferential reading have to be different than the truth conditions on
the domain conditions reading. On the inferential reading the inference
goes through no matter what the semantic function of the relevant instances
is. What truth conditions would give it that inferential role? There is a simple
answer to this question. The simplest truth conditions that give the particular
quantifier the inferential role for which we want it is for sentences in which it
occurs to be equivalent to t!1e disjunction of all the instances that are supposed
to imply it. These truth conditions are simplest in the sense that any other
candidate truth conditions that give the particular quantifier that inferential
role must be weaker: the simplest one would imply them. Similarly, the
simplest truth condition that gives the universal quantifier its inferential role
is the conjunction over all the instances that it is supposed to imply, which is
for us all the instances in our language. Any other candidate truth conditions
that have that inferential role would have to be stronger in that it would itself
have to imply that conjunction. When it comes to the truth conditions that
the quantified sentences have when the quantifiers are used in their inferential
role reading we have an optimal solution, and we thus have some reason to
think that quantifiers in that reading make that contribution to the truth
conditions. 24 As we will see shortly, this leaves one important complication
aside, one we will need to address in the following section. So fo· we have only
given the simple version of the semantics of the inferential reading of
24
The details of this approach to quantification, including how it can be extended to
generalized quantifiers, are spelled out in Chapter 3 of Hofwebcr (20 !Gb). How it relates
to and is different from subsritutional quantification is also discussed there.
154 1!1omas Hofiveber
quantifiers. As we will sec later, a proper version can be given as well, but let's
work with the simple one for now. Whether this is the correct understanding
of quantification, in particular when related to quantification over proposi-
tions, is one question, but it should be clear that this is one natural way in
which the non-referential picture of that-clauses might extend to quantifica-
tion as well. We will not try to settle whether this picture is indeed correct, but
instead focus on what it implies for our discussion of the ineffable.
5.3 lnternalism vs. externalism
To understand talk about propositions involves at least making progress on
the question whether or not that-clauses are referring or denoting expres-
sions, and how to understand qwU1tification into that-clause position. There
are two large-scale options about how such talk can be understood in a
coherent way, which in effect depend on what one thinks about that-clauses
and quantification into that-clause position. If that-clauses pick out things
then quantification into that-clause position should be quantification over
the domain of these things. If that-clauses don't pick out entities then
quantification into that-clause position should be understood along differ-
ent lines. Here, I suggested, the best option is to understand it as being
based on the inferential reading of quantifiers, a reading quantifiers have in
general. These two large-scale views are thus:
(12) a. lnternalism: that-clauses are non-denoting, quantifiers into
that-clause position are used in their internal, inferential, reading.
(12) b. Extemalism: that-clauses are denoting, quantifiers into that-clause
position arc used in their external, domain conditions, reading.
Externalism is naturally connected to an ontology of propositions. If our talk
about the propositional is not completely in error then the domain over which
we quantify must be non-empty, and thus propositions exist. But ifinternalism
is correct then there is no such domain of propositions, and in particular no
propositions exist. I take propositions, if there are any, here simply to be
whatever that-clauses refer to or pick out. They are what we talk about when
we say that Fred believes that p or Sue said that p. If there arc such things as
propositions then they are the things that we talk about when we ascribe
content. But on the internalist picture we do not mlk about any things when
we ascribe content since our that-clauses don't pick out any such things. They
are non-referring and non-denoting expressions. And if propositions just are
whatever we talk about when we ascribe content then it follows from the
internalist picture of such talk that no such things exist. 25 If"that p" as well as
25
More on all this is in Chaptcr4 ofl-lof\vebcr (201Gb).
Are There !neffi1b!e A1pects of Reality? 155
"the proposition that p" docs not refer to anything then whatever there might
be, none of it is the proposition that p. The internalist picture of talk about
propositions guarantees that there are no propositions, so understood. This will
be relevant later. Whether internalism or externalism is correct is a question in
the philosophy oflanguagc that we can't hope to settle here. What we can hope
to see is that this question is cmcial for our debate about the ineffable.
5.4 lnternalism and the effability thesis
Everyone should believe that there is a substantial issue about how we
should understand our talk about the propositional, no matter which side
one eventually ends up on. Suppose, though, that internalism turns out to
be correct. Then quantification over 26 propositions generalizes over the
instances, in my own language. After all, the inferential role I am concerned
with is the inferential behavior that quantifiers have in my language, and to
have that inferential role it needs, on the simplest solution, to be equivalent
to the conjunction or disjunction over all the instances. Thus
(13) I need something
on the internal reading is equivalent to the disjunction
(14) V (l need F)
whereby F can be any of the variety of instances of the sentence: an assistant,
a unicorn, etc. The quantified sentence, on its internal reading, is thus
equivalent ro a disjunction over all the instances in our own language. This
disjunction is infinite, but it is a disjunction of instances nonetheless.
The effability thesis was the thesis that
(I) Everything is effable.
And as we noted above, the relevant way to understand it is that all true
propositions, or truths, are effable, but not necessarily all feelings, objects,
etc. The quantifier "everything" in the statement of the effability thesis is
thus a quantifier over propositions. Assuming the truth of internalism, it is
thus used in its inferential, internal reading. Internal quantifiers have the
simplest truth conditions that give them the inferential role, and so the
effability thesis is equivalent to the conjunction of all the instances in my
own language. Thus, assuming internalism, it is equivalent to the following
conjunction:
26
I take "talk about propositions" and "quantification over propositions" to have a
sense in which they arc neutral between the two options, and I mean it in this neutral
sense here. That there is such a neutral sense is clear using a topical sense of aboumcss, the
sense in which you can talk about aliens all night long, whether or not there arc any.
156 Thomrls Ho/weber
(15) /\ (that p is effable).
Here we have a conjunct for every instance of"p" in our own language. But
for each such instance the sentence "that pis effablc" is true, since we can eff
all the instances in our own language. Thus on the internalist understanding
of t<1lk about propositions the cffability thesis is true. When we say that
everything is cffablc we arc generalizing over the instances m our own
language. And if this is so then the cffability thesis is true.
This answer to the question whether or not any aspect of reality is
ineffable might seem very unsatisfactoty and more like a cheap trick. We
will discuss whether this answer could possibly be correct shortly. Before
that, though, we need to see a bit more about how the intcrnalist picture
deals with the arguments against the cffability thesis. And to see this we first
have to formulate it properly.
5.5 The proper formulation of internalism
The formulation of intcrnalism given above was too simple, in a way that
will matter for us later on for larger questions about the relationship between
what there is and what is the case, and for dealing with the arguments for
ineffable facts. But that it was too simple can be seen quite directly without
having these issues in mind. To formulate internalism properly we need to
make sure that the quantifiers properly interact with all the instances, not
just the simplest ones we have considered so far. All the instances include
cases like "he ate the cookies" and "it is pink." From
(16) John said that it is pink
it follows that
(17) John said something.
But on the account given so far this inference might not be valid. So far we
have not considered how to deal with an instance of a quantifier that
contains context sensitive elements. Implicitly at least, we have simply
ignored them, and ignoring them gives a well-defined semantics for the
quantifiers: the instances arc simply all sentences that have truth conditions
independently of the context in which they are uttered. We could call these
eterrlr1l sentences. 27 But simply having disjunctions or conjunctions of eternal
27
There is a substantial further issue about such sentences, and which ones have that
feature. Many sentences involve contributions from context or speakers intentions besides
filling in values of demonstratives. I will largely put these aside here. For other ways in
which context can affect content the question will be if there is also a way to express the
Are There lneffr1b!e Aspects of Reality? 157
sentences is not enough, since in "it is pink" the "it" might pick out
something that is not picked out by any eternal term, and thus no eternal
sentences would be equivalent to "it is pink." How internalisrn is to be
formulated to overcome this issue is discussed in detail in HofWebcr (2006),
but the main idea is simple enough. The inference from (16) to (17) has to
be _valid no matter ':l~at object might be referred to with "it." Thus frlr any
ob3cct owe need a drs;unct that is true just in case o is pink. And this we can
do by simply adding external quantification over objects on the outside of
our disjunction, and allow new variables to be bound by this quantifier in
these disjunctions. fn our case this would give us the truth conditions of
(17) not as
( 18) V Oohn said that p)
but as
(19) 3xV Qohn said that p[xj).
Herc "p[x]" means that any of the new variables i: may occur in the instances
that replace "p." And since we can't in advance give an upper bound on how
many such variables there might be (after all, John might have said that it is
taller t:1~n that, but shorter than this, etc.) we must allow for infinitely
many. f hus the truth conditions of internalist quantifiers over propositions
are not merely given by infinitary conjunctions and disjunctions, but
involve infinitary quantification as well. All this is at least technically
28
unproblematic. In case of universal quantification the external quantifier
over objects out front of the big conjunction is, of course, a universal
external quantifier. 29
With this new formulation we can now see that the effability thesis
(1) Everything is effable
is not simply equivalent to
same comem without the effects of context. Since our instances are all instances in our
language, we would avoid this issue if we could always find a context insensitive way to
express a content that <~ther.wise was expressed in a context sensitive way. This is certainly
po~st~)le for most co.ntnbunons of context, like dis~m1biguation, various enrichments, ere.
If tt is always posstble for cases other than demonstratives, it will be left as an open
question fr)r now.
28
I n e ff.ect, ·mternaI quanrt·111Ca:1on
· over propositions
· increases t l1e expressive power of
a language to a.small fragment of what is called Lw 1 , w 1 built on top of that language.
1':1ore about tlus ts tn Hof\veber (2006). lnfinitary extensions of first order logic arc
discussed, for example, in Keisler (1971).
29
, How to do this for generalized quantifiers is developed in the appendices to
Chapters 3 and IO ofHofweber (20!6b).
158 Thomas Hofweber
(20) /\ (that p is effa.ble)
but instead to
(21) Vx'/\ (thatp[x] is effable).
And (21) is now not necessarily true any more with just ~my notion of
effability. If what is effable is understood as what is effable by us in the
circumstances we are actually in then this can be false, for example if there
are objects outside of our light cone that we arc thus not able to refer to (to
use an example and assumption discussed above). But (21) is true if we use
the object-permitting notion of effability, as discussed in section 3.5. On
this, object-permitting, notion of effability, missing objects are never a
source of ineffability. It is the notion of the ineffable that passes the
incommunicability test, where reference to objects is always assumed to be
possible and not a relevant source of a limitation. In our case (21), every
instance of the quantifier is in effect equivalent to some sentence
p(a 1, a2 , •.• , a,,) where a; is some parameter standing for an object, and p
is some sentence in our language. On the object-permitting notion of
cffability, every one of these instances is effable.
Thus the proper formulation of internalism guarantees the truth of the
effability thesis using the object-permitting notion of effability. And the
extra resources that are needed to give the proper formulation ofintcrnalism
are the ones that explain why the ineffability thesis nonetheless seems true to
us, as we will sec now.
5.6 Explaining apparent ineffability
If intcrnalism is true, and with it the effability thesis, what then becomes of
our arguments for ineffability? After all, we found them quite compelling
above. But maybe all these arguments implicitly relied on a standard,
externalist picture of propositions and facts? We considered four arguments
for ineffability: built-in limitations, the argument from analogy, cardinality
arguments, and there being no explanation why the effability thesis would
be true. Let's revisit them now.
What could explain that a mind like ours can represent everything there is
to represent? The internalist has a simple answer: "everything" here gener-
alizes over our instances. No wonder we can represent everything, since we
can represent every one of our instances: every instance in our language of
"that p." And relying on an object-permitting notion of eft'lbility we can eff
every instance with parameters. It is no accident and no mystery that we can
represent everything there is to represent. Internalism maintains that it is
Are There lnejjl1ble Aspects of Reality? 159
based on a rnist~lke to wonder here why two separate things coincide: the
propositional and our representational abilities. Instead the propositional
and our representational abilities arc not two unconnected things, if intern-
alism is true. The truth of the effability thesis falls out of how talk about
pr?positions is to be understood. It would be hard to explain why the
cftability thesis is true on an extcrnalist picture, bur it is quite straightfor-
ward on an internalist one.
What about our built-in imitations? Maybe our mind must think a certain
way, maybe we are inflexible and fixed in how we have to think, due to how
our mind evolved or how our brain is structured. This naturally supports that
we are limited in what we can represent on an extcrnalist conception of facts.
If facts are simply there, as part of an independent domain, then we should
expect a mismatch. But on the internalist conception there is no such domain.
So even our lack of flexibility docs not support ineffability, since it does not
support that this lack of flexibility is a limitation. We can explain why the
eftiibility thesis holds even if our minds have a fixed setup.
The cardinality arguments are directly answered by the extra resources
that we get from the proper formulation of internalism. Consider again the
argument that there are more propositions than we can eff, since there are
only countably many sentences in our language, but there are uncountably
many facts about real numbers. The following is true:
(22) For every real number r there is a fact which is the fact that r is a
real number.
This truth seems to be incompatible with the effability thesis as well as with
internalism, since for different real numbers we get different facts, and thus
overall too many facts for the effability thesis ro hold. However, on the
proper formulation of internalist quantification over propositions, (22) is
true, and thus the truth of (22) is compatible with internalism. Here,
crucially, the external quantifier over real numbers interacts with the exter-
nal quantifiers that bind the new variables discussed above. Thus on the
proper formulation of internalism (22) would look like this:
(23) Vr3xV (that p[x] is the fact that r is a real number).
This is true when "p" is instantiated with "x; is a real number," where "x/' is
bound from the outside with an external quantifier that ranges, amongst
others, over real numbers, and thus has the relevant r as an instance. (22)
appears to be in conflict with internalism, but in fact it is only in conflict
with internalism on its naive formulation, but not with internalism on its
proper formulation.
To put it differently: the proper notion of an ineffable fact is one that is
ineffable on an object-permitting notion of cf£'lbility. Cardinality arguments
160 7 homas Hojiveber
like our simple argument using the real numbers argue that there is a certain
cardinality of facts since there arc collections of objects (the real numbers) of
this cardinality. h1rthcrmore, the argument continues, that cardinality is
larger than the cardinality of effable facts and thus most are ineffable. But
on the object-permitting notion of the ineffable, more objects get us more
effable facts. On that notion of the ineffable it is hopeless to tty to show that
there more facts than effable facts, since there are lots of objects involved in
10
all the facts. Such cardinality arguments thus won't get off the ground.
Finally, let's consider the argument from analogy, which is maybe the
most compelling and forceful argument for ineffability. Here the internalist
answer is clear: although it appears to be coherent to imagine aliens or gods
who relate to us like we relate to the honeybee, there in fact can be no such
creatures. There can be no creatures who can represent more facts than we
can, since we can already represent all the facts. We can represent everything
there is to represent, while the honeybee can not. Since we can already
represent everything, there can be no creatures who can represent more.
'fhere can be more powerful creatures with better spaceships, but they
cannot represent any more facts. But whether this answer is at all satisfac-
tory, or merely an endorsement of an absurd consequence of a view, can't
reaily be appreciated without looking more generally at the internalist
picture of the propositional. lnternalism is not merely a view about the
semantics of that-clauses and quantifiers, but it incorporates a completely
different picture of the propositional and of reality understood as the totality
of facts. And only with that picture clearer in view can we see that this
answer is not in fact absurd. Although an externalist will take it to be
extreme bullet-biting, the internalist will take it to be a deep insight into
the nature of the propositional or fact-like aspect of reality. We can only
assess who has the upper hand once the full picture is on the table, to which
we must turn now.
5.7 The internalist picture of the propositional
On the externalise picture of the propositional, propositions form a domain
of entities that we can refer to, and that our quantifiers range over when we
quantify over propositions. Propositions on this view are most naturally
understood as simply being there, as part of reality in addition to regular
objects, waiting for us to refer to them. 31 Facts can be seen as either being
identical to or at least corresponding to true propositions, and so what holds
30 JClow intcrnalism can deal with these and various other cardinality arguments is
discussed in more detail in Hof\vebcr (2006).
31 Again, an exception to this picmrc is SchiHer (2003).
Are There Ineffeble Aspects of Reality' 161
fix propositions will hold for facts as well. Facts, propositions, states of affairs,
;u1ything we would pick out with a that-clause, we can call the propositional.
On the externa.list picture, the propositional is simply there, a part of reality,
ready for us to discover and refer to. If facts are entities then reality as all there
is determines reality as all that is the case. Under this assumption we should
expect, as we saw above, that not all facts or propositions can be represented
by us, and some will remain in principle ineffable. Taking propositions to
form a domain that is simply there, as a part of reality, naturally leads to
accepting ineffable facts as well as the sub-algebra hypothesis as the best
explanation of why the ineffable is so well hidden.
But the internalist account of talk about propositions is not just a
semantic view of talk about propositions, it embodies a completely different
picture of the propositional. On this picture the propositional is not simply
there, and it is not independent of us, in a sense to be worked out. This
difference is crucial when it comes to understanding why the effability thesis
holds, assuming internalism, and how the arguments against effability are to
be answered. In this and the next section I hope to work out more clearly
what the internalist picture of the propositional is and how to understand it
not being independent of us. And to do this we first should contrast the
internalise picture with what it is not.
The internalist does not simply establish the effabiliry thesis with a
semantic trick. Consider, as a contrast, a person who holds that they own
everything. They say that the correct semantics of "everything" is that it
ranges only over their things, and thus they own everything, since they own
all their things. This is a bad view on several grounds, not the least of which
is, of course, that this is not the correct semantics of "everything." But the
crucial difference between this view and the internalist view of talk about
propositions goes beyond that. On the internalise view it is not true that
when we say that we can represent everything we say that we can represent
everything that is in some sense ours. We do not restrict our quantifier to
some subset of the propositions which are related to us, analogous to the
universal owner who restricts their quantifiers to the subset of things that
they own. Internalism does not restrict the quantifier, but instead embodies
a different view of what such quantifiers do, which is tied to a different view
of what singular ascriptions of content do. Such quantifiers are unrestricted
inferential quantifiers. As such their truth conditions give them a certain
inferential role in our language, and the simplest truth conditions that do
this are the ones that are equivalent to generalizations over all the instances.
That these instances are instances in our language, and thus ones we can
represent, is not the result of some sort of a restriction, but simply a
consequence of the simplest truth conditions that give us what we need.
Talk about propositions or facts, on the internalise picture, is not talk about
162 '/Y;omrls Ho/weber
some independent domain, it is not talk about any entities at all. On the
internalist picture, that-clauses do not refer. They do not aim to pick out
any entities. So talk about propositions is very different than talk about
objects.
Still, the internalist defense of the effablity thesis might seem unsatisfactory.
It seems to involve too much language and not enough metaphysics and thus
can't be a defense of the view that all of reality is effablc by us. Reality played
little role in this defense. But that doesn't mean that the defense was defective.
The intcrnalist's explanation of why the effability thesis is true relied on the
connection between the quantifier "everything" that occurred in the state-
ment of the effability thesis and the instances of such quantified sentences in
our own language. That this was enough to see that the effability thesis is true
is surprising, but that doesn't make it incorrect. That the explanation has to
rely on reality in addition to what we do when we talk about all facts or truths
is true on an externalist picture of the propositional, but not on an internalise
one. For the latter, reflection on our language is enough. We can see that once
we talk about the propositional at all, in the internalise way, the truth of
the effability thesis follows. This is smprising, but it might just be true if
internalism is indeed true.
It is tempting to say that on the internalist picture the propositional is not
an independent part of reality, but somehow clue to us. But this on the face
of it doesn't make sense. Which facts obtain is, of course, not in general clue
to us. But there is something right about this, although it's hard to put one's
finger on what precisely it is. I will make a more precise proposal about this
in the next section, but first we should see some more if all this is too shallow
a victmy for the effability thesis.
Let's assume for the moment that internalism is true, and thus the
effability thesis, as stated, is true as well. What this might be tal<e11 to
show is that we need to state the question we wanted to ask differently
than we clicl. As formulated the question has a negative answer, but maybe
we need to reformulate it so that it is more substantial, more about reality,
and less about language. After all, the question we intended to ask was not
supposed to be settled by the semantics of quantifiers and the non-
referentiality of that-clauses. Of course, we can't demand that the questions
we ask are answered the way we expect or intend them to be an~-werecl, but
still, maybe the lesson we should draw from all this is that we need to state
the question we wanted to ask differently. Maybe internalism wins a shallow
victory when it comes to the letter of the effability thesis, but it only pushes
the real issue somewhere else.
This line of thought is indeed tempting, but in the encl it is mistaken. The
truth and recognition of internalism does not motivate that we should ask
the question differently. In fact, there is no better way of asking it. Instead
Are '/here Ineflahle /l;pects of Rerdity? 163
internalism shows us the answer to the question we wanted to ask in the first
place. To see this, let's consider some attempts to ask the question differ-
ently, assuming intcrnalism is true, and why they won't help.
Assuming again internalism, it won't do to ask about the effabiliry of
every fact, or every proposition, or every proposition-like thing. Internal
quantifiers over those make clear that the answer is in line with the effability
thesis. So, maybe we should state the question in a way where we quantify
over something else, something where internalism is not true: sentences, or
inscriptions, or thought tokens, or something along those lines. These are
simply material objects (we can assume) and so internalism doesn't apply to
talk about them (we can also assume). Should we thus ask instead whether
there is some (actual or possible) concrete inscription which can't be
translated into our language, or some thought token which has a different
content than any such token we can in principle have? But this won't help.
If internalism is true then we can conclude that there can be no such
thought token or inscription. If such a token has a content at all then it
has a content which we can think as well. Anything else has no content.
Thus if there are inscriptions, sentences or tokens which we can't translate it
is not because they have a content that is beyond us, but rather because they
have no content at all. And there certainly is no failure of translation if you
fail to be able to translate something devoid of content. This way of trying to
restate the question won't change the issue.
Another attempt could go via truth. Maybe the aliens can say things truly
which we can't say at all? Maybe true things accessible to them go beyond
the truths accessible to us? But this, too, won't help. There is a briclge-
principle that connects things that are true to contents:
(24) x is true if and only if x has a true content.
It is hard to see, maybe inconceivable, how anything could be true, but not
have any content. With this connection, moving the issue to truth doesn't
ch<mge things.
Finally, one might tty to throw in the towel on truth, content, and
propositions, and acknowledge that we are not limited when it comes to
those, but that there is a limitation noneilielcss, but we can't even properly
articulate our limitations. To illustrate with the advanced aliens again, the
idea is that although they are not doing better than we when it comes to
truth and content, they are doing better when it comes to truth* or schmuth
and content* or schmontent. When the aliens look down at us from their
advanced spaceships, they will certainly take us to be limited, and maybe we
can't quite say how, but they might think of us as missing out on some
important truths* or contents*. Now, this is certainly right in many ways,
but is wrong in the crucial way that matters here. We are clearly limited
164 Thomas Hofweber
when it comes to the aliens--we don't have the spaceships, and their
invasion of earth might be a walk in the park for them. But the crucial
issue for us is whether we arc limited in what we can represent conceptually.
And here the thought experiment motivates no such thing. Truth*, what-
ever it is supposed to be, isn't truth, and content* isn't content. Whichever
of our many limitations we consider they are not limitations in what we can
in principle represent conceptually. [f internalism is true then we can reason
conclusively that every fact can in principle be represented conceptually by
us, although other things can't be conceptually represented by us, but these
are not the kinds of things that conceptual representation is supposed to
represent in the first place. Conceptual representation is complete in its
domain, and this indicates that there is no proper separation between
what is to be represented conceptually and the representations that do the
representing. 32
Let us now return to the argument from analogy, which was somewhat
postponed in its assessment above. When the aliens look down at us like we
look clown at the honeybee then we are correct in thinking that they are
superior to us in many ways, but incorrect in thinking that they can
represent more facts than we do. When they think about us as these
primitive creatures, we cannot truly report their thoughts that the primitive
humans don't get all the facts. That simply isn't true, assuming internalism,
of course. We might not be as good as the aliens in all kinds of things, but
when it comes to representing the facts we arc.
There is no question that there is the strong sense that there is a limitation
of us that is analogous to the limitation of the honeybee. Tt might be hard to
state what that limitation is supposed to be more precisely, but there clearly
is the sense of a limitation motivated by analogy. However, assuming
12
' The conclusion here is similar to one drawn by Donald Davidson ( 1984). However,
it is reached in quite different ways. Davidson held that due to his particular theory of
meaning it was impossible for there to be a language that is in principle untranslatable into
our language, and thus that there could not be variation like this across conceptual
schemes, and thus the notion of a conceptual scheme is based on a mistake. The present
view holds that there could not be content that goes beyond what we can say in principle,
not because of a theory of meaning, but because of what we do when we talk about
propositions. A second similarity is to Hilaty Putnam's ( 1981) argument that we arc not
brains in vats, since the question whether we arc, as stated by us, is guaranteed to have a
negative answer. lnternalism is not tied to skepticism, but the fact that the focus on the
question can be a key to its answer is similar in both Putnam's argument as well as here. It
should be noted that the intemalist picture of the propositional is quite different from
McDowell's view on the matter. McDowell explicitly contrasts his view with an "arrogant
anthropomorphism" (1994, 39) which holds that we can represent all facts with our
present conceptual resources. In contrast he holds that what there is to represent about
reality is not influenced by our conceptual resources, but an independent fact about it. Sec
section 8 of Lecture 2 in McDowell (1994) which is devoted to this issue.
Are lhere ineffable Aipects o/Rerility? 165
intcrnalism, we can reason conclusively that there is no relevant limitation.
We will have to weigh the sense of a limitation against the reasons a<,ainst a
limitat~on. ~i~1cc reasons make things more reasonable than merely t~cling a
lurntation, tt JS thus reasonable to conclude that the sense is misguided and
the argument from analogy, although powerful, is in the end mistaken.
We can thus conclude, assuming internalism about our own talk about
~rop'.)si~ions, that there is nothing the aliens can truly say that we can't say
m pnnctple as well. When it comes to conceptual representation we face no
limi.tati.on'. and this is. a consequence of what the propositional aspect of
reality JS ltkc. lnternahsm docs not give us reason to think we are limited
but it gives us a different picture of the propositional. The propositional i~
?ot part of reality in the sense that there is no domain of propositions or
facts of which we might capture more or fewer. Instead, it rcHects our
empl~r,rncnt of talk about propositions, and it doesn't and can't go beyond
that. I he propositional is a reHection of our talk about propositions. 'fhis
might sound like a version of idealism, and it is.
5.8 Idealism vindicated
Ideal'.sm sounds bad, but it doesn't have to be when properly stated.
Idealtsm broadly understood holds that, in some form or other, minds, in
particular minds of creatures like us, are central to reality. The most
common and maybe most natural way to be an idealist is to hold that
what reality is like depends on our minds. And this is naturally spelled out as
ontological idealism: the view that what there is depends on our minds in
some way. Many of the historically significant idealists were idealists of this
sort, but it is rather problematic and rightly widely rejected. It is not clear
how the notion of dependence is supposed to be understood such that
idealism so formulated is compatible with other things we know to be true.
We know that the universe existed before there were any humans and we
know roughly what it was like before we were around. It is not clear how this
is compatible with what there is depending on us, on a natural understand-
ing of dependence. In particular, it is not clear how to even state the idealist
position without it leading immediately to conflict with many other things
we know. Ontological idealism is false, but idealism might still be true. 33
33
A weal<er form of idealism is to hold that even though what there is does not depend
on us, 1:onethelcss,. there being minds like ours is no accident. Any world must contain
some. mmds, and without minds there could be no material world at all. A version of this
pos1t10n was dcfen~!cd by Anton Friedrich Koch with an intriguing argument in (1990)
and (20 I 0). I don r believe that !us argument works, and have tried ro say why not in
 Thomas Ho/weber
166
An alternative way of thinking about idealism is to tie it not to ontology
but rather to the conceptual or propositional aspect of reality. Although
what there is is independent of us, the fact-like or propositional aspect of
reality is not. That is, reality, understood as what there is, is independent of
us, but reality, understood as what is the case, is not. This way of under-
standing idealism faces the same problems at first as ontological idealism.
How can we coherently state a notion of dependence for the propositional
that doesn't immediately conflict with what we know to be the case? Here
we have to distinguish two ways in which the propositional could be
dependent on us. The propositional is truth-dependent on us just in case
which propositions arc true is dependent on us. In this sense it is, of course,
false that the propositional depends on us. Which truths are true is not
something that in general depends on us. We can make some propositions
true or false by affecting the world, but we don't make the true propositions
true in general. Alternatively, the propositional is range-dependent on us just
in case the extent of the propositional is dependent on us, which is to say,
what propositions there are as candidates for being true depends on us,
somehow and in some sense. The natural way to understand this is that what
can be represented conceptually in principle is, somehow, dependent on us
and our conceptual resources. And a natural way to understand that is that it
is guaranteed that all there is to represent about reality can be represente~ by
us, in principle. And this in turn means that the effability thesis is true for a
reason connected to us. The view that this is so was called conceptual idealism
in section 3.4. And this is the form of idealism that might be true.
Conceptual idealism is thus the view that the propositional is range-
depcndent on us. It is a form of idealism, but diflerent from and independent
of ontological idealism. 31 As discussed briefly above, the standard route to
conceptual idealism, and with it the standard defense of the effability thesis, is
via ontological idealism. But conceptual idealism might be true even if
ontological idealism is f:'llse, and internalism about talk about propositions
and facts supports just that option. Ifinternalism is true then the propositional
depends on us, not for its truth, but for its range. lntemalism thus supports
idealism, not ontological idealism, of course, but conceptual idealism.
Conceptual idealism avoids the incompatibility worries that ontological
idealism faces rather directly. It is a promising candidate to provide a coherent
Hofwebcr (2015). Further, completely different, recent versions of idealism include
Adams (2007), whose paper title I borrow for this section, and Smithson (2015).
'l 4 In Chapter VI of Nagel (1986), Thomas Nagel takes something like conceptual
idealism to be the defining mark of idealism. This is slightly unusual, but I believe with
Nagel that the real issue about idealism is just that. Nagel, of course, rejects idealism so
understood.
Are There fnefj(ib!e Aspects of Ret1!ity? 167
formulation of idealism. Although the idealist might be tempted to try to say
more about how the world depends on us, it is not clear how we can do this
coherently. It is not clear how we can on the one hand stick with our picture
of a world that is there for us to discover and which has us as just a small part
of it, while at the same time elaborating on how the world in the end depends
on us. But we can formulate conceptual idealism coherently. And if there is at
least this kind of dependence, range dependence of the propositional, then
this supports a version of idealism that we can make sense of:3 5
If extcrnalism is true for talk about objects, but internalism is true for talk
about propositions then ontological realism and conceptual idealism are
natural consequences. What there is is in general independent of us. But
what can in principle, by anyone, be said about it is not independent of us.
To illustrate this again one last time, consider events. They, as entities, in
general occur without us having anything to do with which ones occur and
which ones don't. They are simply there. But what can be truly said about
them, in the sense of what the possible range of a propositional or conceptual
description of them is, is not independent of us. lnternalism shows why and
how we come in here, since the range of the propositional is not independent
of us. That way ontological realism gets combined with conceptual idealism.
Ontological realism is not compatible with the truth-dependence of proposi-
tions on us, since propositions about what there is can't depend on us for
their truth while it is independent of us what there is. But the propositional
can be range-dependent on us while ontological realism is true. There is no
incoherence here since we are working with an object-permitting notion of
the effable, and thus the range of the propositional can accommodate all
there is, whatever it might be. Even though what things there are is inde-
pendent of us, both in its range and for its existence, and even though these
things can figure in the facts that can in principle obtain, the range of facts
that can obtain is not independent of us. We can in principle eff everyone of
them, on an object-permitting notion of effability, the relevant notion of
35
Hilary Putnam, in (1981) and other places, has defended a view he calls "internal
realism" that goes by the motto that there is no ready-made world. However, Putnam
focuses on ontology and hopes to argue that the world does not come by itself carved into
objects. What there is, for Putnam, is tied to our ways of talking about it, and thus his
view is best understood ;Ls a version of ontological idealism. As Simon Blackburn ( 1994)
h;Ls argued quite successfolly, this view leads to the consequence of a conflict between our
statement of the idealist position with other things we take ourselves to know to be true,
and thus turns the view into an incoherent one, given what we know. Putnam's view
fornses on the world not being ready-made when it comes to the objects that inhabit it.
I find it more fruitfol to consider the conceptual or propositional aspect of the world to be,
usmg the same metaphor, not ready-made. This can, I hope to make clear, be stated
coherently. Even if the world of objects is ready-made, the world of facts is not, on the way
to spell out the metaphor attempted here.
168 Thomrts Hojiueber
effability for our debate. The object-permitting conception of cffability and
intcrnalism about talk about the propositional allow for the combination of
the dependence of the facts on us, on the one hand, with the independence of
the things on us, on the other.
6. CONCLUSION
Whether or not we should accept the ineffability thesis or the effability
thesis depends on what we should think we do when we talk about
propositions. If externalism is true about such talk then we should accept
the ineffability thesis and the sub-algebra hypothesis. This would be insig-
nificant for most of inquiry and ordinary life, since, on the sub-algebra
hypothesis, the ineffable will be there, but it will be hidden from us in a way
that makes clear why it is insignificant for these purposes. But the ineffable
will be significant for metaphysics, in particular debates about what reality as
a whole is like. We should expect that we will naturally be misled to believe
that our sub-algebra is all there is. Here ineffability should lead to modesty
in grand metaphysics.
But if, on the other hand, intcrnalism is true about talk about proposi-
tions then the efE1bility thesis will be true, and we can explain why it seems
to us that some aspects of reality should be ineffable for us. No modesty
would follow for metaphysics from this, but the metaphysical picture of the
propositional that is tied to the internalist view of talk about propositions is
itself a substantial consequence. It combines a version of realism, in that
reality as what there is is independent of us, with a version of idealism, in
that what there is to say about reality can all be said by us, as we are right
now, not by mere accident, but for a reason. The internalist picture of the
propositional makes clear why content cannot be beyond us and thus all
there is to say about reality can be said by us in principle. Internalism thus
implies conceptual idealism, but is compatible with ontological realism.
The question whether internalism or externalism is true about our talk
about propositions is a largely empirical question about what we do when
we talk about propositions. It is a question about our actual use of certain
expressions in natural language, and thus something that we can't settle on
the basis of a priori reflection. The crucial question on which this issue
depends is thus one about language, and a largely empirical question at that.
Idealism, properly understood, and the effability thesis follow if things turn
out one way; ineffability and modesty follow if they turn out another. If the
former then not only would it support idealism, which might sound bad
enough, but furthermore it would support idealism on empirical grounds,
which might sound even worse. Nonetheless, since I have argued in other
Are There Ineffable Aspects rf Reality? 169
work that internalism about the propositional is, as best as I can tell, true,
I side with idealism and the effability thesis. 31'
1homt1S Hofiveber
University of North Ctiro/ina, Chapel Hi!/
REFERENCES
Adams, R. (2007). "Idealism Vindicated." In P. van Inwagcn and D. Zimmerman,
eds, Persons Human and Diuine, Oxford: Oxford University Press, 35-54.
Bach, K. (l 997). "Do Belief Rcpons Report Bcliefa?" i'flcijic Philosophiml Quarterly
78(3): 215--41.
Blackburn, S. ( 1994). "Enchanting Views." In P. Clark and B. Hale, eds, Reading
Putnam, Oxford: Blackwell, 12-30. "
Chomsky, N. (1975). Reflections on Language, New York: Pantheon Books.
Collins,]. (2002). "The Ve1y Idea of a Science Forming Faculty." Dialectica 56(2):
125-51.
Davidson, D. (1984). "The Very Idea of a Conceptual Scheme." In Davidson,
Inquiries into Truth tlnd !nte1prett1tion, Oxford: Clarendon Press, 183-98.
Dretske, F. (l 981). Knowledge mid the Flow of!n/or111t1tion Cambridge, MA: MIT Press.
Field, H. (1994). "Deflationary Views of Meaning and Content." Mind, 103(411):
249-85.
Filcheva, K. (2015). "Can there be ineffable propositional structures?" Unpublished
manuscript.
Fodor, ] . A. ( 1983). The Modu!t1rity of Mind. Cambridge, MA: MIT Press.
Hof\veber, T. (2000). Quantification and Non-Existent Objects.; In A. Everett and
T. Hof\veber, eds, Empty Names, Fiction, t1nd the Puzzles of Non-Existence,
Stanford: CSU Publications, 249-7.3.
Hof\vcbcr, T. (2005). "A Puzzle about Ontology." N01f.s, 39(2): 256-83.
Hof\veber, T. (2006). "Inexpressible Properties and Propositions." In D. Zimmerman,
ed., Oiford Studies in Metaphysics, volume 2, Oxford: Oxford University Press,
155-206.
Hofwebcr, T. (2007). "Innocent Statements and their Metaphysically Loaded
Counterparts." Philosophers' Imprint 7(1): 1-33.
Tha1:ks to David B;:ker, El~~abeth Barnes, M~kc Bertrand, Cian Dorr, Kyle Drig-
36
gers, Matti Eklund, Kras1nma hlcheva, Jeremy Goodman, John Hawthorne, Carrie
Ichikawa Jenkins, Andreas Kemmerling, Anton Friedrich Koch, Matt Kotzen, Dan
Lopez de Sa, Bill Lycan, Ofra Magidor, Anna-Sara Malmgren, A. W. Moore, Alan
Nelson, Ram Neta, David Reeve, Sven Rosenkranz, Geoff Sayre-/vkC:ord, Ted Sider
Rob S~1ithson, Moritz Schulz, Corina Strogncr, and Juhani Yli-Vakkuri for helpful
d1scuss1ons and/or comments on earlier drafts. Special thanks to Karen Bennett for
comments on an earlier draft that led to substantial revisions.
170 Thomas Hofweher
Hofweber, T. (2009). "Ambitious, yet Modest, Metaphysics." In D. Chalmers,
D. Manley, and R. \Xfasscrman, eds, lvletmnetaphysics: New Essays on the FoundtL-
tions of Ontolopy. Oxford: Oxford University Press, 269-89.
Hofwcbcr, T. (2015). "The Place of Subjects in the Metaphysics of Material
Objects." dialectica 69(4): 473-90.
Hofwcbcr, T. (2016a). "From Remnants to Things, and Back Again." In
G. Ostertag, ed., Meanings and Other Things: Li·sdys in Honor of Stephen Schiffer.
Oxford: Oxford University Press, Chapter 2.
Hofweber, T. (2016b). Ontol<!Kf and the Ambitions of Metaphysics. Oxford: Oxford
University Press.
Kant, I. (1781). Kritik der reinen Vemurift. Riga: Johan Friedrich Hartnoch.
Keisler, H.J. (1971 ). Model Theory far !nfinitmy Logic. Amsterdam: North-Holland.
Koch, A. F. ( 1990). Suhjektivitdt in Raum und Zeit. Frankfort am Main: Klostermann.
Koch, A. F. (2010). "Persons as Mirroring the World." In J. O'Shea and
E. Rubinstein, eds, Self; Langw1ge, and WorM Atascadero: Ridgcview Publishing,
232-48.
McDowell, J. (1994). Mind mzd World. Cambridge, MA: Harvard University Press.
McGinn, C. (1989). "Can We Solve the Mind-Body Problem?" Mind 98(391):
349-66.
McGinn, C. (l 993). Problems in Philosophy: The Limits of lnquily. Oxford and
Malden, MA: Blackwell.
Moltmann, F. (2003a). "Nominalizing Quantifiers." journal of Philosophietd Logic
32(5): 44 5-81.
Moltmann, F. (2003b). "Propositional Attitudes without Propositions." Synthese
35(1): 77-118.
Moore, A. (2003a). "Ineffability and Nonsense." Proceedings ofthe Aristotelim1 Society
77: 169-93.
Moore, A. (2003b). "Ineffability and Religion." Europe1111 journal of Philosophy
11(2): 161-76.
Nagel, T. (1986). The View from Nowhere. Oxford: Oxford University Press.
Priest, G. (2002). Beyond the Limits of Thought. Oxford: Oxford University Press.
Putnam, H. (1981). Reason, Truth, and Histo1y. Cambridge: Cambridge University
Press.
Rosefcldt, T. (2008). "'That' -Clauses and Non-nominal Quantification." Philosoph-
ical Studies 137(3): 301-33.
Rosen, G. (1994). "Objectivity and Modern Idealism: What is the Question?" In
). O'Leary-Hawthorne and M. Michael, eds, Metaphysics in Mind. Dordrccht:
Kluwer, 277-319.
Rosenkranz, S. (2007). "Agnosticism as a Third Stance." Mind, 116(461): 55-104.
Schiffer, S. (1987). Remnants of Meaning. Cambridge, MA: MIT Press.
Schiffer, S. (2003). The Things We Mean. Oxford: Oxford University Press.
Smithson, R. (2015). "Edenic Idealism." Unpublished manuscript.
Varley, J. (1977). The Ophiuchi Hotline. New York: The Dial Press.
8
The Metaphysics of Quantities
and Their Dimensions
Bradford Skow
1. INTRODUCTION
Quantities have dimensions. Force, mass, and acceleration are quantities-
for now let this just mean that they may be faithfully measured by
numbers-and the dimension of force is "ML/ T 2 ," the dimension of
mass is "M," and the dimension of acceleration is "L/T 2 ," where "M"
stands for mass, "L" for length, and "T" for time, or duration. While the
notion of a quantity is familiar w those with only a passing acquaintance
with physics, the notion of a quantity's dimension is probably not. 1 Even
though they are unfamiliar, dimensions play a variety of important roles in
physics, and science generally. Perhaps the most important one is the role
they play in the technique of dimensional analysis. One of the aims of
physics is to discover laws, and laws of physics usually take the form of
relations between quantities-Newton's second law, F = ma, is a relation
between the quantities force, mass, and acceleration. Newton's discovery of
this law involved, in part, consulting a lot of observational evidence.
Dimensional analysis is a different technique for discovering law-like rela-
tions between quantities, one that doesn't involve inferring a generalization
from a collection of instances. If you know only that some law or other
relates force, mass, and acceleration, but you also know the dimensions of
these quantities, and a few other facts, you can use dimensional analysis to
conclude that the law is F ma, even if you don't have any of the data that
Newton did. How to use the technique won't matter in what follows. 2 The
1
The notion of the dimension of a quantity has nothing to do with the familiar notion
of a dimension of space.
2
For the record, and simplifying a little: to apply dimensional analysis we need
to know that the force on a body is some function of its mass and acceleration, so that
F g(m, a) for some function g, and we need to know that the fonction g makes this
172 Bmdford Skow
point is that dimensions can help answer important physical questions. But
should metaphysicians care about dimensions? Is the fact that the dimension
of acceleration is J,/ T 2 in any way a metaphysically interesting fact about it?
Does this fact, for example, have anything to do with the nature of
aeceleration?
The physicist R. C:. Tolman thought it did; he wrote in 1917 that a
statement of the dimension of a quantity is "a shortbmd restatement of its
definition and hence [is] an expression of its essential physical nature"
(quoted in Bridgman 1922: 26). "fo say that the dimension of acceleration
is L/ T 2 is certainly to say that there is some connection or other between
acceleration and length and duration. Tolm<rn here says that the connection
has something to do with the essence, or nature, of acceleration. Could that
be right? It seems right to me, at least initially, but it has had powerful
opponents. Percy Bridgman, who won the Nobel Prize in physics in 1946,
scorned views like Tolman's, scorn that comes across in his description of
those views:
It is by many considered that a dimensional formula has some esoteric significance
connected with the "ultimate nature" of an object, and that we arc in some way
getting at the ultimate nature of things in writing their dimensional formulas.
(1922: 24)
Bridgman went on to give several arguments against the idea that dimen-
sional formulas have "some esoteric significance," or, as we might put it, that
they have some metaphysical significance. His view was that dimension
equation "dimensionally consistent." An equation is dimensionally consistent ilf the
dimension of the quantity on the left-hand side is the same as the dimension of the
combination of quantities on the right. Assuming we know these things, we can reason:
force has dimension lvfL/ T 2, so g( m, Ll) must be a combination of mass and acceleration
that also has dimension ML/ T 2. Clearly ifg is multiplication, so g( m,Ll) = ma, we get this
result, since the dimension of nut is the product of the dimension of m and the dimension
of tl, namely ML/ T 2 • The Buckingham IT theorem tells us that, in this case, multiplica-
tion is the only function you can put in for g to make the equation dimensionally
consistent. (For a more detailed explanation of dimensional analysis, including a proof of
the II theorem, see Barenblatt l 996. Technically, applying the technique requires a weaker
assumption th<m dimensional consistency, sometimes called "dimensional homogeneity;"
for more on this distinction see Lange 2009, p. 760.)
Although I have comrasted dimensional analysis with the use of observational data,
dimensional analysis is not an a priori technique; the things one must know to apply it,
namely the dimensions of the quantities and that the quantities stand in a dimensionally
homogeneous relationship, cannot be known a priori. In fact, regarding the example
I have used, the usual route to coming to know that the dimension of force is ML/ T 2
requires prior knowledge of the law that F ma; those of us who know the dimension of
force in this way obviously could not use out knowledge to discover that F = Hltl.
The Metaphysio- of Quantities and Their Dimensions 173
formulas just reflect arbitrary choices we have made when setting up our
scales of measurement.
I'm going to examine this debate. The idea that the dimension of a
quantity has something to do with its nature can be made precise in diffr:rent
ways; I will identify two distinct ways to make it precise, and defend one
of them.
This debate is not one that philosophers discuss, or have ever discussed.
~t least, I have not been able to find anything written on it. 3 The question
1~, nevertheless, interesting, worthy of our attention. I think this is obvious;
further evidence comes from the fact that physicists like Tolman and
Bridgman saw that dimensional analysis raised questions about the meta-
physical import of talk of the dimension of a quantity, and proceded to take
those questions up. For this reason my primary interlocutors will be those
physicists.
If you have never before been exposed to talk of the dimension of
quantity, either through exposure to dimensional analysis or through
some other avenue, you might feel that you are in no position at all to
think about this debate. It's a little like asking someone who has had no
exposure to physics to evaluate hypotheses about the essential properties of
electrons. Fortunately, there is widespread agreement about part of the
answer to the question of what a quantity's dimension tells us about it.
The debate is over whether a quantity's dimension tells us any more than
:i There are certainly articles in philosophy journals that discuss dimensional analysis,
for example Marc Lange's (2009) "Dimensional Explanation" mid Luce's (1971) "Similar
Systems and Dunensionally Invariant Laws." But I haven't found any that discuss the
na;t:r? o'. dimensions, or what dimensions say about the nature of quantities.
I h'.s. is not to say that philosophers have ignored questions about the nature of
quantm<:-'. Ir: '.act some questions about the nature of quantities arc increasingly a focus
of attention. I akc th_e property of being 2 kg in mass. Is this a fondamental, or perfoctly
natu.ra!, property? I~ it 1s non-fundamental, what are the fundamental mass properties or
relations, and how _is bemg 2 kg mms defined in terms of them? Mundy ( 1987) can be
understood as holdrng that the property of being 2 kg mass, and all other particular mass
propem~s, are fun~amen,;al, while Field's views about quantities (Field 1980) suggest that
1r is the mass-less and mass-congruence" relations that arc fundamental (intuitivdv tl
"'.as_s-/ess b holds iff a is less massive thmi b; and a, b tlre mass-congruent to l; d holds ifr'the
;l;Hercnce 111 mass between a and bis the same as the difference in mass between c and d).
l here are other views_ also (surveyed in Eddon 2013). I think that these questions about
the nanire of quammes are orthogonal to the question I take up, the question of what
dune1~s1or,1s say about t!1e nature of quantities. For these questions, the questions Mundy's
:i;1d held~ and others work 1s relevant to, arc questions "internal" to a given quantity.
[he .question they want to answer.about mass-the question of which mass properties or
relations are most fundamcntal--1s a question just about mass, not about its relation to
any ot_her quantities. But the question I'm asking, the question of what the dimension ofa
quannty say.s .about its nanire, is a question about the relations between that quantity and
other quantmcs, as we will sec.
174 Bmdford Skow
that. Cfolman's claim that it tells us something about the quantity's nature
goes beyond the agreed-on part.) Seeing the agreed-on claims about dimen-
sions will, I hope, help the uninitiated get a better grip on what the debate is
about. So that is what I will start with, in the next section.
2. QUANTITIES, SCALES, AND DIMENSION
FORMULAS
Dimensions are had by quantities; so I should start by saying a bit more about
what a quantiry is. Familiar examples are, again, force, mass, and acceleration,
and also velociry, momentum, and energy. I will take a quantity like mass to
be a family of properties, the "specific values" of that quantiry. So the mass
family includes the properties, or values, having 1 kg mass, having 2 kg mass,
and so on.
There is a distinction between "vector" quantities and "scalar" quantities:
all quantities have "magnitude," but vector quantities also have a "direction"
associated with them. Although many quantities of interest are vector
quantities (force and acceleration for example), for the sake of simplicity
I will limit my attention to scalar quantities, and often pretend that some
vector quantities like force are scalar quantities.
An important notion for us is that of a scale of measurement ofa quantity.
A scale associates a number with each value of that quantity; the kilogram
scale, for example, associates the number 2 with the property of having 2 kg
mass. Now of comse there are tons of ways to associate numbers with mass
values, many of them useless-it would be futile to associate the number
3 with eve1y mass value. What I care about are "faithful" scales, scales that
adequately reflect the intrinsic stmcture of the quantities they are scales for.
I won't go into what that intrinsic structure is in a great deal of detail; it will be
enough for our purposes to note that the pro perry of having 4 kg mass is, in
some sense, "double" the properry of having 2 kg mass, and that it is in virtue
of this that a scale is faithful only if the number it assigns the first properry is
double the number it assigns the second. In general, just as the ratio of the
property of having 4 kg mass to the properry of having 2 kg mass is 2, any two
values of a given quantity stand in a definite numerical ratio, and a faithfol
scale will assign to those values numbers that stand in the same ratio.
It follows that to set up a scale for measuring a quantity it is enough to
choose a unit-the value v of that quantity that shall be associated with the
number 1. The number the scale assigns to any other value u is then
determined by the ratio of u to v: if the ratio of u to v is n, then the scale
must assign n to u, if it is to be faithful.
The Metaphysics of Quantities and Their Dimensions 175
It follows, further, that any two faithful scales for a given quantity, like
mass, differ from each other by a positive multiplic,1.tive constant: if we write
S( v) c-cc n to mean that scale S assigns quantity value 1; the number n, then if
Si and S2 are faithful scales for a quantity, there is a fixed positive number K
such that S2(v) == KS1 (v) for every value v. 4 For example, the numbers the
centimeter scale assigns to lengths are all 100 times larger than the numbers
the meter scale assigns to lengths.
Now at the beginning of this paper I said what the dimensions of some
quantities (mass, acceleration) were, and quoted Tolman's claim that the
dimension of quantity has something to do with its nature or definition.
This talk of the "dimension" of a quantity is, however, a little nebulous, and
it is time now to introduce more precise terminology. The term we need is
"the dimension formula of a quantity." Formulas arc in the s<m1e ontological
category as sentences. The dimension formula of mass is "M ," a capital "m,"
and the dimension formula of acceleration is "L/T 2 ," a capital "I" followed
by a slash, etc. Just as we can ask what a sentence means, we can ask what a
formula means. The most uncontroversial claim about dimension formulas
says this about their meanings:
(l) The dimension formula of a quantity defines that quantity's dimen-
sion function.
Great; so what's a dimension function? Here is the answer:
'l I should note that this claim only applies to scales for quantities with a certain
structure, which might be described as "scalar quantities with a ratio structure and a lower
bound." These arc the only kinds of quantities I will be concerned with in this paper.
You might wony that temperature is a counterexample to the claim I make in the text
about scalar quantities. The numbers the Kelvin scale assigns to temperatures are not a
fixed multiple of the numbers the Celsius scale assigns to temperatures (the temperature
value to which Celsius assigns the number 0, Kelvin assigns 273.15, but there is no
positive K with 0 = /(.273.15). This is not inconsistent with my claim, though, since the
Celsius scale is not faithful.
The short explanation of why it is not faithfol is that it fails to adequately reflect that
temperature has a lower bound-to do so it would have to assign O to the lower bound.
The longer, and true, explanation is more complicated. In fact temperature does not
have a lower bound. It is possible for a thermodynamic system to have a negative
temperature on the Kelvin scale. Still, even though temperature docs not have a lower
bound, the Celsius scale is not faithful-and neither is the Kelvin scale. Systems with
negative temperatures (on the Kelvin scale) are hotter than systems with positive temper-
atures (on the Kelvin scale)-thc first kind of system is disposed to transfer energy to the
second kind. But a faithful scale should assign higher numbers to hotter things (this is a
problem for the Celsius scale, since it <Lssigns ne!,>ative numbers to the temperatures to
which the Kelvin scale assigns negative numbers).
Since temperature is not, in the end, a quantity with a lower bound, it technically falls
outside the scope of my claim that any two scales for a quantity differ by a positive
multiple.
176 Bmdford Skow
(2) A quantity s dimension function tells you how the numbers
assigned to values of that quantity change, when you change from
using one jystem of sm!es of measurement to another (in the same
class).
This needs a lot of unpacking, of course; we will want to know what a system
of scales is, what it is for two systems to be in the same class, and how the
dimension function tells you the cited fact-we will want to know what
the function's inputs and outputs arc. But (l) and (2) at least makes one of the
dimension formula's roles clear. If we switch from using the meter scale for
length to using the centimeter scale for length, the numbers assigned to length
values all go up by a factor of 100; somehow this function that the dimension
formula for length defines, the dimension function for length, is going to
deliver up to us this factor.
To a philosopher, the job description in (1) that dimension formulas
answer to may not look very interesting. The question is going to be, the
interesting debate is going to be over, whether there is any other, metaphys-
ically interesting, job that dimension formulas do.
Before getting to that question I should spell out how dimension formu-
las define dimension functions, and how dimension functions work. 5
A system of scales of measurement is a set of related scales for measuring a
set of quantities. A system of scales designates some quantities as "primary"
and some as "secondary." The primary quantities are those measured by
scales in which the unit is not chosen for its relation to the unit of any other
scale. The secondary quantities are those measured by scales in which the
unit is chosen for its relation to the units of other scales. For example, we
usually think in terms of a system in which speed is a secondary quantity,
and length and duration are primary. The unit for speed is chosen by
reference to the units for length and duration: we say that the unit for
speed is the speed at which something travels when it covers one unit of
distance in one unit of time.
That's one way to choose a unit for speed, given units for other quantities.
It is not the only way-we could choose the unit to be the speed at which
something travels when it covers two units of distance in one unit of time. In
general, there is always more than one way to choose a unit for a secondary
quantity, given units for the primaty quantities. 6
Much of the material to follow comes from Chapter l of Barenblatt 1996, though
my terminology is in some places slightly different.
6
Here is a more interesting example: suppose we want force to be a secondary
quantity, and length, mass, and duration to be primary. The usual way to choose a unit
for force is to say that it is the amount of force required to produce unit acceleration in
something with unit mass. But we could choose a unit for force by saying that the unit
The Mettiphysics of Qutintities tind Their Dimensions 177
To se~ up a system of scales, then, we divide the quantities we will
measure '.nto_ those that will be primary and those that will be secondary,
select untts for the primary quantities, and specify how the units for the
secondary quantities are determined by the units for the primary ones.
Suppose two systems of scales X1 and X2 measure all the same quantities,
that the set of q~antitics designated as primaty is the same in both systems,
and that the units for secondary quantities are defined by reference to the
units for primary quantities in the same way. So all differences between X
and X2. Aow, from differences over which values of the primary quantities ar~
the un.1ts. 1 hen X1 and X2 belong to the same chss of systems.
Ag.am, when \;e change from using a scale S 1 for a given quantity Q
to usmg a scale 52, the numbers assigned to values of Q all get multiplied
by some constant K. If S1 and S2 belong to systems of scales that are in the
same class, then even if Q is a secondtiry quantiry, the constant K is
determined by the sctles for the primtiry quantities:
(3) If S1 (the "old" scale) and S2 (the "new" scale) are scales that measure
the same quantity Q, and they belong to systems of scales X1 and X 2
that are in the same class, then the constant K satisfying S2 ( v) cc~
KS1 ( v) (where v is a value of Q) is a function of the ratios of the old
units to the new units of the scales for the primtiry quantities.
This function, which takes some numbers (the ratios of the old units to the
new units) and outputs K, is the dimension function of Q.
To know how the dimension function is defined we look to the dimen-
si~m fo~mula for Q (recall (1)), and interpret the letters that appear in the
d1mens1on formula to denote the numbers that are the ratios of the old units
to the new units of the relevant quantities (and interpret the other symbols
in the dimension formula, like "/," to denote the usual mathematical
~perations). To see how all this works, let us assume that we are working
111 a class of systems of scales in which only length, duration, and mass are
primary, and in which the unit for speed, a secondary quantity, is defined in
the usual way. In this case, there are three primary quantities, and so, since
dimension functions take as input the ratios of old to new units of the
p:ima!J'. quantities, all dimension functions take three arguments. The
dimension formulas that define these functions contain the letters "Iv[ "
"L," and "T," and no other letters; when a dimension formula is used : 0
define a dimension function, "M" is interpreted to name the ratio of the old
unit for mass to the new unit, "L" the ratio of the old unit for length to the
for~e is t:he strength of the gravitational force between two bodies of unit mass that arc one
unit of distance apart.
178 Brr1dfi11d Skow
new unit, and" T" the ratio of the old unit for duration to the new unit. When
Q is a quantity, [Q] is its dimension fonction, and the conversion factor K is
given by K = [Q](M,L,T), the application of the function [Q] to the
arguments M,L, T. 7 •8 ·rhis may be clearer ifl work through some examples:
Example l: Mass. The dimension formula fi)r mass, at least in the class of
systems of scales we arc using, is, again, "M." So [rnass](Nl,L, T) = M,
and M is the number that is the ratio of the old unit for mass to the new
unit for mass. So suppose we arc changing from a system of scales that
uses grams to measure mass, to a system that is othetwise the same except
that it uses kilograms to measure mass. The old unit for mass then is the
gram, and the new unit is the kilogram. The ratio of the gram to the
kilogram is l/1000 (a gram is 111 OOOth of a kilogram). So in this case M =
111000. Then if the old scale, the gram scale, assigns the number n to a
mass value v, we compute that the new scale, the kilogram scale, assigns
the number [mass](M, L, T)n = Mn = (l/lOOO)n to v. The result is
familiar: the numbers assigned to masses change by a factor of l I 1000
when we change from using grams to using kilograms.
Example 2: Speed. The dimension formula for speed, at least in the class of
systems of scales we are using, is "L/ T." So the dimension function for
speed is defined by the equation [speed](M,L, T) = L/ T If we change
from using meters to using centimeters, and use the same scales for the other
primaiy quantities, then L, the ratio of the old unit for length to the new
unit, is equal to l 00, while M = T = 1. So if the old scale for speed assigns
the number n to a speed value v, we compute that the new scale assigns the
number (L/T)n = lOOn to v. Again the result is familiar: the numbers
assigned to speeds change by a factor of 100 when we change from using the
"meters per second scale" to using the "centimeters per second" scale.
7
l should note that the claim that the conversion factor for a quantity in this class is
a fonction, in the mathematical sense, of three arguments, is not the same as the claim
that the conversion factor for a given quantity depend< on each of those arguments. The
function defined by the expression f (x,y,z) = x 2 is a fimction of three arguments, but its
value only depends on the first argument. The examples below illustrate this distinction.
8
I said that everything in this section is common ground in the debate over the nature
of dimensions. This may not quite be right. Bridgman says that on his view "the symbols
in the dimensional formula [are] reminders of the rules of operation which we used
physically in getting the numerical measure of the quantity;" they do not represent "the
facts used in changing from one set of units to another" (30). But he also says that his view
"cannot be distinguished" from the one lam describing "as far <L> any results go." I am not
entirely sure what Bridgman means when he says we should regard dimension formulas as
reminders. Barenblatt clearly endorses the claim that dimension formulas just serve to
define dimension functions, and that the letters "M" and so on denote ratios of old units
to new (1996: 31-2).
The Metdphysics of Quantities rmd Their Dimensions 179
One thing about dimension functions is worth emphasizing. The dimension
function of a quantity gives the factor by which the scale for measuring it
changes when we pass from one system of scales to another system in f1 given
clrtss ofsmles. So dimension fonctions, and dimension formulas too, are class
relative.
For example, we usually take force to be a secondary quantity, and mass a
primary one; the dimension formula for force is then J\1L/7' 2 . But we could
instead use a class of systems of scales in which mass is secondaiy, and force
primary; the dimension formula for force is then just F, while the dimension
2 9
formula for mass is FT / L. The class-relativity of dimension formulas (and
dimension functions) will be important in the next section.
3. THE POSITIVIST THEORY OF DIMENSIONS,
AND ITS OPPONENTS
Herc are the teams. On the one side, Tolman and others, who think that the
dimension of a quantity has something to do with its nature. On the other,
Bridgman and his allies, who deny this. I will call the view of Bridgman and
his allies The Positivist Thcoty of Dimensions; 10 to have it set out explicitly,
if a bit vaguely, the view is this:
THE PosIT!VIST THEORY: The dimension formula of a quantity defines the
dimension function of that quantity (relative to some class of systems of
scale). It docs not (in addition) have anything to do with the nature or
essence of that quantity.
So what, exactly, do those who oppose the positivist theory believe? I will
distinguish between two distinct anti-positivist theses. These theses in turn
9
To figure this out, we assume that the dimension of the right-hand side of F = ma
is t:he s.ame as the. dimension of the left-hand sic!~. Acceleration is st.ill s?cond~:r· ''.nd its
umt still chosen m the customary way, so its cl11nens1on formula rs still L/ 7 . So the
dimension formula Z for mass must satisfy F Z·L/ T 2 , where what can go in for Z are
products of powers of the primary quantities L, T, and F. The only solution is FT 2 / L.
(Why products of powers? I say something about this in the appendix.)
IO The appcar<mce of "positivist" is meant to be suggestive-Bridgman W<L> an opcr-
ationalist, the worst kind of positivist-but should not be taken too seriously. Other
proponents include Max Plank: "to inquire into the 'real' dimension of a quantity has no
more meaning than to inquire into the 'real' nature of an object" (1932: 46; sec also page 8);
Barenblarr (1996); (Langhaar 1951: 5); and (Palacios 1964: xiv). These people may not all
subscribe to the positivist view exactly as I state it, but my interest is in the view I state, not
the interpretive question of whether, and if so just where, its allies' actual views deviate
from it.
180 Bnulford Skow
share two presuppositions, so before stating the theses [ want to articulate
those presuppositions.
The first presupposition is that there is a "naturnl" way to divide quan-
tities into primary and secondary. Return to the two systems of scales I ended
the last section with. One designated mass as primary and force as secondaty,
the other did the reverse. These systems belong to different classes (for, again,
a class of systems of scales is partly determined by a choice of which quantities
arc to be regarded as primaty and which secondary). Now there arc various
criteria we might use to evaluate these classes of scales. One of them might,
for example, be better than another if we arc intcrsted in making certain
kinds of measurements. Anti-positivists think that we can also evaluate them
with respect to whether they designated as primary only quantities that really
are primary. The presupposition here is this:
NoN-REt.ATIVIW: Quantities are not just primary, or secondary, rela-
tive to this or that class of systems of scales; they are also primary, or
seconcla1y, in a non-relative way.
The positivist theory rejects NON-RELATIVITY. However, despite the positiv-
ist theory's opposition to it, I think NON-RELATIVITY is a quite plausible
thing to believe. NoN-REIATIVlTY can appear even more tempting if we use a
different word in place of"primary": doesn't it seem right to say that some
quantities are fimdmnenta! in a non-relative way?
The idea that some properties arc fundamental and others arc derivative
has great currency in metaphysics. t I I certainly think we should accept it.
Since quantities are just families of properties, we should also accept that
some quantities arc fundamental and others derivativc. 12 This gives us a way
to "argue" for NON-RELATIVITY. Some quantities are fundamental, and a
quantity is primary simplicitcr 13 iff it is fundamental; so some quantities arc
primary simplicitcr. I don't expect this argument to convince any skcptics-
that's why "argue" is in scare-quotes-but it at least draws a connection
between being primary simplicitcr and a notion, fundamentality, that is
already well understood. 14
11 Its popularity among contemporary metaphysicians traces back to (Lewis l 983).
12 There arc derails ro work out about the relationship between "fimdamental" as a
predicate of properties and "fundamental" as a predicate of quantities. We could say that a
quantity is fundamental iff every one of it values is a fundamental property; or iff the
disjunction of its values is a fundamental property; and there arc other options. I will not
t1y to work out these details here.
13 "Primary simpliciter" is just a short way to write "primary in a non-relative way."
1'\ A referee objected: Gross Domestic Product (GDP) is surely not a fimdamental
quantity; so I must hold that it is a secondary quantity, and therefore that its dimension
formula, in a class of systems of srnles that designates as primary the quantities that really
Ihe Ji1etaphysics of Quantities and Their Dimensions 181
I cl? not '.11ca:1 to suggest that NoN-REIAIWITY is a completely unprob-
lcmat1c thesis. If we accept that some quant:itics arc primary simpliciter, we
are going to want to know which quantities those arc. How do we go about
figuring this out? Can we know which quantities are primary sirnplicitcr, or
arc we bound to remain ignorant?
. I wil~ only say something brief about these questions. The same questions
ansc with respect to fundamental properties. In that case, one common
suggestion is that we take our best physical theories as guides to which
properties arc fundamental-if a property appears in one of those theories
(is ~xprcsscd by a predicate in one of those theories), that is good evidence it
is for~damcntal. ~'o the extent that this is a good answer to the epistemic
questions about fundamental properties, an analogous suggestion is a good
answer to the epistemic questions about primary quantities: if physicists,
when their aims involve no practical computations, but only the formula-
tion of the laws, use a system of scales in which a quantity is primary, that is
good evidence that that quantity is primary simplicitcr.
A lot more could be said about the epistemology of primary quantities,
and how it does or docs not recapitulate the epistemology of fundamental
properties, but since these epistemic questions are not relevant to what
follows, I will leave the topic here.
Anti-positivists presuppose NON-RELATIVITY because they presuppose
that each quantity has a "true" dimension formula, and a quantity cannot
have a "true" dimension formula unless NoN-RElATIVITY holds. For recall
that a given quantity has tons of dimension formulas, one for each class of
systems of scales that measures that quantity. One of those dimension
formulas can only be its "true" dimension formula if nature singles out
exactly one class of systems of scales as the "true" class; a quantity's true
are primary, is some combination of "Jvf," "/,," "T," and so on (assuming that mass,
length, and dur~tion are primary). This, however, is implausible.
_I am .not cnnrely sure "'.hat t.o think about GDP, or other quantities that appear in
econo.mtcs and other_ special sciences, but I am tempted by this reply. Consider the
quanttty frequency. 1 he unit for frequency is chosen to be the value of frequency had
when .the relevant system completes one "cycle" in one unit of time. But "cycles" is, not a
quantity: 0:1c do?s not choose a unit for measuring cycles, one just counrs cycles. That's
why the chmcns1on formula for frequency contains "T'' (the formula is "l/7'") but
~o letters :hat have anything to do with cycles. The same, I suspect, goes for GDP. We
(111 the US) usually measure GDP in dollars per year, but (my suspicion is) we should
not regard the dollar as a unit for measuring value, but as a thing we count, like cycles.
(If the dollar an~! the euro were both units for measuring value, then one dollar should
always be so.m~ fixed n::rmber of euros. But it's not; the exchange rate fluctuates.) If this is
nght, t?en tt ts not aher aH implausible that GD P's dimension formula contains only
letters for fundamental physical quantities: its dimension formula is the same as that of
frequency, "I /T."
182 Bmrijord Skow
dimension formula is then its dimension formula in the true class. What
would nature have to do to single our exactly one class of systems of scales as
the true class? It certainly must single out one set of quantities as those that
arc primary simpliciter, so it can say that the true class has the property of
designated as primary all and only the quantities that are primary simpliciter.
But it turns out-and I kind of wish this weren't so--that accepting
NoN-REI.ATIVl"IY is not sufficient for defining "true class," and so is not
sufficient for defining "true dimension formula." As I mentioned above, a
class of systems of scales is determined, not just by a choice of which
quantities to designate as prima1y, but also by a way of choosing the units
of the secondary quantities, given units for the primary quantities. And there
is always more than one way to choose a unit for a secondary quantity. So
anti-positivists need a second assumption:
RIGHT UNITS: for any quantity that is secondary simpliciter, only one way
of choosing a unit for that quantity given units for the primary quantities
is the "right" way.
If RIGHT UNITS is correct, then the true class of systems of scales can be
defined as the class that designates as primary exactly the quantities that
arc primary simpliciter, and chooses the units for the secondary quantities
in the right way.
RIGHT UNITS is, I think, harder to swallow than Ncm-REI.ATIVITY. We
usually choose the unit for speed to be the value had by something that
moves unit distance in unit time. But, as I said earlier, we could choose it to
be the value had something that moves two times the unit distance in unit
time. Which is the "right" way for defining a unit for speed? Some might
find it hard to believe that this question has an answer. Different people
might find it hard for different reasons; one reason might be that, if it docs
have an answer, it is hard to see how we could know what that answer is.
T'his particular example, however, doesn't lead me to doubt Rrorr UNITS;
the usual choice for a unit of speed, the choice that designates as the unit
the speed something has when it moves unit distance in unit time, is, at
least in some sense, more natural than any alternative. But other examples
are harder. 15 Still, I will not take an in-depth look at the cases for and
l5 It is tempting to say that the right way for choosing a unit for a secondary quantity
never uses "arbitrarv constants" (as rhe second way for choosing a unit for speed uses the
number 2). Bur rhi~ is nor rhe end of the stoty. Some <1uanriries, like the curl of a vector
field, cannot be defined in terms of more fundamental quantities in any way without
using something like an arbitraty constant (in this case, an arbitrary choice of. an
orientation for space). One thing to say about this case is that curl is nor a scalar quantity,
so is of a kind of quantity rhar I am ignoring. A more substantive response is to say that
curl is nor a "real" quantity, since from a four-dimensional perspective one never needs to
The Metaphysics of Q1tr1ntities and Their Dimensions 183
against RIGHT UNITS here. I will say, though, that one of the theses I will
be interested in, which I will shortly name DEFINITIONAL CONNECTION,
can get by without assuming RIGHT UNITS, even though the most natural
statement of if does require that assumption. (I will explain how it can
avoid RIGHT UNITS in section 6.) So DEFINITIONAi. CoNNECf'ION is still
something you could believe, even if you can't stomach Rica-IT UNITS.
Having flagged that the anti-positivist theses to be discussed rnake pre-
suppositions which could easily be challenged, let's press on. Tolrnan's idea
was that the dimension of quantity has something to do with its essential
nature, that it is a "shorthand restatement of its definition." Now Kit Fine
famously linked essences to definitions in his 1994 paper "Essence and
Modality." The essence of a thing is its "real definition"; it says "what a thing
is," as a "nominal" definition says what a word means. I will read Tolman's
reference to a quantity's "definition" as talk of its real definition, and so also
as talk of its essence. For the sake of precision it is worth regimenting all such
talk using the same locution, so I will use the operator "It is definitional of
X that .... " Using this notion, just what does Tolman means when he says
that the dimension of a quantity is a restatement of its definition? For now
I will only give a vague answer; I will take up the task of making it more
precise in section 5:
DEFINITIONAL CONNECTION: Let Pl, ... , Pn be the symbols for the
primary quantities that appear in quantity Q's dimension formula in
the true class of systems of scales. Then some relationship between
Q and the primary quantities associated with the symbols PI, ... , Pn is
definitional of Q.
"That's very abstract, so it will help to see an instance. Suppose the speed is a
secondary quantity, and length and duration are prima1y, and that the
dimension formula for speed in the true class is "L/T." Then, if DEFIN-
ITIONAL CONNECTION is right, some relationship between speed, length, and
time is definitional of speed. Just what this relationship might be, I will say
more about in section 5.
Before spending more time on DEFINITIONAi. CONNECTION, I want to get
another anti-positivist idea about dimensions on the table. The other one is,
use the curl of a vector field to do physics. (Given rhe theory of relativity, speed is nor a
real.quannty either, bur my argument in the paper is compatible with this; speed, even in
section 5, serves only as an example.) Further problems come from very simple quantities.
!·fore are two. ways to. select a unit for area: let it be the area value had by a square with
umt-lengt.h s1_d~:;. !er .~t be .rhe area value had by an equilateral triangle with unit-length
sides. \'\lh1ch is nght ? Neither uses an arbirra1y constant. !fl had ro take a stand, I would
say that the square is right, but I don't know how to justify this answer.
184 Bradford Skow
in fact, more popular. It is suggested by another expression of anti-positivism,
due to W. W. Williams:
The dimensional formulae may be taken as representing the physical identities of
the various quantities, as indicating, in fact, how our conceptions of their physical
nature ... arc formed, just as the formula of a chemical substance indicates its
composition and chemical identity .... The question then arises, what is the test
of the identity of a physirnl quantity? Plainly it is the manner in which the unit of
that quantity is built up (ultimately) from the fundamental units L, M, and T [ ... J.
(Quoted in Bridgman 1922: 26; the paper quoted from was published in 1892)
Some of what Williams says is close to DEFINITIONAL CONNECTION. But
Williams also draws an analogy between dimensional formulas and chemical
formulas. The idea seems to be that, just as the chemical formula for glucose,
"C6 H 12 0 6 ,'' tells us that a glucose molecule is built out of six carbon atoms,
twelve hydrogen atoms, and six oxygen atoms, the dimension formula for
acceleration, "L/ T 2 ,"tells us that, and how, acceleration values are "built up
from" lengths and durations. This idea identifies a second role dimensional
formulas might play, in addition to that of defining dimension functions. 16
Generalizing gives us our second anti-positivist thesis:
CoNSTRUCI'ION: Let Q be any quantity, and D its dimension formula in
the true class of systems of scales. If Q is non-primary, then D exhibits the
way in which values of Q arc constructed, or built out of, values of
primary quantities.
The picture here is that there are several different ways in which two
quantity values may be combined to create a third. Two quantiry values
may be multiplied together; or one may be divided by the other; or one
quantiry value may be raised to some power. For example, if we assume that
speed is a secondary quantiry in the true class, and that length and duration
are primary, then the dimension formula of speed in the true class is "L/ T."
In this formula the symbols for length and duration flank the division sign.
According to CoNSTRUCrION, this indicates that speed values are con-
structed by taking length values and dividing them by duration values.
The meter per second, for example, which is a speed value, is constructed
by dividing the meter (the value of length assigned the number I when we
choose the meter as our unit) by the second.
16
A formula can define a function without indicating how anything is built up from
other things: the formula "x - y" defines the subtraction function, but even though 12 =
10 - 2, the formula doesn't indicate that 12 is built up from 10 and 2 in any sense.
Conversely, chemical formulas indicate how molecules are built from atoms, but do not
define any functions.
The Metaphysics of Quantities and Their Dimensions 185
It is worth emphasizing that CONSTRl!Cl'!ClN says that some quantity
urtfues-properties--are built, by operations with names like "multiplica-
tion" and "division," out of other l!t!!ues. Some resistance to CoNSTRUCT!ON
is based on failing to appreciate this fact. Bridgman complains of"treat[ing]
the dimensional formula as if it expressed operations actually performed on
physical entities, as if we took a certain number of fret and divided them by
a certain number of seconds" (29). Later on the page he asserts that "We
cannot perform algebraic operations on physical lengths, just as we can
never divide anything by a physical time" (29). 17 If Bridgman is saying that
the following instruction makes no sense, then he is absolutely right:
You've got a meter stick (which is a "physical entity") in your left hand. Please take it
and divided it by the stretch of time during which this table exists. 18
Division just is not an operation defined on meter sticks and stretches of
time. But this observation does not touch CONSTRUcr·IoN, which makes no
claims or presuppositions about dividing physical things by each other.
One could try to make Bridgman-like observations about quantity values.
Pounding the table one could say, "Multiplication is an algebraic operation
defined on numbers, not on quantity values!" But this is only kind of true.
Mathematicians happily talk of multiplying functions together, and of
dividing one group by another. Of course the operations going under the
names "multiplication" and "division" here are different from those going
under those names when we are working with numbers. But the operations
are closely enough related that we use the same name for them. I see no
reason why a set of quantity values could not have an algebraic structure like
that had by sets of numbers, fi.mctions, and groups, so that there were
operations on sets of quantity values that deserved to be called multiplica-
tion and division.
We now have two anti-positivist theses on display, DEFINITIONAL CoN-
NECrION and CoNSTRUCrION (which share the presuppositions NoN-
RELATIVITY and RIGHT UNITS). The first thing I want to emphasize about
them is that they are different theses. Granted, there is probably a way of
understanding talk of one property being constructed out of others on
which DEFINITIONAL CONNECTION, or something quite close to it, follows
from CoNSTRUCl10N. But DEFINITIONAL CONNECTION does not entail CoN-
~TRUCTION; it is logically weaker. You can think that dimension formulas tell
us something about the natures of quantities without thinking that quan-
tities are constructed out of other quantities.
17
Barcnblatt echoes this argument (1996: 30).
18
Suppose substantivalism about time is true, so that there arc such entities as
stretches of time.
 Bradford Skow
186
My main aim in this paper is to say something in favor of DEFfNITIONAL
CONNECl'ION. Distinguishing between DEFINITIONAL CONNECTION and
CoNSTRUCrION is an important part of my argument, since many of the
arguments proponents of the positivist view give make contaC[ only with
CONSTRUCTION, and leave DFF!NITIONAL CoNNF.CrION untouched.
Before discussing DEFINITIONAL CoNNFCf'ION explicitly I want to look at
the best of those arguments, partly to illustrate its irrelevance to DEFIN-
ITIONAL CoNNEGrION, but also because it is worth, at least briefly, corning to
some conclusion about CONSTRUCTION. My conclusion will be that it is
defensible, though I myself see no good motivation for defending it.
4. BUILDING SOME QUANTITIES FROM OTHERS?
One argument against CoNSTRUCl'lON is especially popular; I myself have
endorsed it in the pasr. 19 Herc is a statement of it from Barcnblatt's book
20
Scaling, Se/fSimilarity, and Intermediate Asymptotics:
[I]f the relations for the derived units [that is, units for the secondary quantities]
mentioned above were actually to make sense as products or quotients of the
fundamental [that is, primaty] units, they would have ro be independent of what
we me<lll when carrying out the multiplication or division. For example, according to
the above definition [which I will not quote, but which entails something like
CONSTRUCTION], kgf m is the derived unit for the moment of a force as w~ll as for
work; m 2 /s is the derived unit for the stream function as well as the kmemattc
viscosity, etc. But it is not implied that the stream function is measured in nmltipl~s
of a basic amount of kinematic viscosity or that the moment of a force ts measured ll1
multiples of a basic amount of world In contrast, using our definition [of dimcns'.on
formulas, namely the positivist theory], the fact that the dimensions of two physical
quantities of different nature arc identical docs not seem unnatural. (1996: 33)
I am not entirely sure what Barenblatt's argument here is. But I can find an
interesting argument in here. Let's examine it and not worry about whether
it is exactly what Barenblatt had in mind.
Barenblatt's central observation is that there are distinct quantities with
identical dimension formulas (in some given class of system of scales). He
mentions kinematic viscosity and the stream function. Roughly speaking,
the kinematic viscosity of a fluid is a measure of how hard each bit of the
fluid "pulls" against its neighbors when its neighbors move past it. In a
19
In my (2012), and section 7.4 of my (2015). . First Way. CONSTRUCTION does not say that it is impossible for distinct
Something like this argument also appears in (Palacios 1964: xiii) and (Duncan
20
1953: 123).
The Metapl~ysic( of Quantities and Their Dimensions 187
highly viscous fluid, when a bit of fluid neighboring B moves past it, B pulls
rather hard against that neighbor. (Honey has higher viscosity than water.
As a practical matter, how viscous a fluid is tells you how hard it will be for
you to move through it: as you move you pull the bits of the fluid that touch
you along with you, and this is more difficult to the extent that those bits'
neighbors pull back.) As for the stream function: if you imagine a fluid
(technically an incompressible fluid) flowing in two dimensions, the value of
the stream function at a point P tells you how much fluid flows across the
line from P to an arbitrarily given point A in unit time. Clearly the stream
function and kinematic viscosity are distinct quantities. A blob of honey, all
at the same temperature, has the same viscosity everywhere. 'T'he stream
function, on the other hand, is a property, not of a thing like a blob of
honey, but of an event, the motion of a fluid. It varies from one point in the
flow to another. Nevertheless, in any class of systems of scales that designates
length and duration as primary quantities (and designates the stream func-
tion and kinematic viscosity as secondary), the stream function and kine-
matic viscosity have the same dimension formula, L2 /T.
Why might this be trouble for CONSTRUCTION? Here is one way to spell
out a simple argument against CONSTRUCTION that uses Barenblatt' s obser-
vation as a premise:
(Pl) If CoNSTRUCrION is true, then there cannot be distinct quantities
with the same dimension formula (in the true class of systems of
scales).
(P2) But there are distinct quantities with the same dimension formula.
(C) Therefore, CONSTRUCTION is false.
A reason to believe (Pl) is not hard to find. If CONSTRUCrION is true then
values of kinematic viscosity are built out of values of length and time.
Specifically, if CONSTRUCTION is true, then you get a value of kinematic
viscosity by taking a value of length, multiplying it by itself, and then
dividing the result by a value of duration. But if performing these operations
on values of length and duration gives you a value of kinematic viscosity,
then performing those operations on those same values cannot also give you
a value of a distinct quantity like the stream function. In general: distinct
quantities with the same dimension formula would have to be built up from
the primary quantities in the same way. But it cannot happen that distinct
quantities are built up from the same primaty quantities in the same way.
This might sound convincing, but it goes beyond what CoNSTRUCrION
officially says. There are two ways a proponent of CONSTRUCTION may
resist (Pl).
quantity values to be built up from the same basic values in the same way.
188 Bradford Skow
The argument above for (Pl) smuggled in an extra premise, a "uniqueness"
premise. 1-kre is one way to put it:
UNIQUENESS: For any way W of building a quantity value out of other
quantity values ql ,... ,qn, the application of W ro ql ,... ,qn is unique.
IfUNIQUENf};s is true, and if multiplication and division are ways of building
quantity values, then we do have a good argument for (Pl). But UNIQUENESS
may be rejected. True, multiplication and division of numbers satisfies
uniqueness. But there is plenty of precedent for non-mathematical ana-
logues of mathematical operations to fail to have all the properties of their
mathematical counterparts. Take summation. Lots of metaphysicians hap-
pily embrace the thesis that a collection of material things may have more
than one sum. Plenty think that you can have a bunch of day particles that
have two smns: one of the sums is a lump of clay, the other sum is a statue.
The observation that distinct quantities can have the same dimension
formula, then, does not refute CONSTRUCTION. 21
Second Way. 22 Barenblatt's argument may refute CONSTRUCrION when it is
interpreted one way, but there is a better way to interpret it that escapes the
argument. Officially CoNSTRucnoN just says that Q's dimension formula
"exhibits the way in which values of Qare constructed, or built out of, values
of primary quantities." I have been interpreting this w mean that, ifa qu<mtity
(like viscosity) has for its dimension formula "L 2 / T" then each of its values is
completely specified by saying that it is the square of a given length value,
divided by a given value of duration. On this interpretation, the dimension
formula contains complete information about how a quantity's value,.s are
built up from the basic values. But there is a weaker way to interpret
CoNSTRUCl10N: interpret it to say that a quantity's dimension formula gives
only partial information about how the quantity values are built up. This
weaker interpretation becomes more plausible when you look at the precise
(nominal) definitions of "kinematic viscosity" and of "the stream function"
(they are rather involved, so I won't state them here): the quantities invoked
in the two definitions are quite different. On the weaker interpretation,
CoNSTRUCl'ION does not say that quantities with the same dimension formula
are built up in the same way, so even if UNIQUENESS is true, (Pl) is not.
21
This is not the end of the matter. Fans of uniqueness of summation say their
opponents fucc a "grounding problem": given that the srarne and the clay are sums of the
same particles, how do they manage to have differcnr properties? Fans of UNIQUENFSS
might mount a similar argument. following this thread of the dialectic would take me too
far away from my main line of argument. My guess is that it would parallel the dialectic in
the case of summation, in which the denial of uniqueness remains a viable position. See
Bennett 2004 for more.
22
Thanks to Martin (;lazier and Karen Bennett, who (separately) proposed this way
ro me.
The Metaphysics of' Quantities and Their Dimensions 189
Still, :ven though CoN~TRUC:TION is defensible, I am not tempted to
endorse it, and l agree with Bridgman that the main temptation to endorse it
is. based on a 1:1.istak:. What is that temptation? Consider a law relating
different quanmtcs, for example Newton's second law. What is Newton's
second law? Of course eve1yone knows what it is: F ma. But what does
~his string of symbols mean? How should it be translated into English? !1ere
1s a common translation:
(4) The net force on any material body is equal to the result of
multiplying that body's mass and that body's acceleration.
But if t~is is right then it must make sense to multiply a mass and an
acceleratton. And surely mass and acceleration are not special here. So the
truth. o~Newton's second law supports the thesis that quantity values can be
mulnr:hed together. Again, surely multiplication is not special here; if
quantity values can be multiplied together then one value can also be
divided by an~the.r. Once we're comfortable with this idea it is not a great
leap to sta~t thmkmg of multiplicatio~1 and division as ways of constructing
new quamr:r. v~ues from old ones, of thinking, for example, of speed values
as got by d1v1dmg length values by duration values. 2 :l
~ut I do not t~ink that (4) is what F = ma means in English. In this
demal I follow Bridgman, and I also follow him in his views about what
equations like F = ma do mean. He wrote that 2 /i
· · • X1 [in an equation relating some quantities] might stand for the number which is
th~ measure of a speed, x2 the number which is the measure of a viscosity, etc. By a
s~1 t of shon~and method of statement we may abbreviate this long-winded descrip-
tion mto say mg that x1 is a speed, but of course it really is not, but is only a number
which measures speed. (1922: 17-18)
:x'hat goes for spec.cl goes for force, mass, and acceleration. Bridgman might
JUSt as well have said that while the "a" in "F = ma" stands for the number
which is the measure of a value of acceleration, we might as a sort of
shorthand say that a is a value of the quantity acceleration, or even that a
is an acceleration. But those who did not know this was shorthand would be
mis. -Iecl a bout wh at "a " d enotes. 25· (4) ts· not true when read literally, it is only
u:ue when read as shorthand. When we read it that way, it does not support
CONSTRUCTION.
~: ~or a recent endorsement of this argument, sec van lnwagcn 2014: 104-5.
Sec also Duncan 1953: 6 and Palacios 1964: 18.
25
. S~c al~o pagc.41 in Bridgman 1922 where he complains about people thinking that
somet!11~g like (4) is the reason why all true laws can be written so that the dimensions of
the leh side and the right side are the same.

