Cerebral Cortex August 2007;17:1958--1969
doi:10.1093/cercor/bhl105
Advance Access publication November 13, 2006
Cortical Representation of Verb Processing
in Sentence Comprehension: Number of
Complements, Subcategorization, and
Thematic Frames
Einat Shetreet1, Dafna Palti1,2, Naama Friedmann3 and
Uri Hadar1
The processing of various attributes of verbs is crucial for sentence
comprehension. Verb attributes include the number of complements
the verb selects, the number of different syntactic phrase types
(subcategorization options), and the number of different thematic
roles (thematic options). Two functional magnetic resonance
imaging experiments investigated the cerebral location and pattern
of activation of these attributes. Experiment 1 tested the effect of
number of complements. Experiment 2 tested the number of options
of subcategorization and of thematic frames. A group of mismatch
verbs with different number of options for subcategorization and
thematic frames was included to distinguish between the effects of
these attributes. Fourteen Hebrew speakers performed a semantic
decision task on auditorily presented sentences. Parametric analysis revealed graded activations in the left superior temporal gyrus
and the left inferior frontal gyrus in correlation with the number of
options. By contrast, the areas that correlated with the number of
complements, the right precuneus and the right cingulate, were not
conventionally linguistic. This suggests that processing the number
of options is more specifically linguistic than processing the number
of complements. The mismatch verbs showed a pattern of activation similar to that of the subcategorization group but unlike that
of the thematic frames group. By implication, and contrary to claims
by some linguists, subcategorization seems indispensable in verb
processing.
participating in the event described by the verb, beyond and in
addition to the subject. Both the subject and the complements
act as the verb’s arguments. For example, the verb give describes, as shown in example (1), an event with 3 entities: the
giver—John, the receiver—Anna, and what was given—a
present. In this sentence, there are 3 arguments: John, Anna,
and a present. John is the subject, and Anna and a present are
the 2 complements of the verb.
Keywords: argument structure, fMRI, Hebrew, lexicon, neurolinguistics,
syntax
Introduction
Verbs play a key role in sentence production and comprehension because they specify the relations among words in a
sentence. It is well accepted that some crucial aspects of
sentence processing are determined by the semantic and syntactic attributes of the verbs that appear in the sentence
(Shapiro et al. 1987; Collina et al. 2001; van Valin 2001), but it
is still undetermined which of these are available and accessed
online. The current functional magnetic resonance imaging
(fMRI) study explored the manner in which verb attributes affect sentence processing in auditory comprehension and the
cortical locations that subserve the various attributes of verbs.
We examined particular patterns of activation during the systematic variation of some crucial verb attributes. This, we
hoped, could also bear upon certain linguistic controversies
concerning the lexical representation of verbs. Specifically, we
designed the study to contrast the relative contribution of
semantic and syntactic factors to the various investigated verb
attributes.
The 3 attributes at the center of our study concern the
complements of the verb. The verbal complements are entities
Ó The Author 2006. Published by Oxford University Press. All rights reserved.
For permissions, please e-mail: [email protected]
1
Department of Psychology, Tel Aviv University, Tel Aviv
69978, Israel, 2Wohl Institute for Advanced Imaging, Tel Aviv
Sourasky Medical Center, Tel Aviv, Israel and 3School of
Education, Language and Brain Laboratory, Tel Aviv University,
Tel Aviv, Israel
1. John gave Anna a present
Although the details of this are still debated, linguistic theory
assumes that the representation of the verbal complements
involves several types of information. Access to the lexical
information concerning the complements of the verb was evident during the early stages of sentence processing (e.g.,
Holmes 1987; Shapiro et al. 1987, 1993; Tanenhaus et al. 1989;
Boland et al. 1990; Boland 1993; Trueswell et al. 1993). The
related information included the number of complements of
the verb, the syntactic types of phrases that can serve as complements, and the thematic roles that the complements can
receive.
Number of Complements
Different verbs have different number of complements, depending on the number of entities participating in the event described by the verb. For example, verbs like smile or sneeze
(example 2) have no complements, only a subject. Verbs from
this category are called unergatives; verbs that have one
complement, like lost or ignore (example 3), are called transitives; and verbs that have 2 complements, like give or put
(example 4), are called ditransitives or datives.
2. No complement: John smiled
3. One complement: John lost the keys
4. Two complements: John gave Anna a present
Number of Subcategorization Options
The syntactic types of phrases that can complement the verb
are also assumed to be represented in the lexical entry
(Chomsky 1965; Grimshaw 1979, 1981; Friederici 1995; Collina
et al. 2001). For example, the verb lost can only be complemented with a noun phrase (NP), as illustrated in example (5)
below, and the verb depend can only be complemented with
a prepositional phrase (PP) (example 6). The verb discover can
be complemented with either an NP or a finite clause (complementizer phrase [CP]), as shown in example (7).
5. John lost fthe keysgNP
6. Mary depends on fthe spell checkergPP
7. a. John discovered fthe storygNP
b. John discovered fthat the story is realgCP
The syntactic types of phrases that can be added to the verb
are defined by its subcategorization (van Valin 2001; Carnie
2002). Subcategorization differentiates among types of verbs.
Verbs differ both with respect to the type of subcategorization
they select and with respect to the number of subcategorization
options they allow. As mentioned in the previous examples, lost
and depend have only one subcategorization option, whereas
discover has 2 subcategorization options.
Number of Thematic Options
The third attribute that we consider in the current study is the
thematic frames of a verb. This refers to the set of arguments
associated with the verb in terms of their possible thematic
roles. Thematic roles (also known as theta roles, or h roles) are
the roles performed by the various participants in an event. Very
generally, they describe ‘‘who did what to whom’’ in the sentence. Note that this attribute was termed argument structure
in some studies (Shapiro et al. 1987, 1989, 1993; Shapiro and
Levine 1990; Rubin et al. 1996; Radford 1997; van Valin 2001).
However, argument structure was also used to describe the
number of complements (Haegeman 1995; Thompson et al.
1997; Collina et al. 2001; Carnie 2002) or the lexical information
about the sentential environment of a verb, including syntactic
and semantic information (Trueswell and Kim 1998; Friederici
and Frisch 2000). Therefore, we termed this attribute ‘‘thematic
frames,’’ to emphasize its thematic nature.
Apart from the well-accepted thematic roles such as ‘‘agent’’
(the entity that performs the action or brings about some
change), ‘‘theme’’ (the entity that the action is performed upon),
and ‘‘goal’’ (the participant that is the target of the transfer or
motion in an event denoted by a verb of motion or transfer,
e.g., give), we also consider more complex semantic types such
as ‘‘proposition’’, which stands for sentential complements of
verbs (Grimshaw 1979, 2006; Chomsky 1986; Shapiro et al.
1987, 1989; Rochette 1988; Pesetsky 1991; Carnie 2002).
Like with subcategorization, verbs differ not only by their
possible thematic frames but also by the number of thematic
options that they select: lost has only one thematic option, theme
(example 8), whereas discover has 2 thematic options, theme or
proposition (example 9) (some linguistic theories assume that
both complementation options of discover take the role of
a theme [Haegeman 1995], a claim that will further emphasize
our results with respect to the necessity of subcategorization
frames. We will examine this in the Discussion below).
8. a. <John >agent lost <the keys >theme
9. a. <John >agent discovered <the story >theme
b. <John >agent discovered <that the story is real >proposition
To elucidate the differences between the number of complements and the number of options (of both subcategorization
and thematic frames), let us try to conceptualize the linguistic
difference between number of complements and number of
subcategorization and thematic options. Many syntactic theories represent sentences using ‘‘syntactic trees’’ (or tree diagrams; e.g., van Valin 2001) that represent the syntactic
hierarchy of the various components in the sentence. In the
part of the tree that represents the verb phrase (VP), complements appear as branches and the number of complements is
reflected in the number of branches (Fig. 1). Thus, phrases
Figure 1. Illustration of part of a syntactic tree, the verb phrase (VP), which includes
the verb and its complements. Each complement is represented by a single branch:
verbs with no complements (A) have a branch only for the verb, verbs with 1
complement (B, C) have a branch for the verb and for the single complement, whereas
verbs with 2 complements (D) have a branch for the verb and 2 branches for the
complements. Verbs that select more than one option have more than one tree, and in
each tree the complement type is different (B, C). A ternary branching was used in this
figure for simplification. Following Kayne (1984) and Larson (1988), the accepted
analysis for ditransitives is a strictly binary structure of the form [vp V [vp NP1 [V’ V
NP2]]. In this structure, the VP consists of an empty V taking VP complement whose
head is the verb, with the first object at its specifier position and the second object as
its complement (see also Radford 1997; van Valin 2001).
containing verbs with no complements have no branches for
complements, phrases with transitive verbs have a single
branch, and those with ditransitive verbs have 2 branches. As
for the verb’s options, each option gives rise to a different VP, so
each option can be thought of as represented by a separate
syntactic tree. Hence, access to verbs with 2 options requires
the retrieval of 2 syntactic trees, and access to verbs with 3
options requires the retrieval of 3 syntactic trees. On the whole,
the difference between the number of complements and the
number of options may be represented linguistically as the
difference between, respectively, the number of branches in a
single tree and the number of trees.
