Está en la página 1de 7

See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/283378899

Attention in Language

Chapter · December 2005


DOI: 10.1016/B978-012375731-9/50057-4

CITATIONS READS

13 745

2 authors:

Andriy Myachykov Michael Posner


Northumbria University University of Oregon
59 PUBLICATIONS 308 CITATIONS 439 PUBLICATIONS 78,049 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

An information theory account of cognitive control View project

Human attentional networks View project

All content following this page was uploaded by Andriy Myachykov on 14 November 2017.

The user has requested enhancement of the downloaded file.


INO053 11/11/04 12:54 PM Page 324

CHAPTER

53
Attention in Language
Andriy Myachykov and Michael I. Posner

ABSTRACT networks that are present, at least in full form, only in


the human species. However, prior to imaging, it is
Attention plays an important role in critical aspects important to have tasks that activate selectively key
of the use of grammar and lexicon in human discourse. aspects of grammar and semantics. In this chapter we
We explore methods for the activation of these opera- consider empirical methods and results that trace the
tions during the choice of an adequate syntactic role of attention in processing language and report
structure, referential control, and other grammatical some imaging experiments that have examined the
operations, during the processing of word and sen- mechanisms involved.
tence meaning and in the skill of reading. Neuroimag-
ing has suggested separate systems for syntactic and
semantic processing and has provided some details on II. ATTENTION AND THE
the anatomy of attentional systems related to seman- STRUCTURE OF DISCOURSE
tic analysis.
Two aspects of discourse grammar are usually
ascribed to attentional mechanisms:
I. INTRODUCTION 1. Assignment of syntactic roles in a clause.
2. Referential choice and resolution of ambiguous
The neurobiology of attention and most models dis- reference in discourse.
cussed in this volume have quite naturally involved
rather simple forms of repetitive behavior that can be
studied relatively easily by experimental methods and
A. Assignment of Syntactic Roles
are common to both humans and other animals. From A good example of the research in this area is a set
this perspective, the role of attention in language may of experimental studies by R. Tomlin (1997), in which
seem an odd topic. Sentences are hardly ever repeated, he demonstrated that the choice of syntactic subject
and grammar is a complex set of operations within position in English narrative might result from the
each language. Nonetheless, it is important to under- direction of attention. In a set of experiments, Tomlin
stand how top-down control is exercised over species- used a computer animation program called the “Fish
specific higher mental activities that are involved in Film” (Fig. 53.1). Subjects viewed and described an
aspects of language such as grammar, semantics, or unfolding engagement of the two fish in real time. In
reading. each trial one of the two fish was visually cued (see
Spoken language is a good example of a species- arrow in Fig. 53.1) attracting the subjects’ attention.
specific behavior of human beings. The ability to read The critical part of each trial is the dynamic event, which
represents a high-level skill, closely coordinated with is described by the subject. In their description of the
spoken language, but learned only by a subset of the dynamic event, assignment of the syntactic subject is
human community. The role of attention in reading varied depending on the visual cue. The dynamic
provides an entry into the many complex but arbitrary event was the eating of one of the fish by the other.
skills that human beings can learn to perform. When the dark fish was cued and was then eaten by
Developments in neuroimaging make it possible to the light fish, the subjects said, “the dark fish was eaten
examine the anatomy, circuitry, and plasticity of brain by the light fish.” When the light fish was cued, the

Copyright 2005, Elsevier, Inc.


Neurobiology of Attention 324 All rights reserved.
INO053 11/11/04 12:54 PM Page 325