190 Bmdford Skow
What, exactly, is the correct and fully explicit translation of F = mci into
English? Let Z be the class of systems of scales to which the system we
currently use, the SI system, belongs; then, I think, the correct and fully
explicit translation of F = mtl is
(5) For any material body B, and for any system of scales of measure-
ment S in Z, the number S assigns to the net force on B is equ<tl to
the result of multiplying the number S assigns to B's mass and the
26
number S assigns to B's acccleration.
As I said, I am not tempted to endorse CoNSTRUCrION. Even though I have
not, in this section, defended any arguments that it is false, there is an
argument against it that I like. But since my main aim in this paper is to
defend DEFINITIONAL CoNNECrION, not to refute CoNSTRUCrION, and since
I want to get on to that main aim now, I will save that argument for a later
section (section 6).
5. DEFENDING A DEFINITIONAL CONNECTION
Now that CONSTRUCTION has had its day in the sun I want to turn to
DEFINITIONAL CONNECTION, which has not received nearly as much atten-
tion. Here it is again:
DEFINITIONAL CONNECTION: Let Pl, ... , Pn be the symbols for the
primary quantities that appear in quantity Q's dimension formula in
the true class of systems of scales. Then some relationship between Q and
26 As I have been, I am pretending that force and acceleration are scalars. It is
straightforward to write down the correct vectorial version of (5), but brings in unneces-
sary complications.
One might concede that (5) is the translation of "F = ma" into English and try to
put together a more complicated argument for CONSTRUCTION. (5) makes reference to
numbers. One might argue that the truth of (5) must be grounded in some facts that
do not involve numbers. And perhaps those facts arc facts about the result of
multiplying mass values and acceleration values. But this argument does not succeed.
One need not accept CoNSTRUC!'lON to have facts available to ground the truth of (5).
Instead one could say that the facts that ground (5) are facts like these: if the force on
any material body were doubled, while its mass (and everything else) remained the
same, then its acceleration would double; if the mass of any material body were halved,
while the force on it (and everything else) remained the same, then its acceleration
would double; and so OIL In general: a body's acceleration is directly proportional to
the force on it, indirectly proportional to its mass, and independent of everything else.
(In fact Newton first formulated his second law of motion in terms of proportions; see
Newton 1999: 416.)
11Je Mettiphysics of Quantities and Their Dimensions 191
the primary quantities associated with the symbols Pl, ... , Pn is defin-
itional of Q.
The first :hing .to do is to make DFF!NITIONi\L CONNECTION more specific:
what relationship between Q and the primary quantities is definitional ofQ?
Unfortunately, I do not know how to answer this in a completely general
way. I_ know ho;v to do it for particular quantities, just not for an arbitrary
quantity. So lets focus on a claim about one particular quantity that is
supposed to follow--given an assumption I'm about to state-from the
more-specific-but-unstated version of DEFINITIONAL CONNFC!'ION.
Th~ assumption (one I've made many times) is that speed is a secondary
q~1anttty, and length and duration primary. If this is true, then DEFINITIONAL
CONNECTION is meant to entail the following claims about the essences of
particular values of speed:
• It ~s definitional of the property of moving at l m/s that anything that has
this property throug~Ollt some temporal interval that is 1 second long
moves along a curve ll1 space that is 1 meter long.
• It _is definitional of the property of moving at 1 m/s that anything that has
this property throughout some temporal interval that is 2 seconds Iona
moves along a curve in space that is 2 meters long. b
• It _is definitional of the property of moving at 4 mis that anything that has
this property throughout some temporal interval that is 2 seconds long
moves along a curve in space that is 8 meters long. 27
Using the dimension formula for speed (in the true class of systems of
scales) we can wrap these and all similar claims into one generalization:
(DEFINITIO~AL CONNECTION-Speed): Let ti,b, and c be any positive real
num,?,e:;~ with the fol~owing property: if you interpret "L" to denote b,
and 7 t?, denote c, m ,:he dimension formula for speed, and "compute
the resul_t, you ~et tl. (Smee the dimension formula for speed is "L/ T,"
the requirement is that tl = b/ c.) Let S be any system of scales in the true
class, and let r ~e the speed value that S assigns the number ti, I the length
value that S assigns '.h~ numbe: b, and d the duration value that S assigns
the numb:r c; then It 1s defininonal of r that anything that has r during a
temporal mterval that has d moves along a curve in space that has /.
~or every secondary quantity Q (in the true class), there is a generalization
like (DEFINITIONAL CONNECTION-SPEED), in which Q's dimension formula
27
, . In all .ofd1esc claims, occurrences of descriptions of properties arc to be read de re.
l hat. descnpnvc content, which contains reference to numbers, is not part of the
dcfinmons of these properties.
 Bradford Skow
192
in the true class appears, that advocates of DEFINlTIONAL CONNECTION will
accept. DEFINITIONAL CONNECTION is to be understood to be equivalent to
· · l.1zat.·ions.· 28
· o f t I1ese genera
t l1e conJunctwn
As a reminder, all of this is premised on the assumption that speed is a
secondary quantity. I made that assumption just so I could write down a
model for the generalizations that DEFINITIONAL CONNECTION is the con-
junction of. If speed is not a secondary quantity, then advocates of ~)EF!N­
ITIONAL CoNNECrlON will not accept (DEFINITIONAL CoNNECriON-SPEED),
and will not understand DEFINITIONAL CONNECTION to entail it.
Is DFHNITIONAL CoNNECnON correct? I don't have a knock-clown argu-
ment for it here. My goal is to offer it for your consideration, and try to ~ay a
few things to make it plausible. As l said earlier, just isolating it from
CONSTRUCTION is some defense of it; for a lot of the controversy around
CONSTRUCTION docs not apply to DEFINITIONAL CONNECTION.
The other main thing I have to say in defense of DEFINITIONAL CONNEC-
TION is this: if you like the at-at theory of motion, then you should like
DEFINITIONAL CoNNEcnoN.
28 In any class, including the true class, some quantities are dimensionless: their
dimension function is the constant [QJ = 1. This means that every system of scales in
the aiven class assigns the same numbers to values of Q--no matter what units those
scale~ choose for the primary quantities. Clearly if a quantity is dimensionless in the :rue
class, then DEFINITIONAL CONNE(TION says nothing interesting about what relanons
between that quantity and the primary quantities are dc~initional of that quantity. Is
the existence of dimensionless quantities any kind of problem for DEFINITIONAL
CONNECl"ION?
\'V'cll, if every secondary quantity were dimensionless (only secondary qu<rntities can be
dimensionless), DEFINITIONAL CONNEtTION would be empty. But it is false that every
secondary quantity is dimensionlcss--obviously, if there are primary quantities, which
there are, then some secondary quantities arc not dimensionless (for example the
"squares" of the primary quantities-if length is primary, area is its square-are not
dimensionless).
One idea I have heard is that it may turn out, some hope that it turns out, that_ the
quantities that appear in the laws of physics are all dimensionless. If this is right, and if the
only "important" quantities arc the ones that appear in the laws, then DEHNITl~>~AL
CONNECTION concerns only unimportant quantities. This might be thought to dummsh
its importance. But I find it hard to imagine how the laws could be stated in purely
dimensionless terms: surely acceleration will continue to appear m the laws, and surely
there is no way to make acceleration dimensionless? Nor do I sec why one should hope
that the laws can be stated using only dimensionl~ss quantities. The usual thou~ht is t~ut
then the laws will be independent of our scales of measurement. But even laws mvolv1ng
dimcnsionful quantities, like F =' ma, can be stated in a way that is independent of our
scales of measurement (see footnote 4).
In passing: though DEFINITIONAL CoNNECl"ION says nothin~ about what i.s definitional
of a dimensionless quantity, it docs not follow that dunens1onlcss quantmcs. lack real
definitions. It is just that their dimension formulas play no role 111 stanng those
definitions.
The Metapl~ysics of Quantities and Their Dimensions 193
Why this is so emerges from ttying to get clear on just what the at-at
theory of motion says. Let's start by looking at a classic statement of the
theory, namely Bertrand Russell's:
motion consists merely in the occupation [by some given material thing] of diffrrent
places at different times. (l 937: 473; emphasis added)
Other presentations of the theory use similar language; Frnnk Arntzenius,
for example, states the theory as the view that "there is nothing more to
motion than the occupation of different locations at different times" (2000:
189; emphasis added). What is the "merely" doing in the Russell passage,
and the "nothing more" in the Arntzenius? A natural interpretation has
these words serving to mark that the theoty is stronger than
WFAK Ar-Ar: Necessarily, for any body B, B is in motion during
temporal interval I iff B occupies different places at different times in I.
WEAK Ar-Ar is consistent with the idea that motion, location, and time are
"independent" things that are necessarily connected. But the language
Russell and Amtzenius use suggests that this idea is not supposed to be
consistent with the at-at theory. We can express a stronger version of the at-
at theory (stronger than WEAK Ar-Ar) using talk of real definitions.
Talk of real definitions gives us a way to make sense of the idea of things
that are independent but necessarily connected. For then we can construe
"independent" not as "modally independent" but as "have independent
defintitions": two things have independent definitions iff neither's real
definition places any constraints on what is true of the other. It is part of
the point of recognizing the legitimacy of talk of real definitions, or essences,
or natures, that it makes at least conceptual room for necessaty connections
between things that do not follow from the natures of those things. This
idea is the root of Kit Fine's (1994) well-known arguments against modal
theories of essence.
So one way to do justice to Russell's and Arntzenius's talk of what motion
"consists merely" in, and of what motion is "nothing more" than, is to
interpret them as asserting that some connection between motion, location,
and time, is definitional of motion. More specifically, a better (maybe the
best) statement of the at-at theory looks like this:
STRONG Ar-Ar: Tt is definitional of motion that, for any body B, B is in
motion during temporal interval I iff B occupies different places at
different times in I.
Now the at-at theory is not supposed to be just a theory of the non-
quantitative notion of motion; it is also a theory of the quantitative notions,
speed and velocity. So what docs the analogue of Strong At-At for speed look
194 Bnulford Skow
like? It must say, for every speed, that it is definitional of that speed that
something cover a certain distance in a certain time. But what exactly docs
the theo1y say? Which "certain distance" and "certain time" appear in the
definition of a given speed? The answer comes from having the theory assert
(DEFINITIONAL CONNECTION-SPEED), which I repeat here:
(DEFINITIONAL CoNNECrION-SPFED): Let a,b, and c be any positive real
numbers with the following property: if you interpret "L" to denote b,
and "T" to denote c, in the dimension formula for speed, and "compute
the result," you get a. (Since the dimension formula for speed is "L/ T,"
the requirement is that a = b/ c.) Let S be any system of scales in the true
class, and let r be the speed value that S assigns the number a, I the length
value that S assigns the number b, and d the duration value that S assigns
the number c; then it is definitional of r that anything that has r during a
temporal interval that has d moves along a curve in space that has /.
The thesis DEFINITIONAL CONNECTION-SPEED is part of the at-at theory of
motion, properly understood.
My argument has been that if you like the at-at theory, you should like
DEFINITIONAL CoNNECrTON. That's a bit vague so let me be more specific.
One way to like the at-at theory is to think that it is true; and if you think it is
true, you should definitely like DEFINITIONAL CONNECl'ION, for the quanti-
tative version of the at-at theory is just part of DEFINITIONAL CoNNECJION. Of
course, one can consistently accept (DEFINlTIONAL CONNECTION-SPEED) but
reject the other parts of DEF!N!T!ONAL CoNNECrION. But presumably if you
are drawn to the at-at theory of motion, you will also be drawn to theories of
other secondary quantities (energy, for example, if it is a secondary quantity)
that are similar in spirit. That similarity in spirit amounts to this: they also
assert that it is definitional of values of the secondary quantity in question
that they are related to certain values of the primary quantities.
There is, however, another way to like the at-at theory, one that does not
require you to believe it is true. This is important, since the at-at theory is
mildly controversial even in the context of pre-relativistic and pre-quantum
physics, and, 1 think, more controversial when relativity and quantum
mechanics arc taken into account. 29 You can like the at-at theory even
while thinking it false by thinking that it has the right form. What form is
that? You could think that the at-at theo1y is right to presume that the
correct theory of any secondary quantity will assert that it is definitional of
that secondary quantity that it is related in a certain way to the primary
29
For arguments against the theory see Arntzcnius 2000 and Lange 2005.
The Metaphysics of Quantities and Their Dimensions 195
quantitic~. You .could think that where the at-at theory goes wrong is
(merely) m holdmg that speed is a seconclaty quantity:00
6. DEFINITIONAL CONNECTION WITHOUT
RIGHT UNITS
The main work of this paper is clone, but there is a loose encl I want to tic
up. DEFINITIONAL CoNNECrION, as I stated it, presupposes RIGHT UNITS, a
p:esuppositton which mi_ght seem implausible. I said that DEFINITIONAL
Co~NECTIC:N could be formulated without this presupposition; in this
section I will show how to do this.
The fact that DEFINITIONAL CONNECTION can be formulated without
p:esupposing RICHT UNITS is also relevant to evaluating CONSTRUCrION.
CoM-rnuc:TION, as far as I can tell, must presuppose RIGHT UNITS. So if
Rrc'.HT UNITS is objectionable, its being objectionable provides a reason to
believe DEFINITIONAL CONNECTION rather than CONSTRUCTION.
Without RIGHT UNITS we cannot say that choosing the unit for ai·ea to be
the v~!ue f~r area had by a square with unit length sides is an "objectively
b~tter ch?ice t~an choosi~g the unit to be the value had by an equilateral
mangle with unit length sides. It might seem that DEFINITIONAL CoNNEC
·.rrON requires RICHT UNITS, because stating the generalizations that appear
m DEHN! rro~~L CONNr;;~noN rcqu'.res referring to the true class of systems
of scales, and true class 1s defined m terms of right units. But in fact this is
not so. Here are two ways of stating the definitional connection between
area and length without referring to the true class:
• Let m a~d n be any two positive reals with n = m 2 • Let S be a system ofscales
that designates length as primary and chooses the unit for area to be the area
value had by a square with unit length sides. And let a be the area value S
assign: n, and l the length value S assigns m. Then it is definitional of a that
anythmg that has a has the same area as a square with sides of length /.
• Let m and n b~ any two positive reals with n = If
m 2 . Let S be a system of
scales that designates length as primary and chooses the unit for area to be
the area value had by an equilateral triangle with unit length sides. And let
a be the area value S assigns n, and ! the length value S assigns m. Then it
30
The st~ndard altcrnat.ive to the at-at theo1y of motion says that motion is not
defined by reference to locatton and time. The analogous view of speed has it that speed is
;1::it d.cfincd by .refcre~cc to length and duration, and so that speed is a primary quantity.
I his ts compattblc with DEFlNJTIONAL CONNECTION.