The number of subcategorization options and the number of
thematic options are the same for most verbs (as, e.g., in the case
of the verb discover). Nonetheless, there are few verb categories that allow discrimination between the 2 attributes. One
such verb category is a group of transitive verbs that has 2
subcategorization options but only one thematic option. For
example, as illustrated in examples (10) and (11) below, the
verb taste selects either an NP or a PP and both have the same
thematic role, theme.
10. a.
b.
11. a.
b.
John tasted fthe soupgNP
John tasted ffrom the soupgPP
<John >agent tasted <the soup >theme
<John >agent tasted <from the soup >theme
In the past 2 decades, claims have often been made to the
effect that the lexical entries of verbs should consist only of
thematic information, and not of subcategorization, and that
subcategorization can be reduced to thematic roles (Stowell
1981; Pesetsky 1982, 1991; Chomsky 1986, 1995). For example,
Chomsky (1995) claims that ‘‘subcategorization follows almost
entirely from thematic-role specification’’ (p. 31, see also
Pesetsky 1982). However, it seems that the linguistic jury is
still out on the question of whether or not it is possible to
dispense with subcategorization altogether, as Chomsky (1995)
notes in the end of his discussion of this subject: ‘‘formal
syntactic specifications in lexical entries have not been entirely
eliminated in favor of semantic ones . . . while much of
c-selection (subcategorization) follows from s-selection (semantic properties), there is still a syntactic residue . . . (p. 33)’’.
Transitive verbs that select different number of options for
subcategorization and thematic frames will be used in the
current study to evaluate through brain activations the relation
Cerebral Cortex August 2007, V 17 N 8 1959
between subcategorization and thematic frames in verb processing. This may allow us to establish whether the number of
thematic options by themselves may account for the patterns of
activation during sentence comprehension or whether one also
needs to consider the subcategorization requirements of verbs.
Another distinction that is investigated in the current study is
between 2 types of clausal complements of verbs: finite clauses
(that clauses, clauses that include a finite inflected verb) and
nonfinite clauses (to clauses, with an infinitive verb). These
clausal complements are semantically and syntactically different,
and they are selected by different verbs and appear in different
contexts (e.g., Pesetsky 1991). Whereas some verbs can select
both to and that clauses (such as demand in example 12), other
verbs can be complemented by only one of these clausal types.
For example, start selects only nonfinite clauses (example 13, an
asterisk represents an ungrammatical sentence), whereas deny
selects only finite clauses (example 14). According to Pesetsky
(1991), l-selection (lexical selection, which is different from
both subcategorization and thematic frames) makes reference
to features like [+finite], implying that a distinction between to
and that clauses is represented in the lexicon.
12. a.
b.
13. a.
b.
14. a.
b.
John demanded [to read the book]
John demanded [that Anna would smile]
John started [to read the book]
*John started [that Anna smiled]
*John denied [to read the book]
John denied [that Anna smiled]
Is this distinction semantic, that is, encoded in the thematic
frame, or is it syntactic, pertaining to the verb subcategorization? Some linguists suggested that these 2 types of clauses
form separate semantic types. Whereas finite that clauses are
propositions, the nonfinite to clauses take the semantic type
of ‘‘infinitive’’ or ‘‘event’’ (Pesetsky 1982; Rochette 1988;
Grimshaw, cited in Shapiro and Levine 1990; Grimshaw 2006).
In this paper, we choose to use the term event in order to
distinguish it from terms that we used here for subcategorization options.
According to some other linguistic approaches, the distinction between to and that clauses may also be evident in
their syntactic representation. Bošković’s economic approach
(Bošković 1997) suggests that nonfinite clausal complements
are inflectional phrases (IPs) and finite clauses are CPs, indicating the separation of the clausal subcategorization to IP
and CP (for a suggestion regarding different subcategorization
for that and for-to clauses, also see Bresnan [1972, cited in
Grimshaw 1979]).
If to clauses and that clauses have the same subcategorization
and thematic role, all 3 verbs, deny, demand, and start, select 2
subcategorization and thematic options (because all select NP
and clausal complements). However, if the representation of nonfinite clauses and finite clauses in the lexical entry of a verb is
separate, verbs such as demand have 3 subcategorization options and 3 thematic options, as illustrated in examples (15--16),
whereas start and deny have only 2 options of both attributes.
15. a.
b.
c.
16. a.
b.
c.
John demanded fthe bookgNP
John demanded fto read the bookgIP
John demanded fthat Anna would smilegCP
<John >agent demanded < the book >theme
<John >agent demanded <to read the book >event
<John >agent demanded < that Anna would smile >proposition
1960 The fMRI of Verb Processing
d
Shetreet et al.
All the above attributes—number of complements, subcategorization options, and thematic options—are assumed to be
represented in the verb’s lexical entry. Yet, it is still unclear how
crucial each of them is for sentence processing, which of them
is accessed online, and what the related neural activations
are. Previous studies examined the effects of semantic--syntactic
attributes of verbs on sentence processing and yielded inconsistent results. Fodor et al. (1968) found that sentences with
verbs that selected 2 options caused more errors in paraphrasing and anagram tasks than those with verbs that selected one
option, suggesting that the number of a verb’s options affected
processing load. No significant effect of response time (RT) was
found in this study.
In a series of seminal studies involving online comprehension,
Shapiro and his colleagues (Shapiro et al. 1987, 1989, 1991,
1993) used a cross-modal lexical decision task with different
types of verbs. The time to make a lexical decision was seen as
dependent upon the processing load of the verb. Therefore,
differences in RT could be used to determine the amount of
information that was needed to access the verb. In one experiment, Shapiro et al. (1987) compared the processing load of
the number of subcategorization options with that of the
number of thematic options, using RTs as the measure of processing load. They found increased RTs as a function of the
number of thematic options but not of subcategorization options. Note that in order for the number of thematic options to
increase the processing load, all options of a given verb must be
activated at some stage of processing, regardless of the actual
complement or the related context, and this has been reported
by Shapiro et al. (1989, 1993). Importantly, also for the current
study, Shapiro et al. (1987) tested whether the number of complements had an effect on access to the verb and found that,
unlike the number of thematic options, the number of complements did not affect verb access: verbs with one complement
did not differ significantly from verbs with 2 complements.
Other studies failed to find this effect. Schmauder (1991) and
Schmauder et al. (1991) used fixation time in eye-tracking tasks
and reaction time in interference tasks but found no clear
evidence of either subcategorization or thematic frames effects.
It is unclear what can be deduced from their null results, but
the basic effect found by Shapiro and his colleagues has been
replicated a number of times since the study of Shapiro et al.
(1987). Another study that diverged from Shapiro et al. (1987)
reexamined the verbs from the original study and found that the
number of complements did affect processing load, whereas the
number of thematic options did not (Ahrens and Swinney
1995).
Additional evidence regarding the number of thematic
options came from studies with aphasic patients. Using the
above lexical decision task, Shapiro and Levine (1990) compared 2 groups of patients: one with Broca’s aphasia and one
with fluent aphasia (a heterogeneous group that consisted of 4
patients with Wernicke’s aphasia, 1 with conduction aphasia,
and 1 with anomia) to a control group of healthy subjects.