II. ATTENTION AND THE STRUCTURE OF DISCOURSE 325

very same dynamic event was described as “the light by use of a pronoun or a noun phrase. The choice of a
fish ate the dark fish.” Attention to the cue influenced referent to indicate an entity previously mentioned in
the choice of syntactic subject of the sentence and the discourse is sometimes called anaphora resolution. Con-
grammatical voice, depending on which stimulus was sider the sentence:
cued. The effect demonstrated by Tomlin was robust: (1) A researcher has written a new book recently
the cued stimulus appeared as the subject of the sen- and (Ø) managed to publish it, too.
tence in over 90 percent of the trials. Further analyses What aspects of the antecedents “a book” and “a
showed that these results hold as well for other films researcher” influence the consequent choice of the
and languages other than English. referring anaphoric pronoun “it” for the former and
Tomlin’s approach was extended to the study of the not mentioning the referent (“zero anaphora”), for the
descriptions of static events (Forrest, 1996). L. Forrest later? Usually, both zero and a pronoun anaphora are
manipulated subjects’ attention to one of the quad- chosen if the antecedent has been mentioned relatively
rants of the display with the help of a visual cue and recently. In this case, a common assumption is that the
showed that it influenced the narrative. (see Fig. 53.2). pronoun’s antecedent is a salient entity in the text.
These results show how the choice of syntactic subject More salient entities are accessed more easily and
is determined by the direction of attention. therefore are marked with anaphoric expressions
of a lesser semantic content, for example, a definite
pronoun or a zero anaphora. Less salient entities
B. Reference and Anaphora require more semantically specific forms of reference,
Engaging in discourse requires being able to prop- for example, a full noun phrase.
erly identify and track objects, events, or people that Salience is often described as a combination of the
the discourse concerns. Directing attention to the enti- concept’s degree of activation in the working memory
ties in oral or written text usually involves reference and manipulations of the focus of attention. In
Tomlin’s research, described above, subject assign-
ment was cued by an external event. However, in com-
prehension of discourse, the most common way of
subject assignment is to use linguistic cues to focus
attention. For example, when the antecedent is the
subject of the first clause, then the anaphoric pronoun
will also assume the position of the subject in the
second clause. Another example of a linguistic cue is
that the first mentioned entity has a greater chance of
being referred to by a pronoun rather than a nonphrase
(see Garnham, 1999 for a review).
Both of these linguistic cues have received substan-
tial support. For example, in sentence completion
studies it was found that subject antecedents are
indeed prefered for the reference in the second clause.
However, the preference of the antecedent may also
FIGURE 53.1 Fish Film. depend on other factors. For example, implicit causal-

FIGURE 53.2 Example of stimuli from Forrest (1996).

SECTION II. FUNCTIONS


INO053 11/11/04 12:54 PM Page 326

326 CHAPTER 53. ATTENTION IN LANGUAGE

ity of the verb may bias subjects’ preference for object is to choose a noun phrase in order to refer to the
of the first clause to be refered to by the anaphora. antecedent; the shorter the distance, the more likely
(Garnham, 1987; 1999 for a review). it will be a pronoun (presented or implied as a zero
A more complicated explanation of discourse focus anaphora). If the referent is an anaphoric pronoun,
is a computational model known as centering theory the cognitive search for an antecedent within a file is
(e.g., Grosz et al., 1995). This model hypothesizes the started. This scan is very likely to be short-ranged,
existence of two sets of discourse centers. The forward- localized within a couple of preceding clauses, highly
looking center concerns salience of the discourse enti- automatic, and routine. Finally, a definite noun