196 Bradford S!ww
is definitiord of rl that anything that has rt has the same area as an
equilateral triangle with sides of length /.
We can avoid assuming RICHT UNITS by taking either (or both!) of th.ese to
be conjuncts in DEFINITIONAL CoNNECrlON, and taking the other conjuncts
to he stated in a similar way.
DEFINITIONAL CONNECTION still presupposes NoN-RELi\TIVrTY, for the
quantities that DFFINITIONAL CONNECTION cor:ccn~s. are meant .to be. all
and only the quantities that are secondary sunpltctter. And d1m~ns10~
2 opcrations? 31 Why can't you construct a new quantity by "taking the sine of a length
formulas arc still used to state DEFINITIONAL CoNNEGrION; n = rn JUSt is
(an instance of) the dimension formula for length (in the given system), and
n= VJ m 2 is the dimension formula multiplied by a constant.
It i; harder to see how CoNSTRUCI10N could do without RIGHT UNITS. Is the
product of the property of being l meter with itself identical to the a:ea prop~r?
had by things that are the same area as a square with meter-long sides, '.)r is it
identical ro the area property had by things that are the same area as an eqmlateraI
triangle with meter-long sides? It cannot be identical to both; picking one goes
hand-in-hand with privileging one way of choosing a unit for area over another. If
RIC HT UNITS makes you a lot more nervous than NON-RELATIVITY, you should be
more drawn to DEFfNrfKlNAI. CoNNH.TION than to CONSTRUGflON.
7. CONCLUSION
My goals in this paper have been modest. I have not set ?ut to .prove t~at
DEFINITIONAL CoNNECr!ON is true from premises that are m any mterestmg
sense independent of it. Instead, my aims with respect to this thesis were
these. First, to get DEFINITIONAL CoNNECTlON on the table, and on t~e
agenda. Something close to DEFINITIONAL CoNNECf!ON has surfa~ed m
discussions of the metaphysics of dimensions, but it has not until now
been clearly distinguished from CoNSTRUGrION. And clearly distinguish.in.g
it is a prerequisite to reaching a dear-headed conclusion about whether it is
true. A second aim was to say something in defense of DEPINITIONAL
CONNECTION. l myself found that appreciating the relation between DEl'IN-
ITIONAL CONNEGl'lON and the at-at theory of motion increased my confi-
dence in it, and I suspect it may do the same for others.
I will not pretend that this amounts to a comprehcnsi:e t:e~tment of
DEFINITIONAL CONNECrION-here, as elsewhere in metaphysics, it ts hard to
find good arguments. I hope, instead, that it is a good.start on a worthy project.
In the appendix that follows, I pursue a quest10n that CoNSTRUGI ION
raises that did not fit into the paper's main line of argument.
Brrtdfard Skow
Masstll/;usetts Institute of Technology
The Metapl~ysin of Qurmtities @d Their Dimensions 197
APPENDIX: A B!T MORE ON CONSTRUCTION
CONSTRUCTION, on one reading at least, entails that some quantity values arc got by
"multiplying" other quantity values together, and some arc got by "dividing" one
quantity value by another, and some arc got by "raising a quantity value to a power."
For cx<unplc, if speed is a secondary quantity, the view is that the speed value 4 meters
per second is the result of dividing being 4 meten long by being 1 mmzd in clumtion.
rkre is a pointed question for fans ofCoN5TRUCTION: why arc these the only three
v<1lue," or "subtracting the natural logarithm of a duration value from a mass value"?
It might look at first like this question should embarrass proponents ofCoNSTRUl~
TION. Surely they'll either have to say that it is just a brute fact that these arc the only
operations; or they'll have to say that there is a quanrity got by taking the sine oflength,
it is just nowhere instantiated in the actual world (or of no interest to physicists).
But proponents of CONSTRUCTION do not need ro say dther of these things. There
is a proof in most treatments of dimensional analysis that can be used to explain why
the only operations arc multiplication, division, and exponentiation (for Bridgman's
version see (Bridgman 1922: 21-2)). Herc is the theorem proved:
Let X 1 and X2 be two systems of scales of measurement in the same class. Suppose
also that X 1 and X2 agree on the ratios of numbers <Lssigned to quantity values.
That is, if the quotient of the numbers X1 assigns to values v and u of some given
quantity is r, then the quotient of the numbers X2 assigns to those values is also r.
Then for any quantity Q measured by Xi and Xz, the dimension formula for Q
is a "power law monomial," that is, it has the form "KP;n P!f ...," where symbols for
the primary quantities go in for "P1" and so on, and symbols denoting m1l
numbers go in for "K," "m," and so on.
How is this relevant? Suppose some "quantity" were, say, the sine oflcngrh. Then its
dimension formula (in the true class) would not be a power law monomial. So there
would be systems X 1 and X2 in the true class which did not agree on the ratios of
numbers assigned to quantity values. Since these systems arc faithful to the intrinsic
structure of this "quantity" (we have always restricted ourselves to faithful scales),
there must not be scale-independent facts about the ratios of values of this "quan-
tity." But that is definitional of quantities! This contradiction shows that no quantity
can be the sine oflength, and more generally, that quantities can only be constructed
from other quantities using multiplication, division, and exponentiation. (I leave it
open whether those who like CoNSTRUCrION will want to say that there is something
(which is not a quantity) that is the sine of lcngth.) 32
'll Really we only need two; division can be defined in terms of multiplication and
exponentiation.
Jl Thanks to Martin Glazier, an audience at the University of Rochester, and an
audience at NYU. Thanks especially to Karen Bennett, without whose large investment of
editorial cffons this paper would not be what it is.
198 Bradford Skow

REFERENCES

Arntzcnius, Frank (2000). "Arc There Rc<llly Instantaneous Velocities?" The /v[onist
83(2): 187-208.
Barenblatt, G. I. (1987). Dimensional Amdysis. Trans. Paul Makinen. New York:
Gordon and Breach Science Publishers.
Barenblatt, G. I. (1996). Scrtling, Se/fSimilarity, mu/ Intermediate Asymptotics.
Cambridge: Cambridge University Press.
Bennett, Karen (2004). "Spatio-Temporal Coincidence and the Grounding
MODALITY AND EXISTENCE
Problem." Philosophiml Studies 188(3): 339-71.
Bridgrn<m, Percy (1922). Dimensional Analysis. New Haven: Yale University Press.
Duncan, W. J. (1953). Physical Similarity and Dimensional Analysis. London:
Edward Arnold & Co.
Eddon, Maya (2013). "Quantitative Properties." Philosophy Compass 8(7): 633-45.
Field, Hartry (1980). Science Without Numben. Princeton: Princeton University Press.
Fine, Kit (1994). "Essence and Modality." Philosophiml Perspectives 8: 1-16.
lnwagen, Peter van (2014). /'.1etaphysics. 4th edn. Boulder: Westview Press.
Lange, Marc (2005). "How Can Instantaneous Velocity Fulfill Its Causal Role?" The
Philosophiml Review 114(4): 433-68.
Lange, Marc (2009). "Dimensional Explanations." Notis 43(4): 742-75.
Langhaar, Henry (1951). Dimensional Analysis and Il1eory of Models. New York:
John Wiley & Sons.
Lewis, David (1983). "New Work for a Theory of Universals." Austrahuian joumrtl of
Philosophy 61(4): 343-77.
Luce, R. Duncan (1971 ). "Similar Systems and Dimensionally Invariant Laws."
Philosophy of Science 38(2): 157-69.
Mundy, Brent (1987). "The Metaphysics of Quantity." Philosophical Studies 51 (1):
29-54.
Newton, Isaac (1999). The Principia: Mathematical Principles of Natural Philosophy.
Trans. I. Bernard Cohen and Anne Whitman. Berkeley: University of California
Press.
Palacios, Julio (1964). Dimmsiona!Analysis. Trans. P. Lee and L. Roth. New York:
St. Martin's Press.
Plank, Max (1932). General Mechanics. London: Macmillan.
Russell, Bertrand (1937). Principles of Mathematics. 2nd edn. London: G. Allen &
Unwin.
Skow, Bradford (2012). '"One Second Per Second'." Philosophy and Phenomeno-
logiml Research 85(2): 377-89.
Skow, Bradford (2015). Objective Becoming. Oxford: Oxford University Press.
 9
Vague Existence
Alessandro Torza

Debates about what there is are common and often fascinating. If there were a
black hole close enough to the solar system, we would have reasons to be worried.
Had the Higgs Boson turned out not to exist, that would have meant bad news
for the Standard Model of particle physics. Euclid's proof of the existence of
infinitely many prime numbers was no small feat-and so on and so forth.
Ontology is the study of what there is, unrestrictedly. Ontologists can
argue about the existence of things which are of concern to laypersons
(macroscopical objects, values, fictional characters), scientists (fundamental
particles, fields) and mathematicians (numbers, sets). In other cases, onto-
logical disagreement will turn instead on more exotic items such as sub-
stances, possible worlds, or spatiotemporally disconnected wholes.
Can existence, in the unrestricted sense of ontology, be vague? One
popular construal of vagueness is defined by the method of precisifications.
A precisification is a way of making precise all the terms of a language.
A sentence is vague when some precisification makes it true and some other
makes it false. According to an influential argument due to Sider [25] [26]
[28], in ontology there is no such thing as vague existence, as long as
vagueness is construed precisificationally.
I aim to show that existential vagueness is a coherent notion, albeit in a weaker
form which I will refer to as super-vague existence. Section 1 exposes a gap in the
alleged reductio of vague existence. I will wrap up the section by considering a
potential objection. Section 2 develops and defends a novel framework, dubbed
negative supervaluationary sernantics, which models super-vague existence and its
logic. Two further objections will be anticipated at the end of Section 2.

1. AGAINST "AGAINST VAGUE EXISTENCE"

I. I Sider against vague existence

The question whether existence can be vague is relevant to both ontology
and metaontology. Let us start with the latter.
202 A!essttndro Tor:Zf!
One issue about ontological disputes is that it is often hard to identify the
source of disagreement. This point is the main target of the recent debate in
metaontology, which sees two opposing camps. Metaontological realists
regard ontological disputes as genuine and substantive. When philosophers
argue in the ontology room, it is claimed, their disagreement turns on
different pictures of reality. This position, which took shape in Quine's
work on ontological commitment, has been developed by Peter van Inwa-
gen, Kit Fine, and Ted Sider among others. The antircalist camp contends
instead that ontological disputes arc in some sense semantic or verbal. If the
latter thesis is correct, the kind of disagreement taking place in the ontology
room amounts to a sophisticated version of a dispute on the nature of
tomatoes-whether they are fruits or vegetables. The core of this deflationist
view, which originated with Carnap, was rejected by Quine and then
revived by a group of philosophers including Hilary Putnam, Eli Hirsch,
Amie Thomasson, and, more recently, David Chalmers. 1
Some popular forms of deflationism embrace quttntifier v11riance, the
Putnam-Hirsch view "that quantifiers can mean different things, that there
are multiple candidate meanings for quantifiers" (Sider [29, p. 391]). But if
the quantifier used in the ontology room is semantically vague, there will be
as many existence meanings as there are admissible precisifications of the
quantifier. As a consequence, deflationisrn would likely be true of at least
some ontological disputes. For instance, there may be no determinate fact of
the matter as to whether, say, tables exist, provided that the meaning of
"exist" is compatible with multiple and equally good ways of carving the
domain of the world at the relevant joints, so that tables will exist on some
but not all such existence meanings. 2
Whether existence could be vague has first-order ontological implica-
tions, as well. Restricted composition is the metaphysical view that not all
collections of things have a mercological fusion. According to a particular
brand of restricted composition advocated by van Inwagen [31], organisms
represent the only case in which a collection of objects composes a whole.
On that view, tables and chairs do not exist, whereas pluralities of simples
arranged table-wise and chair-wise do. 3 Van Inwagen 's mereologicttl org11ni-
cism is certainly not the only flavor of restricted composition. On a different
1
A deflationary framework is also developed in Sider (27] on behalf of nco-Fregeans
such as Bob Hale and Crispin Wright.
2 5
Truth be told, it is in principle po"sible that there be vague existence without
ontological deAationism, namely if for every sorta! !', 3x!'x is true (false) on all candidate
meanings for 3. For in this case any existence question would have an objective and
determinate answer, despite the existential vagueness. Thanks to an anonymous referee
for the pointer.
.l On plural quantification, see Boolos [5].
Vague E-dstence 203
way of constraining fusion, a collection of simples composes a whole just in
case the members of the collection are topologically connected. Thus,
organisms, tables, and chairs will exist, whereas table-giraffes will not. This
view is primtt fade more consonant with common sense intuitions about the
conditions under which something can be said to exist. At the encl of the
spectrum we have rnereologiml nihilism, which only accepts the existence of
simples.
It has been noted that restricted composition leads to vague cxistence. 1
Assume that composition is restricted a !d van Inwagcn. On December 7,
43 RC, Marcus Tullius Cicero was on his deathbed. There is a time to at
which he was definitely still alive and a time t 1 at which he was definitely
dead. However, there was arguably no cut-off point in the series starting
with "Cicero is alive" at to and ending with "Cicero is dead" at t 1 . For some
time tk between to and t1 it is vague whether Cicero was dead or alive,
therefore it is vague whether the simples arranged Cicero-wise lying on the
bed would constitute a whole at time tk. Since there were times at which it
was vague whether Cicero still existed, dying Cicero constitutes a case of
vague existence. Or consider the case of viruses. Because it is vague whether
a virus is an organism, it is vague whether it constitutes a whole or it is just a
collection of simples arranged virus-wise.
Existential vagueness also arises if we think, as most people arguably do,
that objects must be connected. For, although the mathematical concept of
connectedness is a sharp one, things get tricky when we apply it to physical
objects. Suppose we are soldering two pieces of metal m 1 and m 2 . What is
the minimum threshold of subatomic interaction that must occur between
m1 and m2 so that the two pieces will count as connected? It docs not look as
though a unique non-arbitrary answer to that question could be provided. If
so, there must be a time at which it is vague whether there exists a whole
composed of m 1 and m 2 • 5
Lewis [18, p. 213] objected to restricted composition on the assumption
that vague existence is incoherent. This strategy was forther developed by
Sider [25, pp. 120-32], who appealed to unrestricted composition in his
4
Or, to be precise, that all interesting ways of constraining the relation of composition
lead to vague existence. Some philosophers, however, beg to differ: Carmichael [8] and
Donnelly (11] have proposed precisificational accounts of composition that do not entail
existential vagueness.
Topology in fact distinguishes between multiple notions of connectedness. That fact
seems to suggest that, if our concept of object is subject to a connectedness constraint,
existential vagueness arises on multiple dimensions: relative to a notion of connectedness
and relative to the application of such a notion to the physical realm. (But it might he
argued, as well, that the first dimension of vagueness should instead be construed as a case
of ambiguiry, rather than vagueness.)
204 Alessandro Tor:z:Lt
proof of four-dimcnsionalism. A full-blown attempt at discharging the
assumption that existence cannot be vague appears only in Sider [26,
pp. 138-43]. The main claim is that
(V) if vagueness is given a precisificational account and existence is
expressed by the unrestricted existential quantifier, then existence
can not be vague.
The first conjunct of the antecedent leaves out some theories of vagueness,
most notably all degree-theoretic constru;1ls (viz. fuzzy logic) but includes
the supervaluationism of Fine [131, as well as the epistemicism of Williamson
[35] and the semantic nihilism of Braun and Sider [6]. Since there arguably
are independent reasons to reject degree-theoretic semantics for vague lan-
guage,6 Sider's proof, if sound, would suffice to cover what happens to be the
de facto standard representation of vague talk.
But what does it mean to precisify a language £? Precisifications are ways
of making precise all of the terms in £. 7 Typically, a precisification is
interpreted extensionally, in the sense that it (i) specifics a domain of
quantification and (ii) assigns extensions over the domain to the non-logical
constants in £.
According to the precisificational framework, a statement of [, is true
(false) ifit is true (false) in every precisification of£; it is vague ifit is neither
true nor false. The precisifications of [, are assumed to be admissib!e-
roughly, they must be compatible with the linguistic practice of the com-
petent speakers of£. For instance, a precisification of English in which the
extension of "hirsute" contains people with zero hairs will be inadmissible.
Likewise for a precisification in which the extension of "bald" contains
people with ten hairs and the extension of "hirsute" contains people with
nine hairs. The task of determining the admissible precisifications of a
language is far from obvious, but for present purpose it will suffice to
employ a primitive notion of admissibility.
The second conjunct of the antecedent in (V) serves to rule out restricted
quantification, which is obviously open to vagueness. For instance, the
statement "There are over 21 million people" will be vague if uttered by
someone referring to the population of the Greater Mexico City, due to the
unsharp nature of its urban sprawl. A quantifier is unrestricted iff it ranges
over absolutely everything that exists. As a consequence, only the unrestricted
6
!t has been shown that degree-theoretic interpretations of vagueness violate classical
logic, misrepresent penumbra! connections (i.e. logical connections among indefinite
sentences) and fail to account for higher-order vagueness. Sec Williamson [38], Keefe [ 161.
7
Vani [32] considers a number of construals of a precisification. 1n rhe present
context, however, such distinctions will not matter.
Vague Existence 205
quantifier, i.e. the one employed in the ontology room, is relevant to Skier's
argument.
Th~ strategy adopted by_ Sider in proving (V) consists in showing that, if
an object-language quantiher 3 is vague and absolute, a contradiction can
~Je derived in its metalanguage. By an application of reductio ad absimlum, it
JS concluded .that object-language quantification cannot be vague. 8 The
argument agamst vague existence can be reconstructed as follows. Assume
by way of reductio that
1. r 3xcp 1 is vague
By the underlying semantics for vagueness, (1) is equivalent to
2. In some precisification r 3x<f> 1 is true and in some precisification
r 3x<f> 1 is false
hence,
3. In some precisification r 3x<f>1 is true.
Truth .in a precisification is truth in a precise language. So, by the standard
extensional truth-conditions for quantified statements in a precise language
(3) is equivalent to '
4. There is something such that, in some precisification, it belongs to
the domain of 3 and satisfies q>
and, a fortiori,
5. There is something such that, in some precisification, it satisfies q,.
Now, </> is intended to be precise. For, otherwise, the main premise (1)
:-voulcl not .amount to the assumption that the quantifier 3 is vague. But the
mterpretauon of a precise expression coincides across all precisifications.
Therefore, (5) entails
8
In Skier's own words:
Suppos~ '3' has two preci~ificarions, '3 1 ' and '3 2 ', in virtue of which '3x<f/ is indcter-
rmnat~ ''.' truth \;alu~, desp_tte the fact that f is not vague. ' 3xf', suppose, comes out true
when 3 means 31 , and false when' 3' means' 3/. How do' 3 1'and' 3/ generate these
truth values? A natural thought is:
Domains '3,' an~l '3/ are associated with difforeru domains; some object in the
domam of one satisfies f, whereas no object in the domain of the other satisfies 'P
Bu~ the natural .thouglu is mistaken. If Domains is asscrtible, it must be determinately
true. But .Domarns,,entatls that some objeer satisfies 'P (if" ... some object in the domain
?f one sausfies f ... , then some object satisfies f). And so' 3xq/ is determinately true not
indcterrmnate as was supposed (Sider [28, pp. 557--8]). '
 Alessandro Torztt Vague Edstence 207
206
6. There is something such that, in every precisification, it satisfies <f>. two simples F and C, the vagueness of the sentence expressing the existence
of a third object: 10 '
Recall that 3 is unrestricted--it ranges over all there is. So, from (6) it can
(E') 3z(x f y !\y f z !\x f z)
be inferred that
7. There is something such that, in every precisification, it belongs to where x and y refer to F and G, respectively. This solution faces two
challenges. If the quantifiers arc truly unrestricted, they will have to range
the domain of 3 and satisfies <f>.
over abstract as well as concrete objects. Since it is necessarily the case that
By the truth-conditions for quantified statements in precise languages, (7) there definitely e~ist infir:itely many abstracta, (E') would co~1c out trivially
entails true at the world Ill quest10n. In order to obtain the desired restriction in the
object language, we must introduce a concreteness predicate "C" and
8. In every precisification r 3x<p' is true
replace (E') with:
which, in a precisificational framework, is tantamount to
(E") 3z( C(x) !\ C(y) !\ C(z) !\x f y /\y f z !\x f z).
9. r 3x</>' is true
B~; then, as I~o~'.~1an [ 17, p.893 J pointed out, the source of vagueness in
Since a statement is vague just in case it is neither true nor false, (l) and (9) (Ee ) could be C rather than the quantifiers. If so, Sider would have failed
are jointly inconsistent. By reductio ad absurdum, we can discharge the main to provide a reductio of vague existence.
premise (1) and infer its negation, namely I find this strategy to block the argument unconvincing. Even if the
concreteness predicate were vague, its vagueness would be irrelevant to
10. r 3x<f>' is not vague
the present argument. For if the vagueness in (E") were due only to the
9
which concludes the proof. vag.ueness of "C," there would have to be an admissible precisification in
which the mereological sum of F and G is concrete and one in which it is
not. But we can make the reasonable assumption that the sum (if any) of a
1.2 Super-vague existence collection of definitely concrete objects is definitely concrete. Since it was
assumed that F and G are concrete simples, Fand G will then have to be
Before we take a closer look at the merits of the argument, some preliminary
definitely concrete and therefore their sum (if any) must also be definitely
remarks are due. As was pointed out, the reductio requires that, whereas 3x<f>
concrete. Hence, relative to the world at which we evaluate (E"), there is
is vague, </> should be precise. Accordingly, there must exist at least one
n,o admi~sible precisification of the language according to which the sum of
instance of</> meeting such conditions. What language could be such that all
F and G-the third object in question-fails to be concrete.
of its terms are sharp except at most the quantifiers? Sieler [26, pp. 139-40 J
A related challenge concerns the semantic status of identity, for there is no
originally formulated the argument in the vocabulary of mereology, so that
point in replacing (E) with (E'), unless identity is precise. This extra
the main premise would be assumption could be discharged by pi&,crybacking on the well-known reduc-
(E) 3x(x is composed of the F and the C). tio of vague identity offered in Evans [12]. Nevertheless, the validity of
Eva.ns' argument has been disputed. 11 Be that as it may, for the time being
But if"F," "G," or "compose" are vague, (E) could be vague without there
I will ~oncede tha~ all logical vocabulary is precise, with the sole possible
being any vagueness at the quantificational level. The use of those terms is
exception of quantifiers, and therefore that any reductio of the va<>ueness of
inessential, however, since (E) can be replaced with a sentence containing
(E') amounts indeed to a reductio of quantifier vagueness. b
only logical vocabulary. Thus, instead of attempting to disprove the vague-
A sep~rate, and more crucial issue concerns the reductio step at the very
ness of (E), we could attempt to disprove, at a world containing exactly the
end of Siders argument. Indeed, reductio ad absurdum is a valid form of

10
Sider [29, p. 390].
9 See Section 2.4 for an attempt at rephrasing Sicler's argument without appealing to 11
Fora discussion, see Williams [33], Barnes [3], Heck (15], Akiba [lJ [2]. Cf. Lewis (19].
reductio.
208 Alessandro Torw
inference for bivalent languages: an inconsistent condition cannot be true,
hence it must be false. When the language is not bivalent, on the other hand,
reductio ad ahsurdum fails, since being non-true is a weaker condition than
being false. In particular, reductio is invalid in the case of vague languages: if
a condition is inconsistent, it must be either false or indeterminate. It
follows that Sider's final reductio step is valid insofar as the language in
which the argument is formulated, viz. the metalanguage of ~xef;.' is perfectly
precise. Since the main premise ( l) is equivalent to its precisihcanonal truth-
condition (2), one is vague just in case the other is. Now, (2) docs two
things: it quantifies over precisifications, and it says of the obje~t-.L~ngua?e
sentence 3xq> that it is true in some but not all of them. By definmon, for
every precisification s, truth-in-s is determinate; therefore, if there is any
vagueness at all in (2), it must come down to what precisifications there are.
The moral is that, if quantification over precisifications is precise, so that
there is exactly one candidate set of precisifications for the object-language,
. goes throug h- L'f not, not. 12
Si. der 's reductw
Let me flesh out this point a bit further. It is not at all uncommon for a
vague term (and, a fortiori, a vague language) to be asso~iat~d ";'}th m~ltip~e
sets of precisifications. Take for instance the color ad3ecuve blue. 1 hts
could refer to the spectrum of visible light with a wavelength between
450 and 490 nanometers-each single value specifying a precisification.
But there is nothing about that very interval that makes it the right one.
If the set of precisifications of "blue" can be 450-490nm, then the intervals
455-495nm or 445-485nm will work just as fine, as well as many other
intervals in that neighborhood. Which is to say, none of them is a better
candidate set of precisifications for the meaning of "blue." We must con-
clude that the term "blue" is second-order vague, insofar as the extent of its
vagueness is itself vague. Could it be the case that second-order vaguen.ess
affects the language of 3xef;? From what we have seen so far, nothmg
prevents such a possibility. In a scenario of this sort, quantifyin.g over
precisifications in Sider's argument would be vague and the reductw step,
therefore, unwarranted.
Let's take stock. As it should now be clear, Sider has conclusively shown
us that
1. r 3xef;' is vague
cannot be true, provided that ef; is precise and 3 is unrestricted. If the
metalanguage of 3 is not perfectly precise, however, there is no guarantee
12 The i<lea that some assumption in Sicler's argument could be indeterminate has also
been explored in Barnes [4].
Vcigue Evistence 209
that ( 1) is false and, therefrire, that vague existence has been disproved. In
particular, even though the quantifier :3 cannot be definitely vague, there is
a prima fade possibility that it might be second-order vague. In this case,
there would be a set of precisifications that makes 3xef; vague, and another
set that makes 3x<fi precise. At this point Sider might reply by offering a
second recluctio, this time around of the vagueness-not the truth-of (I).
The conjunction of this new argument with the original one would then
amount to the desired disproof of vague existence.
The second reductio goes as follows. (Note that the object language in the
present argument coincides with the metalanguage in the previous one.
Accordingly, what "precisification" means in this argument is not what it
used to mean in the previous one. I will use the term "precisification 2 " to
refer to precisifications of the meta-language of 3xef;, i.e. second-order
precisifications of the language of 3xef;.) Assume
1* rr 3xef;' is vague-, is vague
Hence,
2* In some precisification 2 rr 3xef;;' is vague' is true, and in some
precisification2 rr 3x(f>' is vague-, is false
Now, by virtue of Sider's first argument, the first conjunct of (2*) entails
3* In some precisification 2 r l_-, is true
On the other hand, precisifications2 being classical, it is the case that
4* Tn all precisifications2 r l_-, is false
hence a contradiction. If we could now apply reductio ad ahsurdum, we
would be able to infer
5* rr 3xef;' is vague-, is not vague
thus disproving the second-order vagueness of 3xef;. But we are back to
square one. For, the reductio step is licensed as long as the language in
which the argument is formulated (i.e. the meta-metalanguage of 3xef;) is
perfectly precise. In particular, quantification over precisifications 2 in (2*)
will be vague if the quantifier 3 is third-order vague. In such a scenario,
there will be a set of sets of precisifications of the language of ::lxef;
according to which rr ::lxef;' is vague-, is vague, and another set of sets of
precisifications of the language of 3xef; according to which rr 3x(f>' is
vague-, is sharp.
Sider's argumentative strategy could be iterated at any order-and so the
relevant rejoinder. Consequently, as long as quantifying over precisifications,, is
 Alessandro Torzrl
210
vague at every order n, neither side will get the upper hand, and the
possibility of vague existence will remain neither proved nor disproved.
The dialectics prompts the following moral. A precisificational semantics is
a framework in which truth, falsity, and vagueness for a language L arc
formulated, within some relevant metalanguage L , via the notion of
truth-in-a-precisification-of L. Since precisifications arc ways of making a
language precise, it cannot be vague whether an L-statcment is true or false
in a given precisification. This account is perfectly compatible, however,
with there being a vague set of precisifications of L. When that happens,
the prccisificational truth-conditions for ,L-statemcnts can be vague, inso-
far as they arc formulated by means of L -statements of the form "there is
a prccisification .... " Crucially, Sider's reductio of vague existence
requires that the truth conditions for 3xcp be formulated in a perfectly
precise metalanguage. (Or at least that a precise metalanguage could be
found somewhere up in the hierarchy.) The existence of such a language is
neither guaranteed nor required by a precisificational account of vague-
ness. Therefore, pace Sider, if what precisifications there are is vague at all
orders, vague existence will remain an open possibility.
The above discussion is complicated by the use of a hierarchy of meta-
languages, each expressing whether sentences of the relevant object language
are true, false, or indeterminate. Matters can be simplified as follows. Given
sentence q> in an object language L, in lieu of the metalanguage expression
r cp' is true
we will use
where !:-. is an object-language sentential operator with the intended mean-
ing "it is definitely the case that." Also, let's define the expression Icp ("it is
vague whether cp") as •1:-.cp /\ •1:-.•cp. By iterating these newly introduced
sentential operators, it is now possible to reduce the hierarchy of meta-
language truth/falsity/vagueness predicates to the object language L. For
instance, the above condition
l* rr 3xcp' is vague' is vague
translates into L as
II 3xcp.
[n general, we can express that cp is n-th order vague simply by iterating the
I operator n times. Now, let!" be short for the concatenation U· What
n
the generalized reductio has shown is that, for all n, r [" 3xcp' is not true.
Vtigue E·dstence 21 l
Which is to say, existence cannot be said to be definitely vague, no matter
what the vagueness order is (provided that cp is precise). I will henceforth
refer to this condition as Sider-determinacy.
Let us stipulate that an existence statement r 3xcp' is anti-Sider-detennintfte
just in case, for all n, r !'1 3xq>' is not false. I have argued in the foregoing
discussion that precisificational truth conditions need not be given in a
precise metalanguage and, in particular, that the set of precisifications can
in principle be vague at each order in an infinite hierarchy of metalanguages.
Insofar as, for all n, r !" 3xcp' can be vague, it follows that Sider-determinacy
is compatible with anti-Sider-determinacy. When an existence statement
is both Sider-determinate and anti-Sider-determinate, I will say that it
expresses an instance of super-vrigue existence. For example, it might be
that it is vague whether there exists the sum of F and G, and it is vague
whether it is vague whether there exists the sum of F and G, and it is
vague whether it is vague whether there exists the sum of F and G, etc.
In the next section, I will show how super-vague existence can be
accommodated within a precisificational model theory. But before that,
I wish to address a potential objection.
1.3 Metalanguage objection
In the above reconstruction of Skier's argument against vague existence,
I pointed out that the final reductio step requires the metalanguage of 3xcp
(viz. the language in which the argument is formulated) to be perfectly
precise. The same problem cropped up, mutatis mutandis, in the higher-
order generalization of the argument.
Sider touches on the issue of a precise metalanguage as he describes what
it means to give an account of vague statements such as "5 is bald":
When confronted with vagueness, I retreated to a relatively precise background
language to describe the relevant facts. In this background !tzng1Mge I quantified over
the various sets containing persons with different numbers of hairs, and said that the
referent of S was in some but not all of these sets. [... ]Moreover, in principle one could
describe the sets with perfect precision by retreating to a background language employing
only the vocabulary of fondamental physics. (Sider [26, p. 139], emphasis added)
It is tempting to interpret the passage as entailing that the precisificational
truth-conditions of 3x<f> can be assumed to be perfectly sharp and, therefore,
that Sider's reductio of vague existence is valid. Consequently, existence
cannot be super-vague, since it is definitely not vague.
The above passage, however, does not license this conclusion. Recall from
the discussion in Section 1.2 that one thing is to say that (i) "true-in-s" is

212 Alessandro Torvz
sharp, for s a given precisification; and another thing is to say th~t. (ii)
"true in some precisification" is sharp. Indeed, (i) is true by defin1t1on,
insofar as each precisification is a classical interpretation described in the
metalanguage with perfect precision. But (i) can obtain in the absence of
(ii), namdv, if it is indeterminate what precisificuions there <ire. In the
particular ~ase of" S is bald," every precisification for "bald" will be a shar.p
set of persons. Nevertheless, it could still be indeterminate what set of sets is
the extension of the metalanguage term "precisification." And since t~e truth
conditions of a vague statement require quantification over pre~is~fi~atio~s,
the language of Sider's argument can be vague, hence the reductto is 111val1d.
As I argued in Section l .2, the assumption that the language be perfectly
precise is not part and parcel of a precisificational account of vagu~1:ess.
Moreover, the assumption that the precisificational truth-condmons of a
vague statement be formulated in a perfectly precise meta-language can lead
to paradox. The statement
(*) rTed is bald' is vague
is tantamount to
(**) in some precisification rTed is bald' is true, and in some precisi-
fication rTecl is bald' is false
If(**) is perfectly precise, as per hypothesis, it must be either c~efinitely true
or definitely false. It follows that rT ed is bald' is either defi mtely vague or
definitely not vague. Therefore, although Teel-or anybody else, for t~at
matter-could be a borderline case of baldness, he could not be a borderlme
borderline case of baldness. Skier's hypothesis that the metalanguage is
perfectly precise rules out the possibility of higher-o~d~r. vague1~ess. How-
ever, there are independent reasons to admit the poss1b1hty of h1ghe:·-order
vagueness. If "bald," for example, is only first-order vague, .there will b~ a
clear-cut border between the definitely tall and the not definitely tall, which
is no less absurd than there being a clear-cut border between the tall and the
not tall. Moreover, the absence of higher-order vagueness can be exploited
to generate higher-order sorites paradoxes. I conclude that Sider's assump-
tion is unwarranted and unwelcome.
2. SUPER-VAGUE EXISTENCE AND ITS LOGIC
What is the logic of a language whose quantifiers are super-vague? This
Section attempts to provide a model-theoretic answer to that problem.
Vtigue E>:istence 213
2.1 Finean supervaluationism
The most popular precisificational model theory is arguably the supervalua-
tionary semantics put forward in Fine [13J. A specification space is a model
defined by (i) a set of specifict1tion points; (ii) a binary admissibility relation
defined on the set of specification points; and (iii) a selected specification
point called "base point."
Specification points are identified with partial models, each representing a
way of making the language more precise. A partial model is defined by a
domain of individuals and an interpretation function assigning to each
predicate a positive and a negative extension such that the two extensions
are disjoint but do not necessarily exhaust the domain. (Classical models are
the degenerate case of partial models in which the positive and negative
extension of any predicate are jointly exhaustive.) Since quantification is
assumed here to be absolute, a quantifier must range over the whole domain
of a specification point.
The admissibility relation is a reflexive, antisymmetric, and transitive
ordering R such that (1) if ttR.11 and p is true (false) at u, then p is true
(false) at v; (2) every specification point bears the ancestral of the admissi-
bility relation to a complete specification point, which is a classical model
(intuitively, a precisification of the language).
A sentence p is said to be trne in a specification space if it is true at the
base point. It is said to be true at a specification point if it is true at all
accessible specification points. "6p" is said to be true at a specification point
if pis true at the base point. It follows from the definitions that a sentence is
true at the base point if! it is true at all complete specifimtion points.
A sentence ¢ is said to be a supervaluationary consequence of a set r of
sentences if every specification space which makes f true makes cp true.
Sentence ¢ is said to be valid if it is a supervaluationary consequence of the
empty set.
Fine's framework immediately yields some desired results concerning
vagueness phenomena. Bivalence fails, since it is not the case that any
given p is either true or false in a specification space. This is equivalent to
the object-language fact that 6p V 6-p is invalid, as it should be. On the
other hand, p V 'Pis valid (and so is any classical tautology) in virtue of the
classicality of complete specification points.
The feature of Fine's superv<Juationary semantics relevant to the present
discussion is that there is no vague existence at any order (provided
that there definitely arc finitely many objects). To see that, it suffices to
show that (i) existence is definite, i.e. 3x(x = y) ·-> 6 3x(x = y ), and that
(ii) definite statements cannot be indefinitely definite, i.e. 6¢ ~· 66¢.
 Alessandro Torza
214
It follows that Fine's framework, in its basic form, is not suitable for
modeling super-vague existence. u
2.2 Variable domain frames
I have just argued that the hnean supervaluationism yresented here is
unable to account for higher-order vagueness at the object-language level
(cf. Appendix A). Since the failure of higher-~rder va?~1enes_s fam~usly leat~s
to higher-order sorites paradoxes, an alternative prec1s1ficauonal framework
has been proposed by Timothy Williamson in which _statements can be
higher-order vague. ln the remainder of Section 2, [ a11:1 to show that a
suitable generalization of Williamson's semantics_ admits ~f. models for super-
vague existence, i.e. in which both of the followmg condmons are met:
Sider-determinacy: for all n, it is not definitely the case that existence is
n-th order vague.
rmti-Sider-determinacy: for all n, it is not definitely the case that exist-
ence is not n-th order vague
The resulting model-theory will give us an idea of the logic of super-vague
existence.
The semantics developed in Williamson [35] [37] is designed_ fo~ a
sentential language with a definiteness operator. The simplest ~eneral1zat10n
to the first-order case can be defined as follows. A frame :F is a structure
(5, U,R) where Sis a set of points (the specifications), ~a set of in~iv'.d~~ls
(the universe of discourse) and R a relation over S (the adm1ss1b1hty
relation). Let Lt. be a first-order language with a definiteness. operator. 6..
A model /v{ for Lt. is a pair (:F,a) where, a is an interpretation funct10n
such that, for every points E 5, (i) a(=, s) is the identity relation over U and
(ii) for P an n-ary predicate, a(P,s) <;::; U". . .
The truth condition for an atomic formula at a pomt gtven a value
assignment for the variables is classical. Tr~1th conditions ~or connectiv~s.
and quantified formulas are as expected. Given a value as~1gnment, _6.¢ is
true at 5 if, for every point t such that sRt, </> is true at t. A for~mla <p is true
(false) in a model M under a value assignment if it is true (false) at e~ery
point in M. A formula</> is a supervalu~tionarf c?ns~quence of a s~t 1 of
formulas if, given a model and a value assignment, 1ff 1s true then</> ts true.
A formula¢ is valid if it is a supervaluationary consequence of the empty set.
13 The proof is provided in Appendix A. (The finiteness condition can be dropped if
the language is infinitary.)
Vague Existence 215
The admissibility relation R is intended to be reflexive and symmetric (so
as to validate 6.¢ --> </> and </> --> 6.-.6.-.ef>) but intransitive. This choice
allows for the possibility of higher-order vagueness. 11
It will be useful to stress a crucial difference between fine's and Williamson's
precisificational frameworks. Recall that being definitely true means being
true in all precisifications. On the Finean approach, what counts as a
precisification of the language is an absolute matter, since precisifications
are identified with complete specification points (i.e. the points to which the
base point bears the ancestral of the admissibility relation). As a conse-
quence, facts about definiteness and vagueness are absolute, as well-there
can be no instances of higher-order vagueness. In Williamson's models, on
the other hand, there is no such thing as the set of precisifications of
the language. For every specification point s there is a set of admissible
specification points-intuitively, those precisifications that <n·e in the neigh-
borhood of s. Therefore, an expression of the form "there is a precisification
so-and-so" is always relativized. This important feature is meant to capture
the idea, discussed in Section 1.2, that the language can have multiple
sets of precisifications, and that it can be vague which of those sets is
the correct one. Consequently, Williamson's models make room for
second-order vagueness. Likewise, distinct sets of sets of specifications
can be admissible to distinct sets of specifications, thus making room for
third-order vagueness--and so on and so forth. This kind of higher-order
vagueness can be put to use in order to model super-vague existence.
Nevertheless, the Williamson-style semantics just sketched does not
represent a generalization over Fine's framework in the desired direction,
since all specification points in a model have constant domain and, as a
result, constrain existence to be definitely not vague at all orders.
I submit that, in order to be able to model super-vague existence,
we should proceed as follows. A variable domain frame :F* is a quadruple
(5, U,R,Dom) such that
(i) specification points S, universe U and admissibility relation Rare as before;
(ii) Dom is a function mapping each point s ES to a subset of U (intui-
tively, the objects that exist according to the precisification s);
(iii) U = U{ Dom(s) },ES;
(iv) for every n), 1, there is some s ES which is n-determinate.
11
Statements of higher-order vagueness are satisfiable if either !'ief> _, f'if'i<p or •f'i</> -->
f'i•!::,.<f> fails. On the model-theoretic side, the failure of either condition corresponds to a
non-transitive or non-euclidean admissibility relation: see Williamson [37, p. 133]. In
fact, higher-order vagueness has its modal counterpart in the contingency of contingency,
which also takes place only in systems weaker than 55.
216 Alessandro Torza
The notion of n-determinacy is defined recursively as follows:
I . s is \-determinate iff for all s', s": if sRs' and sRs", then
Dom(s') c~ Dom(s'')
n + 1. s is n + I-determinate iff for all s' ,s": if sRs' and sRs", then s' is
n-determinate iff s" is n-detenninate.
As I will discuss in Section 2.3, the property of n-determinacy is crucial,
since it makes models based on a variable domain frame F Sider-
determinate. If (iv) were dropped from the definition of a frame, the
semantics would admit of unintended models, i.e. models that are not
Sider-determinate.
Let us now turn to the problem of finding a suitable notion of local truth,
i.e. truth-at-a-point. In a model based on F, a quantifier evaluated at swill
range over Dom(s ). The nature of the interpretation function a will depend
on the kind of semantics we choose for evaluating formulas with non-
referring terms at a specification point. If x exists at point s, what s~?uld
be the truth-value of "P(x )" at a point t where x docs not exist? 1 hrcc
options are on the table for dealing with local truth.
Positive localsemm1tics. We could allow an atomic formula like "P(x )"to be true
at a point where x does not exist. On this view, for any point t, the interpretation
of a predicate P is a subset of the frame domain, namely a( P, t) c:;; U, and the
value of a free variable <Ul element of U. Accordingly, "P(x)" will be either true or
folse at t depending on whether x is or is not in Dom( P, t).
However, this approach provides a misleading picture of existential vague-
ness. For, if the same set of objects can be referred to according to all
precisifications of the language, existence will be vague. only ~10mi~1ally.
Indeed, whereas quantifiers are restricted to Dom(t), which vanes with t,
free variables will constantly range over U. Hence, there could be a precisi-
fication in which mereological nihilism is true, and yet it is also true of the
sum of this mug and that table that they occupy space in my office. Positive
semantics is therefore inadequate for modeling vague absolute quantification.
Neutral local semantics. A more attractive approach is to (i) let a variable at
a specification point pick something out of the relevant domain, or nothing
at all, and (ii) restrict the interpretation of a predicate in such a way that
a(P,t) c:;; Dom(t). Neutral semantics takes an atomic formula to be inde-
terminate at all points where some of its terms fail to refer. The picture
should then be completed by defining truth conditions for the non:atomic
formulas. As it turns out, neutral semantics is not a viable C<'lndidate for local
truth, either. One of the most attractive features of supervaluationism is that
it preserves classical tautologies. The motivating intuition is that vagueness
facts should not affect the logic of truth-functors. That nice result breaks
Vtigue E'<istence 2l7
down when local truth is defined via neutral semantics. Ifx does not refer at
a specification points, x = x is indeterminate there and so is x 7 x (under
the assumption that local semantics is truth-functional). 'T'hus, x = xVx } x
is indeterminate at s (since, in neutral semantics, a disjunction is indeterrnin-
ate when both disjuncts arc). Hence, some instance of)V•p is untrue in some
model and, therefore, invalid.
To restore the validity of classical tautologies without giving up neutral local
semantics, we could tinker with global truth, i.e. nu th-in-a-model. Let then cf> be
true in a model M* based on F just in case it is true at all points where all of it~
terms are defined. Since in the evaluation of x = x we rule out every world
where xis non-referring, x xVx 7 x must be valid. Out of the frying pan,
into the the fire: now :Jx(x = y) is true in a model iffit is true at all worlds where
y is defined. But in neutral semantics, y is defined at points iff it picks a value in
Dom(s) iffy exists at s. So, :Jx(x = y) is trivially true in evety model and,
therefore, valid. However,!:. 3x(x = y) is invalid. To see that, consider a model
where y is defined at a points but not at a point t where sRt. Then!:. 3x(x = y)
is false at s, since t is an accessible point where :Jx(x = y) is untrue. So,
!:. 3x(x = y) is invalid. It follows that the so-called necessitation rule
(N) If F= cf> then F= !:.fj>
fails under the proposed revision. Since it is a standard desideratum of any
semantics for vague language that validity be closed under definiteness, the
above theory should be rejected.
One might reply that (N) can be restored by making one simple change:
just let !:.cf> be true at a points iff, for every point t such that sRt and all terms
in cf> are defined at t, cf> is true at t. This attempt at validating (N) does more
harm than good, however, because now!:. :Jx(x = y) is true-in fact valid-
even when there are precisifications of the hnguage according to which there
is no y. Since we are looking for a semantics modeling existential vagueness,
I take this to be a reductio of neutral local semantics.
Negative loml semantics. On the third approach for defining truth-at-a-
point, value assignments and the interpretation of the language are exactly as
in neutral local semantics: (i) a variable at s picks something out of Dom(s ),
or nothing at all, and (ii) a( P,t) c;;: Dom(t). These two conditions codify the
reasonable assumptions that reference and predication make sense only
relative to what exists according to a given precisification. The essential
difference is that negative local semantics takes any atomic formula to be
false at a point where some of its terms are non-referring. I claim that a
negative supervaluationary semantics, i.e. a supervaluationary framework
based on variable domains and negative local semantics, yields the correct
account of vague existence phenomena. To corroborate my claim, I will first
 Alessmu:lro Ton:<! Vtigue F<istence 219
218

provide an exact formulation of the theory, and then tease out its main A formula cf> is a NS-consequence of a set r of formulas (i.e. r I= n<;c/J)
iff, for every NS-model /vi*, if M* I= Nd~ then M' I= Nsc/J
semantic and logical features.
A formula cf> is NS-valid iff it is a NS-consequence of the empty set.