Patients with Broca’s aphasia showed response patterns similar
to those of the normal controls. Both Broca’s aphasic patients
and the control group showed online sensitivity to the number
of thematic options. However, this type of sensitivity was not
observed with the fluent aphasic patients, suggesting that
posterior areas, including Wernicke’s area, participate in the
processing of the thematic frames. The involvement of
Wernicke’s area in online processing of thematic frames was
confirmed in a subsequent study by Shapiro et al. (1993), in
which the heterogeneous group of fluent aphasic patients was
replaced with a group of Wernicke’s aphasic patients. The
results were similar to those of the earlier study: patients with
Broca’s aphasia, like normal controls, showed RTs that were
sensitive to the number of thematic options of the verbs,
whereas patients with Wernicke’s aphasia did not. Results in
similar directions come from off-line studies with aphasic
patients. Biran and Friedmann (2006) reported an individual
with severe agrammatic aphasia who had intact knowledge of
subcategorization and thematic frames in tests of production
and grammaticality judgment and another individual who had
conduction aphasia, who had aPASia, impaired access to
predicate argument structure knowledge.
Neuroimaging studies have supported the involvement of
Wernicke’s area (left superior temporal gyrus [left STG]) in the
processing of verbal complements. These studies found an effect for the number of complements using isolated verbs
(Thompson et al. 2005) or verbs in sentential context (BenShachar et al. 2003). Subcategorization effects were revealed in
an event-related potential (ERP) study (Rösler et al. 1993), in
which negativity elicited by the subcategorization violation was
observed in the left anterior field (Broca’s and frontal electrodes). A different ERP study (Rubin et al. 1996) found that the
amplitude of P300 for verbs with 2 thematic options was higher
than for verbs with only one thematic option. Latency of P300
and RT between the 2 conditions did not differ (unlike in
Shapiro et al. 1987, 1991, 1993).
The above studies suggest that all 3 verb attributes are
engaged during sentence comprehension in certain conditions.
In the current study, we have used fMRI to identify the cortical
involvement of each of these attributes. It is important to note
that, as for the number of subcategorization and thematic options, we did not consider the specific semantic type or
syntactic category of the arguments but the number of types
or categories. Our key methodological assumption was that neural involvement would be specific to the investigated attribute if
it reflected not only in simple contrasts between experimental
conditions but also more strongly in a correlation between
levels of activation and the number of complements or options.
We refer to this correlation as ‘‘a graded pattern of activation.’’
The graded pattern of activation could be expected in language areas that were involved in semantic and syntactic processing, as well as in areas that modulated working memory. As
for thematic frames, we expected to find the graded pattern in
Wernicke’s area following Shapiro et al. (1993) and perhaps in
some other frontal regions that were known to be involved with
semantics, such as Brodmann area (BA) 47 (Gold and
Buckner 2002; McDermott et al. 2003). With respect to subcategorization, following Rösler et al. (1993), we expected the
graded pattern to show in frontal areas, including Broca’s area.
Considering the number of options a verb selects, processing
may compute a number of potential alternatives, although only
one is actually realized in the sentence. However, processing
the number of complements requires all items to remain active
online because the whole set of complements is realized each
time the verb appears. Waters et al. (1995) found that the
retention of thematic roles in memory when processing
sentences with movement interfered with the performance of
a secondary task, implying that the holding of thematic information loaded on to working memory. Therefore, because
the number of complements determined the number of the-
matic roles that the verb assigned to its complements, we
assumed that the number of complements would correlate with
working memory load and expected the graded activation
pattern to show in areas that were related to working memory.
In addition, we addressed 2 linguistic issues. The differentiation between subcategorization and thematic frames, together
with the distinction between nonfinite and finite clauses, could
allow us not only to locate brain regions that were involved in
processing the above attributes but also, perhaps, to offer
neurally based arguments that are relevant to some of the linguistic controversies concerning these attributes.
The above considerations led us to design 2 experiments. The
first experiment focused on the effect of the number of complements on brain activation, and the second focused on the
number of subcategorization and thematic options. In both experiments, 3-point scales were used to identify brain regions
that show graded activations.
Experiment 1
The focus of this experiment was to identify brain regions that were
sensitive to the number of complements. We did so by comparing verbs
that took 0, 1, or 2 complements, presented in sentential context. This
created a 3-point scale of the examined attribute. The number of subcategorization and thematic options was kept constant across conditions.
Method
Subjects
The subjects were 12 healthy volunteers (8 females), aged 23--33 years
(mean age 26.2 years). Subjects had normal hearing, no language
impairment, and no psychiatric or neurological history. All subjects
were native speakers of Hebrew, and Hebrew was their sole mother
tongue. They were all right handed (as assessed by Edinburgh handedness inventory, Oldfield 1971). Two other subjects were excluded
from the experiment: one because of an abrupt head movement and
another due to lack of proper responses. Written informed consent was
obtained from all subjects. The Tel Aviv Sourasky Medical Center and Tel
Aviv University ethics committees approved the experimental protocol.
Materials and Procedure
Three psycholinguists judged Hebrew verbs for their number of complements. Verbs that were not agreed by all judges were tested in
sentences with different number of complements. They were given to
17 native speakers for grammaticality judgment. If judged ungrammatical, those 17 judges were asked to offer an alternative correct sentence
based on the same verb.
This procedure created a list of 28 verbs, 7 for each category. The
categories (Table 1) were unergatives (verbs with no complements),
obligatory transitives (verbs with one complement), and obligatory
ditransitives (verbs with 2 complements). In addition, we included
optional ditransitives (verbs with one obligatory complement and one
optional complement, such as send). This category is not discussed in
the present paper. All 28 verbs had only one option of subcategorization
and thematic frames. The only attribute that differed among the verbs
was that of the number of obligatory complements.
Table 1
Example sentences from each condition in Experiment 1
Condition
Example for a sentence
Unergatives
No complements
Transitives
1 complement
Ditransitives
2 complements
Dana shara [etmol] [b-a-miklaxat]
Dana sang [yesterday]adjunct [in the shower]adjunct
Ron shavar [et ha-kos] [b-a-xatuna]
Ron broke [the glass]theme [in the wedding]adjunct
Keren sama [et ha-xulcot] [b-a-aron]
Keren put [the shirts]theme [in the closet]goal
Cerebral Cortex August 2007, V 17 N 8 1961
Each verb appeared in 4 different sentences (example 17), forming
a total of 112 sentences for the entire experiment. The verb was
preceded by a person name (representing the subject) and followed by
2 constituents (Table 1): the sentences with the unergative verbs
included 2 adjuncts or modifiers (‘‘Adjunct’’ stands for any phrase that is
not required by the verb, and can be attached to every verb, usually in
the form of time and place descriptions, e.g., ‘‘at the wedding’’ in Table 1.
‘‘Modifier’’ stands for any phrase added to a complement, such as
‘‘yellow’’ in the sentence ‘‘John touched the yellow shirt’’); the sentences
with the transitive verbs included a complement and an adjunct or
a modifier; and the sentences with the ditransitive verbs included 2
complements after the verb.
17. a.
b.
c.
d.
Dan lost the keys in the office
Ann lost a sock in the laundry
Sean lost the way to the hotel
Liz lost the recipe for the cake
Complements were either NP or PP, and their thematic role was either
theme (for transitives and one of the complements of the ditransitives) or
goal (for the second complement of the ditransitives). In all the
sentences, the verb was inflected for third person singular and past
tense. Half of the sentences of each condition were with a feminine
subject (and hence with a verb inflected for the feminine) and half
masculine. The verbs were controlled for frequency as determined by
the Hebrew Word Frequency Database (http://word-freq.mscc.huji.
ac.il/index.html). Analysis of variance (ANOVA) detected no significant
differences among the various conditions significant (F2,20 = 1.08, P = 0.36
for the combined frequency of masculine and feminine forms) despite
considerable variation (mean frequency = 27, 19, and 6.2 for unergatives,
transitives, and ditransitives, respectively). The number of syllables in
each sentence was controlled for length and ranged between 9 and 13
syllables (mean = 11.5; mean duration [ms] = 1797.4, 1775.5, and 1753.8
for unergatives, transitives, and ditransitives, respectively). There were
no differences among the various conditions (F2,27 = 0.49, P = 0.61).