ties. The backward-looking center is related to references demands a long-ranged, less automatized, and more
to these entities. Forward and backward centers attentionaly demanding search of an antecedent
provide a mechanism of establishing the coherence of (Givón, 1995).
the discourse. Different factors affect the salience (or Psycholinguists also use experimental methods to
“preference”) of forward-looking centers. The most gain evidence supporting the role of referential dis-
frequently mentioned are surface position, grammati- tance. For example, when subjects are asked questions
cal role, and being a backward-looking center of the concerning the referents of the pronoun in the last sen-
utterance. tence of the passage they had just read, the number
Experimental work in the centering theory shows of errors tends to increase with increasing distance
that pronouns referring to more preferred characters between the pronoun and its antecedent. Another way
are understood more quickly than those that refer to is to measure the time it takes to read a target sentence.
less preferred ones. Studies also show that sentences In the standard variant of this task, subjects are given
are read faster when backward-looking center is a a segment of discourse in which the antecedent intro-
pronoun than when it is a noun, and subjects judge duced in the first sentence is later referred to by means
these sentences as more coherent. Also, in reading of different linguistic forms at different textual dis-
short texts consisting of three sentences, processing tances. One result is that it takes less time to read and
time is reduced when the third sentence pronoun understand the sentence that contains a pronominal
refers to the second sentence antecedent than when it referent if its antecedent is at a shorter distance, and
refers to an antecedent in the first sentence (e.g., therefore salient and highly activated (Garnham,
Gordon et al., 1993). 1987).
The choice of anaphoric expression also depends on Priming of lexical decisions can also be used to trace
the maintenance of the referent in memory. Although the degree and time course of activation in referential
other linguistic factors may contribute, this issue is choice. Subjects read the sentence and then decide
closely related to distance separating the anaphora whether a target word (usually one of the two possi-
from its antecedent. One task of psycholinguists is to ble antecedents was in the sentence they have just
trace the change of linguistic forms dynamically as the read. The experimental evidence is that in the case
referential distance varies. Support for the role of dis- of two possible candidate antecedents, the pronoun
tance comes from studies of the frequency of different actually activates both of them, but context serves to
forms of reference in written passages. The results of direct attention to the correct antecedent (Garnham,
one such study are illustrated in Table 53.1 (adapted 1987).
from Givón, 1995). By measuring eye movements, one can examine
A new entity is usually introduced by means of an the time course of anaphoric resolution in more detail.
indefinite noun or a proper name. After that, the rule The time and location of the fixations are both shown
is: the larger the textual distance, the more likely one to reflect the referential distance between the target
pronoun and its antecedent (see Stevenson, 1996, for a
review). Eye movements may also be of help when
TABLE 53.1 Referential Distance and the Choice of trying to disambiguate a sentence in the course of
Linguistic Form
comprehension. When linguistic cues are missing
Referential distance
or ambiguous, eye movements may be triggered to
Construction in # of clauses resolve the ambiguity and attempt to gather the
missing information (Tanenhaus et al., 1995).
1. Zero anaphora 1.0 Syntactic and semantic processing have been
2. Unstressed pronoun 1.0 studied by neuroimaging. Violations of semantic or
3. Stressed pronoun 2.5
4. Definite noun 7.0
syntactic structure of the sentence are commonly used
5. Definite noun with a modifier 10.0 in studies recording scalp electrical potentials (ERP).
These violations give rise to quite distinct wave forms.