2.3 Negative supervalnationary semantics Let us now turn to the key properties of negative supervaluationa1y seman-
tics. Standard prccisificational theories of vagueness, such as Fine's specifi-
A negative superualuationmy (NS) model M* for LL\ .is a pair (.P ,a*) where cation space semantics or Williamson-style constant domain semantics, have
P is a variable domain frame and a* an interpretation function such that, the virtue of being classical in a precise sense: given a purely extensional
for every point t ES, (i) a*(=', t) is the identity relation over Dom( t) and (ii) language, classical consequence (I= c) and supcrvaluationary consequence
for Pan n-ary predicate, a*(P,t) ~ Dom(t)". . . . (I= sv) coincide. Namely, for cf> and r formulated in a first-order language£
Let VAR be the set of variables in £ 6 and S the set of specificauon pomts. without definiteness operator, 15
A value assignment for VAR over M* is a set of partial functions { g,}, Es
such that:
(Eq l) r I= c<P iffT I= sv<,b
l. g, : VAR_, Dom(t) On the other hand, some classical inference rules, which have been regarded as
2. U{t } sis a total function/: VAR_, U being the source of sorites paradoxes, fail in standard supervaluationary semantics.
3. if dx)~nd g,(x) are both defined, then t,(x) = t,(x) Existential instantiation does not hold: from "some number n is the least number
such that n grains of sand constitute a heap" we cannot infer the existence of ~my
The first condition allows a variable to be undefined at some precisifications, particular no such that" no is the least nwnber such that n 0 grains ofsand constitute
whereas the second condition forces a variable to be defined at some a heap." The same applies mutatis mutandis to universal generalization. 16
precisification. Consequently, negative supervaluat.ionary sem.an~ics is r~ot Notice, however, that the result of substituting NS-logical consequence for
a framework for definitely non-referring terms, unlike free logic. fhe thm{ I= sv in (Eqi) does not hold, due to the fact that local semantics for NS is non-
condition guarantees that variable assignments are rigid across specification classical. Here is an intuitive example. (A formal countcrrnodel is provided in
points. . . . Appendix B. l .) We know that "Ted is not a (mereological) simple" classically
Truth-at-a-points (local truth) for cf> under a variable assignment { t,}, Es m entails "Something is not a (mereological) simple." The same inference, on the
an NS-model JVI*, written ( M* ,s, {g,}, Es) I= N5cf>, is defined recursively thus: other hand, is not NS-valid. Suppose that it is vague whether mereological
(at) If cf>= P(x 1 , .. .,x,,), then (M*, s, {0,Es) I= NS</> iff t, is defined nihilism or universalism is tme. Now, in any precisification that allows the
for all i E { 1,..., n} and (t,(x1), .. ., t,(x,,)) E a*(P, s) existence of sums, Teel exists and is not a simple; and in any precisification that
(...,) If cf> = •if1, then ( JVI*, s, {0 s Es) I= N5cf> iff ( .M*, s, {0 , Es) I= Nsifi does not allow the existence of sums, Ted does not exist and therefore (local
(A) If cf>=(i/;Ax), then (M*, s,{0,Es)l=N,cf> iffboth (M*, s,{0,Es) semantics being negative) is not a simple. Hence, Teel is not a simple. But in any
precisification that docs not allow the existence of sums, it is not the case that
l=Nsi/1 and (M*, s,{g,LEs)l=NsX . ,
(V) If cf>= Vxtf;, then (M*,s, {0,Es) I= Nscf> iff, for every {t JsES something is not a simple. Hence, "Something is not a (mereological) simple" is
such that {, is defined on x and differs from t, at most on x, unttue, which shows that classical existential generalization is NS-invalid.
Nevertheless, negative supervaluationary semantics can be shown to validate
(M*, s, {{J,Es) I= NSV1
(!'::..) If cf>= l'::.ifi, then (M*,s, {0,Es) I= Nscf> iff, for every t such that a weaker version of existential generalization, which typically holds in free logic:
sRt, (M*, t, {0,Es) I= NSV1 (3G-) {cf>(x), 3y(x=y)}l=N\'3xcf>(x).
A few more definitions are needed. In fact, a general result can be proven connecting negative supervaluationary
A formula cf> is true in a NS-model M* relative to a variable semantics to negative fee logic: in a purely extensional language, the
assignment { g,}, Es (i.e. ( M*, {g,}, Es) I= Nscf>) iff, for every s ES,
(M*, s, {t,LEs) I= NS</>· 15 See Keefe [16, pp. 174-81].
16
A formula cf> is true in a NS-model M* (i.e. M* I= Nscf>) iff, for every For a discussion and defense of this aspect of supervaluationism, see Kcefo [ 16,
pp. 181-8].
{g,},E5" (M*, {g,},ES) I= NS</>·
220 Alessandro Torut
consequence relation of negative free logic is preserved by negative super-
valuationary semantics. Negative free logic-a first-order logic for languages
with non-referring terms-is sound and complete with respect to negative
semantics, which is the semantics employed here for defining local truth in
NS-models. 17 We can think of a model of negative free logic (NF-model) as
the degenerate case of a NS-model with a single specification point.
A variable assignment over a NF-model is a partial function from free
variables to the model domain. 18 In a first-order extensional language, </> is
a negative-free consequence of r (I' I= NFC/>) if, for every NF-model and
variable assignment, r is true only if 4, is true.
The aforementioned result connecting negative free logic and negative
supervaluationaty semantics is as follows. For 'P and r formulated in a
language £ without "/'.),"
(Eq2) if r I= NF</> then r I= NS'P·
(For a proof, sec Appendix B.2.) On the other hand, the converse of (Eq2)
fails, since in negative free logic but not in NS it can be consistently said of
something that it doesn't exist (cf. Appendix B.2).
We can tease out a few interesting facts concerning the interaction of
existence and identity. First of all, notice that existence is definable via
identity in both negative free logic and negative supervaluationary seman-
tics, because each of the two frameworks validates the biconditional
3y(x =y) +-+ x = x. Since • :3y(x = y) is NF-satisfiable, so is x ye x. On
0
the other hand, in negative supervaluationary semantics nothing is nonexis-
tent, therefore nothing is self-distinct. That is how things should be.
Moreover, the incliscernibiliry of non-existents
(IN) • 3z(x = z) /\ • 3z(y = z)-+ (</>(x) -+ <f>(y))
19
which is valid in negative free logic , fails in negative supervaluationary
semantics (see Appendix B.3).
Although the converse of (Eq 2) fails, a weaker equivalence can be proven.
Namely, negative free logic and negative supervaluationaiy semantics define
the same class of valid formulas in a 6-free language:
(Eq 0) I= NF<P iff I= Ns<p.
(For a proof, see Appendix B.4.)
17
For the relation between negative free logic and ncgativescmanticsscc Nolt [22], Burge [7].
18
Note, however, that a variable assignment over a NS-model with a single specifica-
tion point is a singleton {fl, where ~ is a totfll function.
19
Nolt [22, p. 1033].
Viigue E'xistence 221
Given a language Li\, negative supervaluationary semantics satisfies the
following conditions:
(Tau) F= Ns</>, for every tautology cf>
(K) I= N.'>6(</> ·---+ 1/;) __, (6</> ---> A</1)
(MP) { cp,<f> </1} F Ns</1
(N) ff F= N.I</' then F= N1'1</>
en F Nsfl.<p ·--+ </>
(B) 1= N.'i'P __, 11•6•</>
The two conditions (Tau) and (MP) guarancee the validity of classical
se.ntential logic. By the Kripke schema (K), the "definitely" operator dis-
~nbutes over the material conditional. (T), expressing the facticity of defin-
iteness, holds because the admissibility relation R is assumed to be reflexir
The symmetry of R validates (B). The so-called necessitation rule (~)
ensures tha.t validity is closed under definiteness. In fact, a stronger result
than (N) is provable in negative supervaluationaty semantics (as well
as in most supervaluationary frameworks), viz. the definiteness of truth or
6-introcluction rule: '
I now turn to conditions that fail in the present framework. A non-classical
aspect of negative supervaluationaty semantics, which it inherits from
negative free logic, is that validity is not closed under uniform substitution.
~or P atomic, for instance, P(x) __, x = xis valid whereas -iP(x) -> x = x
ts not.
Since negative supervaluationary semantics admits of variable domain
models, and in particular it could be that Dom(s)c Dom(t) for sRt, the
Barcm formula
(BF) -,/'.)-, :3x<f> _, ::Jx-,f1•</>
has_ ~nvalicl i~stances, such as -,/j_-, :3x(x = y) --+ :3x-i6-i(x = y ). Think
of. I ed, for mstance. In some sense, he exists-namely, the sense of
existence of mereological universalism. But for the nihilist there is no
suc.h th~ng as Teel. Therefore, it is not the case that there is something
which, tn some sense, is Ted. The failure of the Barcan formula is, of
course, not an idiosyncrasy of negative supervaluationary semantics.
~hen "-,/'.)-," ~s substituted with intensional operators of other sorts, it
is not hard to find counterexamples to the schema. In the modal case, the
sentence
If Mary could have had a daughter, somebody could have been Ma1y's
daughter
222 Vr1gue E:dstence 223
20
is false, barring exotic semantic frameworks. The same occurs (unless we bottom, misleading. For, in negative supervaluationary semantics, the rea-
arc eternalists) when the possibility operator is replaced with a tense oper- son why (DI) and (DD) are invalid is that a term may refer at some but not
ator, as in: all precisifications. _On the other hand, the run-of-the-mill counterexamples
to the necessity of identity and distinctness are formulated in counterpart
If Mary will have a daughter, there exists somebody who will be Mary's
theory o: other possible-world semantics relying on non-standard interpret-
daughter. 22
ations of de re truth. Moreover, in such modal frameworks there are true
The Barcan formula also fails in fictional contexts. If the operator "according instances of contingent identiry and distinctness, whereas (DI) and (DD)
to fiction S" is construed as the analog of the necessity operator, we can can at most have untrue instances in negative supervaluationary semantics.
define its dual, "according to fiction S it might be that." Works of fiction _ Incidentally, since variable assignments over NS-models are rigid, it
being typically incomplete, "according to fiction Sit might be that p" is true f-ollows that identity is weakly definite:
just in case the fiction does not entail the falsity of p. Thus, consider
(DI-) x = y ---> !':i( 3z3z'(x = z/\y = z')-+ x = y)
If according to Woyzeck it might be that the Captain has a mistress, then
Distinctness, on the other hand, does not satisfy weak definiteness:
there is somebody who according to Woyzeck might be the Captain's
mistress. (DIY) x f y -> L\( 3z :Jz' (x = z /\y z') --+ x f y)
Since, according to Bilchner's play, the Captain does in fact have a mistress, (See Appendix B.5.)
the antecedent is obviously true. However, the consequent is absurd, unless The importation schema
21
we buy into realism about fictional characters. (IM) 3x!':icf> -+ l':i 3xcf>
Insofar as there can be specification points s, t such that Dom(t )cDom(s)
for sRt, the Converse Barcan formula fails, too. (Proof in Appendix B.6.)
We can finally return to the main point, quantifier vagueness. Our goal is
to find some NS-model of super-vague existence. First of all, we need to
is also invalid. check that the following condition holds:
Further conditions which are invalid in negative supervaluationary
Sider-determinacy: for all A1*, { t,}, Es and n ) 1: ( M*, { t,}, Es) Jz'.' NS
semantics are the definiteness of identity
!'1 :Jx(x = y)
(DI) x = y -> 6.x = y
for otherwise negative supervaluationary semantics would be inconsistent
and the definiteness of distinctness with Sider's result, or with the higher-order generalization of it. Second, we
would have to show
(DD) x f y -+ !':ix f Y·
anti-Sicler-determinae,y: for some M* and { t,} s Es and for all n) 1:
(Proofa in Appendix B.5.) However, (DI) can never be false in a model, for if
it were, that woul;! contradict the fact that x = y I= NS !':ix = y, which is
(M*, {0,Es) Jz'.' •!" 3x(x = y)
an instance of (N ). Likewise for (DD). A proof that both conditions hold in negative supervaluationary semantics
The invalidiry of (DI) and (DD) may bring to mind the analogous case of can be found in Appendix B.7. We can conclude that super-vague existence
the failure of the necessity of identity and distinctness in some versions of is NS-satisfiable.
possible-world semantics. However, the analogy is only superficial and, at Let's recap. Sider argued that

(V) if vagueness is given a precisificational account and existence is

20 One such framework, involving the use of possibilist quantification, is defended in expressed by the unrestricted existential quantifier, then vague
Linsky and Zalta [20 J [21 J, Williamson [36 J. An alternative sem<mtics which validates the existence is incoherent.
Barcan schema is one in which quantifiers range over individual concepts: see Garson [14].
21 Cf. Sainsbury [24, p. 34]. An alternative option is to go Mcinongian and allow
22
quantification over non-existents. For example, sec Lewis [ 18 J.
224 Alessandro l orvt
According to Sider [26, p. 4], the moral of his alleged reductio is that vague
existence "would be radically unlike familiar cases of vagueness. Vague
quantifiers may yet be possible, but such vagueness would require an
entirely different model from the usual one." If I am correct, this moral is
correct only in a qualified manner. In Section l I argued that, although
Skier's (generalized) argument proves the impossibility of n-th order vague
existence, for all n, the result is prirna jacie consistent with the possibility of
super-vague existence. The suggestion has been vindicated in Section 2,
where I have developed a generalization of Williamson-style precisificational
semantics where Sider-determinacy is satisfied and yet super-vague existence
is shown to be a consistent notion. Consequently, Skier's moral holds as
long as vagueness is not super-vagueness, modulo the present choice of
semantic framework.
2.4 Objection from reductio ad absurdum
In Section 1.2, I remarked that Sider's argument against vague existence is
invalid unless stated in a perfectly precise language, since the argument is a
reductio ad absurdum, which is a valid inference form only for bivalent languages.
It might be objected that vagueness in the proof's language doesn't suffice
to rule out the applicability of reductio. For if that language is vague, it
would be reasonable to interpret it within a supervaluationist semantics.
Now, it is true that reductio is invalid in supervaluationism-for instance,
we cannot infer from cf>/\--, tJ. cf> I= .l to I= cf> --> tJ. cf>. However, reductio is
supervaluationarily valid for 6.-free languages. To see that, let I= c and I= sv
denote the relations of consequence for classical logic and standard super-
valuationism, respectively. Now, suppose that f,c/> I= sv .l. By (Eq1 ), we
know that I= c and I= sv are equivalent (cf. Section 2.3). So, f,c/> I= c.l
and, since reductio is classically valid, r F= c•<f>. By (Eq1 ), it follows that
r F= sv•r/>. But Sicler's argument is formulated in the metalanguage of 3xcf>,
which paraphrases away 6. (as well as any other expression defined via tJ.) by
quantifying over precisifications which are extensional, set-theoretic objects.
Thus, the language of Sicler's proof being 6.-free, reductio ad absurdum
appears to be valid after all.
What undermines the above objection is the tacit assumption that, if the
semantics for the language of Sider's proof is supervaluationary, then it has
to be some kind of standmd supervaluationism, such as Fine's specification
space semantics or Williamson-style constant domain semantics. I claim
instead that, if Sider' s proof is coached in a vague language, we should
model it via negative supervaluationism. The reason for this choice is quite
straightforward. I argued in Section 1.2 that if Sider's language lacks
Vague L:·dstence 225
determinacy, this must be due to vague quantification over precisifications.
We need, therefore, a precisificational semantics to deal with vague quan-
tification which, as explained in Sections 2.2 and 2.3, should be negative
supervaluationism, with the relevant consequence relation F= NS· Now, a
feature of negative supervaluationism, which distinguishes it from standard
supervaluationism, is that it fails to validate reductio ad absu1dum even for
Li-free languages. For instance, it is a fact that
Ted doesn't exist F= Ns.l
since names and free variables cannot be definitely non-referring in a NS-
moclel. However, it doesn't follow that
F= NITeel exists
for otherwise existence would always be determinate in negative super-
valuationism, which we know not to be the case clue to the NS-satisfiability
of super-vague existence (Appendix B.7).
It is worth noting that, even though we could give a Sider-style argument
against vague existence which doesn't employ reductio ad absurdum, the new
argument would still have to be formulated in a precise language, in order to
be valid. For example, we could give a proof by cases:
l. r 3xcf>' is either vague or not vague.
ii. Suppose r 3xcf>' is vague. Therefore, it is true at some precisification,
and so true at all precisifications, which means that it is true (cf. steps
3-9 in my reconstruction of Sider's argument, Section 1.1). But if
r 3xcf>' is true, then it is not vague.
iii. Supposer 3xcf>' is not vague. Therefore, it is not vague.
iv. Thus, r 3xcf>' is not vague (proof by cases)
The argument has the form:
i. pV•p
ii. pF=q
iii. -pF=q
iv. F=q
Now, in standard supervaluationism, proof by cases behaves just like
reductio ad absmdum: although invalid in general, it is valid for arguments
stated in a Li-free language. But again, if the language in which the above
proof is stated is vague due to quantification over precisification, the correct
framework is negative supervaluationism, with the relevant consequence
relation F= NS which does not validate proof by cases even for Li-free
languages, as the following simple counterexample will show:
226 Alessandro Torvi
i. a = ilV t1 'F a
ii. a= a F N'i 3x(x =a)
iii. a f a F= NS 3x(x = a)
iv. F NS 3x(x =a)
l ndecd, in some model the three premises arc true, 23 whereas the conclusion
is untrue.
2.5 Inferentialist objection
have modeled super-vague existence by means of precisified quantifiers.
I will now consider an objection, which is an adaptation of an argument
originally formulated by Williamson [34], purporting to show that there
cannot be multiple prccisifications of the existential quantifier.
Given a language £ with vague 3, define a new language £' in which 3 is
replaced with two precisifications 3 1 and 3 2 . For instance, 31 could be the
ontologically sparse quantifier of the mereological nihilist, w~ereas 32 is the
promiscuous quantifier of the universalist. (Likewise, in {-- there will be
"cornposition 1" and "composition 2 "). Now, let cf>(x) be a£ formula. From
31xcf>(x) we can deduce cf>(z) by cxistential 1 instantiation, where z is chosen
so that it does not occur free in cf>(x). By existential2 generalization, 32xcf>(x)
follows from cf>(z). Hence, there exists a deduction of 32xcf>(x) from
31xcf>(x). Since we can produce the same kind of argument running in
the opposite direction, the two quantifiers are equivalent, which contradicts
the initial assumption that 3 1 and :3 2 are distinct precisifications.
To this argument I offer a two-tiered reply. For reasons that will soon
become clear, I take the second part of my reply to be the more enlightening
one. Firstly, the objection assumes that the precisified quantifiers are clas-
sical, in the sense that for each :J,,, the rules of generalization,, and in-
stantiation,, are the classical ones. But this assumption is unwarranted.
Recall that a precisification of the language is identified with a particular
specification point in a model of negative supervaluationary semantics.
Local truth, i.e. truth at a specification point, is defined in terms of negative
semantics. Moreover, negative free logic is sound and complete with respect
to negative semantics. Therefore, the generalization and instantiation rules
for a precisified quantifier are the ones of negative free logic:
• from cf>(z) /\ 3,,x(x z) infer 3,,xcf>(x)
• from rand 3,,xcf>(x) infer cf>(z) /\ 3,,x(x = z), where z does not occur
free in r or cf>(x).
23
a =J a is an NS-inconsistency, hence 3x(x = a) follows from it trivially.
Vrtgue E;;istence 227
With that being said, it is easy to see what goes wrong in the inferentialist
objection. Recall that :31 is the less promiscuous quantifier, which docs
not support compo~ition 1, whereas :32 allows unrestricted cornposition 2 .
Let cf>(x) be the £ formula ".x- is a mereologica[ compound 2 ," where
"compound2" is the univcrs<tlist prccisification of the vague £-term "com-
pound." Since the underlying logic is free, from "there exists2 the fusion 2 of
a table and a giraffe" by existcntia'2 instantiation we can infer "z is the
fusion 2 of a table and a giraffe and z cxists 2 ." But in order to conclude by
existential 1 generalization "there exists 1 the fusion 2 of a table and a giraffe,"
we first need to be able to infer "z is the fusion 2 of a table and a giraffe and z
exists1 ." So, the derivation goes through if "z cxists2" entails "z cxists 1,"
which cannot be assumed without begging the question. 24
I now turn to the second reply. The infercntial!st objection simply
assumes that it is possible to define a new language £ in which the vague
quantifier 3 is replaced with the sharp quantifiers 3 1 and 3 2 . As it turns
out, multiple quantifiers obeying the rules of free logic cannot coexist in the
same language. For if there were such 31 and 32, a new existential
quantifier 3could be defined in £' whose range is the union of the ranges
of 31 and 32:
3x(x z) := :J 1x(x = z)V 32 x(x = z).
Now, if 3 1 is a proper restriction of 3, for some z it is true that
3x(x = z) /\-, 3 lx(x = z). By definition of 3, that is equivalent to
(i) -, 31x(x = z) /\ :l2x(x = z).
Since 31 x(x = z) +-> z = z is a theorem of negative free logic, from the first
conjunct of (i) it follows that
(ii) z f z.
From (ii) and the second conjunct of (i) we can infer
(iii) 32x(x =J x ).
But
(iv) V2x(x = x)
is a theorem of negative free logic. Hence, the claim (i) that 3 1 is a
restriction of 3 is inconsistent, provided that the precisified quantifiers
obey negative free logic. But according to negative supervaluationary seman-
tics, precisified quantifiers do obey negative free logic. We must conclude
21
Cf. Turner [30, pp. 25-6].
228 Alessandro Tor:ztt
that it not possible to define :3 and, therefore, that we cannot use both 31 and
ch within the same language, as the infercntialist objection presupposes. 25
The moral is that quantifiers behave differently from non-logical predicates
in one key respect. If we speak a language where "bald" is vague, we can define
a new language in which the original predicate is replaced with a multiplicity
of precise prcdiGJ.tes "bald 1," "bald2" etc. This cannot be done with quanti--
ficrs on pain of inconsistency. I hope it is now clear that the deeper reason why
the infercntialist objection is unsound is that it assumes that quantifiers
governed by negative free logic can coexist in a single language.
3. CONCLUSION
Whether existence can be vague has consequences both in first-order ontol-
ogy and in metaomology. In the former case, the possibility of vague
existence makes room for vague composition. In the latter, vague existence
may be a symptom of the world lacking a unique quantificational structure.
Sider has famously submitted a reductio of vague existence, on the assump-
tion that vagueness is interpreted precisificationally <md existence is absolute.
In Section 1, I argued that a precisificational framework per se does not
allow us to disprove vague existence, i.e. to prove that it is definitely not
vague. At most it can be proven that existence is not definitely vague. The
same applies to a disproof of higher-order vague existence. The upshot of the
discussion turned out to be that Skier's argument is compatible with
existence being neither definitely vague nor definitely precise, at evety
order. I turned this specific phenomenon super-vague existence. In
Section 2, l provided a precisificational model theoty, dubbed negative
supervaluationaiy semantics, with the aim of modeling super-vague existence
and its logic. Moreover, an objection from reductio ad absurdum and an
inferentialist objection have been taken GJ.re of.
25
It could be objected that a logical constant satisfying some given condition in
a language £ need nor do so in an expanded language £ . For instance, identity satisfies
Leibniz's law if the hrnguagc is extensional, but not ii' we add doxastic or epistemic
operators. Likewise, the various 3,, might n,ot satisf}' the same inference rules in each
precisification £,,of£ and in the expanded£ . This observation overlooks one important
bit of information, namely that the quantifiers 3,, are all precisifications of the original
quantifier 3. I assume the following principle: if c' is a prccisification of the the logical
constant c, then c' satisfies at least the axioms and rules of inference which c satisfies. In
fact, I take that condition to partly define what it mC'ans for c' to he a prccisification ofc. It
follows in particular that 3 1 and 3 2 must satisfy the axioms and rules of negative free
logic.
Vt1gue Edstence 229
If existence is super-vague, we ought to accept that composition might be
super-vague (i.e. vague at <lll orders) and that the world may lack a unique
quantificational structure.
Alessandro Torzrt
Ntttiont1! Autonomous University of Mexico
ACKNOWLEDGMENTS
I would like to thank Dean Zimmerman and one anonymous refrrcc for the
extensive and very helpful comments. l am also in debt to the audiences at the
University of Turin, University of Sussex, Duke University, the XVI International
Congress of the Mexican Philosophical Association, the National Autonomous
University of Mexico, the 2nd Colombian Conforence of Logic, Epistemology and
Philosophy of Science. This work has been made possible in part by the CONACyl'
gram CCB 201 l 166502 and the PAPlIT grant IA400412.
APPENDICES
APPENDIX A: FINEAN SUPERVALUATIONISM
AND EXISTENCE
In Fine's supervaluationary semantics it can be proved that, as long as there definitely
are finitely many objects, there is no vague existence at any order (cf Section 2.1).
For the proof; it suffices to show that (i) existence is definite, i.e.
3x(x = y) ---> l'i. 3x(x = y), and that (ii) definite statements cannot be indefinitely
definite, i.e. l'i.<f> ··-> l'i.l'i.<f>. Let n be the cardinality of the domain of the largest
complete specification point in a space. (If the cardinalities had no upper bound, it
would not be the case that there definitely arc finitely many objects.) Notice that the
domain of the base point@ is a subset of the cardinality ofany complete specification
point, since if 3y(x "= y) is true at <1~ (given an assignment for x), then it is true at
e~ery accessible point. So, there is some m ,,; n which corresponds to the cardinality
of the domain of @. Let Vy(y = x 1Vy XzV ... Vy == x,,,) express which things
exactly exist at @. By the construction of a specification space, that sentence must
be true at all complete specification points, which must therefore have constant
domain. So, 3y(x = y) is true at a complete specification points only if it is true at
all complete specification points. It follows that 3_y(x = y) is true at @, and so is
l'i. 3x(x = y). Therefore, at every sit is true that 3x(x = y) ---> l'i. 3x(x o= y). We can
conclude that existence is not vague. Moreover, Fine's theory does not admit of cases
of higher-order vagueness to be expressed in the object-language: what is true/false/
indeterminate is definitely truc/f.'llse/indetcrminate. Which is to say, both (a) l'i.<p ~-)
l'i.l'i.<p and (b) •l'i.</> ----> l'i.•l'i.<f> arc supervaluationarily valid in Fine's model. As to (a),
 Alessandro Torza Vi1gue hxistence 231
230
suppose that f'..<p is true at a complete specification point s. So, q, is true at all assume that t,(x) =ti, t,(y) == b, a cJ b whereas both t,(x) and t,(y) are undefined.
complete specification points. It trivially follows that it is true at s that every complete Then, the instance of (IN) obtained by substituting !':,,-,[' for q, is false at t under
specification point is such that <p is true at all complete specification points. Which is {t,}sES and therefore untrue in the model.
to say, f'..l'..<p is true at s. Muttttis mutt111dis for (b). As a consequence, since existence is
precise, it must be precise at all orders. Q.E.D.
B.4 NS and negative free logic II
APPENDIX B: ELEMENTS OF NEGATIVE Negative free logic and negative supervaluationary semantics define the same class of
valid formulas in a language without"/'..":
SUPERVALUATIONARY SEMANTICS

B. l NS and existential generalization

\Y/c want to show that, for <p and r formulated in a first-order language[. without 6,
it is not the case that
r F= c</> iff r F= Ns<P
where F= c ( F N.s-l indicates classical (NS-) consequence relation. To see that, let M*
be a NS-model with only two specification points sand t, where Dom(s) = {r1} and
Dom(t) = {a,b}. Also, suppose that a*(P,s) '°" a*(P,t) ={a}. Consider a value
assignment mapping x to b at t and leaving it undefined at s. Since P(x) is false at
both sand t, •P(x) is true in the model. However, 3x•P(x) is true at sand false at t,
therefore indeterminate in the model. Thus, existential generalization does not hold
in general: <f(x) it' NS 3x</>(x). The same reasoning applies mutatis mutandis to
universal instantiation.
The left-to-right direction is an immediate consequence of (Eq 2). As to the
converse, consider a NF-model and a partial function ~ mapping the free variables
c'.f <P to .the domain. Truth in that model is tantamount to local truth at a specifica-
tton point t of some NS-model M* under t, = ~. Since <p is NS-valid, it is true at
M* under {t,}sES and therefore locally true at t under t,. Consequently, 'Pis true in
the original NF-model under the variable assignment (.
B.5 NS and identity
A condition which fails in negative supervaluationa1y semantics is the definiteness of
identity
(DI) x = y ---> /'..x = y