Sentences were divided into 28 blocks, each comprising of 4 sentences with verbs of the same condition. Each condition repeated 7
times. Each verb appeared in a block only once. In each block, half of the
sentences included a subject and a verb in a masculine form and half in
a feminine form. The blocks and the sentences in each block were
presented in a pseudorandom order (which was determined using
Matlab) with no more than 2 subsequent blocks of the same condition.
The presentation of each block lasted 15 s. Sentences were separated by
silence periods of 1300 ms. A tone was heard at the end of each block to
signal a 6 or 9 s of silence. During silence, subjects were instructed to
concentrate on the noises of the magnetic resonance imaging (MRI)
scanner. Stimuli were delivered to the subjects via MRI compatible
headphones using Presentation software (http://www.neurobs.com).
Throughout the experiment, a semantic task was used to ensure that
the subjects attended to the sentences and processed them fully. In this
task, the subjects were asked to decide whether the event described in
the sentence is more likely to happen at home or not. Responses were
given during the intrablock silences (responses were not allowed before
the end of a sentence or after the beginning of the following sentence).
Subjects were requested to press a ‘‘yes’’ or a ‘‘no’’ button with their left
hand fingers after the end of the sentence. For example, for a sentence
like ‘‘Dan slept in the yellow tent,’’ subjects had to respond with no,
whereas for a sentence like ‘‘Jane lent a towel to her flatmate,’’ subjects had
to respond with yes. The number of no response in each block differed
(ranging 1--3), but there were equal number of yes and no responses in
the entire experiment. All responses and reaction times were recorded.
Each subject completed 4 blocks of practice outside and inside the
MRI scanner. Practice blocks included sentences that were similar to
those used in the experiment but with verbs that were not included in
the experiments.
Each experiment lasted 11.3 min, and the entire imaging session
(including practices, anatomical, and functional scans) lasted approximately an hour and a half.
Data Acquisition
MRI scans were conducted in a whole-body 1.5-T, Signa Horizon, LX,
8.25 General Electric scanner, located at the Whol Institute for
1962 The fMRI of Verb Processing
d
Shetreet et al.
Advanced Imaging in the Tel-Aviv Sourasky Medical Center. Anatomical
images for each subject were acquired using a 3-dimensional spoiled
gradient echo sequence with high resolution to allow volume statistical
analyses in single subjects. The whole brain was covered by 70--80 slices,
2 mm thick (no gap). Functional MRI protocols included T2*-weighted
images in runs of 220 volumes. We selected 27 sagittal slices (based
on a midsagittal slice), 4 mm thick (no gap), covering whole of the
cerebrum and most of the cerebellum. We used field of view of 24 cm
and matrix size of 128 3 128, time repetition = 3000 ms, time echo = 55,
and flip angle = 90°.
Data Analysis
Image analysis was performed using SPM2 (Wellcome Department of
Cognitive Neurology, http://www.fil.ion.ucl.ac.uk/spm/). Functional
images from each subject were motion corrected, normalized to a
standard echo planar imaging template, and spatially smoothed using
a Gaussian filter (6-mm kernel).
The analysis assumed the general linear model (Friston et al. 1995) as
implemented in SPM2. The blood oxygen level--dependent response was
modeled with the canonical hemodynamic response function. The time
series in each voxel were high-pass filtered (618 s) to remove lowfrequency noise. Head motion parameters were added as regressors
(Friston et al. 1995; Worsley and Friston 1995), as were the time periods
allocated for responses.
A combined regressor for all conditions was used for modeling all
the experimental conditions together for each subject individually
(first-level analysis) in a parametric design. This allowed us to model the
linear hemodynamic responses (Büchel et al. 1998). For the combined
regressor, a covariate was used to assign weights to the different conditions (1, 2, and 3 for unergatives, transitives, and obligatory ditransitives, respectively). Beta images of the combined regressors in each
subject were used for a random-effect group analysis. In this secondlevel analysis, the parametric design was examined using t-test with P <
0.002 and a minimum cluster size of 50 voxels.
In addition, a second design was defined, in which each condition was
modeled separately. This allowed us to jointly contrast the conditions by
using the specific beta weights of each condition. In this design, we used
ANOVA with P < 0.0005 and a minimum cluster size of 50 voxels.
Results
Assessment of the parametric design in Experiment 1 did not yield
graded activation in any of the canonical language areas (i.e., Broca’s area
and Wernicke’s area; e.g., Bavelier et al. 1997; Karni et al. 2005). Graded
activations were only found in right anterior cingulate (BA 32) and in
right precuneus (BA 7), very close to the midline (Table 2 and Fig. 2). No
other activations were revealed. The same clusters were identified when
contrasting verbs with 2 complements and verbs with no complements.
Reaction Time
Reaction times were analyzed using ANOVA, and, as expected, no
differences were found between the different conditions (P > 0.19).
Experiment 2
The second experiment manipulated the number of options of both
subcategorization and thematic frames. Here, we compared sentences
that included verbs of 1, 2, and 3 options where nonfinite clauses stood
as a different semantic and syntactic category than finite clauses. We did
not consider the specific semantic type or syntactic category of the
arguments but only the number of subcategorization or thematic options. This created a 3-point scale for both subcategorization and
thematic options. Whereas in Experiment 1, we kept the number of
subcategorization and thematic options constant and manipulated the
Table 2
Coordinates of graded activation clusters found in Experiment 1
Region
x
y
z
Cluster size
t-Max
Precuneus
Anterior cingulate
16
10
64
33
33
30
122
93
6.29
6.21
Figure 2. Graded activation clusters found in Experiment 1: right precuneus and right anterior cingulate.
number of complements, in Experiment 2, we kept the number of complements constant and manipulated the number of options: the verb
categories in this experiment differed by the number of subcategorization and thematic options. It is important to note here that the verbs
used in this experiment selected one complement only. This allowed
us to avoid some of the nonlinguistic effects that were evident in
Experiment 1.
In addition, we included verbs with 2 subcategorization options but
only one thematic option (we shall refer to them as ‘‘subcat--thematic
mismatch’’ verbs) in an attempt to differentiate between the effect of
the number of subcategorization options and that of the number of
thematic options. Another distinction, between nonfinite and finite
clauses, was addressed. By making these distinctions, we hoped to be
able to offer a neurally based clarification to the linguistic debate regarding subcategorization and thematic options.
Method
Subjects
Fourteen healthy volunteers (9 females), aged 23--33 years (mean age
26.2 years), participated in the study. Subjects had normal hearing, no
language impairment, and no psychiatric or neurological history. All
subjects were native speakers of Hebrew, and Hebrew was their sole
mother tongue. They were all right handed (as assessed by Edinburgh
handedness inventory, Oldfield 1971).
Materials and Procedure
The same team of psycholinguists as in Experiment 1 judged verbs
according to their number of subcategorization options and thematic
options. The procedure for nonagreed verbs was the same as in
Experiment 1.
Twenty-eight verbs were used, 7 in each category. There were verbs
with 1 option (obligatory transitives), 2 options, and 3 options of both
subcategorization and thematic frames (Table 3). A fourth category,
called subcat--thematic mismatch verbs, consisted of special transitives
that had 2 subcategorization options but only one thematic option.
As in Experiment 1, each verb appeared in 4 different sentences
(example 17 above) and was preceded by a name, but it was always
followed by a complement and an adjunct or a modifier (Table 3). Here
too, the complements that appeared in the presented sentences were
always either NP or PP (both bearing the theta role of theme).
Table 3
Example sentences from each condition in Experiment 2
Condition
Example for a sentence
1-option verbs
Dan hidlik [esh] [b-a-kirayim]
Dan lit [fire]theme [in-the-stove]adjunct
Gal syima [et ha-sefer] [b-a-xacot]
Gal finished [the-book]theme [in-the-midnight]adjunct
Sara nista [et ha-kova] [b-a-xanut]
Sara tried [the-hat]theme [in-the-shop]adjunct
Avi lagam [mic] [b-a-mita]
Avi sipped [juice]theme [in-the-bed]adjunct
2-option verbs
3-option verbs
Subcat--thematic mismatch verbs
In the 2-option verbs, the complement options were NP/PP theme
and either CP/proposition or IP/event. In the 3-option verbs, the
complement options were NP/PP theme, CP/proposition, and IP/event.