SECTION II. FUNCTIONS


INO053 11/11/04 12:54 PM Page 327

III. ATTENTION AND SEMANTICS 327

Semantic violations produce a negative wave with a additional areas of cortical activation that are greater
peak latency of 400 ms (N400). (See Kutas and Van in the generate condition are in the left lateral pre-
Petten, 1994, for a review). In the cases of structural or frontal cortex and posterior temporal cortex. Both of
syntactic violations, an early left anterior negativity these areas have been shown to be involved in many
(e.g., Osterhout and Mobley, 1995) and/or a late posi- tasks dealing with processing the meaning of words or
tive wave with a peak at 600 ms (P600) (e.g., Hagoort sentences.
et al., 1993) are frequently observed. To examine the time course of these activations, it is
Hahne and Friederici (1999) hypothesized that the possible to use a large number of scalp electrodes to
early left anterior negativity is a highly automatic obtain an ERP, which is the scalp signature of the gen-
process, whereas the P600 involves more attention. In erators found active in imaging studies (Abdullaev
a study aimed at testing this hypothesis, the propor- and Posner, 1998). When subjects think of the use of a
tion of correct sentences and sentences with structure noun, there is an area of positive electrical activity over
violations was varied. Incorrect sentences might frontal electrodes starting about 150 msec after the
appear to be either of a low (20% violation) or a high word appears. This early electrical activity is gener-
(80% violation) proportion. The early left negativity ated by the large area of activation in the anterior
was elicited under both conditions. P600 was elicited cingulate. A left prefrontal area (anterior to and
only for a low proportion of incorrect sentences. These overlapping classical Broca’s area) begins to show
results support the idea that early left negativity is an activity about 200 msec after the word occurs. At first
automated first-pass phrase parsing, whereas P600 this area was called semantic because it was more
relates to a second-pass parsing that requires a larger active in the semantic task than in reading aloud.
allocation of subjects’ attention. However, because this lateral area was often involved
A limited number of reports have used fMRI in in working memory and its time course was early, it is
order to identify brain regions associated with syntac- more likely that this lateral frontal area is related to
tic and semantic processes. One study (Newman et al., operations such as holding the lexical item in a brief
2001), found that syntactic violations elicit signifi- store during the time needed to which to look up the
cantly greater activation in superior frontal cortex. associated word use.
Semantic anomalies result in greater activation in infe- The left posterior brain area found to be more active
rior frontal and temporoparietal regions, both of which during the processing of the meaning of visual words
are considered in more detail in the next section. did not appear until a much later time (500 msec). This
activity is near the classical Wernicke’s area, lesions of
which are known to produce a loss of understanding
III. ATTENTION AND SEMANTICS of meaningful speech. These findings suggest this area
is important in finding the semantic association. There
is also evidence of the transfer of information from left
A. Word Association frontal electrodes to the posterior area at about 450
A common task for studying the brain systems millisec into the task. Since the response time for this
involved in semantic processing is to ask subjects to task was about 1,100 msec, this would leave time for
give the use of a common noun (e.g., hammer). In a the generation of related associations needed to solve
typical version of this task for imaging, the subjects are the task.
shown a series of 40 simple nouns (Raichle et al., 1994). Practice on a single list of words reduces the acti-
In the experimental condition, they indicate the use of vation in both the anterior cingulate and lateral corti-
each noun (for example, hammer Æ pound). In the cal areas (Raichle et al., 1994). Thus, the very same task,
control condition, they simply read the word aloud. when it is highly over learned, avoids the circuits
The difference in activation between the two tasks involved in thought and relies upon an entirely differ-
illustrates what happens in the brain when subjects are ent circuitry.
required to develop a very simple thought, in this These studies provide a start in understanding the
case how to use a hammer. Results illustrate that the functional roles of different brain areas in carrying out
anatomy of this high-level cognitive activity is similar executive control. The medial frontal area appears
enough among individuals to produce focal average most related to the executive attention network and
activations that are both statistically significant and is active when there is conflict among stimuli and
reproducible. responses. It may be serving as a monitor of conflict,
One area that is more strongly activated when gen- but it is possible that it plays other roles as well.
erating the use of a word is on the midline of the The lateral prefrontal area seems to be important in
frontal lobe in the anterior cingulate gyrus. Two holding in mind the information relevant to the task.