B.2 NS and negative free logic I To see that, just cc:nsider a NS-model and a variable assignment in which x and y
co-refer to a at pomt s, whereas tis a point such that sRt and a~ Dom(t). Since
For <f and r formulated in a language [. without '6', x == y is true at s and false at t, (DI) is false at s and therefore untrue in the
model. A symmetrical scenario yields a counterexample to the definiteness of
(Eq2) if r F NFef' then r F NSrP·
distinctness:
To see that, let r be true in the NS-model M* given a variable assignment {t,},ES'
Then, for every t,,
r is locally true at t. Since local truth for a 6-free language in a NS- (DD) x =f y __., /'..x =f y.
modd is t~mtamount to truth in a NF-model, and since q, is an NF-consequence of I',
However, (DI) can never be false in a model, for ifit were, that would contradict the
it follows that <pis locally true at t under t,.
Hence, <pis true at M* under {t, },ES'
fact that x = y F= Nsll>: = y, which is an instance of (N\ Likewise for (DD).
However, the converse docs not hold. Because a variable assignment over a NS-
Since variable assignments over NS-models arc rigid, it follows that identity is
modcl maps each variable to an object at some specification point,• 3y(x = y) can
weakly definite:
only be false or indeterminate in a model. Hence, it must be that• 3y(x ~~ y) F= Nsl..
On the other hand, • 3y(x = y) is NF-satisfiable. (DI ) x = y-> 6(3z3z'(x = z/\y "'' z') -> x = y)
Distinctness, on the other hand, docs not satisfy weak definiteness:
B.3 NS and the indiscernibility of non-existents
(DD ) x cf y-> 6(3z3z'(x = zJ\y = z')-> x cf y)
The indiscernibility of non-existents:
For suppose a variable assignment maps x and y to the same object a at point t,
(IN) • 3z(x = z) J\ • 3z(y = z) -> ( <f(x) -> <p(y)) and let sRt, where a~Dom(s ). Since the assignment is rigid, x and y will fail to refer at
fails in negative supervaluationary semantics. In order to see that, consider a NF- s, and so the antecedent of (DD ) must be true at that point. The consequent, on
model with two specification points s, t such that tRs and a*(P,s) = { b}. Now, the other hand, is false at s.
232 Alessandro Tor:zd Vtigue hY:istence 233
B.6 NS and the importation schema (IM) Akiba, Ken (2004). "Vagueness in the World," Notis 38(3): 407-29.
Barnes, Elizabeth (2009). "Indeterminacy, Identity and Counterparts: Evans Recon-
The importation schema: sidered," Synthese 168(1): 81-96.
(IM) 3xf..</> /.\ 3xrp Barnes, Elizabeth (2012). "Metaphysically Indeterminate Existence." Philosophical
Studies 166(3): 495-510.
is not NS-valid. To construct a counterexample, let /vi* be a NS-model with specifi- Boolos, George (1984). "'I'o Be ls To Be a Value of a Variable (or rn Be Some V11lues
cation points sand t where sRt, Dom(s) cc= {a} <rnd Dom(t) {b}. Suppose that of Some Variables)," Journrd ofl'hifosophy 8 l: 430-50.
a*(P, s) = (i) and a*(P,t) {b}. Then, it is mte at s that 3x6•P(x), since •P(x) is Braun, David and Theodore Sider (2007). "Vague, So Untrue," Notis 41(2):
true at sunder t,= {(x,rt)} and trne at t under t, ,..
(/).But 3;,--,P(x) is fi1lse at t and 133-56.
therefore 6 3x•P(x) is false at s. So, 3xf1•P(x) 6 3x•P(x) is untrue in;\//*. Burge, Tyler (1974). "Truth and Singular Terms," Nous 8: 309-25.
Carmichael, Chad (2011). "Vague Composition without Vague Existence," Notis 4 5
(2): 315-27.
B.7 NS and super-vague existence
Carnap, Rudolf ( 1950). "Empiricism, Semantics and Ontology." Reprinted (1956)
In order ro show that negative supervaluationary semantics can model super-vague in Meaning rmd Necessity, 2nd cdn. Chicago: University of Chicago Press: 205-21.
existence, it suffices to show that the following conditions hold: Chalmers, David, David Manley and Ryan Wasserman, eds. (2009). Metametaphy-
sics: New Essays on the Foundations of Ontology. Oxford: Oxford University Press.
Sider-determinacy: for alLVJ*, { tJ . ES,and n? 1: (M *,{t,}sES)JCN\ /"3x(.F-~y)
Donnelly, Maureen (2009). "Mcrcological Vagueness and Existential Vagueness."
anti-Sider-determinacy: for some M and {tJsES and for all n? I: (M*,
Synthese 168(1): 53-79.
{tJ,Es) JC Ns'I" 3x(x = y)
Evans, Gareth (1978). "Can there be Vague Objects?", Analysis 38(4): 208.
The first part, Sider-determinacy, follows immediately from the fact that every Fine, Kit (1975). "Vagueness, 'fruth, and Logic," Synthese 30 (3-4), 265-300.
variable domain frame F* of a NS-model is by definition n-determinate, for every Carson, James (2001). "Quantification in Modal Logic," in Dov Gabbay and Franz
n? 1 (cf Section 2.2). Guenthner (eds.) Handbook of Philosophical Logic, second edition, vol. 3. Dor-
An NS-model satisfying anti-Sidcr-determinac)' can be constructed as follows. drecht: Reidel: 267-324.
Let .M* o= (S, U,R,Dom,a) be a model for a Lt, language without non-logical Heck, Richard (1998). "That There Might Be Vague Objects (So Far As Concerns
constants, where Logic)," Monist 81 (2): 27 4-96.
Keefe, Rosanna (2000). Theories of Vagueness. Cambridge: Cambridge University Press.
• S ~, {s0 ,s 1 ,s2 ,. . ., s,,,. .. } (the specification points)
Korman, Daniel (2010). "The Argument from Vagueness," Philosophy Compms 5
• U = {a,b} (the individuals) (10): 891-901.
• R is the reflexive and symmetric closure of { (s,,,s,, ! 1) },,,, 0 Lewis, David (1986). On the Plum!ity of Worlds. Oxford: Basil Blackwell.
• Dom(so) = {a,b}; Dom(s,, 11) ={a} Lewis, David (1988). "Vague Identity: Evans Misunderstood," Analysis 48(3):
• a(=, s,,) is the identity relation overs,, 128-30.
Now, let {e,},ES be an assignment over ;VI* such that t,0 (y) = b. It is easy to show Linsky, Bernard and Edward Zalta (1994). "In Defense of the Simplest Quantified
that, for all n? I, (M*,s,,, {t,}sES) I= ,ysf" 3x(x = y). Hence, (M*, UJsES) Modal Logic," Philosophical Perspectives 8: Philosophy of Logic and Lt1ngu,1ge,
JC ,ys•f" 3x(x = y), for all 11? 1. Atascadero, CA: Ridgcview: 431-58.
We can conclude that super-vague existence is possible in negative supervalua- Linsky, Bernard and Edward Zalta ( 1996). "In Defense of the Contingently Non-
. . 26 concretc," Phiwsophicril Studies 84(2-3): 283-94.
ttonary semantics.
Nolt, John (2006). "Free Logics" in Dale Jacquette (ed.), Philmophy of Logic.
Ht1ndbook of the Philosophy of Science. Amsterdam: North Holland: 1023-60.
REFERENCES Quine, W. V. 0. (1951). "Two Dogmas ofEmpiricism." Philosophiet1l Review60(1):
20-43.
Sainsbury, Richard (2009). Fiction and Fictionalism. London: Routledge.
Akiba, Ken (2000). "Vagueness as a Modality," The Philosophical Quarterly, 50(200):
Sider, Theodore (2001). Four-Dimensionalism. Oxford: Oxford University Press.
359-70. Sider, Theodore (2003). "Against Vague Existence," Philosophical Studies 114(1):
135-46.
26 Notice, however, that the tautology /" 3x(x y)V•I" 3x(x = y) remains NS- Sider, Theodore (2007). "Nco-Fregcanism and Quantifier Variance," Proceedings of
valid, for all 11. the Aristotelian Society 81 (l ): 201-32.
234 Alesscmdro 7'orw
Sider, Theodore (2009a). "Against Vague and Unnatural Existence: Reply to Liebes-
man and Eklund," Nmis 43(3): 557-67.
Sider, Theodore (200%). "Ontological Realism," in Chalmers eta!. [lOJ: 384-423.
Turner, Jason (2010). "Ontological Pluralism." }011T1litl off'hilosophy 107(1): 5--34.
van lnwagen, Peter (1990). Mt1terial Beings. Ithaca: Cornell University Press.
Varzi, Achille (2007). "On Supervaluationism and Its Logics," Mind 116(463):
633-75.
Williams, Robert (2008). "Multiple Actualities and Ontically Vague Identity,"
Philosophiml Q11t1rter£y 58(230): 134-54.
Williamson, Timothy (1987). "Equivocation and Existence," Proceedings of the
Aristotelim1 Society 88: I 09-27.
Williamson, Timothy (1994). Vt1g11eness. London: Routledge.
Williamson, Timothy (1998). "Bare Possibilia," Erkermtnis 48(2-3): 257-73.
Williamson, Timothy (1999). "On the Structure of Higher-Order Vagueness,"
Mind 108(429): 127-43.
Williamson, Timothy (2005). "Vagueness in Reality," in Michael Loux and Dean
W. Zimmerman (eds.) The O:x.jord Handbook of Metaphysics, Oxford: Oxford
University Press: 690-715.
10
Ersatz Counterparts
Richard Woodward
We find it congenial to talk not only of what goes on at possible worlds, but
also of their inhabitants, the possibilia. Thus we talk not only of worlds
where donkeys talk but ~tlso of talkative possible donkeys. And when we talk
this way, we can think of ourselves as speaking a certain language: the
possibilist language.
We speak the possibilist language to talk about possibilities, by means of
talking about possible things. But we also talk about possibilities in a more
familiar way, by talking about what might have been and what might not, by
talking about what is contingent and what is necessary. Here, we can think
of ourselves as speaking a different language, one that is readily amenable to
regimentation using boxes and diamonds: the modalist language.
These languages share a subject matter, modal space, and the sentences of
each express claims about this subject matter. It's not so surprising, then,
that we can pair sentences of the one language with sentences of the other.
The modalist sentence 'there might have been talking donkeys' can, e.g. be
paired with the possibilist sentence 'there is a world where donkeys talk' and
the modalist sentence 'round squares are impossible' can be paired with the
possibilist sentence 'there is no world where squares are round'. Of course,
there are questions about whether everything you can say in the one
language you can also say in the other, and there are questions about
whether one of these languages is more fundamental than the other. But
set these issues aside.
Our focus is on some delicate issues relating to de re possibilities. When
we tty to pair a modal claim like 'Messi might have won' with some sentence
of the possibilist language, it's natural to appeal to worlds where Messi wins:
worlds which represent Leo Messi as winning. But how does a world
represent Messi as winning? You might think it does so by containing that
man. But nearly everyone thinks that this is false. Even Alvin Plantinga
(1974), a card-carrying believer in transworld identity, thinks that talk of
possibilia isn't to be understood in terms of flesh and blood individuals like
236 Richard Woodward
you and me, but in terms of special abstract objects: individual essences.
Whatever are the constituents of other worlds, it seems, Messi is not one of
them. 1 And David Lewis, who does think that talk of possibilia should be
understood in terms ofobjects like you and me, explicitly tells us that 'nothing
is in more than one world' ( l 968, p. l l 4). So how is de re representation to
be understood?
A natural answer to this question begins by appealing to representatives of
Messi: objects that, though numerically distinct from Mcssi, stand in an
important relation to the man himself. The thought is then that Messi' s
representative, Messi*, ca.n go where he cannot. To sec how this helps,
consider how things play out for Plantinga. Plantinga has an easy time
identifying Messi's representative: it's his individual essence, the property
of being Messi. This property is distinct from Messi and stands in an
important relation to the man himself. Plantinga cannot hold that Messi's
essence wins at any world: essences don't do that kind of thing. But that just
means that he needs to be creative when interpreting sentences like 'Messi*
wins'. Plantinga's preferred option is to take 'wins' to express the property
of being co-instantiated with the property of being rt winner. In this way,
Plantinga can tell a story about how it can be true at a world that Messi wins.
Lewis appeals to representatives too. But his representatives are the real
thing: flesh and blood concrete objects just like Messi. And whereas Plantinga
only has only one representative, Lewis has many. So whereas Plantinga can
hold that Messi' s actual representative is identical to his representatives
elsewhere-it's always Mcssi's essence-Lewis needs a different story.
Lewis's story is familiar. Messi' s representatives are his counterparts:
possible objects that are similar to him in some contextually salient way.
Speaking loosely, we might say that Messi's counterparts in other worlds are
him in those worlds, but Lewis (I 968, p. l l 4) insists that this is not identity
in any literal sense, and that it would be better to say that Mcssi's counter-
parts are the men he would have been, had things been different. But though
Messi's counterparts in other worlds are not reril/y him, the distinctness of
Mcssi and his representatives is not a unique feature of counterpart theory.
Even someone who believes in transworld identity can accept that Mcssi's
representatives are not really him. Plantinga is a case in point.2
1 Phrasing the point in terms of'constituents' is perhaps misleading. It's not clear that
Plantinga believes that possible worlds have constituents in any sense. What's really at
stake here is the semantic point that in a standard Kripke semantics each world is
associated with a domain of possible individuals. And on Plantinga's view, these domains
arc sets of essences.
2 formal defin1t1on of the counterpart relation, but he does his best to explain the notion.
For more on this point, sec Woodward (2012).
Ew1tz Counterparts 237
What is distinctively councerpart-thcoretic, and what transworld identity
theorists cannot accept, is Lewis's account of when something counts as a
representative of Messi. For one thing, whether one object is a counterpart
of another is a context-sensitive matter: what counts as a counterpart in one
context need not count as a counterpart in another. For another thing, the
counterpart relation is not an equivalence relation: though reflexive, it is
neither symmetric nor transitive. And as is well-known, these two features
arc beneficial, allowing us to account for the inconstancy of our de re modal
judgements, solve puzzles generated by the confliccing modal properties of
statues and lumps of clay, and avoid commitment to the idea chat the logic
of de re modality is S5 (see, e.g. Lewis (1968, 1971, 1986)). 3
Is this beneficial package Lewis's and Lewis's alone? Could you jettison
Lewis's robustly realist interpretation of the possibilist language and still be a
counterpart theorist? Those attracted to an 'ersatz' counterpart theory must
answer two questions. They must tell us what counterparts arc, if not the
flesh and blood things that Lewis takes them to be. And once they have told
us that, they must tell us in virtue of what these things are counterparts of
each other. That is, they must tell us two stories: one about the relata of the
counterpart relation <md one about the relation itself 4
l. ERSATZ COUNTERPARTS
The benefits that [just associated with counterpart theory were semantical,
conceptual, and logical in character. The theoty gives us a neat semantic
treatment of de re modal predications; it 'fits' our de re modal concepts and
gives us a nice logic. A further benefit is more metaphysical in character:
counterpart theory allows Lewis to reduce de re representation to de dicto
representation. This point is often lost, partially because there is a tendency
3
There are two ways of doing counterpart-theoretic accountancy, and I have just
played fast and loose with the distinction. Going one way, we might count many relations
<L' counterpart relations and then think that (ideally) one of these is selected in a context,
with the other counterpart relations being ignored or excluded. Going this way, what we
should say is that counterpart relations in general are not equivalence relations. Going the
other way, we might count (in a context) a single relation as the counterpart relation, and
then think that which relation counts as the counterpart relation shifts around from
context to context. Going this way, what we should say is there is no semantic guarantee
that the counterpart relation is an equivalence relation, even though (in some contexts)
the counterpart relation will be an equivalence relation. Thanks to John Divers for
discussion.
4
Question: Can't the actualist refuse to tell a story about the counterpart relation, and
take it a> :1 !?~imitive? Answer: The story needn't be an analysis. Lewis never gives us a
238 Richard Woodward
in the literature to contrast Lewis's theory with 'representational' theories
like Plaminga's. But don't be misled: Lewis needs a story about representa-
tion just as much as anyone else. And he gives us two, one for the de dicto
case and one for the de re.
What a Lewis world represents de dicto--which purely qualitative facts it
represents as holding, which purely general propositions it represents as
being true-is analysed in mereological terms. So, for example, a Lewis
world represents the existence of talking donkeys just in case it contains
talking donkeys and has talking donkey as parts. And once we've settled
what each Lewis world de dicto represents, we can introduce de re represen-
tation by appealing to the qualitative features of the possibilia that populate
Lewis's worlds and the similarity relations that they bear to each other.
Some dislike this way of setting things up. Representation, I'm often told,
is a notion that only has real application in connection with ersatz views of
worlds and doesn't do much work for someone like Lewis. He has his worlds
and that is enough to get his analysis up and running. Well, what's true is
that representation isn't part of the fundamental ideology of Lewis's view:
it's reduced to a combination of mereology and similarity. I find it difficult
to see how this reduction means that representation doesn't do any work
for Lewis, but in any case, Lewis (l 983a, 1986) is clear that a possibility is
not always a possible world: individuals represent possibilities too. And it
bears emphasis that which possibility an individual represents depends on
which counterpart relation we have in mind. Thought of as one of my
counterparts, a given object may represent a possibility for me; thought of
as one of yours, the very same thing may represent a possibility for you.
Moral: forgo words like 'representation' if you want, but don't pretend
that Lewis himself didn't make more than enough room for them, and
don't pretend that Lewis doesn't appeal to the notion of representation to
do important work.
Ersatz counterpart theorists hope to reduce de re representation to its de
dicto cousin too. The first thing to remember is that when the ersatzer speaks
of the actual possible world, she is not speaking about concrete reality but a
special abstract world: the one that is actualized. And concrete reality is the
thing that does the actualizing, not the thing that is actualized. 5 A similar
point appeals to possible individuals. When the ersatzer speaks of actual
possible individuals, she is not speaking about concrete things like you and
me, but about their abstract representatives at the actual world. Real things
like you and me are the things which are represented, not the things which
do the representing. We are not medium but message.
5
See Divers (2002, pp. 228-9) for more on this distinction.
EiJatz Counterparts 239
At this point, the challenge to specify the relata of the ersatz, counterpart
relation becomes pressing. How should it be met? Well, think of all the
purely qualitative predicates that Lionel Messi satisfies. 6 Then think of his
representative at the actual world, Messi\ as the set of those predicates:
Messi* = !x is male, ,\· is a fi:.)otballer, x is Argentinian , .. }
This generalizes to give us representatives for each actual individual. We can
introduce representatives for the merely possible individuals by remember-
ing th.at ~Y. ~rsatzer aims to reduce de re representation to de dicto repre-
sentation. I hmk of what a world represents de dicto as the purely qualitative
sentences that it makes true. Each of the objects that exist at w can then be
associated with the set of qualitative predicates that w says they satisfy. So if
w says that there is a talking donkey called 'Robin', then our representative
will be the set of qualitative predicates that Robin satisfies at w:
Robin* = {xis a donkey, x talks, x has a tail, ... )
In this way, our ersatzer is able to construct representatives for not only
actual individuals, but merely possible ones too.7
When I asked you to think of all of the qualitative predicates that Messi
satisfies, I really did mean all of them. So our set-theoretic representation
doesn't only include predicates like 'xis male' but also ones like 'xis part of
world containing no talking donkeys' and 'x is such that there are exactly n
stars'. As Lewis puts it:
The crsat-l individual is by no means a purely intrinsic description; the description is
extrinsic in a big way. By the time we are done describing an individual completely,
we have en passrmt described the world wherein it is situated. (1986, p. 149)
Our setting thus has an important consequence: it entails that Messi* and
Robin* are worlclbound. 1'his isn't obvious, since we've identified them
with sets of open sentences ;rnd one might think that such things exist at
every world. What's going on?
Notice that to each ersatz individual there corresponds a very complex
qualitative property. Let's say that something plays the Robin-role just in case
it satisfies each of the predicates in Robin*, that is, iff it is a donkey, who
talks, and has a tail, and so on. Our Leo Messi, then, has the qualitative
property of being such that there is no player of the Robin-role. Indeed, for
each role associated with each merely possible object, Messi has the property
1
' See Lewis (1986, pp. 148-50).
7
There are fumously a number of issues regarding the descriptive adequacy of this kind
of ersatz setting, but I think it has become dear that they can be addressed: sec, e.g. Melia
(2001) and Sider (2002).
240 Richard Woodward
of being such that there is no player of that role. Moreover, for each role
associated with each actually existing object, Messi has the property of being
such that there is a player of that role.
The point is that Messi* is bound to the actual world in the sense that
8
only the actu<tl world represents the existence of a player of the Messi-role.
[fa world w represents the existence of a player of the Messi-role, then w
represents that existence of a player of each qualitative role associated with
each actual individual. And this is because part of what's involved in playing
the Messi-role is being worldmates with a player of the me-role, a player of
the you-role, and so on. So if w contains Messi* and thereby represents the
existence of a player of the Messi-role, w must also represent the existence of
a player of the you-role and so must contain you*. Moreover, if w represents
the existence of a player of the Messi-role, then w must also represent the
non-existence of a player of each qualitative role associated with each non-
actual individual. And this is because part of what's involved in playing the
Messi-role is being worldmates with no player of the Robin-role, no player
of the Alice-role, and so on. So if w contains Messi* and thereby represents
the existence of a player of the Messi-role, UJ cannot also represent that there
is a player of the Robin-role and so cannot contain Robin*.
To be clear: this is not to say that only the actual world represents Messi,
nor that there is a bizarre necessary connection between Messi and me, nor
that there is a bizarre necessa1y disconnection between Messi and Robin.
The point is rather that ersatz worlds don't represent the existence of Messi
by representing the existence of a player of the Messi-role. Questions of de re
representation-and in turn questions of de re modality itself-are answered
by looking to the facts about counterparts, just as on Lewis's account. In
essence, all we have is an ersatz surrogate for Lewis's original demand that
nothing is in more than one world.
So much for specifying the relata of the ersatz counterpart relation. What
can the ersatzer say about the relation itself?
2. ERSATZ COUNTERPART RELATIONS
Remember that Lewis explains his counterpart relation in terms of similarity:
whether two things are counterparts depends on whether they are similar to
8 Setting aside worries about whether there could be distinct worlds that were
nonetheless indiscernible in terms of what they de re represent. Compare Lewis (1986,
p. 224) and see Divers ( 1996) for discussion. A few seasons back, the Messi-rolc was called
False Nine.
Ersatz Counterparts 241
each other in some contextually important way. When asked to explain her
counterpart relation, can't the ersatzer just say the same thing?
No. Lewis's explanation has it that two individuals are counterparts iff
those very things ;tre similar in some contextually salient way. If the ersatzer
parroted this, she' cl end up saying that Messi* and Robin* are counterparts
iffthose very things are relevantly similar. And this seems to me to get things
back to front: whether Messi* and Robin* count as counterparts shouldn't,
I think, turn on whether or those two sets-for that's what they are--are
similar. Rather, whether these representations are counterparts should tum
on whether the things they represent are similar. 9
The trouble is that the ersatzer can't accept this. For Robin* represents a
player of the Robin-role, and the ersatz.er doesn't believe that there is any
such thing. The present proposal works when two actual individuals are
counterparts: the ersatzer can hold that whether Messi* is a counterpart of
me* depends on whether he is similar to me in some contextually relevant
way. But the proposal simply breaks down when one of the relata represents
a merely possible individual.
When it comes to explaining the counterpart relation, the ersatzer faces
troubles that Lewis does not. But though she has fewer ontological resources
than Lewis, the ersatzer has extra ideology: primitive modality. Her worlds
are maximally consistent sets of sentences or maximally possible states of
affairs (or whatever), where consistency and possibility are either taken as
primitive or defined in primitively modal terms. So whereas Lewis uses his
plentiful ontology to explain the counterpart relation, couldn't the ersatzer
deploy her extra ideological resources to do the same thing?
This strategy runs into trouble too. Suppose that the ersatzer tried to
explain her ersatz counterpart relation as follows:
Messi* is a counterpart of Robin* iffNecessarily, for any two things x
and y, if x plays the Messi-role and y plays the Robin-role then x is
relevantly similar toy
The problem, as Lewis notes, is that the right-hand side of this biconditional
is trivial. 10 There is no possible world containing both a player of the Messi-
role and a player of the Robin-role. For remember that it's built into the
Messi-role that anything that plays the role is such that there is no player of
9
Sieler (2002, p. 303) agrees. The poinr cm be made a little more vivid by remembering
that we could have associated each qualitative predicate with a unique number, and then
taken Messi* and Robin* to be sets of numbers. But whether Mcssi* and Robin* arc
counterparts shouldn't turn on whether {I, 2, 3, ... } is similar to {4, 5, 6, ... }. It should
turn on wbcther the things that these sets represent arc similar to each other.
10
Lewis (1986, p. 238). Compare Sieler (2002, p. 304).
242 Richrml Woodward
the Robin-role, due to the fact that the ersatz descriptions are 'extrinsic in a
big way'. Our modal explanation of the counterpart relation is a failure and
the ersatzer seems to be running out of options pretty quickly.
Cfhe problem with the modal analysis considered above arises because it
requires an intra-world similarity relation between a player of the Messi-role
and a player of the Robin-role. And so one might naturally think that the
counterpart theorist should require an inter-world similarity relation instead.
That is, one might think that the counterpart theorist should endorse the
following modal analysis of the counterpart relation:
Messi* is a counterpart of Robin* iff Necessarily, for any thing x, if x
plays the Messi-role then, necess<uily, for any thing y, if y plays the
Robin-role then x is relevantly similar toy
Certainly this analysis fares better than its predecessor since its adequacy
does not require the existence of a single world containing players of both
roles. But the result comes at a price: it commits the ersatz counterpart
theorist to primitive modality that is de re in character. And remember that
the counterpart theorist's project, as it is being conceived here, is similar to
Lewis's: she aims to reduce representation de re to representation de dicto. In
this setting, the appeal to primitive de re modality is off-limits. To put that
point otherwise, the counterpart theorist aims to reduce the de re modal facts
to facts about ersatz counterparts and and our present question is whether
these latter facts can themselves be further reduced. Within this context, it is
clear that the ersatz counterpart theorist cannot explicate the ersatz coun-
terpart relation in the manner suggested.)
At this stage, it's worth repeating a point that I made earlier. Even if we grant
that no world contains both a player of the Messi-role and a player of the
Robin-role, we are not thereby forced to deny the Humean thought that
anything can co-exist with or without anything else. For one thing, the ersatz
counterpart theorist, like Lewis, thinks that the question of whether a could
exist without b turns on whether or not there is a world containing a counter-
part ofa but not b. For another thing, and as Phillip Bricker (2001) stresses, the
Humean denial of necessary connections is best articulated in terms of dupli-
cation: there are no necessary connections between distinct things iff for any
two distinct things, it is possible for a duplicate of the one to exist without a
duplicate of the other. Cmcially, however, duplication is cashed out in terms of
intrinsic properties: x is a duplicate of y iff x and y have the same intrinsic
properties. Even if the extrinsic information contained within the Messi-role
and the Robin-role means that there is no world containing both a player of the
Messi-role and a player of the Robin-role, the Humean picture is in place as
there are worlds containing duplicates ofMessi and Robin: things which satisfy
the intrinsic information contained within the Messi-role and the Robin-role.
Emuz Co11nterp11rts 243
Next, consider the following explanation of the counterpart relation:
Messi* is a counterpart of Robin* iff Necessarily, for any two things x
and y, if x is a Mcssi-·duplicate and y is a Robin-duplicate, then x is
relevantly similar toy.
This explanation is better than the original modal analysis insofar as some
worlds contain both Messi-duplicates and Robin-duplicates. The problem is
that it is crucial to counterpart theorists that things can count as counterparts
because they are similar in highly extrinsic respects: you may count as a
counterpart of mine by being twelve feet from a cat, for instance. Be that as it
may, the explanation nonetheless hints at how the counterpart theorist can
explain her relation. For what we have in effect done is restricted the Messi-
role and the Robin-role so as to ensure that there is a world containing players
of both roles. This idea was a good one, but by focusing solely on the intrinsic
information contained within the roles, we went too far. What we are after,
then, is a way of constructing worlds that contain copies of Messi and Robin,
and we know that 'copying' needs to be handled carefully. If we are too
libe~al, our resultant explanation of the counterpart relation will go trivial,
but rf we are too conservative, our explanation will be overly restrictive.
At this point, we need to make an assumption about the extent of modal
space, '.1.nd grant that disjoint spacetimes are possible so that some worlds represent
the exrstence of a plurality of disjoint 'island' universes. I take it that this
assumption is legitimate in context: even Lewis accepts that it is a cost of his
account that it cannot allow for the possibility of disjoint spacetirnes. Lewis of
course thinks the cost is negotiable and outweighed by other lovely features of his
view but, be that as it may, the point is just that it is generally accepted that it is a
benefit of ersatz proposals that they can allow for the possibility of disjoint
spacetimes. Moreover, suppose that w 1 and w 2 each represent the existence of a
single univ~rse. Then I assume that there is a third world w 3 that represents the
existence of two disjoint spacetimes v1 and v2 such that v1 is a duplicate of w 1
and v2 is a duplicate of Wz. 11 Figure l 0.1 shows what we have.
Figure I 0.1 The copying process
11
This is the 'hefty metaphysical assumption' mentioned in Sider (MS). I'll mount a
defence of the assumption in due course.
244 Richard \,Yloodwrtrcl
Now, if w 1 contains a player of the Messi-role and w2 contains a player of
the Robin-role, then w3 will contain duplicates of each. The trick is now ro
understand the arrows in such a way as to ensure that the extrinsic infor-
mation about Messi and Robin carries over into w3 too.
Theodore Sider (MS) suggests a neat way to pull off this trick. Earlier
we saw that an object o plays the Mcssi-role only if o satisfies the open
sentence 'x is such that there arc no talking donkeys' and that an object o*
plays the Robin-role only if o* satisfies the open sentence 'x is such that
there are talking donkeys'. Suppose, however, that we restrict the quanti-
fiers of these open sentences by different predicates. In our toy example,
say that an object is on the right iff it is spatio-temporally connected to a
talking donkey and that an object is on the left iff it is not on the right.
Then even though it is impossible for there to be two objects o and o* that
satisfy the predicates 'x is such that there are no talking donkeys' and 'xis
such that there are talking donkeys' respectively, it is possible for there to
be two objects that satisfy the predicates 'x is such that there are (on the
left) no talking donkeys' and 'xis such that there arc (on the right) talking
donkeys' respectively.
Moreover, suppose that we restrict all of the quantifiers of all of the open
sentences that arc members of Messi* by the predicate 'is on the left' and all
of the quantifiers of all of the sentences that are members of Robin* by the
predicate 'is on the right'. This allows us to delineate two complex qualita-
tive roles: the Messi-copy role and the Robin-copy role. We can then explain
the ersatz counterpart relation as follows:
Mcssi* is a counterpart of Robin* iffNecessarily, for any two things x and
y, ifx is a Messi-copy and y is a Robin-copy then xis relevantly similar toy.
Given the assumption about the possibility of disjoint spacctimcs, that for any
two worlds w 1 and w 2 there is a third world w 3 containing copies of Wt and w2,
it turns out that the right-hand side is non-trivial. But the copying exports into
w, not only the intrinsic information contained within Mcssi* and Robin* but
the extrinsic information too: if Wt represents the player of the Messi-rolc as
being twelve feet from a cat and w2 represents the player of the Robin-role as
being twelve feet from a C.1.t, then w3 will represent both the Mcssi-copy and the
Robin-copy as being twelve feet from a cat. In these ways, our explanation of
the counterpart relation avoids the problems that afflicted its predecessors.
Before we move on, I want to forestall a possible misunderstanding about
the role that disjoint spacctimes play within the proposed account. The
worry can be illustrated if we suppose that Mcssi could have been lonely and
thereby could have existed in isolation from any distinct thing. On the
present account, representing this possibility requires that Messi* has a
lonely counterpart-call this counterpart Loner*. And Messi* and Loner*
Ersatz Countetparts 245
being ~ountcrp<:ns requires_ (in part) that there is a world w which represents
the existence of a player of the Mcssi-copy role and a player of the Loner-
copy role. One might now worry that this means that we have in fact not
represented the possibility of Messi being lonely; what we have instead
represented is the possibility of Messi and Loner co-existing.
What is true on the present proposal is that Messi* and Loner* being
counterparts requires (in pan) that there is a world w which represents th~
co-existence of players of the Mcssi-copy role and the Loner-copy role. But
even though the theory says that whether Messi* and Loner* arc counter-
parts depends on what goes on in worlds containing copies of both,
representing the possibility of Mcssi being lonely docs not require that any
player of the Loner-copy role is lonely. What is instead required is that Mcssi*
has a counterpart who is lonely*. And given (a) that Loner* and Mcssi* arc
counterparts and (b) that Loner* is lonely* insofar as any player of the Loncr-
rolc is lonely, we thereby represent the possibility ofMessi being lonely. So
note that even though the theory appeals at a crucial stage to the existence of a
world w _chat represents the de clicto possibility of the copies co-existing, w
does not itself represent the cle re possibility ofMessi being lonely. Given that
1\1~ssi* ar:d Loner* arc counterparts, the worlds which represent that possi-
bility arc mstcad those which contain a player of the Loner-role.
N_ow, as Sider (MS) points out, the explication of the ersatz counterpart
relauon that we arc considering requires a rather hefty assumption about
what possibilities there arc. But as hefty as it is, the assumption seems well
motivated given a Humean picture according to which there arc no neces-
sary connections between distinct things.
3. THE PRINCIPLE OF SOLITUDE
Intuitively, there are necessa1y connections between distinct things just in
case there is an x such that, necessarily, x coexists with some y that is distinct
from x. But as Bricker (2001, p. 37) points out, this claim is ambiguous.
Ifwc treat 'some y' as having wide-scope, we get one anti-Humcan principle:
3x3y (xis distinct from y /\ D (x exists ::J y exists))
But if 'some y' has narrow-scope, we get another anti-Humcan principle:
3xO(x exists ::J 3y(x is distinct fromy)
(These principles arc somewhat sloppy: remember that what's really at stake
is not the necessity of x coexisting with y but the necessity of a duplicate of x
coexisting with a duplicate of y. I'll take this qualification as read.)
246 Richard Woodward
Now, when Lewis (1986, p. 88), as a good Humean, says that anything
can fail to coexist with anything else, he is denying the first principle
above. Take any two distinct things, and it is possible for (a duplicate of)
the one to exist in isolation from (a duplicate of) the other. But that docs not
entail that, for any x, it is possible for (a duplicate of) x to exist all by itself.
So one could deny the first anti-Humean principle whilst accepting the
second.
With Bricker, I think that rejecting only the first anti-Humean principle
is unattractive: its denial is too weak to capture the full range of possibilities.
Consider a red ball and a blue candle. Given that we are rejecting the first
anti-Humean principle, we know that it is possible for the red ball to exist
without the blue candle. Maybe the blue candle is replaced by a blue pencil?
But, intuitively, it is also possible for the red ball to exist without anything
blue existing. Maybe the blue pencil is replaced by a green one? But surely it
is also possible for the red ball to exist without any other coloured thing
existing, and also possible for the red ball to exist without any other extended
thing existing, and also possible for the red ball to exist without .... We are
sliding down a slippety slope, and the denial of the second anti-Humean
principle lies at the end of our journey.
To deny the second of the anti-Humcan principles outlined above is to
endorse the Principle of Solitude:
Vx<>(x exists !\•3y (y is distinct from x))
Properly understanding this principle requires us to be quite specific about
the range of the quantifiers. After all, we do not want it to turn out that the
Principle of Solitude rules out the necessary existence of things like numbers
and sets, and to avoid this result it's crucial that the quantifiers are taken to
range only over objects that are located in space and time. Moreover,
existing in insolation is not quite the same as being surrounded by empty
spacetime. In the world where Messi exists all by himself, spacetime has
whatever shape Messi has.
Even within a Lewisian setting, endorsing the Principle of Solitude does
not by itself deliver the possibility of disjoint spacetimes. For even though
the quantifiers range only over objects that are located in spacetime, it is still
natural to think that the quantifiers range over what John Divers (1999)
calls ordinary individuals: individuals that are wholly located in a single
world. But suppose that we understood the quantifiers in the Principle of
Solitude as ranging over not only ordinaty individuals but also over extm-
ordinrtry or transworld individuals that are partially located in different
worlds. Call this the Generalized Principle of Solitude, or GPS for short.
Then the possibility of disjoint spacetimes follows immediately by instan-
tiating the quantifier to any transworld individual.
Enatz Counterparts 247
Now, Bricker's defence of GPS is based on the thought that there should
be no restriction on what can be actualized. He invites us to look at things
from God's perspective. To insist that only single worlds can be actualized
'would be to put a limitation on God's power to choose, one not grounded
in any logical necessity' (2001, p. 38). He continues:
Suppose that, in surveying the worlds prior to actualization, Cod found that two or
more worlds were tied for best. Why must God choose between actualizing one
world, or the other? He's all-powcrfol! He can simply say: 'Actualize those!'
It's important to see that Bricker isn't proposing to redefine the Lewisian
concept of a world so that there arc worlds which contain spatiotemporally
disconnected parts. Rather, what he is in effect proposing is that there is
another sense in which Lewis was right to say that a possibility is not always a
possible world: a possibility is sometimes some possible worlds, where the
'some' is plural. Two worlds taken together can represent the possibility of
disjoint spacetimes even if no world taken in isolation can do the same thing.
So whilst embracing Bricker's picture might force the Lewisian to tweak her
analysis of modality, the Lewisian concept of a world can remain fixed.
Though Bricker mounts his defence of GPS in a Lewisian setting, his
point can be developed in an ersatz setting too. But we need to tread
carefully. To begin, note that there is a certain sense in which it doesn't
matter, from (Bricker's) Lewisian perspective, which worlds are actualized.
Consider two scenarios. In the first, God actualizes a single world containing
talking donkeys; in the second, He actualizes a single world containing no
talking donkeys. Either way, it is still true that there are, quantifiers wide
open, talking donkeys: the scenarios only generate different answers to the
question of whether talking donkeys are actualized. That is just to say that
when we assume a possibilist metaphysic, the class of things which exist
simpliciter includes but is not exhausted by the class of things which are
actual simpliciter. Actualization will therefore make a metaphysical differ-
ence, not an ontological one.
When we assume an actualist metaphysic, however, the class of existing
things just is the class of actual things. So it makes an ontological difference
whether God actualizes a world where donkeys talk as opposed to a world
lacking talking donkeys. [n the former scenario, God makes it the case that
there are, quantifiers wide open, talking donkeys; in the second, He makes it
the case that there are, quantifiers wide open, no talking donkeys.
Next, consider a third scenario in which God actualizes two worlds, one
which contains talking donkeys and one which doesn't. This scenario is
troubling for the actualist in a way that it is not troubling for the possibilist.
For in actualizing a world where donkeys talk, God seems to make it the case
that there are, quantifiers wide open, talking donkeys. But in actualizing a
248 Richmd Woodward
world containing no talking donkeys, God seems to make it the case that
there are, quantifiers wide open, no talking donkeys. But God can't make it
the case that there are and are not talking donkeys.
A second difference between the Lewisian setting and the ersatz setting is
that Lewisian worlds are concrete objects whereas ersatz worlds are at best
abstract representations of concrete objects. Suppose that God looks down
on the worlds (be they concrete or abstract) and finds that two of them, WI
and Wz, are equally good candidates to be actualized. If the worlds God
looks down upon are Lewisian, He cannot tty to solve His problem by
creating a third world containing copies of WI and 10 2 . Lewisian worlds are
spatiotemporally unified by definition, after all. Moreover, He doesn't need
to do this: I-le can just take w1 and w2 together and actualize both. But if the
worlds He looks down on are abstract representations of worlds, God can tty
to solve His problem by creating a third world that represents the existence
of two disjoint spacetimes. Nothing in the definition of abstract worlds
requires that they represent spatiotemporally unified wholes. Moreover,
there is at least a prima fade case for thinking that God has to solve His
problem in this way: for, as we saw, if God were to actualize both w 1 and w 2 ,
there is a worry that He would make a contradiction the case, by making it
the case both that talking donkeys exist and that talking donkeys don't exist.
Time to take stock. We have seen that there are good reasons to think that
Humeans should accept the Generalized Principle of Solitude. And we have
also seen that CPS suggests that it is possible for more than one Lewis world
to be actualized simpliciter. Exactly how to transpose CPS into an ersatz
setting is a delicate issue, however. Suppose that WI and w 2 are ersatz worlds
that represent the existence of single unified spacetimes v1 and v2 • Then
CPS does not require that it be possible for both WI and w 2 to be actualized:
CPS will be upheld if it is possible for both VI and v2 to coexist in isolation
from each other. This would be underwritten if the ersatzer were to accept
that it is possible for more than one ersatz world to be actualized. But there is
a dear logical problem with this idea, and the ersatzer doesn't need to pursue
this strategy in any case. She can instead hold that CPS is satisfied because
there is a third world, w 5 that contains copies of v1and v2 •
Insofar as the ersatz counterpart theorist is a Humean, then, I submit that
CPS is both well motivated and supports the hefty metaphysical assumption
required to underpin our explanation of the ersatz counterpart relation.
(Question: What if the logical problems can be solved? Won't it then turn
out that CPS is underwritten not by the hefty metaphysical assumption
required for the explanation of the counterpart relation but by the possibility
of more than one ersatz world being actualized?
Answer: Suppose that w 1 represents the existence of a player of the Messi-
role and w 2 represents the existence of a player of the Robin-role. Even if the
Ersatz Counterparts 249
logical problems can be solved, so that 10 1 and w 2 can be jointly actualized, it
does not follow that it is possible frir a player of the Messi-role to coexist with a
player of the Robin-role. Indeed, the set of sentences that would have been
true had W1 and W2 been jointly actualized is seemingly identical to the set of
sentences that would have been true had w,
been actualized, where w 1 is the
world constructed by the copying process described earlier. In this set;se, the
possibilities generated via multiple actualization just are the possibilities
generated by the copying process. Put otherwise, there is no difference,
within an actualist metaphysic, between actualizing w 3 and jointly actualiz-·
ing WI and Wz. Either way, we get the same range of possibilities required to
underpin our explanation of the ersatz counterpart relation.)
4. SEMANTICS VERSUS METAPHYSICS
Ersatz counterpart theorists hope to enjoy the benefits of counterpart theo1y
without footing the ontological bill. But Trenton Merricks (2003) has
argued that ersatz counterpart theories simply don't work. He is owed an
wswer, and I shall give mine. 12
Merrick's criticisms, surprisingly, focus not on the ersatzer' s story about
the counterpart relation itself, but on her story about its relata. He, objects
that the method by which we constructed ersatz individuals is just one of
many. Instead of taking possibilia to be sets of open sentences, we could
equally well have associated each sentence with a number and taken ersatz
individuals to be sets of numbers. Or we could have used different sentences,
written in a different language. But remember that the theory reduces the fact
that Messi is possibly Fto facts about Messi* and his counterparts. That Messi
is possibly F might, for example, be reduced to something like:
There is some set of open sentences x such that xis a counterpart of
Messi* and x is F*
(Where xis F* iff it has 'xis F' as a member.) But why, Merricks wonders, is
this any better than reducing the target modal fact to facts about some other
set-theoretic candidate? The multitude of candidates is thus problematic as
it 'absurdly implies that no single analysis of a modal property is better than
some incompatible analyses' (2003, p. 533). 13
12
Mcrrick~'s ri_i~i1_1 target is the ersatz counterpart theory developed by Heller (1998),
and some of hrs cnt1c1sms focus upon quite specific details of Heller's proposal. But others
arc ~no re gcner'.11, and can be levelled against any ersatz counterpart theory.
1
' Worse still, there 1s a qualitative problem: ersatz counterparts arc not cut out to
represent the possibilities that they arc designed to represent. For they fail to represent
250 Richmd \Voodward
Notice that Merricks thinks of the counterpart theorist as offering us an
analysis of modal properties like being possibly F. But even if we grant this,
it's far from clear exactly what kind of analytic project she's engaged in. And
to see that, just observe that philosophers use possible worlds and individ-
uals for a variety of different purposes. lli
Most obviously, some philosophers cast possible worlds and individuals
in a foundational role, and this is precisely the key role that counterparts
play in their original Lewisian setting: on Lewis's view, to be possibly F just is
to have a counterpart who is F. Perhaps the ersatz counterpart theorist
agrees. If so, let's say that she is deploying counterpart theoty to give a
metttphysical analysis of de re modality.
Possible worlds and individuals aren't only used to ground modal truths:
they're also used as semantical tools. Thus possible worlds can be used to
specify the semantic values of modal operators, functions from worlds to sets
of objects can be used to specify the intensions of predicates, and sets of
worlds can be used to specify the conditions under which a sentence is true.
When we give a semantical theory of a language which incorporates claims
such as these, we need entities for our (metalinguistic) quantifiers to range
over, and possible worlds, individuals, and other things constructed out of
them can serve as those entities. Perhaps the ersatz counterpart theorist sees
her counterparts as playing this role. If so, let's say that she is deploying
counterpart theory to give a semantical mzalysis of de re modality.
It's important to realize that it is one thing to hold that ersatz counter-
parts play a foundational role and quite another to hold that they play a
semantical role. This point isn't new: Allen Hazen (1979, p. 319) made it
over thirty years ago, telling us that the question of what grounds de re
modal truths is 'irrelevant' to the assessment of counterpart theory qua
semantic theoty. There has been a tendency to overlook this, however, a
tendency traceable to the fact that counterparts play both roles within the
context of Lewis's modal philosophy. But the two roles should be sharply
distinguished, and just as one might use possible worlds in one's semantics
without thinking that what it is for 4J to be possible is for </> to be true at
some possible world, one might use counterparts in one's semantics without
intrinsically since, being sets, they 'just sit there' (p. 535). Good candidates, Merricks
thinks, must have their representational properties 'in and of themselves'. I won't say
much about this worry here, since I think that it's been adequately dealt with by Sider
(MS). The basic point is that what matters is that Messi* represents Messi re!tztive to the
structure in terms of which the counterpart theorist's semantic theory is given. And this
relative notion of representation is perfectly objective, even if there is no 'intrinsic' sense in
which Mcssi* represents Mcssi.
14 Herc I agree with Agustin Rayo (2012), from whom the terminology introduced
below is borrowed.
h:ratz Counterparts 251
thinking that de re modality can be metaphysically analysed in counterpart-
thcoretic terms.
Once foundational issues are divorced from semantical ones, Merricks's
complaint can be understood in two ways: either as a complaint about the
adequacy of ersatz counterpart theory qua semantical analysis or as a
complaint about its adequacy qua metaphysical analysis.
The latter complaint can be answered quickly. For whether or not some
ersatz counterpart theorists are aiming to provide a foundation from which de re
modality can be constructed, that's certainly not the role which ersatz coun-
terparts play within the theoty developed earlier. I distinguish between the
counterpart* relation, which holds between ersatz individuals, and the counter-
part relation, which holds between the things represented by ersatz individuals.
What's true is that the theory tells us that 'Messi is possibly F' is true iffMessi*
has a counterpart* who is F*. But that's not the end of the story: whether this
truth condition obtains depen(fa on whether certain de dicta modal facts obtain.
And so what's ultimately going on, foundationally speaking, is that we have a de
dicto reduction of de re modality. Facts concerning Messi* and his counter-
parts* are, in this setting, not being appealed ro in order to provide a foundation
to which de re modality can be reduced. The foundational version ofMerricks' s
objection simply doesn't arisc. 15
What ifMerricks is objecting to the deployment of ersatz counterparts in
a semantical analysis of de re modality? Here, it's worth remembering that
the distinction between foundational and semantical projects wasn't lost on
Lewis. He tells us early on in Plurality that when we are doing semantics:
... we need no possible worlds. We need sets of entities which, for heuristic guidance,
may be regarded as possible worlds, but which in truth may be anything you please. We
are doing mathematics, not metaphysics. ( 1986, p. 17)
So even Lewis-arch enemy of all things ersatz-is carefol not to complain
that sets of sentences (or other abstract whatnots) are incapable of serving as
adequate semantic values within an intensional semantics. His complaint is
rather that ersatz constructions are inadequate tools when our project is
metaphysical in character, and the anti-ersatzer diatribe in Plurality is meant
to convince us only that ersatz constructions are not up to the foundational
15
As Jason Turner pointed out to me, there is a h1rther aspect to this issue. When we
accept a certain metaphysical analysis, there might be different routes to the analysis.
Ultimately, my counterpart theorist offers a de dicto analysis of de re modality. Bur the
route to the analysis is indirect: we first analyse de re modality in terms of counterpart*
relations and then offer a de dicto analysis of counterpart* relations. That's not the only
route: I could have instead analysed de re modality in terms of counterpart** relations and
then offered an analysis of those. But the end result is the same either way, and the fact
that there arc different routes to the ultimate analysis isn't important.
252 Richtlld Woodwtlld
task. Their semantic adequacy is granted from the beginning, and Lewis
(l 986, pp. 143-4) is careful to point out that their foundational inadequa-
cies need not render ersatz possibilia inadequate semantical tools. 16
But hang on, how can the fact that there are different ways to construct
ersatz possibilia not have a negative effect on the semantic project? Won't we
end up accepting that no single semantical analysis of a language is better
than some incompatible an:tlysis based on some other way of constructing
ersatz possibilia? Isn't that what the semantical version of Merricks's worry
amounts to?
}low we answer these questions will depend on how we conceive of the
explanatory ambitions of semantic theorizing. 17 On what we can think of as
the 'folklore' conception of (model-theoretic) semantics, the model theory
ranges over a class of interpretations of a language, amongst which we hope
to find the 'intended' one: the one that 'gets the interpretation right'. The
role of a semantic theory emerges as that of providing a constitutive account
of both logical consequence (truth-preservation under every interpretation)
and truth simpliciter (true on the intended interpretation). But this con-
ception of the explanatory ambitions of a semantic theory is not mandatory.
On what we can think of as the 'instrumentalist' conception of semantics-
a view recently defended by Hartry Field (2008)-the goal of a semantic
theory is not that of providing constitutive accounts of truth or consequence
but rather that of providing an extensionally correct account of logical
consequence. What we want to know is what follows from what? and, for
the Fieldian instrumentalist, any value that we might associate with a
semantic theory lies in its provision of extensionally correct answers to this
question. (Which is a relief, given that Field's take on logical consequence is
that it is best viewed as a primitive notion, meaning that semanticists trying
to provide a constitutive account of it are barking up the wrong tree.)
If the explanatory ambitions of semantic theorizing are understood along
instrumentalist lines, the existence of many equally good semantic theories
is harmless. For the goodness of a semantical analysis (of a language l)
depends, from the instrumentalist's point of view, on how well that analysis
16
A similar point is made in a different context by Sider (1996, §3), who points out
that the fact that ordered pairs can be constructed in different ways isn't a problem unless
the constructions are being used to give a metaphysical analysis of relations. But that
doesn't mean that other uses of pairs arc illegitimate because on those uses no metaphys-
ical significance is claimed: the pairs arc only used to get a job done.
17
This way of putting the point is perhaps a little tendentious. Rather than think that
there is this single project, semantics, and a debate about its cxplanat01y ambitions, one
might instead think that there arc a bunch of different projects, each of which might be
called 'sem;rntics' and each of which has different explanatory ambitions. Compare the
discussion of the role Pbmtinga's semantics serves in Woodward (201 l).
Hrsrltz Counterparts 253
tracks the extension of the consequence relation (of[). And so the existence
of more than one equally good semantic theory would only show that we
have more than one way to provide an extensionally accurate specification of
logical consequence. And the mere fact that the entities these semantic
theories quantify over are not intrinsically suited to represent possibilities
does not prevent them from serving this instrumental end. We are, after all,
doing mathematics, not metaphysics.
If we conceive of semantic theorizing as being more ambitious-if we
accept something like the folklore conception-then the instrumemalist's
reply to Merricks is unavailable: now the goodness of a semantic theory
ar'.1ounts to more than its mere ability to offer correct answers to questions
of consequence. But saying that isn't saying very much: there is still a
question of what the ambitions of semantics are. Lewis's (1986,
pp. 40-50) discussion of the role of possible worlds and individuals in the
an;tlysis of language is particularly revealing in this regard. 18
Lewis's starting point is that language is used to convey information: you
know something; you want me to know it too; I take you to my source of
knowledge and to speak truthfully; via this mechanism I come to possess the
piece of information you wanted me to possess. But though I rely on you to
speak truly, the words you use will be true on some interpretations and false
on others. The right interpretation is the one that specifies those conditions
under which you are indeed truthful and my trust is well-placed. So a
semantics must associate each sentence of a language with a truth condition,
conceived of as a set of worlds. And the assignment must be systematic, and
tell us 'which speakers at which times at which worlds are in a position to
utter which sentences truly' (p. 40).
This means that we must assign, by finite means, truth conditions to an
infinite number of sentences, and Lewis's story about how this can be done
is familiar: we list the vocabulary of the language, assign to each expression a
syntactic category and a semantic value, list the rules for building new
expressions from old ones in such a way as to ensure that the semantic
values of the new are a function of the semantic values of the old, and specify
truth conditions for sentences as a function of the semantic values of their
subsentential constituents.
Lewis then gives us a job description for semantic values, and it's at this
point where things get interesting. Semantic values are there to do two
things and two things only: to generate other semantic values and to
18
Even before we get into the details, we can see that there are crucial diHerenccs
between the Ficldian picture and the Lewisian one. For whereas Field thinks that semantic
theories serve an instrumental role in the analysis oflogical consequence, Lewis thinks that
they serve a genuinely explanatory role in the analysis of linguistic communication.
254 Richflrd Wooclwflrd
generate truth conditions for sentences. The object, Lewis (p. 41) insists, is
not that we should find entities capable of deserving names 'from the
established jargon of semantics', but just that we find entities capable of
doing these two things. It's for this reason that Lewis uses the colourless
term semflntic value rather than a more established piece of jargon like referent.
After all, the thought that my name refers to me rather than to a set of open
sentences is a Moorean fact. But 'there is no reason not to say both that my
name has me as its referent and also that it has a certain [set] as its semantic
value' (p. 42). So long as a set of open sentences carries enough information to
generate other semantic values, it will be deserving of the name.
Notice, then, that the mere observation that semantic values come cheap
wouldn't have bothered Lewis a great deal. For semantic theories are judged
to the extent that they associate sentences with the right truth conditions
and so long as the truth conditions are generated systematically, which
entities are used as semantic values is unimportant. In a sense, subsentential
matters arc irrelevant if sentential questions are answered systematically.
As Lewis put it much earlier: 'Semantic values may be anything, so long as
their job gets done' (1980, p. 26). And again, their only job is to generate
semantic values and truth conditions. 19
This isn't instrumentalism. As far as the Fieldian instrumentalist is
concerned, a semantic theory can assign to sentences whatever truth condi-
tions it likes so long as it answers questions of consequence correctly. And to
see that, note that the notion of an 'intended' interpretation doesn't play any
role for the instrumentalist. On the Lewisian picture, by contrast, semantics
does aim to specify the conditions in which speakers are in a position to utter
a sentence truly and a semantic theo1y which assigned silly truth conditions
would be a silly theory even if it delivered extensionally correct answers to
questions of consequence.
What is true, admittedly, is that the Lewisian picture involves an instru-
mentalism of sorts about semantic values: semantic theories which assign
silly semantic values might still assign the right truth conditions after all.
True enough, we might have pause for thought were we to think that the
semantic project was to specify something like 'what we were referring to all
19
Doesn't naturalness constrain subsentcntial interpretation too? Well, Lewis (1984)
does accept naturalness as a constraint on interpretation, but he prefaces that entire
discussion by conceding to Putnam that semantics takes ordinary things like you and
me to be semantic values. If the relevant pages of Plurality (pp. 40-50) are anything to go
by, Lewis doesn't believe this for a second. And though Lewis docs mention naturalness in
relation to his preferred convention-based approach to language-see, e.g. Lewis_
( 1992)-it's far from clear that he endorses anything like the magnetic conception of
subsentential reference that he's ofren associated with (cf. Schwai-L (2014)) and
Wcathcrson (2013)). Thanks to Robbie Williams for discussion of these issues.
Enfltz Counterpflrts 255
along'. But it's hard to see what the problem is once the notion ofa semantic
value is divorced from more everyday semantic jargon. Moreover, even if we
are happy with something like the picture Lewis sketches, it's still an open
question exactly how instrumental our account of semantic values needs to
be. For instance, even if we arc carefol to distinguish the referent of a term
from its semantic value, it doesn't automatically follow that there is no
connection between the two. Consider, for example, the values assigned to
ordinary names in a Momagovian setting: sets of properties. Though these
sets clearly aren't what ordinary uses of proper names refer to, they can be
used to specify ordinary referents: thus whereas the semantic value of 'Leo
Messi' is a set of properties, that name refers to the object that instantiates
those properties. And this point has an analogue in the context ofMcrricks's
complaint against ersatz counterpart theory. For though the present picture
involves the thought that 'Leo Messi' has a set of open sentences as its
semantic value, we can at least define up an intuitive referent in terms of it:
the name refers in the ordinaty sense to that object which in fact satisfies all
of those open sentences.
These issues are rich and deserving of more attention. But I hope that
they can be set aside in the present context. For remember that Merricks
objected to ersatz counterpart theories on the grounds that there is no good
reason why ersatz individuals should be identified with sets of open sen-
tences rather than something else. But whether the fact that ersatz individ-
uals can be constructed in many ways is problematic depends on which roles
they are playing within our account of de re modality. Perhaps the problems
Mcrricks sees are genuine if the ersatzer were trying to deploy ersatz
counterparts to provide a metaphysical reduction of de re modality. But
on the version of the view I've defended, ersatz counterparts are not playing
a foundational role. Rather, ersatz counterparts are semantical tools, to be
used as semantic values. And the semantic adequacy of ersatz counterparts is
neither vitiated by the fact that they can be constructed in many ways nor by
the fact that they fail to be intrinsically representational. What matters is just
that they get the job done.
CONCLUSION
Lewis (1986, p. 4) calls his realist account of modal space 'a paradise for
philosophers' and challenges his opponents to deliver the theoretical benefits
that realism offers more cheaply. Whether she can enjoy all of the benefits of
Lewisian realism is doubtfol, but I contend that the ersatzer can enjoy those
associated with his treatment of de re modality. For the counterpart-theoretic
256 Richard Woodward
analysis of de re modality Lewis oflers can be made acceptable by the lights of
those of who reject Lewis's ontology. ·ro those of us who find counterpart
theory attractive, this should come as something of a relief.
Richmd Woodwmd
University ofHmnburg
ACKNOWLEDGEMENTS
Thanks to Elizabeth Barnes, Ross Cameron, Sharnik Dasgupta, John Divers,
Thomas Kri:idcl, Ted Sider, Tatjana von SolodkofF, Jason Turner, Isabel van dcr
Linde, Barbara Vetter, Jennifer Wang, and Robbie Williams. Earlier versions of this
paper were presented at the U niversitics of Barcelona, Leeds, and Hamburg, and at
Humboldr-Univcrsitat zu Berlin-many thanks to my audiences on those occasions
for discussion. My research on this paper was conducted within the context of the
DFG Emmy Noether Research Group Ontology After Quine and was supported by
my involvement in the Nature of Assertion: Comequences for Re!Ativism and Fiction-
tt!ism project (FFI2010-169049); the Vagueness t1nd Physics, Metaphysics, rtnd Mett1-
Metr1physics project (FFl2008-06153); and the l'ERSP-Philosophy of Perspectiuri!
T'lioughts and Facts project (CSD2009-00056). Many thanks to the DFG, the
DGI, MICINN, and the Spanish Government for supporting these projects.
Ersatz Counterparts 257
Lewis, David. 1983a. 'Individuation by Acquaintance and by Stipulation'. The Phih;-
sophiml Reuietu 92(1): 3-32.
Lewis, David. 1984. 'Putnam's Paradox'. A11stmh1sim1 jounuz! of Philosophy 62(.3):
221-36.
Lewis, David. 1986. On the Plurality of Worlds. Oxford: Basil Blackwell.
Lewis, David. 1992. 'Meaning Without Use: Reply to Hawthorne'. Austrahisirm
joumal of Philosophy 70(1): 106-10.
Melia, Joseph. 200 l. 'Reducing Possibilities to Language'. Anil/ysis 61 (269): 19-29.
Merricks, Trenton. 2003. The End of Counterpart Theory'. }01m1r1l of Philosophy
100(10): 521-49.
Plantinga, Alvin. 1974. The Nature of Necessity. Oxford: Oxford University Press.
Rayo, Agustin. 2012. 'An Actualist's Guide to Quantifying-In'. Critica 44(132): 3-34.
Schwar1,, Wolfgang. 2014. 'Against Magnetism'. Austrtt!tzsirzn joumal of Philosophy
92(1): 17-:36.
Sider, Theodore. 1996. 'Naturalness and Arbitrariness'. Philosophical Studies 81 (2):
283-301.
Sider, Theodore. 2002. 'The Ersatz Pluriverse'. jozmwl ofPhilosophy 99( 6): 279-315.
Sider, Theodore. MS. 'Beyond the Humphrey Objection'. Unpublished manuscript.
Weatherson, Brian. 2013. 'The Role of Naturalness in Lewis's Theory of Meaning'.
journal far the Histo1y ofAnafytiml Philosophy 1( 1O): 1-19.
Woodward, Richard. 2011. 'The Things That Aren't Actually There'. Philosophical
Studies 152(2): 155-66.
Woodward, Richard. 2012. "Fictionalism and Incompleteness." Notis 46(4): 781-90.