No other options, such as interrogatives or exclamations (Grimshaw
1979), were possible as complements. All other parameters (inflection,
length, etc.) were the same as in Experiment 1. Frequency was tested as
in Experiment 1, and the 4 verb categories did not differ on frequency
(F3,27 = 0.35, P = 0.79 for the combined frequency of masculine and
feminine forms) despite considerable variation (mean frequency = 10,
30, 24, and 16 for 1-, 2-, 3-option, and mismatch verbs, respectively). In
addition, no differences of duration were found in the various conditions (F3,27 = 1.04, P = 0.37; mean duration = 1810.4, 1784.3, 1779, and
1867.3 for 1-, 2-, 3-option, and mismatch verbs, respectively).
The experimental design, the procedure, and the semantic task were
identical to those of Experiment 1.
Data Acquisition
Parameters of data acquisition were the same as in Experiment 1.
Data Analysis
Preprocessing was the same as in Experiment 1. Similar to Experiment 1,
a combined regressor was used with a covariate assigning weights to 3
of the 4 conditions (1, 2, and 3 for 1-, 2-, and 3-option verb, respectively).
We will refer to this as the original parametric design. Here too, beta
images were used for a random-effect group analysis and t-test, with P <
0.002 and a minimum cluster size of 50 voxels. Two more parametric
analyses were used to test the differences between subcategorization
and thematic options (see Results).
Cerebral Cortex August 2007, V 17 N 8 1963
In addition, areas that showed graded activation in the original parametric design were defined as regions of interest (ROIs) using MarsBar
(http://marsbar.sourceforge.net). The ROIs were entered to a second
design in which each of the 4 conditions (1, 2, and 3 options verbs and
the subcat--thematic mismatch group) was modeled separately. This
allowed us to jointly contrast subcategorization with thematic frames
and nonfinite clauses with finite clauses by using the specific beta
weights of each condition. In this design, we used ANOVA with P <
0.0005 and a minimum cluster size of 50 voxels.
Results
The number of subcategorization and thematic options yielded significant activations in language areas, unlike the number of complements.
The assessment of the parametric design in Experiment 2 revealed
activations in left STG (BA 22) and in 2 clusters in left inferior frontal
gyrus (left IFG): one in BA 9 and the other in BA 47 (Table 4 and Fig. 3).
These areas were used as functional ROIs in some of the following
analyses.
Subcategorization Options versus Thematic Options
We first used a separate-conditions design in which each condition was
defined separately and the ROIs were defined by the parametric design.
Using MarsBar, we extracted the average beta weights for each of the 3
clusters (Fig. 4), so there were 3 3 4 beta weights for each subject
(clusters 3 conditions). Using t-tests, we compared the average weights
of the mismatch verbs to both the 1-option and the 2-option groups. In
all clusters, the difference between the beta weights of the subcat-thematic mismatch verbs and the 1-option verbs was significant (t13 =
1.85, P = 0.001; t13 = 9.66, P = 0.003; t13 = 2.08, P = 0.002 in left STG, left
IFG BA 9, and left IFG BA 47, respectively), whereas the difference
between the beta weights of the subcat--thematic mismatch verbs and
Table 4
Coordinates of graded activation clusters found in Experiment 2
Region
Original parametric design
STG
IFG—BA 47
IFG—BA 9
Design A1
Precuneus
Design A2
STG
IFG—BA 47
IFG—BA 9
Design B1
Precuneus
Design B2
STG
IFG—BA 47
IFG—BA 9
x
y
z
Cluster size
t-Max
59
36
48
52
19
11
12
8
29
52
60
108
4.76
5.83
5.62
8
58
49
87
6.19
59
36
48
52
19
11
12
8
29
54
58
108
4.66
5.71
5.57
8
58
49
57
4.41
59
36
48
52
19
11
12
8
29
53
69
106
4.71
6.01
5.62
the 2-option verbs was not (P = 0.6; P = 0.1; P = 0.08 in left STG, left IFG
BA 9, and left IFG BA 47, respectively).
Next we performed another analysis, Analysis A, that included 2 parametric calculations in which the subcat--thematic mismatch verbs
replaced either the 1-option verbs (design A1) or the 2-option verbs
(design A2). The weights of the replaced condition were assigned to the
subcat--thematic mismatch verbs (Table 5). This was done in order to
examine the similarity between the mismatch verbs (which included 2
subcategorization options but only one thematic option) and its 2 comparison groups: the 1-option verbs and the 2-option verbs. We then
compared the results of the original parametric design to each of these 2
additional designs. In our reasoning, similarity of the original results to
the results of design A1 would imply that it was thematic frames that
played the critical role in processing, whereas similarity to the results of
design A2 would imply that subcategorization played the critical role.
The results showed that the original parametric design resembled design A2: both showed graded activations in the same 3 clusters (left STG,
left IFG at BA 47, and left IFG at BA 9; Table 4). By contrast, in design A1,
only one cluster was identified in the left precuneus.
Finally, we approached the same issue from a slightly different angle.
In this analysis, all 4 conditions were included (Analysis B in Table 5).
Again we used 2 designs: in the first (design B1), we assigned the weight
1 to the subcat--thematic mismatch verbs (identical to the weight given
to the 1-option verbs), whereas in the second (design B2), we assigned
the weight 2 to the mismatch verbs (identical to the weights given to
the 2-option verbs). Again, the results of the 2 designs were compared
with the results of the original parametric design. Similarity to design B1
would imply that it was thematic frames that played the critical processing role, whereas similarity to the design B2 would imply that
subcategorization played the critical role. Like with the earlier analysis
(above), the results of design B1 identified only the precuneus cluster
for graded activation, whereas design B2 was similar to the original
parametric design in identifying graded activations in left STG, BA 47 and
BA 9 (Table 4).
Nonfinite versus Finite Clauses
The distinction between nonfinite and finite clauses was part of the
parametric design because the verbs in the 3-option category selected
both finite and nonfinite clauses. The graded activations found in this
design indicated that 2-option verbs (which selected either IP/event or
CP/proposition) and 3-option verbs (selecting both IP/event and CP/
proposition) were indeed different.
To address this question more directly, we used the separate-conditions design and the beta weights that were extracted from the ROIs
of the original parametric design. Three-option verbs included both
finite and nonfinite clauses, whereas 2-option verbs included only one of
the clause types. The pattern of activation of the 2 types of clauses
would be similar if their lexical representations were identical. However, an activational difference between them would imply that their
lexical representations were distinct. Therefore, we compared the
average beta weights of the 3-option verbs with those of the 2-option
verbs (using t-tests). In BA 47, the beta weights of the 3-option verbs
were significantly greater than the beta weights of the 2-option verbs
(t13 = 75, P = 0.038). In the 2 remaining clusters (left STG and BA 9), the
beta weights of the 2 categories did not significantly differ.
Figure 3. Graded activation clusters found in Experiment 2: left STG and left IFG (both BA 47 and BA 9).
1964 The fMRI of Verb Processing
d
Shetreet et al.
Figure 4. Beta weights extracted from graded activation clusters in left IFG BA 9 (left) and in left STG (right).
Table 5
The weights assigned to the different conditions in the original parametric design, analysis
A and analysis B
Weights
Original parametric
design
Analysis A
Analysis B
A1 design
A2 design
B1 design
B2 design
1
1 option
Mismatch
1 option
1 option
2
2 options
2 options
Mismatch
1 option
Mismatch
2 options
3
3 options
3 options
3 options
3 options
2 options
Mismatch
3 options
Reaction Times
Reaction times were analyzed using ANOVA. A Scheffé post hoc test
found that the reaction times for the mismatch condition were significantly longer than the reaction times of the other conditions (P =
0.02). Note that this difference went contrary to both our expectations
and the imaging results, so it could not explain the identified activations.
Discussion
The current study explored the cortical locations and the
patterns of activation of 3 semantic--syntactic attributes of verbs
that were believed to participate in online sentence processing:
the number of complements, the number of subcategorization
options, and the number of thematic options. The main finding
was that whereas increasing the number of subcategorization
and thematic options yielded graded activation in language
areas, increasing the number of complements did not. Our
results also bear on open questions in linguistics with respect to
the representation of subcategorization and the separate
representation of finite and nonfinite clause complements.