SECTION II. FUNCTIONS


INO053 11/11/04 12:54 PM Page 328

328 CHAPTER 53. ATTENTION IN LANGUAGE

Even when a single item is presented, it may still References


be necessary to hold it in some temporary area while
Abdullaev, Y. G., and Posner, M. I. (1998). Event-related brain poten-
other brain areas retrieve information relevant to the
tial imaging of semantic encoding during processing single
response. Together these two areas are needed to solve words. Neuroimage 7, 1–13.
nearly any problem, which depends upon retrieval of Duncan, J., Seitz, R. J., Kolodny, J., Bor, D., Herzog, H., Ahmed, A.,
stored information (Duncan et al., 2000). Both of these Newell, F. N., and Emslie, H. (2000). A neural basis for general
areas could be said to be related to attention, or one intelligence. Science 289, 457–460.
Forrest, L. B. (1996). Discourse goals and attentional processes in
might identify only the medial area with attention and
sentence production: the dynamic construal of events. In “Con-
the lateral area with working memory. In either case ceptual Structure, Discourse and Language.” (A. E. Goldberg,
they begin to give us a handle on how the brain parses Ed.), Stanford, CA, CSLI Publications, 149–162.
high-level tasks into individual operations that are Garnham, A. (1987). Understanding anaphora. In “Progress in the
carried out in separate parts of the network. Psychology of Language.” (A. W. Allis, Ed.), Vol. 3. Lawrence
Erlbaum Associates, London, 253–300.
Garnham, A. (1999). Reference and anaphora. In “Language Pro-
B. Reading cessing.” (S. Garrod and M. J. Pickering, Eds.), Psychology Press,
Hove, England, 335–362.
The ability of fluent adult reading seems to rely
Givón, T. (1995). Coherence in text versus coherence in mind.
upon two areas of posterior cortex that in fluent In “Coherence in spontaneous text.” (M. A. Gernsbacher and
readers automatically organize visual words. The T. Givon, Eds.), TSL#31. John Benjamins, Amsterdam, 59–
visual word form area lies within the visual system in 115.
the fusiform gyrus of the left hemisphere. In skilled Gordon, P. C., Grosz, B. J., and Gilliom, L. A. (1993). Pronouns,
names, and the centering of attention in discourse. Cognitive
adult readers, this area appears to package the letters
Science 17, 311–347.
of words into visual units. This system allows the Grosz, B. J., Joshi, A. K., and Weinstein, S. (1995). Centering: a frame-
reader to avoid scanning the individual letters. A work for modelling the local coherence of discourse. Computa-
second area that lies close to auditory cortex allows the tional Linguistics 21, 203–225.
fluent reader to retrieve a phonological word unit from Hagoort, P., Brown, C., and Groothusen, J. (1993). The syntactic pos-
itive shift as an ERP measure of syntactic processing. Language
the visual input. Children with difficulties related to
and Cognitive Processes 8, 439–484.
the word form can read slowly but lack fluency, while Hahne, A., and Frederici, A. D. (1999). Electrophysiological evidence
those with phonological difficulties cannot sound out for two steps in syntactic analysis: early automatic and late
new words and are limited to reading familiar words controlled processes. Journal of Cognitive Neuroscience 11, 194–
and guessing at those they do not know. Although in 205.
Harm, M. W., and Seidenberg, M. S. (1999). Phonology, reading
a logical sense the visual word form precedes the
acquisition, and dyslexia: Insights from connectionist models.
phonological system, there is evidence that the visual Psychological Review 106, 491–528.
word form may develop later, and the visual and Kutas, M., and Van Petten, C. (1994). Psycholinguistics electrified:
phonological system function in close interaction Event-related brain potential investigations. In “Handbook of
(Harm & Seidenberg, 1999). Psycholinguistics.” (M. Gernsbacher, Ed.), Academic Press, New
York, 83–143.
Little or no attention is needed for these systems to
Newman, A. J., Pancheva, R., Ozawa, K., Neville, H. J., and Ullman,
operate in the skilled reader, but attention is needed M. T. (2001). An event-related fMRI study of syntactic and
for these systems to develop properly. Children with semantic violations. Journal of Psycholinguist Resources 30,
difficulty in learning to read employ the executive 339–364.
attention network (including both lateral and medial Osterhout, L., and Mobley, L. A. (1995). Event-related brain poten-
tials elicited by failure to agree. Journal of Memory and Language
frontal areas) to compensate for insufficient develop-
34, 739–773.
ment of the more automatic posterior visual word Posner, M. I., and McCandliss, B. D. (1999). Brain circuitry during
form and phonological systems (Shaywitz et al., 1999). reading. In “Converging methods for understanding reading and
The word form system seems to develop relatively dyslexia.” (R. Klein and P. McMullen, Eds.), MIT Press, Cam-
late. While one can read without an efficient word bridge, 305–337.
Raichle, M. E., Fiez, J. A., Videen, T. O., McCleod, A. M. K., Pardo,
form system, it apparently improves the speed and
J. V., Fox, P. T., and Petersen, S. E. (1994). Practice-related changes
fluency of reading. Studies show that at age 10, in the human brain: functional anatomy during nonmotor learn-
activation of the visual word form system requires ing. Cerebral Cortex 4, 8–26.
familiar words that children have already learned. Shaywitz, S. E., Shaywitz, B. A., Pugh, K. R., Fulbright, R. K.,
However, in skilled adults the visual word form is Constable, R. T., Mencl, W. E., Shankweiler, D. P., Liberman, A.
M., Skudlarski, P., Fletcher, J. M., Katz, L., Marchione, K. E.,
activated by letter strings never before seen, provided
Lacadie, C., Gatenby, C., and Gore, J. C. (1998). Functional dis-
that the strings obey the rules of the written language ruption in the organization of the brain for reading in dyslexia.
and thus are orthographically regular (Posner & Proceedings of the National Academy of Sciences of the United States
McCandliss, 1999). of America 95, 2636–2641.

SECTION II. FUNCTIONS


INO053 11/11/04 12:54 PM Page 329

III. ATTENTION AND SEMANTICS 329

Stevenson, R. J. (1996). Mental models, propositions and the tion in spoken language comprehension. Science 268, 1632–
comprehension of pronouns. In “Mental Models in Cognitive 1634.
Science.” (J. Oakhill and A. Garnham, Eds.), Psychology Press, Tomlin, R. S. (1997). Mapping conceptual representations into lin-
Hove, England, 53–76. guistic representations: The role of attention in grammar. In
Tanenhaus, M. K., Spivey-Knowlton, M. J., Eberhard, K. M., and “Language and Conceptualization” (J. Nuyts and E. Pederson.),
Sedivy, J. E. (l995). Integration of visual and linguistic informa- 162–189.

SECTION II. FUNCTIONS

View publication stats