REFERENCES

Bricker, Phillip. 200 I. 'Island Universes and the Analysis of Modality'. In Gerhard
Preyer and Frank Siebelt (eds.), Reality tlnd Humean Superuenience: Ersays on the
Philosophy of David Lewis, 27-55. Lanham, MD: Rowman & Littlefield.
Divers, John. 1996. 'On the Prohibitive Cost of Indiscernible Concrete Possible
Worlds'. Australasian journal of Philosophy 72(3): 384-9.
Divers, John. 1999. 'A Genuine Realist Theory of Advanced Modalizing'. Mind 108
(430): 217-39.
Divers, John. 2002. Possible Worlds. London: Routledge.
Field, Hartry. 2008. Saving fruth from Paradox. Oxford: Oxford University Press.
1-LrLen, Allen. 1979. 'Counterpart-Theoretic Semantics for Modal Logic'. journal of
Phi/{)sophy 76(6): 319-38.
Heller, Mark. 1998. 'Property Counterparts in Ersatz W'.>rlds'.Journal ofPhilosophy
95(6): 293-316.
Lewis, David. 1968. 'Counterpart Theory and Quantified Modal Logic'. journal of
Philosophy 65(5): 113-26.
Lewis, David. 1971. 'Counterparts of Persons and their Boclies'.joumal ofPhilosophy
68(7): 203-11.
Lewis, David. 1980. 'Index, Context, and Content'. In Stig Kanger and Sven
Ohman (eds.), Philmophy and Grammar, 79-100. Dordrecht: D. Reidel.
 11
The Principle of Sufficient Reason
and Probability
Alexander Pruss

l. THE PRINCIPLE OF SUFFICIENT REASON

I will take the Principle of Sufficient Reason (PSR) to be the claim that
necessarily every contingent truth has an explanation (Pruss 2006).
It has been often noted that many of our ordinaty epistemic practices
presuppose the existence of explanations. Rescher, for instance, talks of how
the investigators of an airplane crash do not conclude that there is no
explanation from their inability to find an explanation (Rescher l 995,
p. 2). But it is one thing to agree that the PSR holds for propositions
about everyday localized matters, and another to generalize to cosmic cases,
such as the origination of the universe. Of course it would be difficult to
precisely define what counts as "localized." Still, we have a rough grasp of
localization: an LED lighting up is localized, while an infinite regress of
events or the complete state of our large universe presumably are not.
I will argue, however, that in order to make sense of our scientific
epistemic practices, we need a principle like the PSR that applies to global
matters. In doing so, I will draw on recent mathematical work on laws of
large numbers for nonmeasurable processes (Pruss 2013).
The aq,111rnent begins with the local cases and generalizes to the global
ones.
I need to note that I understand the PSR to be compatible with
indeterministic phenomena. The PSR is a claim that explanations exist,
not that deterministic explanations exist. The explanations might well be
probabilistic in nature, and might even involve low probabilities, as is
widely accepted in the philosophy of science (Salmon 1989). After all,
giving explanations is largely about making events understandable, and we
understand low probability chancy events as well as high probability ones
(Jeffrey 1969).
2()2 Alexander Pmss
2. LOCAL CHANCES
2.1 Frequency to chance inferences
l will consistently use the word "chance" to indicate probabilities that are the
tendencies of stochastic processes. Consider now this simple inference.
A coin has been independently flipped a thousand times, and about 750
times the coin has landed heads. There is a very natural inference: the coin is
loaded in such a way as to have approximately a 3/ 4 chance oflanding heads.
This natural inference leads to further predictions of the coin's behavior.
Frequency-to-chance inferences like the above are everywhere, and
inductive reasoning to a universal generalization is arguably just a limiting
case.
Let Cp be the hypothesis that the chance of the coin's landing heads is p.
Plausibly, the inference in the coin example was based on the fact that given
c~ with p close to 3 / 4, it is very likely that about 750 times we will get
heads, while given Cp with p far from 3 / 4, this is unlikely. The reason for
this is a Law of Large Numbers: given a large number of independent and
identically distributed chancy trials, the frequency of an outcome among the
trials is likely to be close to the chance of the outcome.
There are difficulties here, of course, with how we determine that the coin
flips are independent and identically distributed. Independence perhaps is
backed by the fact that we just cannot find any memory mechanism for the
coin, and identical distribution by the fact that the coin flips appear to be
identical in all relevant respects. The questions of how exactly one cashes out
these considerations are difficult, but they are not the questions I want to
focus on.
Instead, I want to focus on why there are probabilities at all. Our intuitive
Bayesian-flavored reasoning was based on comparing the probabiliry of our
observed frequencies on the different hypotheses Cp. But what about
hypotheses on which there are no probabilities of frequencies? In the next
subsection, I will sketch a picture of how such hypotheses might look by
making use of nonmeasurablc sets. A reader who wishes to avoid mathem-
atical complication may wish to skip that subsection.
The basic idea behind the technicalities is to imagine a case where we pick
out a series of heads and tails results by throwing a dart at a circular target,
and deeming heads to have occurred when the clan lands within some set A
and tails to occur otherwise. But we will take the set to be utterly non-
measurable ("saturated nonmeasurable"): it has no area, and doesn't even
have a range of areas. I will then argue on technical grounds that the
hypothesis that coin toss results were generated in such a way can neither
The Principle of Sufficient Reason and Probability 263
be confirmed nor disconfirmed by observation. But in order to confirm an
alternate hypothesis-say that a series of heads/tails results was the result of a
sequence of fair coin tosses--we need to rule out the hypothesis that it was
generated in some "nonmeasurable" way like this.
2.2 Nonmeasurable sets
So suppose c.oins are being flipped in this roundabout way. A dart with a
pet:f~ctly defined tip (say, infinitely sharp or perfectly symmetrical) is
urnformly randor:i-ly thrown at a circular target. There is a region A of the
targe.t marked. of-~, and a detector generates a heads toss (maybe in a veiy
physical way: tt picks up the coin and places it heads up) whenever the dart
lands in A; otherwise, it generates a tails toss.
The chance of the coin landing heads now should be equal to the
proportion of the area of the target lying within A, and we can elaborate
CP to the hypothesis that the area of A is p T, where T is the area of the
whole target. Given the observation of approximately 750 heads, it seems
reasonable to infer that probably the area ofA is approximately (3/4)T.
. But what ~fan are: ca~rnot be assigned to the marked region A? Famously,
given the Axiom of Choice, there are sets that have no area-not in the sense
that. they have zero area (like the empry set, a singleton or a line-segment),
but m the sense that our standard Lebesgue area measure cannot assign them
any area, not even zero. In such a case, there will also be no well-clefinccl 1
chance of the clan hitting A, and hence no well-defined chance of heads. Let
N be the hypothesis that A has no area, i.e. is non-measurable.
Now, here is a fascinating question. If N were true, what would we be
likely to observe? Of course, if we perform 1000 tries, we will get some
number n of heads, and n/ 1000 will then be a frequency between O and 1.
We might ~low think as follows. This frequency e<rn equally well be any of
the 1001 different numbers in the sequence 0.000, 0.001, .. ., 0.999, 1.000.
It ~eems unl~kely.' then, that it's going to be near 0.750, and so Ci;4 (and its
neighbors) ts still the best hypothesis given that the actual frequency
is 0.750.
But this reasoning is mistaken. To see this, we need to sharpen our
hypothesis N a little more. There are non-measurable sets A where we can
say things about the frequency with which A will be hit. For instance, it
could be that although A is non-measurable, it contains a measurable set A 1
of area (0.7 4) T and is contained in a measurable set A 2 of area (0.76) T.
1
.One .wants to say "no chance," but that suggests "zero chance," which is not what
one is saymg.
264 Alexander Pmss
(Think of A as of the target plus a nonmeasurable set localized to an
area containing only 2% of the target.) But the dart will hit A 1 about 74<YfJ of
the time, and A 2 about 76% of the time. Whenever the dart hits A 1, it hits
A, and whenever it hits A, it hits rh, so we would expect the dart to hit A
approximately 74%A to 76% of the time. And so our observed frequency
would be no surprise.
The mere fact that A is nonmeasurable does not rule out probabilistic
predictions about frequencies, because a nonmeasurable set might be "quite
close" to measurable sets like A 1 and Az that bracket A from below and above.
However some sets are not only nonmcasurable, but saturated nonmeasur-
able. A set A is saturated nonmeasurable provided that all of A's measurable
subsets have measure zero and all measurable subsets of the complement of A
are also of measure zero. Given the Axiom of Choice, for Lebesgue measure
on the real line, there not only are nonmeasurable sets, but there are saturated
2 our presumption that a student coming late to class was not struck by
nonrneasurable sets (Halperin 1951).
When A is saturated nonmeasurable, no method of generating predic-
tions by bracketing a the probabilities into an interval, like the one from
74% to 76%, will work. The only measurable subsets of A will have zero
area and the only measurable supersets of A will have area T. So our
bracketing will only tell us the trivial fact that the frequency will be between
0 and 1, inclusive.
Let M then be the hypothesis that A is saturated nonmeasurable. Can we
say that given M, the frequency is unlikely to be near 0.750?
The answer turns out to be negative even if we have infinitely many
observations. In Pruss (2013), I gave has given a plausible mathematical
model of an infinite sequence of independem identically distributed satur-
ated nonmeasurable random variables, and it follows from Theorem 1.3
there that for any nonempty interval I which is a proper subset of [O, l ], the
event that the limiting frequency of the events is in I is itself saturated
nonmeasurable. Thus, even with infinitely many independent shots, we
could say nothing probabilistic about the observed frequency of hits of our
saturated nonmeasurablc set being near 0. 750: the probability would neither
be large nor small. And analogous claims can be argued for in finite cases by
building on the tools in that paper (see Theorem 2 in the Appendix).
So in our case above, if I is some small interval like [0.740, 0.760]
centered on 0.750, there is nothing probabilistic we can say about the
2 If Ao is a saturated nonmcasurable subset of the interval _[O, I], then the ~~a1;esian
product Ao x [O, 1] will be a saturated nonmeasurablc subset of the square [O, 1 J-. ( fherc
is also a directly two-dimensional construction in Sierpir\ski 1938.) Intersecting this with
a disc of radius 1/2' centered on (I /2, 1/2) yields a saturated nonmcasurable region for a
disc-shaped area, like in our example.
The Principle of Sufficient Reason and Probability 265
observed frequency being in 1 when the target set is saturated nonmeasur-
ablc as the event of the frequency being in I is then saturated nonmeasur-
ablc. In particular, we cannot say that the frequency is unlikely to be near
0.750 (nor that it's likely to be near). The observation ofa frequency close to
0.750 is neither surprising nor to be expected given that the target set is
saturated nonmeasurable.
Our probabilistic reasoning thus cannot disconfirm hypotheses of satur-
ated nonmcasurability. Such hypotheses endanger all our local scientific
inferences from observed frequencies to chancy dispositions, inferences
central to our epistemic practices. Yet our local scientific inferences are,
surely, good. If we cannot disconfirm saturated nonmeasurability hypoth-
eses a posteriori, then we need to do so a priori.
Kleinschmidt (2013) defends a presumption of explanation principle
(EP) on which we are justified in presuming explanation, by analogy with
lightning, even though people are sometimes struck by lightning. She
then writes: "In the lightning case, we can explain our reluctance by noting
that we justifiably believe, partly due to induction, that lightning strikes
rarely. We might hope to give a similar explanation of our endorsement of
EP." But while we can say what it would look like if lightning was more
common than we think it to be-and thus we can say on empirical grounds
that it is not so-the upshot of our argument is that we cannot say what it
would look like if there were more unexplained events than we think, and
hence cannot say on empirical grounds that it is not so. 3
Another approach would be to assume low prior credences for hypotheses
of saturated nonmeasurability. But it is difficult to justify this in a way that is
not ad hoc. Given the Axiom of Choice, there are just as many saturated
3
Kleinschmidt (2013) also attempts to give counterexamples to the PSR on the basis of
cases involving the diachronic identity and synchronic composition of persons. Why, for
instance, does a cloud of atoms in group GI compose a person while the cloud of atoms in
group G2 , which differs only by a single atom, docs not) (The fission case can be handled
analogously.) A foll response to her intricate arguments would take us roo far afield, but the
facts underwriting the composition claims arc either necessary or contingent. Since I am
only defending a PSR for contingent truths, it is only the contingent option that I need
consider. But even without assuming the !'SR, simply using Klcinschmidt's own El', it will
count ve1y strongly against a theory on which it is contingent that an arrangement of atoms
composes a person if the theory can give no explanation of why it happens to do so.
Consider, for instance, dualist theories on which souls attach to some groups of atoms with
no explanation as to why they attach to the ones they attach to. Those theories are
implausible precisely because of the explanatory failure. And the same will be true of
those materialist theories on which it is a bn1te unexplained fact that some atoms compose
a person. Kleinschmidt has not shown that there is a bmte-composition theory that is still
plausible after taking into account her own EP.
266 Alexander Pruss
nonmeasurable subsets of, say, [0, l] as there are measurable ones (see the
Corollary in the Appendix). Granted, subjective Bayesians may not worry
too much about the t1c! hoc in the choice of priors, but the reason they do not
worry is because of the hope that evidence will swamp the priors and make
them irrelevant. But evidence cannot affect the credence of the nonmeasur-
ability hypotheses, since those hypotheses do not generate the kinds of
likelihoods that arc needed for confirmation or disconfirmation, and so
one would simply be stuck with the 11c! hoc priors.
Note, however, that the credences of such hypotheses would have to be
very low indeed: they would have to be zero or infinitesimal. I sketch the
argument in a simple case. Suppose that N is a hypothesis according to which
the coin tosses are independent (in the sense of Pmss 2013) saturated
nonrncasurable sets. Suppose that P(N) > 0 and that the only alternative
to N is the hypothesis H that the coin tosses are fair and independent so that
P(H) = l - P(N). Let Ebe the observed evidence-a sequence of n coin
toss results (say HTHHHTH if n = 7). We would like to be able to use
Bayes' Theorem to say that the probability of H given the evidence is:
P(EIH)P(H)
P(HIE) = P(EIH)P(H) + P(EIN)P(N)
Unfortunately, the term P( EI N) in the denominator is not a number, since
Eis saturated nonmeasurable conditionally on N. But we can, nonetheless,
use Bayes' Theorem. We can model the probability ofa nonmeasurable set
by an interval of numbers. In the case of P( EI N), the interval will range all
the way from 0 to 1, i.e. will be [O, 1]. fortunately, only one of the quantities
on the right-hand-side corresponds to an interval: the other quantities,
P(E IH), P(H) and P(N) are ordinary classical probabilities. We can
then use the range of the interval P(E IN) and Bayes' Theorem to define
an interval-valued probability for P(H IE).
Since P(E IN) is the interval [O, 1], one endpoint for the interval corres-
ponding to P( HIE) will be given by replacing P(E IN) in Bayes' Theorem
with zero:
_ _P_(E_'l_H_)P_(_H_)__ =
1
P(E!H)P(H) + (O)·P(N) .
This turns out to be the upper endpoint. The second endpoint-the lower
one-will correspond to replacing P(E IN) with one, the other endpoint of
its interval:
P(EIH)P(H)
P(EIH)P(H) + (l)·P(N).
The Principle of Sufficient Rerlson and Probability 267
Now, P(EIH) = (l/2)". It f(illows that this lower endpoint of the
interval given by P(E IH) is less than or equal to:( l /2)" / P(N). Therd(ire,
the interval-valued probability P(H j £) will contain the interval
[(1/2)"/P(N), l]. 4
As long as P(N) is strictly bigger than zero and not infinitesimal, it
follows that as n goes to infinity--i.e. as the amount of evidence gathered
increases-the hypothesis H gets to have a probability range closer and
closer to [O, l ]. But in order to get confirmed, to become believable, H
would have to be an interval of probability like [a, b] where rt is close to one,
not close to zero. So the more evidence we gather, the broader our prob-
ability interval for H. This is Bayesian divergence rather than convergence.
As we observe more coin-flips, the nasty hypothesis N infects our probabil-
ities more and more.
One could tty to claim that P(N) is infinitesimal, but that would seem
to be ac! hoc. Alternately, one might try to completely a priori rule out
pathological hypotheses, ruling them to have probability zero. For instance,
one might simply deny any version of the Axiom of Choice strong enough
to imply the existence of nonmeasurable sets. Or one might deny the
possibility of physical processes sensitive to whether an input falls into a
pathological set.
2.3 No-explanation hypotheses
But clever mathematical constructions arc not the only potential source of
probabilistic nonmeasurability in the physical world. Suppose a coin is
tossed, but there is a hypothesis H that there will be no explanation at all
of how the coin lands. Simplify by assuming that the coin must land heads
or tails. Then I contend that the event of the coin landing heads has no
probability conditionally on H.
For suppose it has some probability. What is it? The only non-arbitrary
answer seems to be the one given by symmetry or the Principle of Indiffer-
ence: 1/2. And, generalizing, the only non-arbitrary probability for any
particular fixed sequence of n tosses on a no-explanation hypothesis will be
2-n. But if that is the answer, then the no-explanation hypothesis cannot be
4
This argument is simpler than the usual kind of reasoning with interval-valued
probabilities (e.g. Whitcomb 2005), because only one of the quantities involved is an
interval. We can justify the argument more rigorously by thinking of interval valued
probabilities as determining a family of classical probability functions, and then applying
Bayes' Theorem to each member of the family. The only relevant variation between
members of the family will be as to the value assigned to E given N, and that will range
between 0 and I, thereby generating a range for the posterior probability containing
[(1/2)"/P(N), I].
268 Alexander Pruss
empirically distinguished from the hypothesis that the coin tosses form an
independent sequence of stochastic events each with probability I /2 of
heads, as both hypotheses predict the same behavior.
In fact, although 1/2 is the only non-arbitrary answer, the arbimuy answers
face exactly the s;une problem. For each possible answer, arbitrary or not, to
the question of the probability that the coin lands heads conditionally on the
no-exphrnation hypothesis, there is a possible stochastic process hypothesizing
which yields the same prediction as the no-explanation hypothesis does. And
then no amount of observation can rule out the no-explanation hypothesis.
Whether this is a serious problem will depend on whether we can rule out the
hypothesis a priori, an option we will discuss later in this section.
Formal results on nonmeasurable sets like those discussed in Section 2.2
model not only those events that are saturated nonmeasurable due to their
pathological set theoretic construction but also those that are nonmeasur-
able simply because no probabilities can be assigned to them.
Within a probabilistic framework, there are now three possibilities. Either
(a) conditioning on a no-explanation hypothesis yields an exact probability,
or (b) it yields no probability at all, or (c) it yields a probability range or
interval. If it yields an exact probability, then we cannot empirically distin-
guish the no-explanation hypothesis from an explanato1y chance hypothesis
whose chance assignment matches that exact probability. If it yields no
probability at all, not even a range of probabilities, we are dealing with a
saturated nonmeasurable event, and our formal results tell us that we cannot
make any probabilistic predictions about frequencies, and in particular
cannot say whether it is likelier or less likely that we would get the same
observations from an explanatory chance hypothesis.
Now consider the third option, that of a probability range. Perhaps when
we condition on the no-explanation hypothesis, we get an interval [a, b] of
probabilities with a<b and [a, b] ;F [O, I]. In such a case we can make some
probabilistic predictions by the Law of Large Numbers (Pruss 2013,
inequality (1)), namely that limiting frequencies are almost surely going to
be between ti and b (both inclusive). It can be shown that nothing stronger
can be shown about the limiting frequencies than that they are between
a and b (Pruss 2013, Theorem 1.3). But in fact surely a no-explanation
hypothesis assigns at best the full range [O, 1] to the probability, and so the
only thing that follows from the no-explanation hypothesis is that the
frequencies range between 0 and 1, which is trivially true. We cannot,
thus, say that a particular set of observed frequencies fits worse or better
with the no-explanation hypothesis than with some exphrnarory hypotheses:
all frequencies fit equally well with the no-explanation hypothesis.
Thus, whether or not the no-explanation hypothesis has a probability,
and whether this probability is exact or an interval, the no-explanation
7he Principle of Sufficient Reason and Probttbi!ity 269
hypothesis cannot be ruled out empirically. lt is simply not amenable to
statistical methods, or if it is amenable, it is amenable in the wrong way: the
argument based on an interval-based interpretation of Bayes' Theorem
given at the end of Section 2.2 shows that once we have a non-zero non-
infinitesimal probability of a no-explanation hypothesis, we end up with
poorer and poorer confirmation of serious explanatory hypotheses the more
data we gather.
So the no-explanation hypothesis must be ruled out ff priori. There are two
initially attractive proposals for doing so. We can say that no-explanation
hypotheses simply cannot hold. That would be to subscribe to a localized
version of the PSR. Or we could simply say that no-explanation hypotheses
have very low prior probabilities.
But as in the case of saturated nonmeasurability, it is unsatisfactory to set
low priors for no-explanation hypothesis. Any values for these priors would
be arbitrary once we allowed for the possibility of their truth, and the usual
subjective Bayesi;m hope that arbitrary priors will eventually be swamped by
evidence (a classic statement of this is in Edwards, Lindman and Savage
1963, p. 197) is not applicable given the non-amenability of no-explanation
hypotheses to empirical study.
One might compare the low priors for no-explanation hypotheses to the
low priors we assign some other wacky hypotheses. For instance, our prior
for the aberrant hypothesis that the gravitational force between two objects
10
of mass m1 and m2 at distance r apart has magnitude Gm 1m2 / r 2 + IO " will
presumably be ve1y small, since othe1wise we could never conclude that the
actual magnitude of the force is Gm 1ml/ r2 , as the aberrant hypothesis
cannot be empirically distinguished from the Newtonian hypothesis.
0
On its face, the analogy fails because we reject the Gm 1 m2fr2 + 10 "'
hypothesis on the grounds that it does not have the kind of simplicity we
expect in explanatory hypotheses, while bald no-explanation hypotheses are
actually simpler than any explanato1y hypotheses.
But perhaps the analogy is this: in both the no-explanation and wacky
gravity cases, our only reason for assigning low priors is that if we assigned
high priors, we couldn't do science. We cannot justify the assignment oflow
priors in any other way, and do not need to.
Neither case of assignment of low priors without further justification is
satisfactory, but it is better to be in one unsatisfactory situation than two.
Furthermore, one could hope that an inductive logic of a Carnapian variety
might eventually non-arbitrarily justify low probability assignments to more
complex hypotheses-but this won't help with, and in fact will exacerbate,
the problems with very simple no-explanation hypotheses.
Finally, the defender of a global PSR has some hope of being in neither
unsatisfactory situation. For not only has she ruled out all no-explanation
270 Alexander Pruss
hypotheses, but she has some hope of arguing to a global explanation that
assigns lower probabilities to more complex scientific hypotheses. For it may
be that the global explanation is one that, like the theistic and optimalist/
axiarchist global explanation proposals, makes better states of af£'1irs more
likely. 5 And simpler laws are better, both aesthetically and instrumentally.
Thus far I have argued that we need some kind of localized PSR as a
metaphysical principle to rule out no-explanation hypotheses on which our
observations simply have no probability. A central claim is that if nothing can
be said about the probabilities of individual events, then nothing can be said
about the probabilities of observing frequencies of events in a sequence, even
an infinite sequence, of observations. Thus, the no-explanation hypotheses
cannot be ruled out observationally, since to do that we would have to say
that our observations are improbable on the no-explanation hypothesis.
A tempting response to the arguments of this section is to say that just as
it counts a posteriori against a hypothesis when our observations have low
probability on it but have high probability on an alternative, it counts fl
posteriori against a hypothesis when our observations have no probability on
the hypothesis but have high probability on an alternative. But if no possible
sequence of observations has any meaningful probability on a no-explanation
hypothesis, then by this principle every possible sequence of observations
would disfavor our no-explanation hypothesis (since every sequence is prob-
able on some explanatory hypothesis). But if E is evidence against a hypoth-
esis, tl1c:n the: negation of E will be some evidence for the hypothesis. But if
evidence has no probability on tl1e no-explanation hypothesis, the negation of
that evidence abo has no probability on the no-explanation hypothesis, and so
by the suggested principle both E and its negation would count as evidence
against the no-explanation hypothesis, which appears absurd unless the no-
explanation hypothesis is impossible. 6
3. GLOBALIZING
3.1 Small worlds
Can we have a local PSR without a global one? One can tty to give restricted
P5R-like principles that can be applied to local states of affairs without being
5 Optimalism/axiarchism as formulated by Leslie (1979) and Rescher (201 O) make the
best option be necessary. But an interesting variation would be a theory on which value
simply makes a world more likely.
6 7
If a hypothesis is impossible, we may be able to give an argument for its impossibiliry
starting with F and another argument for its impossibility starting with the negation of E.
The Principle o/Sufjicient Reason and Probr1bi!ity 271
applicable to global ones (for a critical discussion, see Koons 2006). This
would allow one to secure: the: advantages of the PSR for scientific reasoning
while avoiding the disadvantages of (a) significant ontological commitment--
after all, the PSR has been used to argue for grand hypotheses like: the
existence of God-and (b) van Inwagc:n-style arguments against a global
PSR (van lnwagen 1983, pp. 202-4).
It is not dear, however, whether a localized PSR avoids the problems, if
it's taken to be a metaphysical principle and hence metaphysically necessary.
First of all, a localized P5R applied to a small world can produce the same
kinds of ontological commitment as a global P5R applied to a big world.
Plausibly, it is possible to have a walnut world-a world with a finite past
whose contingent components have always consisted of a single basically
unchanging but otherwise ordinaty walnut7 and its parts (cC the ball
thought experiment in Taylor 197 4, pp. 105-6). When we apply the
localized PSR to the walnut-which is as local as something can be-we
conclude that there is an explanation of the walnut. The existence of our
walnut is not explained by the activity of its parts, so this walnut's explan-
ation would have to go beyond it. But as there is nothing contingent outside
the walnut in the walnut world, the explanation would have: to invoke
something that exists necessarily or some kind of necessary metaphysical
priniciple. And this being or principle that is necessary at the walnut world
would also be necessary at the actual world by the axiom 55 of the most
popular system of modal logic (55 says that if a proposition is possibly
necessaty, it is necessary; necessities cannot vary between possible worlds),
so we have not avoided significant ontological commitment by supposing a
localized P5R. Granted, this argument uses 55, but having to deny 55 would
also be a case of significant metaphysical commitment. So a localized PSR
has similar commitment-engendering features as its grander relatives.
Van Inwagen's argument against the P5R is based on forming the
conjunction of all contingent truths and asking for the explanation of this
conjunction. If this explanation is itself a contingent truth, then we absurdly
have a case of self-explanation, since in explaining the conjunction of all
contingent truths, this contingent truth explains itself, being one of the
conjuncts. If, on the other hand, the explanation is a necessary truth, then
we have a puzzle about how a necessary truth can explain a contingent
one. One way to set up the puzzle is by invoking the principle that if p
explains q, then p emails q. Van Inwagen explicitly affirms this principle
(van Inwagen 1983, p. 203), though he uses "sufficient reason" in place of
Or, more precisely, an exact duplicate of an ordinary walnut. It may be an essential
property of a walnut per se that it comes from a walnut tree.
272 Alexander Pruss
"explanation." But in the case of explanation, this principle is widely denied
in the philosophy of science in the light of the existence of stochastic
explanations (see, for instance, Salmon 1989), so it is better to leave the
argument as simply based on the bare intuition that a necessary truth cannot
explain a contingent one, which intuition I will call the "Van Inwagen
Principle" (VlP).
But our best storv about how the walnut in a walnut world can be explained
involved the explar~ation of the walnut's existence by means of a necessary
being or a necessary metaphysical principle. The principle-based explanation
could perhaps involve something like a stochastic axiarchism, where the chance
of a universe arising is proportional to its value, but whatever that would look
like, it would be a violation of VIP. And the most obvious way to take the
explanation of the walnut's existence by means of a ncccss;uy being would be
to take it that necessary truths about the existence and essential properties of
8
the being explain why the walnut exists, in violation ofVIP.
So as long as the localized PSR is a metaphysical principle, we do not
actually gain much if anything from eschewing a global principle, since we
get results similar to those of a global PSR when we apply the localized PSR
in a small world.
Dean Zimmerman has suggested to me that locality could also be taken
relative to the size of the whole universe. On the reading, a localized PSR
would apply to, say, the molecules of the walnut in the walnut world, but
not to the walnut as a whole. But it would be implausible to think that a
walnut does not need an explanation when it is all alone, so that its existence
is not localized, but needs one when it is but one among many things.
3.2 Global probabilities
But there is further reason to accept a global PSR. A localized PSR is
insufficient to solve the kinds of probabilistic problems that led us to it.
8 Someone who resists a global PSR but accepts a local one might also try to expl~in
the walnut's existence by supposing an infinite regress of contingent properties or a
necessary being. Perhaps the necessary being is a person who decided at to to create the
walnut, and did so on the basis of a contingent mental state that was caused by the
person's contingent thoughts at L 1; these thoughts were then caused by yet earlier
contingent thoughts at t ~z, and so on. In this case, a localized PSR is satisfied: each
localized evcnt--the existence of the walnut, the choice at to, the thoughts at L1, and so
on-is explained. The whole sequence of thoughts is not explained, however, which
blocks the van lnwagen argument. Notice, however, that we only avoid violation of VIP
by further ontological commitment, namely supposing an infinite regress of contingent
states of a necessary being. Furthermore, this ontological commitment runs afoul of the
modal intuition that there is a possible world containing the walnut and only at most
finitely many contingent states that aren't states of the walnut.
The Principle ofSzdficient Reason and Probability 273
Consider universes which have a finite past but exist over an interval of
times open at the lower end, say an interval modeled by the semi-infinite
interval { t : t > 0}. For each instantaneous state of such a universe, there is
an earlier state, but there is no earliest state. It has been argued by Grifobaum
(1993) that our universe is like that. Let us forther assume that each state is
explained by an earlier state.
A localized PSR will not apply to the infinite regresses of states we find
in such universes. 'T'lms, a localized PSR will be compatible with an
unexplained no-beginning finite-past universe. 9 But then there will be
no meaningful chances for the existence of such universes, and hence no
non-arbitrary reason to prefer the hypothesis we think is true-a universe
that begins with a Big Bang about 13.8 billion years ago-over aberrant
hypotheses such as Russell's five-minute hypothesis on which the universe is
five minutes old with the twist that there is no initial moment (compare
Koons 2006, Section 2. l).
This was too quick. Although there are no chances associated with the two
hypotheses, maybe there are epistemic probabilities. But what reason do we
have to take the precise limiting sequence of states that we think obtained
after the Big Bang to be more probable than the limiting sequence of states in
the five minute hypothesis? Again, we would have to assign an arbitrary
epistemic probability. One might think that of all possible unexplained no-
beginning finite-past universes, the five-minute universe (i.e. the one that
coincides with what we rake to be our universe over the past five minutes,
without an initial time) has an initial segment that is too elaborate-
festooned with galaxies and stars as it is, and with at least one planet of very
complex life-to come into existence without an explanation, while the Big
Bang universe's initial segment is not as elaborate. Bur both initial sequences
of states live in the same configuration space, and there does not appear to be
a reasonable way to privilege the Big Bang sequence over the five-minutes-ago
sequence as an option f-(lr coming into existence causelessly ex nihilo. For what
it's worth there is even an intuition preferring the five-minutes-ago sequence.
For one would intuitively expect an unexplained universe would have high
entropy, and the states in the five-minutes-ago sequence have higher entropy
than the states in the post-Big-Bang sequence do.
But suppose that we have some way of arguing that the Big Bang
hypothesis is simpler than the five-minute hypothesis. We can beat the
Big Bang hypothesis in respect of simplicity. Consider the hypothesis that
reality consists of a short truncated backwards light cone (say, extending
9
Hume (1779, Part IX) has argued that an infinite causal regress is explained by the
causes within it. However, the discussion of the beginningless cannonball flight thought
experiment in Pruss (2006, Section 3. l.4) makes this implausible.
274 Alexander Pruss
backwards one minute), with no initial state, centered on the present state of
my brain. 'This hypothesis supposes only a tiny fraction of the complexity of
our gigantic universe (indeed even of our solar system), and there will be a
lot more simplicity than in the Big Bang story. After all, it is a part of the Big
Bang story that the world evolved in the kinds of complex ways that resulted
in a life full of earth, many galaxies, and so on. All the details of this
complexity will have free parameters, either determined at the time of the
Big Bang or set by later chancy events.
Thus a priori simplicity arguments do not rule out our aberrant hypoth-
eses. We really do seem to need a PSR.
4. PSR-COMPATIBLE ABERRANT HYPOTHESES
Our argument for the PSR was that it allows us to reject aberrant local and
global hypotheses that we have no a posteriori way of rejecting. A natural
response is that there are aberrant hypotheses that are compatible with the
PSR as well, and the PSR only pushes the problem back a step.
For consider the hypothesis that a necessarily existing being with the
essential property of being a science-hater creates a universe that appears
very much other than how it is. If the PSR is restricted to contingent states
of affairs, as in Pruss (2006), then it is compatible with this aberrant
hypothesis. Compare this to the hypothesis that a necessarily existing
being with the essential property of being a science-lover creates a universe
that allows us to make correct scientific predictions. We cannot compare the
science-lover and science-hater hypotheses by means of chances. By S5, each
hypothesis is either necessarily true or necessarily false, and hence has chance
1 or chance 0, but we don't have access to these chances except by figuring
out which hypothesis is true.
However, in the realm of PSR-compatible hypotheses, we can make use
of simplicity reasoning to rule out aberrance. For instance, the science-hater
hypothesis has many free parameters: just how much does the science-hater
hate science and in what respects, how powerful is it, etc. And there does not
seem to be any canonical set of values for these free parameters. On the other
hand, there is a canonical science-lover hypothesis in the literature, namely
the theistic hypothesis that there exists a perfect being, who has maximal
power, maximal knowledge, and loves all that is fine and good precisely in
proportion to how fine and good it is. And of course science getting at the
truth is fine and good.
And while there will be many variants of the stochastic axiarchic hypoth-
esis that universes are likely to come into existence to the extent that they are
The Principle of Sufficient Reason and Probability 275
good, depending on just how the direct relationship between value and
chance works, it could turn out that a broad range of them would produce
non-aberrant stories.
5. CONCLUSIONS
\'Vhere there are no explanations, there arc no chances. Where there are no
chances, it is difficult to have non-arbitrary probabilities. In some cases, we
can have non-arbitrary probabilities based on simplicity considerations. In
others, perhaps by indifference considerations. But neither simplicity nor
indifference considerations help with rejecting aberrant no-explanation
hypotheses that undercut the scientific enterprise. We need a PSR, and not
just one restricted to local situations, but a global one.
Finally, while our arguments above took for granted that chances were
governed by the axioms of classical probability theory, it may be that we
should relax this assumption. For instance, while a rational agent tends to
act in proportion to the strength of her reasons, it would be hasty to suppose
that this tendency can be quantified by classical probability theoty. More-
over, global explanatory hypotheses are unlikely to give rise to numerical
chances, if only for cardinality reasons-likely, there is no set of all possible
worlds (e.g. Pruss 2001), and classical probability theory is defined over sets
of possibilities. But the main point of the paper should remain: where there
are no explanations, there are no chances-whether the chances are under-
stood as classical probabilities or are governed by some more complex
calculus-and hypotheses that do not give rise to some sort of chances
need a priori refutation.
Alexander Pruss
Baylor University
APPENDIX: SOME RESULTS ABOUT
NONMEASURABLE SETS
A set is saturated non measurable with respect to a probability measure provided that all
measurable subsets have measure zero and all measurable supersets have measure one.
Theorem 1 Assuming the Axiom of Choice, the cardinality ofthe set ofall saturated
non measurable subsets of[O, I) is equal to the cardinality of the set of
subsets of[O, l ).
For the proof, we need the following easy Lemma. Let ALJB be the symmetric
difference (A B) U ( B ~~A) of the sets A and B. Recall that every subset of a set of
Lebesgue measure zero is measurable and has measure zero.
276 Alexander Pruss
Lemma If A is d satttrdted nonmetwmib!e subset of [O, l) mzd AL1 B is d set of
Lebesgue met1sure zero, then lJ is Stltttrtlted nonmettsurttb!e.
l'roofofiemm,1 1. For any set C, the set B () C differs from An Cat most by a set
of measure zero since AL1 !3 has measure zero, and any two sets that differ by a set of
measure zero must either both be measurable or both nonmcasurable.
Thus, if C is a measurable subset of 13, it follows that A() C differs from C by a set
of at most measure zero.'] 'hus, A n C is a mc;Lrnrablc subset of A, hence has measure
zero, and hence C also has measure zero since it differs from it by at most a set of
measure zero. This argument shows that if all measurable subsets of A have measure
zero, the same is true for B. Applying this argument to the complements [O, I) - A
and [O, I) -- 13, and using the observation that all measurable supersets of U that are
subsets of [O, I) have measure I if ;rnd only if all measurable subsets of [O, l) U such qualifications from now on) and ( Yk )*
have measure zero, we conclude that all measurable subsets of [O, l) B have
measure zero, from which we conclude that all measurable supersets of B that arc
subsets of [O, I) have measure l. Thus, B is saturated nonmeasurablc.
Proof of Theorem I. Famously, the C'lntor middle-thirds set CC::: [O, l) is a set of
zero measure that has the same cardinality c as all of [O, I). Given any one saturated
non measurable subset S of [O, l) (these exist by Halperin 1951), for each subset A of C
let S11 = (S ·--A) U A. 'Tben S1i/'1S c;;: C, and hence S11 is saturated nonrncasurable.
There arc as rn;rny sets S11 as subsets of C, and there arc as many subsets of C as of[O, I).
The fact that the number of subsets of the Cm tor set equals the number of subsets
of [O, l) and that every subset of a set of zero measure is measurable also yields:
Corollary Assuming the Axiom of Choice, the cardinality of the set of st1turt1ted
11onmet1s1mzble sets is equal to the ct1rdinality of the set of met1surab!e
subsets of [O, I).
Following Pruss (2013), I will model independent sequences of nonmeasurablc
events E1 , ••• , L~, as follows. We have a probability space (D, :!, P) which is a product
of probability spaces (Dk, Jk, Pk) for l ~ k ~ n. Then for each k, we suppose that
the event E1: depends only on the factor D1: (this is how one models independence).
This means that E1.. can be written as
for some subset u:1: of Dk.
Let N be a "nonmeasurable random nriable" representing the number of events
Ek that happen, i.e. N is a fonction from Q to the natural numbers such that N( w)
equ:1ls the cardinality of the set { k : w E I~}.
Then we have:
Theorem 2 Suppose E1 , ••• , En are satumted nonmeas111·t1b!e. LetJ be ttny non-empty
proper subset of {O, ... , n }. 7hen the event N;={ wED:N( w )El} is
st1tumted nonmeasurable.
The Principle ofSujficient Rertson anti Pmbtlbi!ity 277
Hence nothing non-trivial can be probabilistically said about the number of the LI:
that occur.
The proofof'T'hcorcm 2 will use the terminology in Pruss (2013). In particular,
for any function J, measurable or not, on a probability space there is a mt1ximal
nonmet1sumb!e minomnt .h and a minimr1! nonmeasumb!e maiomnt / ' such that
J; ~ J ~ J* and such that for any measurable 1-i.rnctions g and h such that g ~ f
and J ~ h we have g S £ almost everywhere (i.e. outside of some sct--pcrhaps
empry--of probability zero) and h 2: f * almost everywhere.
Proof of Theorem 2. Deline the function '¥I· on Qk by 'l' k ( cv) == l if w E Uk and
'I' k( w) cc. 0 othe1wise. Define the function Y1. on Q by Yk( {w1, ... , wn)) '¥;.( cv1J
Thus, Yi,( <v) is 1 iF w E l'.i: and 0 othe1wisc.
Because E1: is saturated nonmcasurablc, ( Yk ), 0 (almost everywhere-] will omit
L It follows from Pruss (2013, Prop-
osition L2) that ('¥1:) = 0 and ('I'k)*
0 I. It follows that Uk is a saturated non-
mcasurablc subset off.?:. Let Pk and f> k be probability measures on £21: extending P to a
a-field that includes Ei, and such that [>1. (Uk) 0 and 1\( U1:) == l. These exist by
0
Pruss (2013, Lemma 1.5): just let/ in the Lemma be 'I'k·
Now let j be any member of}. Let Qbe the probability measure on Q formed as
the product of the probability measures!' 1 , ... , l)J' P_ J+ 1, ... , f>n· Then Q(l'.ie) = l if
k Sj <llld otherwise Q(E1:) =·= 0. Thus with Q-probability l, exactly j of the El,
occur, and since j Ej, we have Q(N;) == I.
Let l be any member of {0, ... , n} -- I Let R be the product of the probability
measures J>1, .. .1>1,IJ 1 1 1, ... J'n· Then R(E1J = l if k SI and otherwise R(E1:) """' L
Hence, with /?-probability one exactly l of the Ei: occur, and since lr/J, we have
R(N;)=O.
Thus there is an extension of l' on which N; has measure zero and an extension of
Pon which it has measure one, from which it follows (Pruss 2013, Lemma 2.3) that
N; is saturated nonmeasurable with respect to P.
REFERENCES
Edwards, \Y/., Lindman, H. and Savage, L.]. ( 1963). "Bayesian Statistical Inference
for Psychological Rcsc;trch," PsychologiCt1l Review 70(3): 193-242.
Griinbaum, Adolf. (1993). "Narlikar's 'Creation' of the Big Bang Universe was a
Mere Origination," Philosophy ofSrience 60(4): 638--46.
Halperin, Israel. (1951). "Non-Measurable Sets and the Equation f(x +y) =c
f(x) + /(y)." Proceedings ofthe Americmz Mathematim! Society 2(2): 221-4.
Hume, David. (1779). Dit1!ogues Concerning Nt1tuml Religion.
lnwagcn, Peter van ( 1983). An Esst1y on Free Will. New York: Oxford University Press.
Jeffrey, Richard C (1969). "Statistical Explanation vs. Statistical Inference." In Nicho-
h> Rescher (ed.), Essays in Honor of Carl C7. Hempel, Dordrccht: Reidel, 104--13.
278 Alexander Pruss
Kleinschmidt, Shieva. (2013). ''Reasoning without the Principle of Sufficient Rea-
son." In Tyron Goldschmidt (ed.), The Puzzle oj'Existence: Why Is There Something
R11ther Tlvm Nothing? New York: Routledge, 64-79.
Koons, Robert. (2006). "Oppy and Pruss on Cosmologic;1l Arguments." Presented
ar rhe Central Division Meeting of the American Philosophical Association,
Chicago, April 2006.
Leslie, John. (l 979). Vdue 1md E>:istmce. Totowa, NJ: Rowman and Littlefield.
Pruss, Alexander R. (200 l). "The Cardinality Objection to David Lewis's Modal
Realism." Philosophical Studies 104(2): 167-76.
Pruss, Alexander R. (2006). The Principle of Sufficient Reason: A Re11ssessment. New
York: C.unbridge University Press.
Pruss, Alexander R. (2013). "On the Law of Large Numbers for Nonmeasurable
Identically Distributed Random Variables." Bulletin of the Polish Academy of
Sciences--Mathematics 61 (2): 161-8.
Rescher, Nicholas N. (1995). Sritisfjing Re11son: Studies in the Theo1y of Knowledge.
DordrechtlBosron/London: Kluwer.
Rcscher, Nicholas N. (2010). Axiogenesis: An Essay in Metaphysical Optima/ism.
Lanham, MD: Lexington.
Salmon, Wesley C. (1989). Four Decades of Scientific Explanation. Minneapolis:
University of Minnesota Press.
Sierpit1ski, Wadaw. (1938). "Sur un problcmc conccrnant lcs famille d'ensembles
parfaits." Fundamenta Mathematime 31 ( 1): 1-3.
Taylor, Richard. (1974). Metaphysics. Englewood Cliffs, NJ: Prentice-Hall.
Whitcomb, Kathleen M. (2005). "Quasi-Bayesian Analysis Using Imprecise Prob-
ability Asscsscmenrs and the Gcncmlized Bayes' Ruic." Theo1y and Decision 58(2):
209-38.
12
Grounding Ground
Jon Erling Lit/and
1. INTRODUCTION
Ifr's being the case grounds ef/s being the case, what grounds that r's being
the case grounds ¢' s being the case? 1 This is the Problem of Iterated Ground.
Dasgupta (2014b), Bennett (2011), and deRosset (2013) have grappled
with this problem from the point of view of metaphysics. But iterated
ground is a problem not just for metaphysicians: the existing logics of
ground 2 have had nothing to say about such iterated grounding claims. In
this paper I propose a novel account of iterated ground and develop a logic
of iterated ground. The account-what I will call the Zero-Grounding
Account (ZGA for short)-is based on three mutually supporting ideas: (i)
taking non-factive ground as a primitive notion of ground; (ii) tying non-
factive ground to explanatory arguments; and (iii) holding that true non-factive
grounding claims are zero-grounded (in Fine's sense).
A notion of ground is factive if the truth of"f grounds <p" entails that each
y E r as well as </> is true; the notion is non-factive otherwise. Most authors
take a factive notion of ground as their primitive; I adopt a non-factive
notion as primitive. Taking a non-factive notion of ground as basic allows
one to solve the Problem ofiterated Ground for factive ground: if 6 factively
grounds</> then this is grounded in 6's non-factively grounding cf> together
with 6's being the case. 3 This, of course, just shifts the bump under the rug:
what grounds that 6 non-factively grounds cf>?
1
Herc f arc some (true) propositions and r/> is a (true) proposition. For the official
formulation of claims of ground, see §2 below. In the interest of readability I will nor
distinguish carefully between use and mention throughout.
2
Fine 2012b; Correia 2010, 2014; Schnieder 2011; Poggiolcsi 2015.
3
Ir is not strictly speaking neccssa1y to hold that true facrive grounding claims arc
partially grounded in non-facrive grounding claims (§10). However, assuming this allows
for a smoother presentation; and, as we will see, it docs no harm.
280 Jon Erling Litlmzd
Fine (2012a, pp. 47-8) distinguished between a truth's being
ungrounded, on the one hand, and having the empty ground or being zero-
grounded on the other. Crucially, being zero-grounded is a way of being
grounded. 1 show that if !':i's non-factivcly grounding</> is zero-grounded we
have a formally satisfactory solution to the Problem ofitcrated Ground. To
go beyond a merely formal solution we must answer two questions: (i) What
does it mean to say that a truth is zero-grounded? (ii) Why should we believe
that (true) non-factivc grounding claims are zero-grounded?
We answer these questions by tying non-factive ground to explanation.
The basic idea is that for !':i to non-factively ground (p just is for there to be a
special type of argument from premisses (exactly) !':i to conclusion (/;--what
we can call a metaphysiml!y explttnat01y argument. If one accepts this con-
nection between ground and metaphysically explanatory arguments, the
notion of zero-grounding is unproblematic: a truth is zero-grounded if it
is the conclusion of an explanatoty argument from the empty collection of
premisses. The seemingly mysterious distinction between being ungrounded
and being zero-grounded is a special case of the familiar distinction between
not being derivable and being derivable from the empty collection of
premisses.
[n response to the second question, I do not simply postulate that non-
factive grounding claims are zero-grounded. If the claim that !':i non-factivcly
grounds if> just is the claim that there exists an explanatoty argument from !':i
to if> there are compelling reasons for holding that the claim that !':i non-
factivcly grounds cfr-if true-is zero-grounded. To substantiate this I show
how to develop a logic of iterated ground-the Pure Logic ofiterated Strict
Full Ground (PUSFC). A novel feature of PLISFG is that its deductive system
distinguishes between explanatory arguments and what we may call "plain"
arguments. This allows us to equip factive and non-factive grounding
operators with natural introduction and elimination rules. (In fact, the
rules are proof-theoretically harmonious.) Together these rules entail that
true non-factive grounding claims arc zero-grounded.
1.1 Overview of the paper
§2 explains how the various notions of ground are to be understood. §3
rehearses a serious problem posed by claims of iterated ground. §4 formally
states the ZCA. §5 sketches a graph-theoretic account of ground, discusses
how the graphs are to be understood, and shows how the zero-grounding of
non-factive grounding claims is a natural consequence. §6 show how we can
understand ground in terms of explanato1y arguments and develops a
deductive system distinguishing between explanatory and merely plain
Grounding Ground 281
arguments. §7 shows how to find introduction rules for the grounding
operators. §8 uses an inversion principle to find matching elimination
rules; these rules have the consequence that non-factive grounding claims,
if true, arc zero-grounded. §9 defends the ZCA against the objection that
every true 11011-factivc grounding claim has the same (empty) ground. § 10
compares the ZGA with the "Straightfotward Account" (SFA) offered by
Bennett (2011) and deRosset (2013) and argues that even the SF/\ needs
zero-grounding. The paper concludes with some issues for further work in
§11. There are two technical appendices. Appendix A states introduction
and elimination rules for the Pure Logic of Iterated Strict Full Ground
(PLISFG) in an "amalgamation-friendly" form to facilitate comparison with
Fine's Pure Logic of Ground (PLG). Appendix B develops a graph-theoretic
semantics and uses it to show that PLISFC: is a conservative extension of a
subsystem of PLG. 4
2. GROUND AND EXPLANATION
I take ground to be an explanatory notion. As I will understand ground, to
say that <p 0 ,<p 1 , •.. ground </> just is to say that ef> 0 ,ef> 1, ... explain if> in a
distinctively metaphysical way. 5 The explanatory connection between the
grounds and what they ground is vety intimate; following Fine, I take the
grounds to explain the grounded in the sense "that there is no stricter or
foller account of that in virtue of which the explanandum holds. If there is a
gap between the grounds and what is grounded, then it is not an explanatory
gap." (Fine 2012a, p. 39) 6 Here are some plausible cases of ground:
4
The graph-theoretic semantics can be extended to a semantics for all the grounding
operators of Fine's Pure Logic of Ground-and more besides. Ir is also possible to find
introduction and elimination rules for these operators-and more besides. This is a task
for another occasion.
5 I should flag a controversy here. That grounding is intimately connected with
explanation is widely accepted; that r's non-factively grounding 4, just consists in r's
explaining rf> in a distinctive way is not uncontroversial, though it is accepted by Fine
(2001, 2012a), and Dasgupta (2014a, 2016). An alternative view would take grounding
to be a (the?) distinctive relation of determination that "underwrites" such metaphysical
explanations. On this view grounding stands to metaphysical explanation as causation
stands to causal explanation. (Such a view is held by Audi (20 I 2b, p. 688), Audi (20 l 2a)
and Schaffer (2012, 2016).) I will not attempt to refute this position here. Many of the
claims made in this paper can, in any case, be appropriated by the defenders of this other
vic\v.
6
Fine (2012a, pp. 38-40) distinguishes between metaphysical, normative, and natural
ground. Here we will only discuss metaphysical ground: it is only with metaphysical
ground that the connection between the grounds and that which they ground is this
intimate.
282 Jon Erling Litland
(1) A conjunction is grounded in its conjuncts.
7
(2) A disjunction is grounded in its true disjuncts.
(3) a's being G is grounded in a's being F, where Fis a determinate of
the determinable G.
Several notions of ground have been distinguished in the literature. The one
that has been the frKus of the debate over iterated ground is factive, fit!!,
mediate, strict ground. Adopting the notation of (Fine 20 l 2a, 2012b) we
express claims of ground using a sentential operator "<". 8
"<"has variable arity on the left: if 6 is any set 9 of sentences and <pis a
sentence then the result of concatenating the sentences in 6 (in any order)
with < and </> is a sentence. 10 Since nobody thinks that the order of the
sentences in 6 matters to whether 6 grounds <p, we disregard order and
(ambiguously) write 6 < <p for the resulting sentence. 6 are here the grounds
while</> is the grounded. We can pronounce a grounding claim 6 < </>as"</>
because 6". Note that we allow both infinite and empty 6; the latter means
that < </> is well-formed. Since the sentential operator locution can be
cumbersome we often nominalize and, unofficially, speak of grounding as
a relation between truths.
7
Though this is problematic because of the paradoxes of ground discussed by Fine
(2010a). We set the paradoxes of ground aside for the purposes of this paper.
8
I here follow Fine (2001, 2012a, 2012b), Schnieder (2011), Dasgupta (2014b), and
Correia (201 O) in expressing ground by means of sentential operators. The alternatives
discussed in the literature are (i) to treat grounding as a relation between facts (Rosen
2010; Audi 2012a, 2012b; dcRosset 2013; Bennett 20 l l; and Trogdon 20 l3a; and (ii) to
treat grounding as a relation that can hold between objects in arbitrary ontological
categories (Schaffer 2009). The reasons for expressing ground by means of a sentential
operator are frankly pragmatic. Erst, by expressing ground using sentential operators one
can remain neutral on some vexed issues concerning the existence and nature of facts (cf.
Fine 2001, 20 l 2a, 20 l 2b; and Correia 20 l 0). Secondly, it is easier to formulate a logic of
ground if one expresses ground using sentential operators. Thirdly, the crucial notion of
zero-grounding is also easier to make sense of if we express claims of ground using
sentential operators. That being said, it would be possible to reformulate much of what
follows if one favored treating grounding as a relation between facts.
9 It might be better to let /::,. be a multiset (see footnote 54). In the main text I have
opted for simplicity.
10 This is a mttrzy-one notion of ground. One might want to allow< to have variable
arity on the right as well: when r and /::,. are any two sets the result of concatenating the
sentences in r with < <rnd /::,. is a grounding claim. Dasgupta (20 l 4a) argues that there is a
non-distributive notion of many-many ground. r < 6. can be the case without there being
o
an I such that I'= U;E/r;, /';. = U;E/O; and I';< 1 for each iEI. Lidand (2016b)
extends Fine's truthmaker semantics to develop a logic of many-many ground. While
such a notion of ground is of considerable interest we will not discuss it further in this
paper-though see footnote 35.
Grounding Ground 283
The following logical features are commonly taken to hold of<. We will
take them as adequacy constraints on any logic of the operator <: if a
proposed logic of ground fails to validate these features we will not accept it.
< is factive in the sense that if 6 < <p, then ef> and each 8 E /:,. is true. It is
fit!! in the sense that if 6 < </>is the case then nothing need be added to 6 in
order to explain why it is the case that ef>: its being the case that L~ fully
accounts for its being the case that <p. It is mediate in the sense that we allow
6 to ground </> by way of grounding some t/; that also grounds ef>. 1 1 It is strict
in the sense that if 6 < </> then it is impossible for each 8 E /::., to be the case
while </> helps explain a o' E 6. In the special case of one-one grounding
daims-i.e. claims of the form </> < t/1-this has the consequence that
grounding is im:_fiexive, transitive, and asymmetric. 12
The claims of ground in (1), (2), and (3) above are all plausible cases of
factive, strict, full, mediate ground.
Central to the ZGA is a notion of non-factive (full, mediate, strict) ground.
Whereas only truths can factively ground, even falsehoods can non-factively
ground. We use"=?" as a sentential operator for this notion; it has the same
grammar as<. Much more will be said about=? later; for now it suffices to
know that if (each y in) r is the case and r =} </> is the case then r < 4,. u
The claims (1), (2), and (3) are also true non-factivc grounding claims.
3. A STATUS PROBLEM
While claims of iterated ground are interesting in their own right they also
give rise to serious problems. Since problems of this sort are now well-
known, we consider only a simple case. 14 (For ease of expression we talk as if
11 More generally: we allow that 6. grounds <p by way of there being a decomposition
6.0,6.1, .. .,6.,,, of 6. and some if; 0,,P 1,... such that 6.1 grounds if1 1 for each i and •/io,•/1 1,... ,6.,,,
taken together groun<l q,.
12
I should note that these principles have been contested. Jenkins (2011), Wilson
(2014), Correia (2014, §7.3), and Kramer (2013) doubt that grounding is irreflexive;
Dasgupta (20 I 4b) doubts that grounding is asymmetric; and Schaffer (2012), contm
Schaffor (2009, pp. 375-6), argues that grounding is not transitive. For a defense of the
principles, sec Raven (2013) and Litland (2013).
13
For more on the distinction between factive an<l non-factivc ground, sec Fine
(2012a, pp. 48-50).
1'I Essentially this problem is discussed by Bennett (2011), deRosset (2013), and
Dasgupta (2014b). The problem does not essentially turn on every truth being grounded
in some ungrounded truths. Problems like this arise even if one allows infinitely descend-
ing chains of ground. Sider (2012, pp. 143-5), for example, formulates a Status Problem
turning on Fine's notions of a fact's being constitutive rf reality and its bcingfactual that p
(for which notions, see Fine 200 l, 201 Ob).
284 Jon Erling Litumd Grounding Ground 285