Experiment 1 investigated the activation patterns that were
associated with the number of complements. We identified 2
clusters that showed graded activation according to a 3-point
gradient of the tested attribute. These were located in the right
precuneus and in the right anterior cingulate but not in
language areas. Experiment 2 investigated the patterns of activation associated with subcategorization and thematic options.
Three clusters of graded activation were identified, 1 in left STG
and 2 in left IFG: in BA 47 and in BA 9. Note that, by contrast to
Experiment 1, all 3 clusters were in the left hemisphere. Below,
we discuss possible implications of these results and their
relevance for the understanding of sentence processing.
The areas found in Experiment 1, right precuneus and right
anterior cingulate, had not traditionally been considered language areas. The precuneus seemed to be involved in visual
imagery (Fletcher et al. 1996), suggesting the possibility that the
graded activation found in this area concerned the imagistic
retention of the entities included in the sentences. The number
of entities that participated in an event described by a verb was
represented by the number of complements. Because most
entities included in the sentences used in Experiment 1 were
imageable, the number of imageable entities in these sentences
correlated with the number of complements. Hence, most of
the sentences that included unergative verbs consisted of only
one imageable entity; most of the sentences that included transitive verbs consisted of 2 imageable entities; and so on. Thus, it
may well be that, while processing these sentences, the related
entities were held in imagistic form and required graded imagistic resources. This could give rise to a correlation between
levels of neuronal activation and the number of complements in
the right precuneus.
In addition, we expected the number of complements to load
on to working memory resources due to the need to retain a
greater amount of information as the number of complements
increased. Indeed, activations in the anterior cingulate were observed in relation to working memory load (Awh et al. 1996;
Schumacher et al. 1996; Smith et al. 1996; Braver et al. 1997).
Remarkably, the part of the anterior cingulate that was activated
in the present study (BA 32) was said to play a role in speech
due to its connections with the auditory cortex (Paus et al.
1993). However, the absence of similar activations in frontal
areas that were found to sustain working memory might restrain this argument and might indicate the involvement of
other functions associated with this region, such as response
monitoring (Paus et al. 1993; Carter et al. 1998; Barch et al.
2000) or, more likely, decision making (Elliott and Dolan 1998;
Paulus et al. 2001, 2002). The available data could not resolve
this matter.
By contrast to Experiment 1, the regions found in Experiment
2 comprised the classical language regions. Left STG forms a
part of Wernicke’s area and was expected to show activation
when manipulating the number of a verb’s options. This
prediction arose directly from the study of Shapiro et al.
(1993), which investigated the performance on a lexical decision task of Broca’s aphasic patients, Wernicke’s aphasic
patients, and normal controls. Using RTs as a measure of processing load, Shapiro et al. (1993) found that normal controls
and patients with Broca’s aphasia were sensitive to the number
of thematic options, whereas patients with Wernicke’s aphasia
were not. They inferred that the number of complementation
options of verbs was processed in Wernicke’s area.
Wernicke’s area was also suggested, based on language
deficits of patients with Wernicke’s aphasia, to be involved in
Cerebral Cortex August 2007, V 17 N 8 1965
access to the semantic properties of words (Milberg and
Blumstein 1981; Blumstein et al. 1982; Mesulam 1998). Because
more pieces of information needed to be handled, retrieval of
verbs with greater number of options was assumed to generate
a greater processing load on both word and sentence levels.
This could explain the graded activation found in left STG.
Additionally, Wernicke’s area had also been associated with the
semantic processing of sentences (Fletcher et al. 1995; Sakai
et al. 2001; Vigneau et al. 2006). Indeed, activations in this area
were found while contrasting well-formed sentences with
semantically ill-formed sentences (Luke et al. 2002), as well as
while contrasting sentences in different languages (Schlosser
et al. 1998). In a meta-analysis of similar studies, Vigneau et al.
(2006) suggested that the left STG was involved with semantic
integration of complex linguistic information. Semantic integration was expected to play a role in the current study because
the number of options of a verb could indicate that a larger
amount of semantic properties needed to be integrated in order
to generate a stable sentence meaning. However, it is not selfevident that the graded activation in left STG originated in
thematic options, usually considered a semantic attribute. Based
on the experimental design, this activation could also arise from
the number of subcategorization options. It might just be that
left STG is where thematic frames and/or subcategorization are
stored.
The 2 additional clusters identified in Experiment 2 are
located in the left IFG, one in BA 47 and one in BA 9. Whereas BA
9 is mostly associated with working memory (Cabeza and
Nyberg 2000), BA 47 has shown repeated involvement in semantic processing across a range of tasks (Poldrack et al. 1999;
Gold and Buckner 2002; McDermott et al. 2003; for review, see
Bookheimer 2002). Both areas are believed to be involved with
semantic memory (Demb et al. 1995; Gabrieli et al. 1996; Lee
et al. 2002; Shivde and Thompson-Schill 2004; for review, see
Bookheimer 2002).
The nature of the semantic processes subserved in these
areas of the left IFG had been investigated by Thompson-Schill
et al. (1997). They argued that the left IFG activity did not
concern semantic retrieval per se but rather the selection of
competing alternatives from semantic memory (also see
Thompson-Schill et al 1999, 2006; Fletcher and Henson 2001).
Indeed, BA 9 was one of the areas within the left IFG that lighted
up in a high-selection versus low-selection comparison performed by Thompson-Schill et al. (1997). By inference, regions
that had a role in the selection of semantic information were
likely to load according to the number of options a verb could
select.
However, other studies suggested that the selection mechanism in the left IFG also applied to functions other than semantic processing (Rowe et al 2000; Zhang et al. 2004). One
such approach was suggested by Grodzinsky et al. (1993)
following studies that investigated individuals with agrammatic
Broca’s aphasia, who had damage in the left IFG (Swinney and
Prather 1989; Grodzinsky et al. 1993). Grodzinsky et al. (ibid)
claimed that the patients were impaired in the ability to jointly
hold several options in order to perform a computation that
would lead to the selection of one of them. They referred to this
ability as a part of working memory. This implied that the left
IFG, or specific areas within it, was involved in holding several
activated options to allow selection.
Although Grodzinsky et al. (1993) did not discuss the exact
areas within the left IFG that were involved in option selection,
1966 The fMRI of Verb Processing
d
Shetreet et al.
later studies indicated that they were similar to those activated
in our study. For example, Petrides’ model of frontal function
(Petrides 1995) proposed that the ventrolateral regions (which
included BA 47) were responsible for selecting and deciding the
relevance of information for a given task, and as mentioned
before, BA 9 was identified by Thompson-Schill et al. (1997).
Interestingly, the Broca’s patients in Shapiro and Levine
(1990) and Shapiro et al. (1993) showed normal sensitivity to
thematic options (referred in their study as argument structure). This might be explained by splitting the processing of
complement options between Broca’s and Wernicke’s areas.
Access to lexical information about the number of complement
options could be performed in the left STG that was spared in
the Broca’s patients. The spared left STG could suffice to create
RT sensitivity to the number of complementation options, even
if the subsequent processes of holding the options activated and
selecting among them were impaired.
In addition to identifying the cortical areas that were involved
in processing the investigated verb attributes, we also considered
2 open questions in linguistics. Both concerned the distinction
between opposing linguistic approaches, one regarding nonfinite clauses and one regarding subcategorization. The pattern
of activation revealed in Experiment 2 allowed us to present
neural-based arguments appertaining to the linguistic debates.
Regarding the finite and nonfinite clauses, we speculated that
the lexical representation of finite clausal complements of
a verb would differ from that of nonfinite clausal complements,
based on their syntactic and semantic differences (Pesetsky
1991). The results of Experiment 2 supported this speculation.
We compared 3-option verbs that included both clause types
with 2-option verbs that included only one type, expecting that
distinct lexical representations would manifest in a difference
of activational patterns. Although the difference between 2- and
3-option verbs was significant only in BA 47, in all 3 regions
a graded activation was revealed. These graded activations indicated that the representation of the 2 clausal complement types
was different due to the inclusion of the finite--nonfinite distinction in the parametric design. Our results did not indicate
whether this distinction was based on subcategorization (CP vs.
IP) or on thematic frames (proposition vs. event) as the stimuli
could be described as differing in both attributes.
In order to differentiate between subcategorization and thematic frames, we investigated another verb category that
formed a subset of transitive verbs and subcat--thematic mismatch verbs. These mismatch verbs selected 2 subcategorization options and 1 thematic option and showed activations
similar to another group of verbs that selected 2 options of both
subcategorization and thematic frames. Because similar number
of subcategorization options showed similar activations, we
inferred that subcategorization played a critical role in the
online processing of sentences. Thus, our findings support the
idea that the representation of subcategorization is separate
from the representation of thematic frames (e.g., Grimshaw
1979) and that it affects online sentence processing. By inference, our results ran contrary to claims that the representation of complementation options could be reduced to the
representation of thematic roles, as had been suggested by
Pesetsky (1982, 1991) and Chomsky (1986, 1995).
Additional support for the role of subcategorization in
processing load came from linguistic theories that allowed
only a limited set of basic thematic roles (e.g., Haegeman 1995).
Whereas we assumed that sentential complements carried
a thematic role of proposition or event, these theories rejected
such complex semantic types, and the thematic role they
assigned to the clausal complement was theme (example 21).
According to such analysis, our 2-option and 3-option verbs
would be viewed as having only one thematic option (example
21) but differential subcategorization options (example 22).
Therefore, if one held that all 3 options carried the same
thematic role, the graded activations found in Experiment 2
between the 3 types of verbs would also be ascribed to
subcategorization differences.
21. a.
b.
c.
22. a.
b.
c.
<John >agent demanded <the book >theme
<John >agent demanded <to read the book >theme
<John >agent demanded < that Anna would smile >theme
John demanded [the book]PP
John demanded [to read the book]IP
John demanded [that Anna would smile]CP
Note that the identification of subcategorization role in verb
access does not exclude online access to information about
thematic frames. The latter possibility is supported by the fact
that the 3 regions identified in Experiment 2 are known to have
a role in semantic processing. Indeed, as regards the IFG
regions, semantic processing may be their prime connection
with language processing (Fiez 1997; Poldrack et al. 1999;
Burton 2001). In linguistic terms, thematic frames are related to
semantics, whereas subcategorization is considered a formal
syntactic attribute. Thus, it seems that our results could indicate
either the additional lexical-syntactic nature of these areas or
the involvement of thematic frames in the access to the verb’s
lexical information (Botwinik-Rotem [forthcoming] argued that
the 2 options of complements that the verb taste selects have
different thematic roles. She suggests that the NP complement
[example 11a] is assigned with the thematic role theme and the
PP complement [example 11b] is assigned with the thematic
role ‘‘subject matter.’’ The current results may be taken as
support for her view).
Other findings could also bear upon the manner in which
thematic frames influenced processing load (Shapiro et al. 1987,
1991, 1993; Rubin et al. 1996). Of special importance was the
study by Shapiro et al. (1987), which also compared subcategorization with thematic frames and concluded that the
latter was more crucially tied up with processing load. In that
study, RTs for lexical decision were found to increase as
a function of the number of thematic options but not as
a function of subcategorization options. The discrepancy
between the present findings and those of Shapiro et al.
(1987) could arise in the differences between the stimulus
verbs: Shapiro et al. (1987) used alternating datives and nonalternating datives.
Whereas nonalternating datives can take only NP PP complements, alternating datives can take both NP PP (gave the flowers
to Mary) and NP NP complements (gave Mary the flowers). It
might be that no difference was detected between these 2 types
of datives not because subcategorization does not affect
processing but rather because alternating and nonalternating
datives actually do not differ with respect to their subcategorization. According to some theories, the 2 complementation
options of the alternating datives actually form only one
subcategorization option, NP PP. The NP NP option is taken to
be derived from NP PP through movement (Larson 1988;
Belletti and Shlonsky 1995; Ben-Shachar et al. 2004). Such an
account explains why alternating and nonalternating datives did
not yield any difference in RTs in Shapiro et al. (1987), as they
both have only one subcategorization option.
The above considerations lead us to suggest that the processing of the number of a verb’s options is essentially linguistic
because it engages specifically linguistic regions. By contrast,
the number of complements seems to load onto general
cognitive resources and is therefore processed in networks
that are not specifically linguistic. Note that our findings do not
imply that information about the number of complements is not
being used during sentence comprehension but only that it
does not specifically load onto conventional linguistic areas and,
by inference, not onto linguistic processing.
In our review of the literature, we found inconsistent
accounts for the involvement of the different verb attributes
in comprehension. Our results were in line with other evidence
suggesting that lexical information about the verbal complements, such as subcategorization, was accessed during sentence
processing (Holmes 1987; Tanenhaus et al. 1989; Boland et al.
1990; Boland 1993; Trueswell et al. 1993). In particular, our
results supported the findings of Shapiro et al. (1987), as against
those of Ahrens and Swinney (1995). Ben-Shachar et al. (2003)
and Thompson et al. (2005) found, using fMRI, activations that
corresponded to a manipulation in the number of complements.
However, control for the number of subcategorization and
thematic options was not reported in these studies. It could well
be that the observed results in both of the above studies were
mediated by the number of complementation options.
Let us now return to the syntactic tree representation of complements and options (Fig. 1). As mentioned, verbal complements are represented by branches within the VP, whereas
complementation options are represented by different trees.
This metaphor of branches and trees may be used to explain the
results of our experiments. We found that the number of
options, but not the number of complements, gives differential
activations in language areas. By inference, only the number of
complementation options can be regarded as a specifically
linguistic property. It thus seems that when a verb is accessed,
the number of branches that should be constructed within the
same tree does not tax linguistic resources, whereas the
number of VP trees does. We suggest that ‘‘neural linguisticity,’’
to use a slight neologism, concerns syntactic trees rather than
syntactic branches.
To conclude, this study indicates that the number of complementation options (the number of trees) that are represented in the lexical entry of the verb are accessed in the left
STG, and further held for selection in the left IFG, in BA 9 and BA
47. We further suggest the neurolinguistic validity of the
distinction between 2 types of clauses and support the role of
subcategorization in sentence processing. Studies such as the
present emphasize the importance of interfacing linguistics
with neurolinguistics. These 2 disciplines should enhance
and enrich each other, as well as impose constrains on one
another. In this effort, neurolinguistic results can contribute to
linguistic theory and linguistic theory can inform neurolinguistics, to the benefit of our understanding of both language and
the brain.
Notes
The research was supported by the Joint German-Israeli Research
Program grant GR01791 (NF) and by Adams Super Center for Brain
Cerebral Cortex August 2007, V 17 N 8 1967
Studies (NF). Many thanks to Lew Shapiro for his comments on the
manuscript. Conflict of Interest: None declared.
Address correspondence to email: [email protected].
References
Ahrens K, Swinney D. 1995. Participant roles and the processing of verbs
during sentence comprehension. J Psycholinguist Res. 24:533--547.
Awh E, Jonides J, Smith EE, Schumacher EH, Koeppe RA, Katz S. 1996.
Dissociation of storage and rehearsal in verbal working memory:
evidence from positron emission tomography. Physiol Sci. 7:25--31.
Barch DM, Braver TS, Sabb FW, Noll DC. 2000. Anterior cingulate and the
monitoring of response conflict: evidence from an fMRI study of
overt verb generation. J Cogn Neurosci. 12:298--309.
Bavelier D, Corina D, Jezzard P, Padmanabhan S, Clark VP, Karni A,
Prinster A, Braun A, Lalwani A, Rauschecker JP, et al. 1997. Sentence
reading: a functional MRI study at 4 tesla. J Cogn Neurosci. 9:664--686.
Belletti A, Shlonsky U. 1995. The order of verbal complements:
a comparative study. Nat Lang Linguist Theor. 13:489--526.
Ben-Shachar M, Hendler T, Kahn I, Ben-Bashat D, Grodzinsky Y. 2003.
The neural reality of syntactic transformations—evidence from fMRI.
Psychol Sci. 14:433--440.
Ben-Shachar M, Palti D, Grodzinsky Y. 2004. Neural correlates of
syntactic movement: converging evidence from two fMRI experiments. Neuroimage. 21:1320--1336.