grounding was a relation between truths. Note that the official formulation the above trivial manner (cf Dasgupta 20!4b). Second, whether or not it is
would require quantification into "sequence-of-sentences" position.) a substantive matter whether every objecr is 0-fundamcntal the other views
Consider first the following principle: on iterated ground do avoid the conclusion that all objects are 0-
fundamental. For dialectical purposes the ZCt\, too, had better not have
(Foundation) There are some truths /'I,. such that each o
EL\ is this consequence.
ungrounded and such that each truth <P is either in /'I,. or else is grounded
in some /'1,. 0 <;;; 1'1. Let us call such a collection of truths a foundation.
Many philosophical disputes can be construed as disputes about whether 4. THE ZERO-GROUNDING ACCOUNT
there could be foundations consisting only of certain types of truths. For
instance, a physicalist might think that there is a foundation consisting only Recall that when /'I,. is any number of sentences and <Pis a sentence, then /'I,. < <P
of physical truths: no mental truth need be entered into the foundation. and /'I,. =? <P are sentences; in particular, if /'I,. is 0, the empty collection of
A metaethical naturalist might think that there is a foundation consisting sentences, then < <P and =? <P arc sentences. If < cp, then <P is grounded:
only of non-normative truths: no normative truth need be entered into the there is a collection /'I,., namely the empty collection, of sentences such that
foundation. /'I,. < <Pis true. Let us, for now, take the notion of zero-grounding for granted
A physicalist might want to go further. Not only is there no need for and see how it allows us formally to solve the problem of Iterated Ground.
mental trnths in the foundation; distinctively mental objects and properties The ZCA holds that /'I,. < <P is grounded in /'I,. together with /'I,. =? <P and
arc not required either: 15 there is a foundation :F such that no mental that /'I,. =? <P is strictly fully zero-grounded. That is, it is the case that
objects and properties arc constituents of the truths of :F. 0 < (/'I,. =? <P). We avoid the conclusion that every object is fundamental.
What a physicalist might want to say is that no mental object is If /'I,. < <P is the case then this is grounded in 6 together with /'I,. =? cp. The
0-f\rndamental in the following sense: latter is zero-grounded and hence grounded. If the claim <P concerns the object
(Object-Fundamcntality) An object a is Ofundamental iff the object a a, a will not be 0-fund<unental just on account of occurring in 6 =? cp.
figures in an ungrounded truth. What about the claim that /'I,. =? <P is zero-grounded? If this claim is
ungrounded, the problem is only pushed back. 'T'he solution is obvious: if
If that is right, the physicalist is committed to: a truth i/1 is zero-grounded then the truth that 1/1 is zero-grounded is itself
zero-grounded.
(Derivative Objects) There is an object a that is not 0-fundamental.
We then get the following sequence of grounding claims:
The problem is that if truths about ground are themselves ungrounded then
every object is 0-fundamental. For consider some object a: either there is /'I,. < <P (/'1, 6 =? <P) < (/'I,. < <P)
nothing that grounds the truth that a exists or there is something that [(1'1,!'.l =? q,), ((ll,6 =? <P) =? (1'1 < 4'))] < ((1'1,1'1 =? <P) < (6 < 4')) ...
grounds this truth. If the former, then a is 0-fundamental since the truth
that a exists is fundamental. If the latter, suppose that the truth that a exists
Since 0 < (1'1 =? <P) we simplify and get: 16
is grounded in the truths r. Then this truth, namely the truth that a's 1'1<(1'1<4') /'I,.< (1'1 < (6 < <P))
existence is grounded in r, is a further truth. If this truth is ungrounded then
we again get that a is 0-fundamental since the truth that the existence of a is The ZGA gives rise to infinitely many grounding claims involving q,. We do
grounded in r will be in every foundation. not, however, have an infinitely descending chain of ground with <P on top:
We will take it as a constraint on an account of iterated ground that it not <P is grounded in /'1, but since <P is not grounded in /'I,. < <P there is no
commit us to every object's being 0-fimdamental. There arc two reasons regress. 17
for imposing the constraint. First, the question whether every object is
0-fundamental seems like a substantive one: it should not be settled in 16
I assume the Cut principle: if I'< <P and 6,</i < ifl, then f,/:\ < q,. (The relevant case
is r == 0.)
17
For more discussion of why there is no problem, sec Bennett (2011, pp. 33-5),
15
See dcRossct (2013, pp. 3--6) for some reasons for thinking this. deRossct (2013, pp. 19-20), and Rabin and Rabern (2015). In any case, if there were a
286 Jon Erling litland
There is a superfici<tl similarity with the accounts of deRosset (2013) and
Bennett (2011). Their view is that if/::,, < <p then what grounds this is just/::,,
itself. 18 Let us call this the Straightfarwrtrd Account (SFA). We will discuss
the SFi\ in greater detail in § l O; here just note that the ZGA and the SFA differ
even though they agree that !'i. < <p only if !'i. < (/: , < </>).The ZCA holds that
!'i. < </> is partially grounded in /::,, =? ef>; the SFA denies this. 19
To move beyond the merely frmnal we must answer the following two
questions.
(4) What exactly does it mean to say that something is zero-grounded?
(5) How can we make it intelligible that truths of the form !'i. =--? </> are
zero-grounded?
The question in (5) is not a request for the grounds for something's being
zero-grounded: it is part of the view that if</> is zero-grounded, then the
truth that</> is zero-grounded is itself zero-grounded (and so on). Rather, one
is asking for a story making it comprehensible that something is zero-
groundcd. It is true that by taking non-factivc grounding claims to be
zero-grounded (if true) we solve the Status Problem. \Xlhile this provides
some reason to postulate that they are zero-grounded (if true), it does not
help us understand why they should be zero-grounded. The key to answer-
ing these questions lies in answering the following one:
(6) How should the notion of non-factive ground be understood?
We want to end up in the following situation. Consider conjunction: there
is no mystery about why a true conjunction </>/\if; is fully grounded in its
conjuncts </>,if1 taken together. Once one understands what conjunction is
one understands that a (true) conjunction is grounded in its conjuncts.
Similarly, once we get clear on what non-factive grounding is there will be
no mystery why a true claim of non-factive ground is zero-grounded.
problematic regress here a similarly problematic regress would arise for the other accounts
of hcrated Ground: the ZGA is no worse off
18
A similar view is tentatively suggested by Raven (2009).
19
If we avail ourselves of the notion of immetlir1te strict full ground we can state the
difference perspicuously. (For the distinction between mediate and immediate ground, sec
Fine (2012a, pp. 50-1).) According to the ZGA, 6 < <P is not immediately strictly folly
grounded in 6-the immediate grounds for 6 <<Pare 6, (6 =?¢)taken together; 6 < <P
is only mediatedly strictly fully grounded in 6. According to the SFA, on the other hand,
6 < <P would be immediately strictly fully grounded in 6. Bennett and deRossct admittedly
do not discuss the problem in terms of immediate strict foll ground, but it seems clear that
the natural way of developing their view is by insisting that 6 is the immediate full ground
for 6 < rp.
Grounding Ground 287
5. GROUND, MACHINES, AND GRAPHS
To sec both what zero-grounding is and why non-factive grounding claims
would be zero-grounded (if true) it is useful to begin with a picture of grounding.
Think of a machine generating truths from other truths. 20 The machine
is fed truths, churning out truths grounded in the truths it is fed. A truth is
ungrounded if the machine never churns it out; a truth is zero-grounded if the
machine churns it out when it is fed no input. In terms of this picture, why
would the machine give the verdict that !'i. =? </> is zero-grounded if true?
Think of it like this. When the machine is fed no input the machine, instead
of remaining idle, "simulates" the results of being fed various input. In
simulating what happens when it is fed the propositions !'i. the machine
proceeds just as it would have if it in fact had been fed/::,, as input. If, when
running the simulation, the machine churns out <f;, the machine ends the
simulation and churns out 6 =? <f>. Since the machine was fed no input this
means that !'i. =? </> is zero-grounded if true.
The machine picture is closely related to a graph-theoretical picture often
employed in discussions of ground. Let us develop this picture both for non-
factive and factive ground. 21 A directed hypergraph is a tuple g = \V,A,t,h).
Here V is a collection of vertices-think of these as propositions. A is a
collection of hyperarcs. t,h arc functions t,h: A--> P( V). If A EA, t(A) is
the tail of A and h(A) is the head of A. We demand that h(A) is a singleton. 22
Assume that the vertices are propositions; we can then speak of an arc A as being
from the propositions </> 0 ,</> 1, ••. to the proposition <p. In terms of the machine
picture, an arc A EA corresponds to the application of a mechanism inside the
machine. t(A) are the propositions the mechanism operates on; h(A) is the
result of the mechanism's operating on t(A). 23 For simpliciry, we assume that
A is chained: if Ao,A1, ... ,B are arcs such that t(B) = v{v0 ,v1,... ,w0 ,wi, ... }
and h(Ao) = vo,h(A1) = V1 ,... , then there is an arc C such that
t( C) t(Ao) U t(A1) U ··· U { wo,w 1,... } and h( C) = h(B). 24
The following two graphical representation of a hyperarc with tail
</> 0 ,</> 1 , •.. and head</> are usefully kept in mind.
20
The analogy is from Fine (2012a, pp. 47-8). The extension to non-factive ground
as well as the idea of "simulation" is new here.
21
Schafler (2009) and deRossct (2015) have also used graph-theoretic ideas in
connection with grounding. (See also Schaffer (2016) for a related approach.)
22
To capture a many-many notion of ground we would lift this restriction.
23
Note that A represents not the mechanism itself but rather an application of the
mechanism. We will see why this might matter in §9.
24
Without this assumption the graphs would capture immediate and not mediate
ground. In many ways it would be better not to assume that A is chained; but this raises
technical problems that arc best set aside here.
288 }on Erling Litumd
cf>o
cp
c/>o </>1
-----~--~---·~
cp
We can now say that some propositions ef> 0,cf> 1,... non-fizctiveiy ground a
proposition cf> iff there is an arc A such that h(A) is {cf>} and t(A) is
{ cf> 0 ,cf> 1, ... }. A proposition cf> is (non-factively) zero-grounded if there is an
arc A with the empty tail the head of which contains the proposition cf>.
A proposition cf> is ungrounded if there is no arc A with h(A) = {cp}.
To deal with factive ground we need the notion of a pointed directed
hypergraph. This is a tuple ( V,F,A,h,t). Herc ( V,A,h,t) is a directed
hypergraph and F <;;;; V is a set of vertices dosed under taking heads; that
is, if Fo <;;;; F and A EA is such that t(A) = F0 then h(A) <;;;; F. 25 Think of
F as the actually obtaining propositions-the facts. We now say that
cf>0 ,cf> 1 , ..• factiveiy ground cf> if cf> 0 ,cf> 1 , ... <;;;; F and c/> 0 ,c/> 1 , .•• non-factively
ground cf>. Since F is dosed under taking heads every truth that is non-
factively zero-grounded is in F.
Graph theory gives us nice formal models of ground, but how does it help
explain why truths of the form f.. =? cf> are zero-grounded?
Consider an arc A from some propositions c/> 0 ,c/> 1 , .•• to a proposition cf>.
What does this arc represent? One possibility is to take A to represent the
truth that cf>(Jlc/> 1,... non-factively ground cf>. I would like to defend a different
view. What represents that cf> 0 ,cp 1 , ... ground cf> is not an arc with tail
{ cf>o ,cf> 1 , ... } and head cf>, but rather the proposition that there is such an
arc. What, then, do the arcs represent?
If we adopt the machine picture we can say that an arc from c/> 0 ,c/> 1 , ... to cf>
represents an application of a mechanism to the propositions q>0 ,c/> 1 , •.. , an
application that yields the proposition q>. (And so: what represents that
cf>0 ,cf> 1 , ... ground cf> is not the application of a mechanism taking us from
cf> 0 ,cf> 1 , ... to cf> but rather the fact that there is an application of a mechanism
taking us from input cf> 0,c/> 1, ... to output cf>.)
If one thinks of the arcs as applications of mechanisms the arcs are not
propositional; and it does not make sense to <L~k what grounds them. However,
even if the application of a mechanism taking us from input c/>0 ,c/> 1 ,. .. to output
cf> is not propositional-and so not apt to be grounded·-there is still the
proposition that there is such an application of a mechanism. The question
what grounds this proposition is legitimate. If we think of the proposition
25 Some further conditions need to be imposed, but we do not have to
now. See Appendix B.
Grounding Ground 289
cf>0 ,cf> 1 ,. •. =? cf> as the proposition that there is an application of a mechanism
with input cf> 0 ,cp 1 , ••• and output q> the above story about simulation suggests
why (true) propositions of the form cf> 0 ,cf> 1 , ... =? cf> should be zero-grounded.
Suggestive though the machine picture is, it is not to be taken literally.
We should provide a less pictorial account of what the arcs represent and
what claim is being made by a non-factive grounding claim c/> 0 ,c/> 1 , ... =?cf>.
6. GROUND AND EXPLANATORY ARGUMENTS
Above I wrote that we may think of an arc between some propositions
cf> 0,cf> 1,... and a proposition cp as the application of a mechanism to the
propositions cf> 0 ,cf> 1 , ... ; in the introduction I wrote that I would tie ground
closely to explanatory arguments. These pictures are related. Think of an arc
between cf> 0,cf> 1, ... and cf> as an explanatory inference from c/> 0,c/> 1, •.• to cf>. We
may represent this graphically as follows: c/>o c/>i ··· . More
· genera 11 y, th'm I<
of a hyperarc A as an explantttory c1rgumentfrom~he propositions in the tail of
A to the proposition in the head of 11. 26 One can now think of the machine
as encoding the explanatory arguments, with different explanatory argu-
ments corresponding to different mechanisms.
We can now say that cf> 0 ,cf> 1 , ..• non-factively ground cf> if there is an
explanatory argument from c/>(p</> 1 , ... to cf>. And we can say that c/> 0 ,c/> 1 , ...
factively ground cf> if there is an explanatory argument from c/> 0 ,c/> 1 , ... to cf>
and cf> 0 ,cf> 1 , ... are the casc. 27
A main contribution of this paper is showing how we can develop a
mathematically rigorous theory of metaphysically explanatory arguments;
having done this a satisfactory logic of iterated ground drops out naturally.
I will not attempt to give a non-circular account of what makes some-
thing an explanatory argument: the notion is taken as a primitive. That is
not to say that the notion cannot be elucidated. The intuitive idea is that the
explanatory arguments are composed from basic explanatory inferences.
Plausible cases of explanatory inference are conjunction-introduction,
disjunction-introduction and the inferenze from a is F to tl is G-where
F is a determinate of the determinable G. (These correspond to the
uncontroversi<tl cases of ground mentioned in §2 above.) More generally,
whenever one thinks that [ immediately grounds cp, the ZGA holds that the
inference from r to cf> is explanatory. Flippantly put, one obtains basic
26
It is in order to sustain this interpretation that we require that the hyperarcs be
closed under Cut.
go into this 27
deRosset (2013, pp. 12-13) comes quite dose to this idea, bur he focuses on factive
grounding and is not (there) trying to develop a logic of ground.
290 Jon Erling Litland Grounding Ground 291

explanatory inferences as follows. Take a claim of strict full immediate labeling the nodes in D(s) is the set of (undischarged) premisses of T.
ground--r < </> for instance-rotate it 90° clockwise and replace < with If</> labels nodes we often abuse notation and write D( </>) for D(s).
a horizontal line: the result is an explanatory inference 5. When we write
It is appropriate for a logic of iterated ground to rcniain silent on which
arc the basic explanatory inferences: which particular inferences arc explana-
tory is (by and large) a material and not a formal matter and so one on which
logic remains silent. That being said, for the purpose of illustration, let us we mean that E is an argument with conclusion ef>. It is often important to
assume that conjunction and disjunction introduction arc explanatory note what the premisses of an argument E are. The notation
inferences in this sense. If one thinks that a conjunction is not grounded
in the conjuncts and a disjunction is not always grounded in the true r
disjuncts one should substitute some inferences that are obtained from E
one's favorite claims of (immediate) ground in the way indicated above. </>
What logic can do is tell us how exphnato1y inferences interact and how
they arc chained together to form explanatory arguments.To do this we need a is to mean that E is an argument with conclusion </> where each occurrence
deductive system that distinguishes between two rypes of argument. (This is of a proposition y Er labels some node in D( </>). (We do not demand that
the first technical innovation of this paper.) Let us use uppercase calligraphic each node in D( </>) is labeled with a proposition in r.)
letters E,V,F,. .. (possibly with subscripts) as variables over arguments. One Rules of inference that involve discharge will be written.
rype of argument is the explanatory argument: if there is an explanatory - - - - - o, 1, 2, ...
argument E from premisses t:. to conclusion ef>, then if 6 is the case its being (/>o, <Pi, </>2,
the case that 6 fully explains its being the case that ef>. We also have the plain
arguments: if there is a plain argmnent E from 6 to ef>, then if 6 is true, ef>, too, is E
true. With a plain argument, however, there is no guarantee that f..., if true, </>
explains ef>. 28 To indicate that an argument Eis cxplanato1y we may write£( e); ~0 1 11 2, ...

similarly, we may write E(p) to indicate that the argument Eis merely plain.
Let us be more rigorous about what exactly arguments are. An argument is
This is to be understood as follows. E is an argument with conclusion (/> where
a quadruple T ( T, ~, L,D) where ( T, ~, L) is a labeled rooted tree. ( T is
D( </>)are labeled with (amongst others) the propositions ef>0 ,</> 1 , •••• In passing to
the set of nodes, ~ is the tree-order and L : T _, P is a function assigning
the conclusion if; we can discharge any node labeled with one of the </>;· More
labels to the nodes of T. We may think of P as the class of propositions.) We
precisely, if S <::; D(ef>) is any set of nodes with labels only from ef> 0 ,ef> 1 , .. ., any
demand that there is T(l <::; T such that each t E T0 is a top node and such that
argument of the above form where D( 1') = D( </>)\Sis an application of the rule.
for each sE T there is tE T0 such thats~ t. 29 Dis a function T -> P(T).
(This means that unless otherwise specified we allow both vacuous and
For each sE T, D assigns a set D(s) <::; {t 2: sit is atop node}. This
multiple discharge: if a rule allows the discharge of premisses of a certain
"discharge function" allows us to keep track of which assumptions a line
form we can discharge any number of premisses of that form.)
in the argument depends on. (This is required since we will consider argu-
When r = y 0 ,y 1 , ... we often use the following more compact notation
ments that discharge assumptions.) The root of the (sub)tree is the conclusion
of the (sub)argument.m Ifs is the root ofa (sub)tree T, then the propositions

28
I should note that there is nothing in the formalism forcing plain (or even strict)
arguments to be, say, classically valid arguments. This is important: one should not rule
out that 6. grounds ef>, even though 6. docs not logically entail <p.
29
Because of examples of the sort given in Dixon (2016), Rabin and Rabern (2015),
and Litland (20 ! Ga) we cannot require that there <U"C no infinite branches in T.
30
Rather: the proposition labeling the root of the tree is the conclusion of the
argument. We will be sloppy about this since there should be no cause for confusion. to indicate that WC can discharge any of the y Er.
292 Jon Erling !,it/and Grounding Grottnd 293

lNCUJSION Any explanatory argument is a plain argument. that rg;

~~ 18~;, IliJp ~ where (1S~,l!iJ;) is closed under the constraints in
Figure 12. l. Some explanation of these principles are in order.
AssUMPT!ON For any (p, (p is a plain argument fron1 <f to cf
(INCLUSION) is straighd(Hward. If its being the case that r would explain
NoN-C:IRCULARITY If Eis an explanatory argument frorn prerr1isscs (p, Do, 31, ••• why <fa is the case, then it is certainly true that if r is the case then <p is the
to 4J, and f) i.s a plain argument fron1 f' to(/> and f)i is a plain argurnent case. (ASSUMPTION) is standard natural deduction: we can write down any
to 8; for each i, then for any 1/1 the following is a plain argument from assumption we like, with the result being a plain argument. (PLAIN CrIAINTNc;)
!', 2' 0 , 6t,.·· to 1/,: is also unproblematic: it simply tells us that we can chain together plain
arguments to get plain arguments.
[" ... 6; .. .
(OIAININC) tells us that the result of chaining together explanations is
lJ ... D, .. .
itself an explanation. One might be concerned about this if one thought that
rp ... ~i···
explanation is not transitive. Whatever one thinks about that in general
£ there should be no problem here: explanatory arguments correspond to full
q,
ground. If f grounds <fa then r provides a full account of why it is the case
if1
that <fa. It is hard to see how there could be counterexamples to transitivity
for this notion of ground. 31 In any case, since we want to capture mediate
PLA1N C!tAINING If E; is a plain argument from 6; to rP; for each i, and Dis a
ground we can always insist on closure under (CIIA!NINC).
plain argument to <p from 1p 0 , ¢ 1,. • ., !"then
(NoN-CIRCUIJ\RlTY) requires further explanation. (We take take up some
more philosophical issues in §§6.1 and 6.2.) The goal is to have explanato1y
2'o 61
arguments correspond to strict ground in the sense that <Po ,<fa 1 ,. .. < <fa iff
Eo £1
... i- each of the <P; is the case and there is an explanatory argument from </y0 ,4> 1 ,. ••
<Po 1
1'1
D
to <fa. Since (partial strict) ground is irreflexive it cannot be possible for some
<p </J,rflipif1 1,... to be the case and for there to be an explanato1y argument from
</J,r/io,rP1 ,... to <fa.
is a plain argument from 2' 0 , /1. 1,. . ., I" to rf>. It will not do simply to say that there arc no explanatory arguments from
</y,r to <fa, for any r. The problem is that this tells us nothing about what
If E; is an explanatory argument from 6; to rP; f(ir each i, and Dis an
happens under the supposition that 4, (partly) strictly grounds itself. And we
CHAtNING

explanatory argument to <p from ¢ 0 , 'PI ... ., f' then

need to use the assumption that <fa contributes to grounding itself in
subordinate arguments. (NoN-C!RCUIJ\RITY) gets around this problem by
60 61
expressing the irreflexivity of ground as a closure-condition on the class of
Eo E1
... I'
strict and plain arguments. If, per impossibile, <fa did contribute to explaining
r/>n 1
P1 32
<fa, then we can conclude anything-albeit only plainly.
])

"' 31
In view of this, it is significant that the putative counterexamples to transitivity in
is an explanatory argument from /1. 0 , /1. 1, .. ., ['tor/>. Schaffer (2012) are all couched in terms of partial ground. For responses to the counter-
examples see Raven (2013) and Litland (2013).
32
Figure 12.1 Arguments explanatory aud plain Note that (NON-C:IRCLJl.AR!'JY) docs not simply take the form:

f',r/>
The principles governing how explanatory and plain arguments interact E(e)
are depicted in Figure 12.1. Officially, Figure 12. l is understood as follows. -r/>N on-ercu
. Iartty
. '·
Assume given a collection of basic explanatory and plain arguments (ll<~,,ll<~p). if1
(One may think of these as collections of explanatory and plain inferences.) For suppose that there is an explanatory argument from r/> (and some further premisses to
The explanatory and plain arguments over (!E 0 !Ep) is the least (!E~,!E;) such r/> itself). Suppose fonhcr that r/> follows plainly from l'i. \Y/c should be able to conclude
294 Jon Erling Litland
Finally, the discharge function D behaves as one would expect. 33
Three observations about these rules. First, none of these principles
ensure that there are any explanatory arguments: they only tell us how
explanatory arguments combine to form further explanatoty arguments.
This is as it should be: it is not part of the job of a pure logic of ground to
ensure that there arc cases of strict ground.
Second, note that the principles in Figure 12. l respect the non-monotonicity
of ground: if there is an explanatory argument from 6 to</> there need not be an
explanatoty argument from 6,1/; to ef>.
Third, the principles deal exclusively with explanatory arguments; no
mention is made of explanatory inference. By developing a theoty of
explanatory inference we would be able to develop a logic of immediate
ground. W c refrain from doing so for two reasons. First, the previous
discussions of iterated ground have focused on mediate ground; it is con-
venient to follow suit. Second, there are some technical problems in express-
ing that mediate ground is the closure of immediate ground under Cut.
These problems can be solved, but the most convenient way of solving them
involve both weak34 and many-many ground. 35 Developing this machinety
would be a distraction.
6.1 Is non-factive ground non-circular?
Explanatory arguments correspond to non-factive full ground in the sense
that </>rp</> 1 , ... non-factively ground </> iff there is an explanatory argument
anything from 6,L But since the argument from Li to q, is merely plain if we use (l'l,\IN
CHAlNlNG) we get the argument
Li
q, r non-factively) it has to be possible for the propositions in r to be jointly
E(p)
q,
This argument is only plain, which means we cannot apply Non-circularity*. (NoN-
ClRCULARlTY) gets around this by building some chaining into the rule.
33 To be pedantic. If q, is an instance of (AssuMHION) D( q,) = q,. In an instance of
(CllAlNlNG) D(</>) = U;D(ef>;) U D(I'), where D(r) is the set of nodes decorated by the
propositions in I'. In an instance of (NoN-CIRCULARlTY) D(~1) = U;i)(o;) U D(ef>).
3
" For the notion of weak ground, see Fine (2012a, pp. 51-3; 2012b, pp. 3-4).
35 Since the notion of many-many ground is of considerable interest it is of some
importance that the z,;A can be extended to accomodate it. This can be done by treating
arguments as certain directed, acJ'clic hypergraphs, where we now allow the head of an arc
to be of any cardinality. This allows us to model inferences with several simultaneous
conclusions-where the conclusions are read conjunctively. This differs from wmdard
multiple-conclusion natural deduction-see e.g. Read (2000)-where one reads the
multiple conclusions disjunctively (as in multiple conclusion sequent calculus). I hope
to return to these matters elsewhere.
Grounding Ground 295
from </> 0 ,</> 1, ... to</>. While (NoN-C:IRCULARITY) ensures that factive ground is
asymmetric, (NoN-C:IRC:UL1\RITY) does not ensure that non-factive foll
ground is asymmetric. The reason is that the (NoN-C:IRCULARrIY) does not
discharge the premisses on which </> depends. y;
This is not an oversight: there is, arguably, nothing wrong with explana-
tory arguments from </> (and some further) premisses to </> itself: To see this
consider the following situation. Suppose that a is part of b but it is possible
that b instead is part of a. (a might be an organism that has entered into the
body of organism b where a is now fulfilling some function inside b's body;
but it might equally well have been b that had entered into the body of a.)
Then, while the existence ofa (partly) strictly grounds the existence of b, it is
possible that the existence of b (partly) strictly grounds the existence of a.
There would then be an explanatory argument from the claim that tt exists
(and some further premisses I'a) to the claim that b exists. But there is also an
explanatoty argument from the claim that b exists (and some further
premisses rb) to the claim that a exists. But then, by (CHAINING), there is
also an explanatoty argument from the claim that a exists (and the premisses
rll ,rb) to the claim that {/ ex.ists.
While (NoN-CIRCULARTIY) docs not rule out arguments from</> (and some
I') to</> itself, (NoN-C:IRCUI.t\RnY) ensures that</> and I' cannot jointly be the
case. In particular, (NON-CIRCULARITY) ensures that if there is an explanatory
argument from just</> to </> itself then </> is not the case.
6.2 Impossible explanations
We have taken the line that in cases like the above we have a genuinely
explanatory argument; it is just that it is impossible for its premisses to be
jointly true. One might balk at this, insisting that for r to ground </> (even
true. If one docs so insist, the above case shows that one has to reject
(CHAINING) for explanatory arguments. Should we so insist?
I think this would be a mistake. In fact, not only do I think that there are
explanatory arguments with impossible premisses; I also think there are
explanatory inferences with impossible premisses. Above I mentioned that
I take conjunction introduction to result in explanatory inferences: this is so
even in a case like </> •</> where both the premisses and the conclusion
of the inference are i~~o~tble.The reason is that I take the explanatoriness
of an inference to be a matter of the farm of the inference; all inferences of
36
I <tm grateful to two anonymous reviewers for making me realize that I had to
address the issues discussed in this and the next subsection.
296 Jon Erling Lit!mui
that form are explanatory. It would take us too far afiel~1 fully to defend this
view here, so let me offer a more concessive response.v
We could agree to reserve the words "explanatory argument" ("infer-
ence") for an argument (inference) where the premisses can be jointly true.
(Arbitrarily chaining together explanatory inferences gives us the shmexp!a-
natmy arguments.) While there cannot be an explanatory argument from </>
(and some premisses r) to ~h itself there might be a shmexplanatoty argu-
ment. What matters for my purposes is that if there is a shrnexplanatory
argument from <p (and r) to </> something has gone wrong and we can
conclude that not all the propositions in I' U { </>} are true. Non-factive
ground in the present sense will then be correlated not with explanatory
arguments but with shmexplanatory arguments. Whenever a shmexplana-
tory argument has true premisses, however, it will be explanatory and so
the relationship between factive ground and explanatory arguments
is preserved.
I will continue to refer to explanatory arguments; if one has scruples one
should mentally substitute "shmcxplanatory argument" for "explanatory
argument".
7. GROUNDING OPERATORS:
INTRODUCTION RULES
We can now give introduction rules for the non-factive grounding operator
=>. Since cf>0 ,</> 1, ••. => </> is meant to report that there is an explanatory
argument to</> from cf> 0 ,cf> 1,. •• its introduction-rule looks like this:
Jl Let me briefly mention some connections and issues. first, the notion offomz is
much broader than logical form narrowly construed. For instance, I hold that the
. c is burgundy . . . . . .
mfercnce . I · (where c 1s a particular name) 1s explanatory m virtue of its
c 1s rec
form. Secondly, there arc important connections between the view that explanatoriness
is a matter of form and the principle of Fonna!ity (Rosen 2010, pp. 131-2; see also Audi
2012b, pp. 697-8) and Fine's idea of "generic ground" (Fine 2016). Thirdly, for a
particular inference, we obtain its forms (plural) by replacing particular constituents in
the inference with schematic lecters. The above burgundy-reel inference has a form
x is burgundy . explanatory inference. We then ask the machine to simulate the result of
. . Fourthly, a familiar point: an inference has many forms, and it need
x is red
not be explanatory under all its forms. (For instance, the burgundy-red inference is not
c
exp 1anatory unc1er the 10rm: E(x))
- ( ) . Can we maI<e sense o f a notion o f a "canonica l'" 10rm
c
I; x
of an inference? And can the explanatoriness of a particular inference always be explained
by reference to the cxplanatoriness of its canonical form? I hope to discuss these issues at
greater length elsewhere. Thanks to Ralf Bader for discussion of these issues.
Grounding Ground 297
0. I, 2, ...
</>o' </>1 '· · · ,
[
J A ·-·'> , o, 1, 2, ... , =>··-Introduction
ip0 , 1p, ... _.,, c1 _J
Herc [ is an explanatory argument and D( 4>) is labeled by all and only the
propositions </> 0 ,</> 1 ,. •.• In this case, we demand that we discharge all
the propositions on which </> depends; that is, D( cf> 0 ,<p 1 , ... => <p) == 0.
(Note the similarity to the introduction rule for a strict conditional: =>
stands to the explanatory arguments as the strict conditional stands to
deductively V<llid arguments (Scott 1971) .)
We have to discharge all and on(y the premisses on which <p depends
because => is to capture non-factive full ground. If 6 =? </> is the case then 6
(if true) provides a full explanation for why <p is the case; moreover, every
13 E 6 is relevant to explaining why </> is the case. This is captured by
discharging all the 13 E 6.
Since we arc working with a deductive system with two types of argument
it does not suffice to specify that arguments of a certain form are to be valid;
we also have to specify whether arguments of the relevant form are to be
explanatory or merely plain. We will treat arguments of the form depicted in
the =>-I rule as explanatory. What justifies us in so treating them?
The only reasonable alternative view would require more than an
explanatory argument [ from 6 to </> in order to allow us to conclude
6 => cf>; in addition, one would require the premiss that [ is explanato1y.
(If one adopted such a view the question would naturally arise what grounds
the truth that [ is explanatory.)
We should resist this view. What is needed to conclude 6 => cf> is just an
explanatory argument [ from 6 to cf>; there is no need for the fi.trther truth
that [ is explanatory. The requirement that we need this further truth is as
inappropriate as the demand that in order to apply conditional proof we
need not just a valid argument V from </> (and some further premisses) to if1,
we need, in addition, the premiss that V is valid.
It might be helpful to think about this in terms of the machine picture.
To determine whether 6 => </> we go to a machine that encodes every
being fed input 6. The machine then procedcs to run the simulation. If the
machine churns out </> it also churns out 6 => cf> and ends the simulation. At
no step in this process is it necessary for the machine to check whether the
inferences it carried out were explanatory.
Similarly, in order to check whether we can conclude the conditional
</> -> 1/1 we might go to a machine that encodes all the logically valid
298 Jon Erling Lit!rmd
inferences and ask it to simulate the result of being fed the proposition (p. If
the machine churns out 1/;, the machine stops its simulation and also chums
out</> --; 1/;. At no point is it necessaty for the machine to check whether the
inferences it carried out are logically valid.
Having =? in place we can state introduction rules for the factive
grounding operator <:
<-·Introduction
Arguments of this form, too, will be explanatory. Note the similarity to
conjunction-introduction; note also that the following is an instance of<-
introduction: ;: <-r
In this setting there is nothing mysterious about the notion of zcro-
grounding. A truth ¢ is zero-grounded if there is an explanatory argument
from the empty collection of premisses to the conclusion <f>. If there is an
explanatoiy argument £ from I' to ¢, we can now show that the non-factive
grounding claim r =} </> is (factively and non-factively) zero-grounded.
-]
r therefore treat all elimination rules as giving rise to plain arguments.
We cannot yet show, however, that if r =? </> is true, then I'=? </> is zero-
grounded. To do this we need elimination rules for=?.
8. GROUNDING OPERATORS:
ELIMINATION RULES
We use a proof-theoretic inversion principle (see e.g. Read 201 O) to find
elimination rules. This principle says that the elimination rule(s) for an
operator ,\ should be such that if ¢ follows from each of the conditions
(given by the introduction rules) allowing us to assert A(tjJ0 ,. .. ,t/J,,), then</>
should follow from ,\(1/; 0 , ... ,tj;,,) by an elimination rule. And conversely: if</>
follows from,\( tjJ 0 ,. . .,tf;,,) by an elimination rule, then </> has to follow from
any of the conditions allowing us to assert ,\(t/J0 ,. . .,t/J,,). (Where these latter
Grounding Ground 299
conditions arc given by the introduction rules.) Let us apply the principle to
the easy case of<.
According to the <-introduction rule, we are entitled to conclude 6 < </>
from premisses /1 and 6 =} <f>. Anything that follows from those premisses
must therefore follow from /1 < <f>. The elimination rule for< takes the form:
--! -----2
6 11=?</>
£
6<(/> 1/1
-------1,2: <-Elimination
1/1
This is to be read as follows. If£ is an argument (explanatory or plain) to
conclusion 1/; and we, in the course of£, have used the assumptions 6 and
/:,. =} </>some number of times, we can conclude t/J from 6 < <f>, discharging
any number of the assumptions /1 and /1 =? <p.
Arguments of this form are plain. The premisses of an application of <-E
do not explain its conclusion (or if they do this is not by dint of their being
the premisses of an application of <-E). In general, the premisses of an
application of an elimination rule do not explain its conclusion. We will
The elimination rule for =} is more interesting; a second technical
innovation is required to find elimination rules here. The introduction
rule for =? tells us that we are entitled to assert /1 =? ¢ if there is an
explanatory argument with premisses (all and only) 6 and conclusion <f>.
Anything that follows from the existence of such an argument should follow
from /1 =} <f>. How can we express the assumption that there exists an
explanatory argument from premisses 6 to conclusion ¢?
We cannot-not without extending our conception of natural deduction:
we have to be able to assume and discharge arguments as well as propositions.
To do this we introduce the notion of a hypothetical argument.
An expression of the form /1 II-- e </> is to stand for a hypothetical explana-
toty argument with conclusion </>and premisses (exactly) 6. 38 Such hypo-
thetical arguments 6 II-- e </> only occur in contexts of the form:
6[ /1 11--e </>J'
ef>
58
For an application of, in effect, hypothetical arguments in a different context sec
Schroeder-Heister (1984).
300 Jon fa-ling Ut!,md Grounding Ground 301

Arguments of this form arc explanatory. What this says is that by using a We say that cp follows from r in the PUS!·(; if there is l'o <;: ; r such that
hypothetical explanatory argument from 6 to cp one explanatorily infers cp rhcre is an argument in Pl.ISFC with premisses 1'0 and conclusion cf. \Y./c write
from 6. 1' f- cp if this is the case. PLISFC suffices to prove (almost) alJ the basic facts
Even though hypothetical arguments can be assumed and discharged in about non-iterated factivc ground. ;,o
the course of an argument, they ;1re not premisses on which the (sub)-
Proposition 8.1.
conclusions depend. (We assume an argument, not the existence of an
argument.) One may think of the hypothetical arguments as assumed rules o,
(i) Left factivity: 6 < cp ~- for rd! E 6. o
of inference. lf </>follows from </; by a rule of inference R, then if R were not (ii) Rightfitctivity: 6 < cp f-- </>.
valid, </> would not have followed from </1, but that does not mean that the (iii) Non-circulm-ity: (6,cp) < 4, f-- </1 fin· all </1.
vrdidity of the rule R is a premiss which has to be added to 1/1 in order to (iv) Cut: 60 < cp0 , 61 < cp 1 , ••• , (cp 0 ,cp 1 , ... ,I' < </>) 1- 6 0,6 1 , ••• ,r < cp.
derive cp. (We assume the rule, not the validity of the rule.) 39
We can now write down the elimination rule for=?:
Proof Given the unfamiliarity of the system let us prove Proposition 8.1. (ii)
and Proposition 8. l (iii), leaving the rest as an exercise for the reader. (To
-----] prove Proposition 8.1 (i) and Proposition 8.1 (iv) we require vacuous
6 If- r cp discharge.)
[
---2
6 =? cp 1/1 rl· . . ----1
6
- -- JA If- e c/>1 1
- - - - - - - - 1,=?-!',1mmat1on
</1 /\ => c/> c/>
- - - - '· => -E
- - - - - - 1.1 < -E
'P
This is read as follows. Suppose we have an argument [with conclusion 1/1 in
'P
the course of which we have relied on some instances of the hypothetirnl
argument 6 lf-e cp. =?-E allows us to conclude •fa from 6 =? cp discharging ---2
any number of occurrences of the hypothetical argument 6 If- e cp.
~ [l\, lf-e cp]l
For any given collection of explanatory and plain arguments (Ee,lEp) we
'P
define the explanatory and plain arguments over (ll<~e,IEp) as the least class of - - Non-circularity
</1
arguments (18~;,,l~) such that Ee <;: ; IE; and IEp <;::;IE; and such that - - - - - - - - - - - " ' => -E
(!8~'.,, IE;) is closed under the rules in Figure 12. l and the introduction and
elimination rules given above. The arguments of the Pure Logic of Iterated if;
Strict Full Ground (PLISFG) are the explanatory and plain arguments over We can, moreover, prove the following two crucial principles about iterated
(0,0). ground. (For readability, we write 0 =? cp instead of =? cp; and similarly
for <.)

Proposition 8.2.
(i) 7 6 =? cp 'r 0 < (6 =? cp)
39
Two reasons for insisting that we assume the argument and not its validity: first, if (ii) 6 < cp ~ 6 < (6 < cp)
we had to assume the validity of the argument /\ If-, cf> in addition to the argument
L\ If-, <p we would be off on a regress a la Carroll (1895). (In fact this regress was noted
already in 1837! (Bolzano 1972, vol. 2, §199).) Second, the idea behind the hypothetical
arguments is that to assume/\ If-, cf> is to assume that we can explain <p from/\; ifwc had
to assume the validity of the argument/\ If-, cf> and not just the argument/\ If-, cf> this
would not work. For suppose we derive cf> from /\ using the hypothetical argument
/\If-, <p, and that</> now depends not just on/\ but also on the validity of/\ If-, cf>. We
40
would now nor have an explanation of</> from /\; we would rather have an explanation of It docs not prove the principle of Amalgamation. Sec Appendix A for a way of fixing
</>from/\ together with the validity of/\ If-, <p. rhis.
302 Jon Erling Litland Grounding Ground 303

Proof The following derivation establishes Proposition 8.2 (i): Clearly, the explanation for why p => p /\ p is the case differs from the
explanation for why p =? p V p is the case?"<\ 2
-l
This criticism is misguided: while every true non-factive grounding claim
~[/\If-- ¢]2
'P " /', =? cf> has the same (immediate) strict fi_ill ground-the empty ground-
---],=?-! diffcrent true non-factivc grounding claims are (immediately) zcro-
A =;, 'P
=} -! grounded in different ways. This is easiest to see by looking at the machine
----cl picture. When the machine is asked to simulate being feel pit will churn out
0<(1\=?c/>) both p =? p /\ p and p =? p V p. But there is no reason to think that the
- - - - - - - - 2, =;, -E
0 < (i\ =? 'Pl
simulations the machine runs in order to reach these outputs arc the same;
in particular, the two simulations might comprise the application of differ-
Note here how the hypothetical argument !'., 11-e cf> (labeled "2") is not
ent mechanisms.
counted as a premiss on which the sub-conclusion cf> depends. This is
In the framework of the explanatory arguments we can see that the
required in order for the applications of =?-I to be justified: we have to
objection trades on an ambiguity in what is meant by "explanation." ln
discharge all assumptions on which cf> depends. Note also how we use <-I in
this framework there are two things one can mean by an explanation of cf>:
the case where /', is empty.
one can mean a collection of propositions cf>0 ,cf> 1, ... from which cf> can be
The following establishes Proposition 8.2 (ii):
derived in an explanatory way; alternatively, one can mean an argument
-3 witnessing that cf> can be derived in an explanato1y way. In the former sense
cf> [i\ If-- e </>1
4 p =? p /\ p and p =? p V p have the S<tme explanation; in the latter sense
they have different explanations.
5 - - - - 3 , =?-!
/\ =} cf> To see this consider Figure 12.2. This depicts the arguments establishing
-------<-!
/\ < (p that=? (p =? pAp) and=? (p =? pvp). These arguments differ, amongst
- - - - 5 , =?-! other things, in that the first contains an application of /\-introduction
i\ =} (i\ <(/>)
----2 ----------<-! while the second contains instead an application of V-introduction. (\Ve
i\ =} /\ < (i\ «p ) assume that V-introduction and /\-introduction are rules of explanatory
4,=? -E
«p
i\ i\<(i\<cp) inference.)
- - - - - - - - - - - - - - l,2 < -E What is depicted in Figure 12.2 is a situation where two distinct proposi-
/\ < (i\ «p )
tions have exactly the same grounds but are grounded in different ways. But
These results show that factive ground ( <) behaves as it should; in particu- there are also situations where distinct propositions have the same grounds
lar, all the constraints on factive ground laid down in §2 above are satisfied and are grounded in the same way. Figure 12.3 depicts such a situation. In
by <.<il this case not only do=? (p,q =? pAq) and=? (r,s =? r/\s) have the same
For more technical detail about PUSFG, see Appendices A and B.

-I --1 -1
p p p
9. EVERY (TRUE) NON-FACTIVE GROUNDING /\-[ ---V-l
p J\p pvp
CLAIM HAS THE SAME GROUND I, =?-I I,=?-!
p=?p!\p =}
=}-[ =?-!
=} (p =} p J\f') =} (p =} p v p)
The following objection comes naturally to mind. "According to the ZGA
every (true) non-factive grounding claim has the same ground-indeed, the Figure 12.2 Different ways
same immediate ground-namely the empty one. How could this be?

42
41 Note that we do not have (r,q, ~-> 1) f-- .;,, for each •/J. This, as we argued in §6.l Dasgupta (2014b, p.573) criticizes the views of deRosset (2013) and Bennett
above, is as it should be. (2011) along these lines.
304 Jon Erling Lit/and Grounding Ground 305

--2 --2 presupposttton: on my favored way of conceiving of the relationship

~----~/\-! _ _ _-'_/\-!
between the SFA and the 7CA they are not in competition.
__/'_0_'1_ _____ 1 7 =>-I --~--1,2,=>-l
p, '/ => p J\ '/ ,-~-I __.!~s_=::__~!:_!_ ____ =>-I Let us first observe that the logical techniques developed for the 7GA can
=>(p, '/ ='> p J\ q) => (1; s => r J\ s) equally well be employed in developing a logic of iterated ground that is in
accord with the SFA. A defender of the SFA could, for example, give an
Figure 12.3 Sarne way
introduction rule for an operator < + as follows:

ground-the empty one--but they also seem to be grounded in the empty

ground in the same way. We can make sense of this in terms of the machine
picture. In the two cases depicted in Figure 12.3 the same mechrmisms are 6
applied (and they are applied in the same order) but the applications of the ------ I
/'!.<I ify ,
< 1- - I
mechanisms differ since p,q are different propositions from r,s. (It is for this
reason that we have taken an arc A to represent the application of a
Here [ is an explanatory argument and 6 are all and only the premisses on
mechanism not the mechanism itself.)
which¢ depends. This rule in effect compresses the==?---/ and <-I rules into
An advantage of the framework of explanatory arguments is that it
one rule. '1/i Is there something wrong such an introduction rule for < t ?
promises us the means for defining the notion of a way of grounding. For
In my view there is nothing wrong with an operator governed by such an
consider a particular explanatory argument. Uniformly replace items in that
introduction rule. We should just insist that the operators <+ and <
argument with schematic letters. This gives us an argument farm. If every
capture different notions of factive ground. The propositions 6 < if, and
argument of that form is explanatory the argument form is an explanritory
6 < t cp are, afrer all, readily distinguished in terms of their grounds: 6 < if,,
argument form. We may identify the ways of grounding with the explana-
but not 6 < f cp, is partially grounded in 6 ==? if,. We may also describe the
tory argument forms. As is easily seen, the two arguments in Figure 12.3
difference in terms of explanatory arguments: while one gives an explanatory
have the same explanatory form.
argument from 6 to if, as part of explanatorily inferring 6 <f- if,, the claim
We can now deal with a further objection to the 7GA. Dasgupta (20 I 4b,
that there is such an explanato1y argument (that is: 6 ==? if,) is not part of the
pp. 531-2) observes that there are patterns in grounding. (For instance,
grounds for 6 < + if,.
all conjunctions are alike in terms of how they are grounded.) An account
One might, however, worry that the possibility of introduction rules like
of ground should provide us with an account of these patterns. Armed with
the ones for < 1- above present a problem for the ZGA: could one, by holding
the notion of a way of grounding we have a ve1y simple explanation: the
that <+ was the "real" notion of factive ground, give a satisfactory solution
patterns in grounding are the result of different propositions being grounded
to the Status Problem without invoking either non-factive ground or zero-
in the same way.'u
grounding?
This would be the wrong conclusion to draw-for two reasons. First, if
one accepts the framework of explanatory arguments it is hard to reject the
10. COMPARISON WITH THE notion of non-factive ground. And if one accepts the notion of non-factive
STRAIGHTFORWARD ACCOUNT
44
One might wonder why the mlc should not be given as follows instead:
The SFA, recall, holds that when r < q,, then r < (r < q,); unlike the 7CA,
/':,
however, the SFA does not hold that f < if, is partly grounded in I' ==? if,.
Why should we prefer the 7GA to the SFA? The question has a false £
--'"-!.<'-!
/':, <' ¢ '
43
I believe the norion of a way of ground is of great importance both for the theory of (Herc /':, arc all and only the premisses on which ¢ depends and £ is explanatory.) The
ground and for the applications of ground. More has to be done to put the talk of ways of difference between the two rules is that in the latter one, /':, is not discharged. The reasons
grounding on a rigorous footing, but this is not the place to do this. The issues that arise for this arc purely technical: it makes finding elimination rules easier and it avoids certain
arc vety closely related to the issues of famz mentioned in footnote 37. complications in the statement of (CHAINING).
306 Jon Erling Lit!and
ground, a Status Problem like the one in §3 arises for non-factivc ground.
This presents the defender of the SFA with a ch,1llenge: if true non-factive
grounding claims arc not zero-grounded, in what are they grounded?
Second, it turns out that even if we reject the notion of non-factive
ground and treat factive ground as < + we still need zero-grounding. To
sec this, consider negated grounding claims. 45
Suppose that the object a occurs in the truth <p and that while each
o E I:!. is the case, •(I:!. < </>) is also the case. (For a concrete example: let a
be Socrates, <p the truth that Socrates was Greek, and I:!. the truth that the
Holy Roman Empire was dissolved in 1806.) If-.(!:!.<</>) is true and
ungrounded then the object a would be 0-fundamental. If the Status
Problem is to be satisfactorily solved, true negated grounding claims, too,
have to be grounded.
The defenders of SFA have had nothing to say about such negated
grounding claims. 16 Let us first consider what a defender of the /GA could
say. For a defender of the ZGA the immediate full grounds of I:!. < </> are I:!.
and I:!. =? <p taken together. It is plausible to hold that to ground •(I:!. < </>) is
either to ground •o o
for a E I:!. or to ground -.(I:!. =? </>) .47 In the envisaged
o
scenario, each EI:!. is true; what has to be grounded, then, is •( L'I =? <p).
The natural move is to say that such true negated non-factive grounding
claims are zero-grounded. One is, to be sure, owed a justification for treating
such negated non-factive grounding claims as zero-grounded; but the
machine and graph-theoretical pictures of §5 provide a natural justification.
In terms of the machine picture: when the machine is fed no input, it
simulates being given the propositions L'I as input. The machine then tries to
find an arc from L'I to ef>. When it has tested all the arcs A with t(A) = L'I and
has noted that none of them have h(A) = ef>, the machine reports back that
•(L'I =? </>). Since the machine was fed no input •(L'I =? </>) will be zero-
grounded.48
45
I should stress that negated-more generally, embedded-grounding claims cause
problems even if one docs not accept the present framework of explanatory arguments.
46
As admitted by deRosset (2013, p. 16).
47
The simplest way of dealing with negation in the logic of ground is to follow a
broadly "bilatcralist" strategy and give separate introduction and elimination rules for
negated and unnegated propositions; for this strategy, sec, e.g. Fine (20 l 2a, p. 63)
A different, perhaps preferable, treatment of negation is pursued in Fine (2016).
48
How docs the machine know that it has inspected all the arcs? We might imagine
that in running the simulation the arcs arc ordered in such a way that each arc occurs
infinitely ofren but such that if an arc A with tail /I,. occurs both at position t and at a later
position /, then all arcs with tail D. have occurred at least once before/. If, having found
no arc B with tail /I,. and head <f>, the machine inspects an arc it h<ts previously inspected, it
aborts the simulation and outpms -,(fl,.~'> <p).
Grounding Ground
If one accepts the connection between grounding and explanatory argu-
ments developed in this paper one would have to exhibit natural introduc-
tion and elimination rules for negated non-factive grounding claims ensuring
that true non-factivc grounding claims are zero-grounded. This can in fact
be done but we cannot go into the details hcrc. 49
One can say something similar about < 1 : if L'I is true and •(I:!. < 1 c/>) is
also the case, then •(L'I <+ </>)is (immediately) zero-grounded. In terms of
the machine picture the account is much the same. The machine is fed no
input and simulates the result of being fed L'I as input. It runs through all
arcs with tail L'I. After observing that there is no arc with tail I:!. and head ef>,
the machine churns out the truth •( L'I < t- ef>).
Holding that factive ground behaves like < 1 does not obviate the need
for zero-grounding. 50
11. CONCLUDING REMARKS
In this paper I have developed a novel account of iterated grounding-the
/GA-and I have showed how-by taking a notion of explanatory argument
as basic-one can develop a logic of iterated ground. No matter what one
thinks of the metaphysics of the ZCA, the logical techniques developed here
should prove useful for others interested in the logic of ground. In closing let
me mention some areas where more work needs to be done.
First, more needs to be said about explanatory inference (as opposed to
argument). Explanatory inference is of interest because of its connection to
immediate ground and having a logic of immediate ground would be of
considerable interest.
Second, more needs to be said about the notions of a way of grounding
and the fonn of an explanatory argument. I believe that it is only by geting
clearer on these notions that we, in a principled manner, can develop fine-
49
In this paper we have limited ourselves to simple hypothetical arguments from some
propositions to a proposition. To give a rule for negated grounding claims we would have
to be able to assume rules. In particular, if we know that ..,(fl,.=> </>)we should be entitled
to assume a rule that lets us conclude the absurd from any explanatory argument from /I,.
to ¢. That is, we have to assume not just arguments that allow us to pass from (sets or)
propositions to propositions, but also rules that allow us to pass from arguments (or more
generally: rules) to propositions. By extending the machinery of higher-order rules in
Schroeder-Heister (1984) to a setting where we have both explanatory and plain argu-
ments this can be done.
50
Thanks to an anonymous reviewer for pressing me to get clearer on the relationship
between the ZCA and the SFA.
308 Jon Erling Ut!mzd
grained conceptions of ground-ones, for example, where a proposition p
strictly grounds the conjunction p /\ p.
Third, we have restricted our attention to the pure logic of ground. To fully
vindicate the framework we have to be able to extend the framework of explana-
tory arguments ro deal with the impure logic of ground. Of particular philosoph-
ical interest is extensions dealing with negation and the conditional. The case for
d1e ZCA would be greatly strcnbrthencd if this could be done in a natural way.
Fourth, more must be said about the relationship between essence and
ground. Many philosophers have held that there is a deep connection
between essence and ground. 51 I believe that the ZCA allows us to adopt a
distinctive view on this relationship; I end by baldly stating that view.
Those who accept that there is a connection between essence and ground
have held that it is certain truths about grounding that arc essentially true.
For instance, Rosen (2010, p. 130) holds that it is essential to disjunction
that for all propositions p, if pis true then p grounds p Y q. In my view this
is a mistake: what is essential to disjunction is not a truth, but rather an
explanatory inference. (ln this case, the explanatory inferences from p to
p Y q and from p to q Y p.) 52
Jon Erling lit/and
University of Texas, Austin
ACKNOWLEDGMENTS
Material from this paper has been presented at a reading group at Harvard Univer-
sity, the University of Texas at Austin, a conference on grounding at the University
of Nottingham, a workshop at C')MN, a workshop on Ground and Groundcclness at
Munich and at the 2012 meeting of the Eastern division of the APA. I'm gratefol to
members of the audience at all those places. I am very grateful to Michael Raven,
Shamik Dasgupta, 0ystein Linncbo, and ]Cinne Kriener for comments on earlier
drafts of this material. Special thanks to Louis deRosset for extended discussions of
the logics sketched in this paper and to Kit Fine for several suggestions that led to
technical improvements. I am also very grateful for the detailed and very helpful
comments I received from several anonymous reviewers.
51
Their number include: Audi (2012b), Fine (2012a), and Trogdon (2013b). Rosen
(20 l O) and Dasgupta (20 I 4b) have even held that truths about ground are always partly
gmzmded in trnths about essence.
52
The idea that what is essential might be an inference is not unprecedented: Fine
suggested that we should think of "the nature of the logical concepts as being given, not
by certain logical truths, but by certain inferences" (Fine 1994, pp. 57-8). The idea is
developed in greater derail in Correia (2012). What is new in the present setting is the
distinction between explanatoty and merely plain inference and the claim that what is
essential arc certain exp!antttory inferences.
Grounding Ground 309
APPENDIX A Pl.ViFC AND PiSFC
T'he Pure Logic of Strict Full Ground-PLSFC·-is the subsystem of Fine's Pure Logic
of Ground-PLC--that concerns only strict foll ground. !ts rules are depicted in
Figure 12.4.
l'LSFl; is a sequent system: in PlSFC an expression of the form I' < </>is a sequent not
a sentence and so a claim of iterated ground would be ill-formed.
There is an obvious translation of PLSFG into the language of l'USFC: we translate
the sequent/';, < 'P by the formula L\ < </>. Proposition 8.1 ensures that l'LISFG proves
the translations of Non-circularity and Cut. It rurns out that we cannot derive the
translation of the Amalgamation principle. In fact, Amalgamation is arguably not
correct. To sec this, consider the following situation. Suppose the inferences from tj1
(;md D) to t/1 V D arc cxplanarory; and suppose that these are the only explanatory
inferences with conclusion </; V D. Suppose further that there is an explanatory
inference from </> to t/1 but no explanatory inference from </> to D. Then there is an
cxplanatoty argument from 'P to tj;VO and an explanatory argument from tj; to tj;VO,
but there is no explanatory argument from <f>,t/1 to </;VD. So while</> =} •/ND and tj1 =>
</;VO we do not have <f>,t/1 ,,> tj; v D.5.i. 54
To facilitate comparison with Fine's system we will, however, insist on AmaJ,
gamation. We validate Amalg,unation by using "amalgamation-friendly" =} mies-·
sec Figure 12.5. Say that /'J.o,/'J. 1 ,. . .,i E I is a covering of/';, ifU;E 1/'J.; "-~/';,,In the
amalgamation-friendly =?-E rule the io,ii ,... J0 J1 , ... enumerate the coverings of /'J..
We now (re)dcfinc PLISFG as follows. The ttrguments of PLISFC arc the least classes of
cxplanarory and plain arguments closed under the constmints in Figure 12. l, the
amalgamation-friendly=> mies and the previous rules for<. Let~- be the provability
relation in PLISFG and kt f-rLSFG be the provabiliry relation in Pl5F<;.
A,•/> < 1>
l. Non-circularity
i'lo < 'Po i'l 1 < 'P1 . ..
------------------Cut
'Po· ¢,,. . ., I'< 'P
i'lo,i'l1, ... ,f'< 'P
Amalgamation
Figure 12.4 Pure Logic of Strict Full Ground
53
Correia (2014, n. 17) makes the same observation, using a different exam pk.
54
Not all applications of Amalgamation are problematic. If </>,</; are as above, we
should have </>,t/1 => <(1 J\ tj;. The simplest way of getting a satisfactory treatment is by
having the grounding operators take multisets on the left. Then since if1,1/1 -=> tj; J\ if1 we
can derive </>,</; =? t/1 J\ t/1 using Cut; in contrast, since we do not have tj;,tj; -=> tj;VD we
cannot derive </>,</; =} tjNO using Cut.
310 Jon Erling Litland Grounding Ground 311

60 61 graph 9 ( V, A, t, h) be given and let V0 e V be some vertices. The closure of V0

Eo E1 under A is the least W ;> Vii such that if Vi <;:; Wand t(A) = Vi then h(A) <;:; W.
_j__ _'fi_.. _~ =?-I A pointed directed hypergraph is a tuple g ~= ( V, F, A, t, h) where ( V,A,t,h) is a
60,f'..1,. .. =? <p directed hypergraph and Fe V is a set of designated vertices such that
(i) Fis closed under taking heads, that is, if F0 <;:; F and t(A) -- F0 , then h(/;(>) E F.
(ii) 9 restricted to F is acyclic. 57
----Jo ------Ji···
f'..Jo If- e ¢
We say that F is the 11ctual wor/£1of9. Note that we do not require that g is
acyclic: we only require that 9 is acyclic on the actual world. From now onwards, we
~cfrr to acycl'.~· amalgamating, chained, pointed, directed hypcrgraphs simply as
Figure 12.5 Amalgamation-Friendly Rules hypergraphs.
\Y/c will interpret l'lSFC over hypergraphs. More precisely, let the sequcnts of PLSFC
be formed from the set of atomic letters IP ~~ {p 0 ,p 1 ,. .. } • A model for l'LSFC is a pair
It is easy to show that the translation of the rules of Pl.SFc; are derivable in Pl.ISF(;: (9,[]) where g == \Vg,Fg,Ar;;,tg,hQ) is a hypcrgraph and []:IP--> fg is function
PUSH; is an extension of PLSFG. W c rnn show that PUSFG is a conservative extension of from the atomic letters into F. If r is a collection of atomic letters then [r] is
the (relevant subsystem of) Fine's Pure Logic of Ground. This provides some UyEry{[y]}. We call such a (9,[)) agmph-mode!.
support for the ZGA. Since it is plausible that PLSFG is the correct logic of strict full Truth and consequence are defined as follows:
ground, this conservativeness result shows that by accepting the ZCA one is not • 9 F r < <p iff there is A EA such that t(A) <:;; [q, and h(A) = [</>}.
thereby forced to accept any claims of non-iterated ground that one was not • Let S be a set ofsequents and let a be a sequent. We say that a is a consequence of
independently committed to accepting. S (SF a) iffor all graph-models (9,[ ]) if each (3 ES is true in (9,[ ]) then r1 is true
Jn order to establish this conservativeness result we develop a graph-theoretic in (9,[]).
semantics for PISFG and show that any model for PL5FC not verifying a certain sequent
ti < </> can be extended to a model for Pl.ISFG not verifying the corresponding It is straightforward to establish the following:
statement ti < rp. 55
Proposition B. l. PLSFC is sound with respect to graph-models.
Proof Obvious.

APPENDIX B GRAPHICAL SEMANTICS Proposition B.2. PLSFC is complete with respect to graph-models.
Proof Let S be a collection of scquents and a a sequent such that SY:a. We define a
The semantics will be based on the directed, pointed hypergraphs introduced in §5. model (9,[]) as follows. We let V={p:pisanatom}. We put
\Y/ c impose the following demands on the hypergraphs. A= {r <</>:Sf-- r < </>}. We put t(r < </>) = r and h(r <</>)=cc• <f>. We put
A directed hypergraph g = (V,A,t,h) is ch11inediffor all arcs B,Ao,A1,. .. such [</>]=</>.Let g = \V,F,A,h,t). It is straightforward to check that (9,[]) is a model
that t(B) =c• { v0 ,v1 ,. . .,w0 ,w 1 ,. .. } and h(A;) = { v; }, t(ir each i, there is an arc C with of S in which a. is not true.
t(C) = t(Ao) U t(A1) U ... U {w0,w1,. .. } and h(B). 56 When we deal with iterated ground cardinality problems arise. We will assume
A directed hypergraph 9 = ( V,A,t,h) is 11m11lgmn11ting if whenever {A; };EleA that the set of IP of propositional atoms is strongly inaccessible. Say that IP has
arc such that h(A;) = h(Aj), for all iJ E I there is an A E A such that cardinality K. The sentences of l'LISFG over IP' arc generated as follows. Whenever r is
U;Eit(A;) = t(A) and h(A;) = h(A), for each i. We say that a directed hypcrgraph a set of sentences of cardinality < K and </> is a sentence, then r => </>and r < </> arc
9 = ( V,A, t, h) is cyclic if there arc hyperarcs A0 ,Ai ,. ..,A,, in A such that h(Ao) <:;; sentences.
t(A 1 ), .. .,h(A,,_ 1) <:;; t(A,,) and such that h(A,,) E t(A 0 ). It is acyclic otherwise. Let a To give a semantics for iterated ground we introduce the following notion.
Agmph with operators is a tuple g \V,F,A,h,t,<, =>)where \V,F,A,h,t) is a
hypergraph satisfying the following constraints:
55
I should stress that the graph-theoretic semantics has some artificial features-for
instance, the treatment of< and =?- is "written in by hand." I hope to provide a more (i) V is a strongly inaccessible cardinal.
illuminating semantics elsewhere; for the purposes of establishing the conservativity
results these artificial features are, in any case, unproblematic.
56 57
This ensures that Cut and Chaining are taken care of in a simple way. That is, the subgraph generated by A over F is acyclic.
312 Jon Erling Lit/and
(ii) lt(A)I <IV! forallAEA.
< and =·> are one-one functions ···+ V such that
(iii) if A is an arc then there is BE A such that t(B) = 0 and h(B) t(A) ==> h(A),
(iv) if X =-=> v E F then there is A EA such that t(A) ==· X and h(A) v,
(v) if t(A) C:: F then t(A)<h(A) E F and there is B such that
t(li) == t(A) U t(A) ==> h(A) with h(B) t(A)<h(A),
(vi) if X <v E F then X C:: F and there is A such that t(A) == X and h(A) tJ.
Remark B.3. The reason for requiring V to have strongly inaccessible cardinality is
that it is very natural to assume the (Distinctness Principle):
(Distinctness Principle) If Vo =f Vi or v0 =f 111 then ( Vri =? vo h,i: (Vi=? v1).
(The truth that Vo non-factively ground v0 and the truth that Vi non-factivcly
ground u1 seem distinct.)
If there arc 1<-many vertices a graph-model might have as many as 2"-many arcs.
The (Distinctness Principle) would then ensure that there arc 2K many vertices after
all. We therefore impose some constraints on the graphs.
A model for PLISFG is a p:lir (Q,[]) such that Q is a graph with operators and [] :
IP' ~-> \.'f; is a fonction from the propositional atoms into the vertices of Q. We here
demand that V has greater cardinality than the set of sentence letters.
We extend the interpretation function from the atomic letters to arbitrary
formulae in the obvious way.
• [r => </>] = [IJ => [</>]
• [r < r,6] [r]<[</>]
Hypothetical arguments arc dealt with as follows: HT 11-p 'Pis a plain hypothetical
argument WC say that g validates r 11-p </>if whenever [r) c:: F, [</>] c:: F. !ff II-, </>is a
hypothetical explanatory argument we say that Q validates I'll-,</> if there is an arc
A EAg with [r] = t(A) and[</>) h(A).
The following soundness theorem is easily established:
Proposition B.4. Let explanatory and plain arguments (IE,,!Ep) be given.
(i) If there is a plain argument fimn r to 4, in PLISFG over (lffi, ,lEp) then jiJr all models Q
ualidrtting (lffi,,lEp) we hrwe that if[r] C:: F then [</>] E F.
(i) I/there is an explanatory argument from r to <p in PI.ISFC over (lE,,lEp) then far all
models Q 11t1lidating (E,,Ep) there ium arc A such that t(A) = [r) and h(A) '·"' [1>].
Proof We prove the claims simultaneously by induction on the complexity of the
arguments witnessing that there is a plain (explanatory) argument from r to </>.
One can also establish the following completeness theorem:
Theorem B.5. Let (lE,,IEp) be some explanato1y and plain arguments. If a is
a statement that is not derivable from (lE,,IEp) then there is a graph with
operators Q and an evaluation [] such that a is not true over (Q,[]).
Grounding G'rounci 313
Proof By a straightfotward extension of the proof of Proposition B.2.
To establish that PIJSFG is a conservative extension of l'LSFG we establish the
following proposition. (As stated we require a proper class of inaccessible cardinals.)
Propositon B.6. let Q == ( V,F,A,t,h) be rt hype1gmph. There is rt graph with
operrttors Q 1· (V 1, F 1, A+,=..'- ,<,t+, h') such tht1t
(i) V C V 1 ,A C:: A' ,h s;: h' ,t C:: t 1
(ii) A 1 restricted to V is A
(iii) // ,t' restricted to V is h,t.
Sketch: Let1< be the cardinality of V. Let A be the least strongly inaccessible cardinal
> W c extend Q to a graph with operators in ,\-many stages Q ·= Yo ,91 , ... ,QY" .. in
1<.
the obvious way.
Theorem B.7. Let S be a collection ofsequents in the language ofPLSFG and Let
r < cf> be a sequent in the language of PI5FG. If SJ.LprsFC r < cf> then
SJ.LpusFG r < cf>.
Proof Let (Q,[]) be a model witnessing that SY' r < 'P· Say that Vg has cardinality 1<.
We can extend Q to a graph with operators Q' = (V', F', A 1, =>,<,t 1, /i 1 )
satisfying the conditions in Proposition B.6. The interpretation [] is extended to
r
an interpretation [ of the language of l'l.!SFG in the obvious way. The restriction of
A 1 to v is A. It follows that r < </> is not true in (QI ,[ ]').
REFERENCES
Audi, Paul (2012a). "A Clarification and Defense of the Notion of Grounding."
In Metaphysical Grounding, ed. Fabrice Correia and Benjamin Schnieder.
Cambridge: Cambridge University Press, pp. 101-21.
Audi, Paul (20 l 2b). "Grounding: Towards a Theory of the In-Virtue-of Relation."
Journal of Philosophy 109. 12, pp. 68 5-711.
Bennett, K. (2011). "By Our Bootstraps." Philosophical Perspectives 25.1, pp. 27-4 l.
Bolzano, Bernhard (1972). Theoiy of Science. Attempt at a Detailed and in the main
Novel E>:position of Logic with Constant Attention to Em·lia Authors. Trans. Rolf
George. Berkeley: University of California Press.
Carroll, L. (1895). "What the Tortoise said to Achilles." Mind 4.14, pp. 278-80.
Correia, Fabrice (2010). "Grounding and Truth-functions." Logique et Analyse
53.211, pp. 251-79.
Correia, Fabrice (2012). "On the Reduction of Necessity to Essence." Philosophy and
Phenomenological Research 84.3, pp. 639-53.
Correia, Fabrice (2014). "Logical Grounds." Review of Symbolic Logic 7.1,
pp. 31-59.
Dasgupta, Shamik (2014a). "On the Plurality of Grounds." Philosophers Imprint
14.20, pp. 1-28.
314 Jon Erling Lithmd
Dasgupta, Sharnik (2014b). "The Possibility of Physicalism." joumt!I of Philosophy
l l l.9/10, pp. 557-92.
Dasgupta, Shamik (2016). "Metaphysical Rationalism." N011s, 50.2, pp. 379-418.
dcRossct, Louis (2013). "Grounding Explanations." !'hilosophm-' Imprint 13.7,
pp. 1-26.
dcRosset, Louis (2015). "Better Semantics for the Pure Logic of Ground." A11t1lytic
Philosophy 56.3, pp. 229-52.
Dixon, T. Scott (2016). "What is the Well-Foundedness of Grounding?" Mind
125.498, pp. 439-68.
Fine, Kit (1994). "Senses of I·~sscncc." In Motblity, Mortl!ity, 11nd Belie/ ESSttys in
Honor of Ruth Barctm Marcus, ed. N. Ascher, D. Raffman, and W. Sinnott-
Armstrong. Chicago: University of Chicago Press, pp. 53-73.
Fine, Kit (2001). "The Question of Realism." Philosophers Imprint I.I, pp. l-30.
Fine, Kit (201 Oa). "Some Puzzles of Ground." Notre Dame journal ofFormtd l.ogic
51.l, pp. 97-118.
Fine, Kit (2010b). "The Question of Ontology." In Metametaphysics: New Essays on
the Foundations of Ontology, ed. D. J. Chalmers, D. Manley, and R. Wasserman.
Oxford: Clarendon Press, pp. 157-77.
Fine, Kit (2012a). "Guide to Ground." In Metaphysical Grounding, ed. Fabrice
Correia and Benjamin Schnieder. Cambridge: C1mbridge University Press,
pp. 37-80.
Fine, Kit (2012b). "The Pure Logic of Ground." Review of Symbolic Logic 5.1,
pp. 1-25.
Fine, Kit (2014). "Truth-maker Semantics for lntuitionistic Logic." Journal of
Philosophical Logic 43.2-3, pp. 549-77.
Fine, Kit (2016). "Identity Criteria and Ground." Philosophical Studies 173.1,
pp. 1-19.
Jenkins, C. (2011). "Is Metaphysical Dependence Irreflexive?" Monist 94.2,
pp. 267-76.
Kramer, Stephan (2013). "A Simpler Puzzle of Ground." Thought 2.2, pp. 85-9.
Lithmd, Jon Erling (2013). "On Some Counterexamples to the Transitivity of
Grounding." Essays in Philosophy 14.1: Grounding Relation(s). Ed. Paul Hovda
and Troy Cross, pp. 19-32.
Litland, Jon Erling (2016a). "An Infinitely Descending Chain of Ground without a
Lower Bound." Philosophical Studies 173.5, pp. 1361-9.
Litland, Jon Erling (2016b). "Pure Logic of Many-Many Ground." journal of
Philosophical Logic, DOI l 0.1007 /s I 0992-015-9386-2, pp. 1--47.
Poggiolesi, Francesca (2015). "On Defining the Notion of Complete and Immediate
Formal Grounding." Synthese DOI 1O.l007/s11229-015-0923-x, pp. 1-21.
Rabin, Gabriel and Brian Rabern (2015). "\X'ell Founding Grounding Grounding."
journal of Philosophical Logic DOI l0.1007/s10992-015-9376-4, pp. 1-31.
Raven, Michael J. (2009). "Ontology, From a Fundamentalist Point of View." PhD
thesis. New York: New York University.
Raven, Michael]. (2013). "Is Ground a Strict Partial Order." Amerimn Philosophical
Quarterly 50.2, pp. 191-9.
Grounding Ground 3l5
Read, Stephen (2000). "Harmony and Autonomy in Classical Logic." Journal of
!'hilosophiett! Logic 29.2, pp. 123-54.
Read, Stephen (2010). "General-elimination Harmony and the Meaning of the
Logical Constants." Journal of Philosophical Logic 39.5, pp. 557-76.
Rosen, G. (2010). "Metaphysical Dependence: Grounding and Reduction." In
Modtt!ity: Metilphysics, Logic, and E"pistemolog:;<, ed. Bob Hale and Aviv Hoffmann.
Oxford: Oxford University Press, pp. 109-35.
Schaffer, Jonathan (2009). "On What Grounds What." In Metametaphysics: New
Essays on the Found11tions of Ontology, ed. D. ]. Chalmers, D. Manley, and
R. Wasserman. Oxford: Oxford University Press, pp. 347--83.
Schaller, Jonathan (2012). "Grounding, Transitivity, and Contrastivity." In lvfettt-
physicti/ Grounding, ed. Fabrice Correia and Benjamin Schnieder. Cambridge:
Cambridge University Press, pp. 122-38.
Schaffer, Jonathan (2016). "Grounding in the Image of Causation." Philosophiutf
Studies 173.l, pp. 1-52.
Schnieder, Benjamin (2011 ). "A Logic for 'Because'." Review of Symbolic Logic 4.3,
pp. 445-65.
Schroeder-Heister, Peter (1984). "A Natural Extension of Natural Deduction."
Journal of Symbolic Logic49.4, pp. 1284-300.
Scott, Dana (1971). "On Engendering an Illusion of Understanding." journal of
Philosophy 68.21, pp. 787-807.
Sider, T. (2012). Writing the Book of the World. Oxford: Oxford University Press.
Trogdon, Kelly (2013a). "An Introduction to Grounding." In Varieties of Depend-
ence, ed. M. Hoeltje, A. Steinberg, and B. Schnieder. Munich: Philosophia Verlag,
pp. 97-122.
Trogdon, Kelly (2013b). "Grounding: Necessa1y or Contingent?" Pacific Philosoph-
ical Quarterly 94.4, pp. 465-85.
Wilson, Jessica M. (2014). "No Work for a Theory of Grounding." Inquily 57.5-6,
pp. 535-79.
 Author Index

Adams, R. 166, 169 Chalmers, D. 63, 72, 77, 81, 85, 89, 120,
Akiba, K. 207, 232 121, 170, 202, 23:3, 314
Allen, K. 86, 88, l09, l 12, 1 l4, 120 Cicero, M. T. 203
Anjum, R. 38 Clarke, R. 38
Armstrong, D. M. 43 Crm~ l~ 14, 36, 314
Arntzenius, F. 193, 194, 198 Cohen, J. 90, 97, 110, 121
Asher, N. 23, 35 Contessa, C. 8, 9, 36
Audi, P. 281, 282, 2%, 308, 3 l 3 Correia, F. 279, 282, 283, 308, 309
Austin, J. L. 99, 120 313, 314

Bach, K. l51, 169 Dancy, J. I 15, 121

Bacon, F. 22, 33 Dasgupta, S. 279, 281, 282, 283, 285,
Barcnblatt, G. 172, 176, 178, 179, 185, 303, 304, 308, 313, 314
186, 187, 188, 198 Davidson, D. 164, 169
Barnes, E. 207, 208, 233, 256 Davies, M. 70, 72
Belnap, N. 67, 68, 72 deRossct, L. 64-72, 74, 81, 279, 281,
Benbaji, H. 108, 120 282, 282, 284, 285, 286, 287, 289,
Bennert, K. 188, 197, 198, 279, 281, 303, 306, 308, 314
282, 283, 285, 286, 303, 313 Divers,]. 237, 238, 240, 246, 256
Bcnncrr, M. I 08, 120 Dixon, T. 290, 314
Bird, A. 5, 29, 32, 34, 35, 39, 43, 45 Donnellv, M. 203, 233
Blackbum, S. 167, 169 Dowry, b. 23, 36
Bolzano, B. 300, 313 Drctskc, F. 89, 134, 169
Bonevac, D. 11, 35 Duncan, W. 186, 189, 198
Boghossian, P. 65, 72, 88, 89, 111, 113,
114, 115, 120 Eddon, M. 173, 198
Bonomi, A. 23, 35 Ellis, B. 29, 34, 3Ci
Boolos, G. 202, 233 Evans, C. 70, 72, 85, 108, I 09, 121,
Braun, D. 204, 233 207, 233
Brentano, F. 32, 35 Evcrcrt, A. 39, 169
Brewer, B. 90, 120 Euclid 201
Bricker, P. 242, 245, 246, 247, 256
Bridgman, P. 172, 173, 178, 179, 184, Fara, M. 4, 9, IO, II, 28, 36, 51, 59
185, 189, 197, 198 Field, H. 68, 72, 140, 169, 173, 198,
Broackcs, J. 85, I 04, 113, 120 252, 256
Byrne, A. 85, 86, 88, 114, 115, 120, 121 Filchcva, K. 142, 169
Fine, F. 193, 198, 202, 204, 215, 219,
Campbell, J. 85, 86, 89, 90, 108, 114, 229, 233, 280, 281, 282, 283, 286,
119, 120, 121 287, 294, 296, 306,308, 314
Carmichael, C. 203, 233 Frege, G. 66, 67, 72
Carnap, R. 63, 66, 72, 73, 77, 202, 233, Fodor, J. 133, 169
269
Carroll, L. 300, 313 Garson, J. 222, 233
Child, W. 108, ll6, 121 Gcrt,J. 85, 119, 121
Choi, S. 5, 36, 39 Goodman, N. 4, 14, 28, 36, 169
Chomsky, N. 133, 169 Gundersen, L. 11, 36, .39
318 Author Index Author Index· 319
Hale, B. 63, 67, 68, 72, 73, 1(,9, 202, 315 Macpherson, F. 96, 122 Rabern, B. 285, 290, 31 Ii Taylor, R. 2'71, 278
lfallman, P. 23, 36 Makin, S. 18, l 9, 26, 36, 44 Rabin, G. 285, 290, 314 Thomasson, A. 63, 65, 73, 202
Halperin, I. 264, 276 Manley, D. 8, l l, 12, U, 14, 29, 30, 34, Raven, M. 283, 286, 29J, 308, 314 Tolman, R. C. 172, 173, 174, 175,
llamari, K. 67, 73 35, 36, 49, 51, 53, 54, 56, 59, 170, Rayo, A. 250, 257 179, 183
l-lauska, J. 10, 36 233, 314, 315 Re~d, S. 67, 73, 294, 298, 315 Trogdon, K. 282, 308, 315
Hazen, /\.. 250, 256 Martin, C. B. 5, 31, 36, 38, 51 . 59 Rescher, N. 261, 270, 277, 278 Turner, J. 227, 234, 251, 25(>
!leek, R. 73, 207, 233 Maurreau, M. 11, :l7 Roscfddt, T. 151, 170 Tye, M. 89
Heil, J. 31, 36 McDowcll,J. 85,87, 110, 111, 115, 122, Rosen, C. 148, 170, 282, 2%,
Heller, M. 249, 256 164, 170 308, 315 Urbaniak, R. 67, 73
Higginbotham, J. 23, 24, % McGinn, C. 85, 89-91, 107, 110, 114, Rosenkranz, S. 148, 169, l 70
Hilbert, D.R. 85, 86, 88, 114, 120, 121 115, 122, 131, 170 Russell, B. 119, l 22, 193, l 98, 273 van [nwagen, P. 189, 198, 202, 203, 234,
Hirsch, E. 63, 7.1, 202 McKitrick, J. 6, 43 Russell, C. 65, 77 271, 272, 277
Hofweber, T. 63, 73, 149, 150, 151, 152, Melia, J. 239, 257 Ryle, C. 4, 37, 45, 85, 122 Varley,]. 139, 170
153, 154, 157, 160, 166, 169 Mellor, D. H. 20, 37 V;mi, /\.. 204, 234
Hordcn, J. 74-81 Merricks, T. 249, 250, 251, 252, 253, Sainsbury, R. 222, 233 Velleman, J. D. 63, 73, 88, 89, 111, 113,
Humberstonc, L. 70, 72 255, 257 Salmon, W. 272, 278, 281 114, 115, 120
Hume, D. 67, 273, 277 Miscevic, N. 89, 122 Schaffer, J. 52, 59, 281, 282, 283, 287, Vetter, B. 38, 45, 56, 59, 256
Moline, J. 10, l 1, 18, 37 293, 315
Jenkins, C. S. L 18, 36, 169, 283, 314 Molnar, G. 4, 20, 29, 31, 37, 39 Schellenberg, S. 97, 98, 99, 122 Watkins, M. 85, 104, 122
Johnston, M. 5, 6, 36, 85, 86, 89, 90, Molnnann, F. 151, 170 Schiffer, S. 63, 69, 73, 150, 151, Wasserman, R. 8, 11, 12, 13,
104, 105, 110, 114, 115, 121 Moore,/\.. 131, 169, 170 160, 170 14, 28, 34, 36, 37, 48, 49,
Mundy, B. 173, 198 Schnieder, B. 279, 282, 313, 314, 315 50, 51, 53, 54, 56, 59, 170,
Kant, I. 136, 143, 170 Mumford, S. 10, 37, 38, 43 Schrenk, M. 39 233,314, 315
Kaplan, D. 77, 81 Schroeder-Heister, P. 299, 307, 315 Weathcrson, B. 254, 257
Kalderon, M. 85, 89, 90, 91, 111, 112, Nagel, T. 133, 134, 166, 170 Schwarz, W. 254, 257 Westphal, J. 85, 122
114, 121 Newton, I. 171, 189, 198, 269 Sellars, W. l 18, 122 Whitcomb, K. 267, 278
Keefe, R. 204, 219, 233 Noe,/\.. 86, 97, 98, 102, I 11, 113, Shoemaker, S. I 04, 122 Wiggins, D. 85, 1I1, 123
Kelly, S. 95, 121 117, 122 Shapiro, S. 67, 73 Williams, R. 207, 234
Keisler, H. 157, 170 Nolan, D. 18, 36 Sider, T. 202-12, 214, 216, 223, 224, Williams, W. 184
Kennedy, C. 34, 36, 49, 59 228, 232, 233, 234, 239, 241, 244, Williamson, T. 203, 214, 215, 219, 222,
Kleinschmidt, S. 265, 278 ()'Shaughnessy, B. 92, 122 252,257, 315 224, 226, 234
Kramer, S. 283, 314 Siebert, C. 95, 97, l 22 Woodward, R. 236, 252, 257
Koch,/\.. 165, 169, 170 Palacios, J. 179, 186, I 89, 198 Skow, B. 186 Wright, C. 63, 67, 68, 72, 73, 202
Koons, R. 271, 273, 278 Parsons, T. 24, 37 Smith, /\.. D. 38, 95, I 22
Korman, D. 207, 233 Pettit, l'. 86, 122 Smithson, R. 166, 169, 170 Yablo, S. 85, 108, 123
Kratzer, /\.. 24, 36 Peacocke, C. 67, 72, 73, 85, 122 Steinberg, /\.. I 0, 37, 63, 73, 315
Kroll, N. 28, 29, 36, 38-59 Phillips, I. 86, 92, 122 Stevenson, J. 66, 67, 68, 73, 76, 80 Zalta, E. 222, 233
Kripke, S. 221, 236 Place, U. 31, 37 Szabo, Z. 26, 29, 34, 37 Zimmerman, D. 229, 272
Kulvicki, J. 116, 121 Plank, M. 179, 198
Plantinga, /\.. 235, 236, 238, 252, 257
Landman, F. 23, 25, 36 Poggiolesi, F. 279, 314
Lange, M. 172, 173, 194, 198 Portner, P. 23, 37
Langhaar, !-I. 179, 198 Prawitz, D. 67, 73
Leslie, J. 270, 278 Priest, C. 127, 170
Lewis, D. 6, 8, 15, 16, 36, 51, 59, 180, Prior, J\. 67, 73
198, 203, 207, 223, 233, 236-43, Pruss, A. 264, 266, 268, 273, 275, 276,
246-8, 250-7, 278 277, 278
l.iersc, C. 29, 36 Putnam, !-I. 164, 167, 169, 170, 202,
Linsky, B. 222, 233 254, 257
Lidand, J. 282, 283, 290, 293, 314
Locke, J. 85 Quine, W. V. 0. 4, 37, 63, 65, 73, 202,
Luce, R. 173, 198 233, 256