Biran M, Friedmann N. 2006. Selective impairments of syntactic information in the lexicon. A Paper Presented at the 7th Science of
Aphasia Meeting, Neurocognition of Language: Innovative Aspects.
Sardegna (Italy).
Blumstein SE, Milberg W, Shrier R. 1982. Semantic processing in aphasia:
evidence from an auditory lexical decision task. Brain Lang. 17:
301--315.
Boland J. 1993. The role of verb argument structure in sentence
processing: distinguishing between syntactic and semantic effects.
J Psycholinguist Res. 22:133--152.
Boland JE, Tanenhaus MK, Garnsey SM. 1990. Evidence for the
immediate use of verb control information in sentence processing.
J Mem Lang. 29:413--432.
Bookheimer S. 2002. Functional MRI of language: new approaches to
understanding the cortical organization of semantic processing.
Annu Rev Neurosci. 25:151--188.
Bošković Ž. 1997. The syntax of nonfinite complementation: an
economy approach. Cambridge (MA): The MIT Press.
Botwinik-Rotem I. Forthcoming. Musa akif umilot yaxas mucraxot
[indirect object and obligatory prepositions]. In: Hatav G, editor.
Theoretical Hebrew linguistics. Jerusalem (Israel): Magnes.
Braver TS, Cohen JD, Nystrom LE, Jonides J, Smith EE, Noll DC. 1997. A
parametric study of prefrontal cortex involvement in human
working memory. Neuroimage. 5:49--62.
Bresnan J. 1972. Theory of complementation in english. [Doctoral
dissertation]. Cambridge (MA): MIT.
Büchel C, Holmes AP, Rees G, Friston KJ. 1998. Characterizing stimulusresponse functions using nonlinear regressors in parametric fMRI
experiments. Neuroimage. 8:140--148.
Burton M. 2001. The role of the inferior frontal cortex in phonological
processing. Cognit Sci. 25:695--709.
Cabeza R, Nyberg L. 2000. Imaging cognition II: an empirical review of
275 PET and fMRI studies. J Cogn Neurosci. 12:1--47.
Carnie A. 2002. Syntax: a generative introduction. Oxford: Blackwell.
Carter CS, Braver TS, Barch DM, Botvinick M, Noll D, Cohen JD. 1998.
Anterior cingulate cortex, error detection, and the on line monitoring of performance. Science. 280:747--749.
Chomsky N. 1965. Aspects of the theory of syntax. Cambridge (MA): The
MIT Press.
Chomsky N. 1986. Knowledge of language: its nature, origin, and use.
New York: Praeger.
Chomsky N. 1995. The minimalist program. Cambridge (MA): The MIT
Press.
Collina S, Marangolo P, Tabossi P. 2001. The role of argument structure
in the production of nouns and verbs. Neuropsychologia. 39:
1125--1137.
1968 The fMRI of Verb Processing
d
Shetreet et al.
Demb J, Desmond J, Wagner A, Vaidya C, Glover G, Gabrieli J. 1995.
Semantic encoding and retrieval in the left inferior prefrontal cortex:
a functional MRI study of task difficulty and process specificity. J
Neurosci. 15:5870--5878.
Elliott R, Dolan RJ. 1998. Activation of different anterior cingulate foci in
association with hypothesis testing and response selection. Neuroimage. 8:17--29.
Fiez JA. 1997. Phonology, semantics, and the role of the left inferior
prefrontal cortex. Hum Brain Mapp. 5:79--83.
Fletcher PC, Happe F, Frith U, Baker SC, Dolan RJ, Frackowiak RSJ, Frith
CD. 1995. Other minds in the brain: a functional imaging study of
‘theory of mind’ in story comprehension. Cognition. 57:109--128.
Fletcher PC, Henson RNA. 2001. Frontal lobes and human memory:
insights from functional neuroimaging. Brain. 124:849--881.
Fletcher PC, Shallice T, Frith CD, Frackowiak RS, Dolan RJ. 1996. Brain
activity during memory retrieval: the influence of imagery and
semantic cueing. Brain. 119:1587--1596.
Fodor JA, Garrett MF, Bever TG. 1968. Some syntactic determinants of
sentential complexity, II: verb structure. Percept Psychophys. 3:
453--461.
Friederici AD. 1995. The time course of syntactic activation during
language processing: a model based on neuropsychological and
neurophysiological data. Brain Lang. 50:259--281.
Friederici AD, Frisch S. 2000. Verb argument structure processing: the
role of verb-specific and argument-specific information. J Mem Lang.
43:476--507.
Friston KJ, Holmes AP, Worsley KJ, Poline JB, Frith C, Frackowiak RSJ.
1995. Statistical parametric maps in functional imaging: a general
linear approach. Hum Brain Mapp. 2:189--210.
Gabrieli JDE, Desmond JE, Demb JB, Wagner AD. 1996. Functional
magnetic resonance imaging of semantic memory processes in the
frontal lobes. Psychol sci. 7:278--283.
Gold BT, Buckner RL. 2002. Common prefrontal regions co-activate
with dissociable posterior regions during controlled semantic and
phonological tasks. Neuron. 35:803--812.
Grimshaw J. 1979. Complement selection and the lexicon. Linguist Inq.
10:279--326.
Grimshaw J. 1981. Form, function, and the language acquisition device.
In: Baker CL, McCarthy JJ, editors. The logical problem of language
acquisition. Cambridge (MA): MIT Press. p. 165--182.
Grimshaw J. 2006. Words and structure. Cambridge (MA): MIT Press.
Grodzinsky Y, Wexler K, Chien YC, Marakovitz S, Solomon J. 1993. The
breakdown of binding relations. Brain Lang. 45:396--422.
Haegeman L. 1995. Introduction to government and binding theory.
London: Blackwell.
Holmes VM. 1987. Syntactic parsing: in search of the garden path. In:
Coltheart M, editor. Attention and performance XII. Hillsdale (NJ):
Erlbaum. p. 587--599.
Karni A, Morocz IA, Bitan T, Shaul S, Kushnir T, Breznitz Z. 2005. An fMRI
study of the differential effects of word presentation rates (reading
acceleration) on dyslexic readers’ brain activity patterns. J Neurolinguistics. 18:197--219.
Kayne R. 1984. Connectedness and binary branching. Foris (NL):
Dordrecht.
Larson R. 1988. On the double object construction. Linguist Inq. 19:
335--391.
Lee ACH, Robbins TW, Graham KS, Owen AM. 2002. ‘Prey or pray?’
Semantic memory investigated using a novel episodic memory task.
Neuroimage. 16:724--735.
Luke KK, Liu HL, Wai YY, Wan YL, Tan LH. 2002. The functional anatomy
of syntactic and semantic processing in language comprehension.
Hum Brain Mapp. 16:133--145.
McDermott KB, Petersen SE, Watson JM, Ojemann JG. 2003. A procedure
for identifying regions preferentially activated by attention to
semantic and phonological relations using functional magnetic
resonance imaging. Neuropsychologia. 41:293--303.
Mesulam MM. 1998. From sensation to cognition. Brain. 121:1013--1052.
Milberg W, Blumstein SE. 1981. Lexical decision and aphasia: evidence
for semantic processing. Brain Lang. 14:371--385.
Oldfield RC. 1971. The assessment and analysis of handedness: The
Edinburgh inventory. Neuropsychologia. 9:97--113.
Paulus MP, Hozack N, Frank L, Brown GG. 2002. Error rate and
outcome predictability affect neural activation in prefrontal cortex
and anterior cingulate during decision-making. Neuroimage. 15:
836--846.
Paulus MP, Hozack N, Zauscher B, McDowell JE, Frank L, Brown GG,
Braff DL. 2001. Prefrontal, parietal, and temporal cortex networks
underlie decision-making in the presence of uncertainty. Neuroimage. 13:91--100.
Paus T, Petrides M, Evans AC, Meyer E. 1993. Role of the human anterior
cingulate cortex in the control of oculomotor, manual, and speech
responses: a positron emission tomography study. J Neurophysiol.
70:453--469.
Pesetsky D. 1982. Paths and categories [doctoral dissertation]. Cambridge (MA): MIT.
Pesetsky D. 1991. Zero syntax: Vol. 2: infinitives. Material intended for
zero syntax but not included in the final publication [Internet].
Available from: http://web.mit.edu/linguistics/www/pesetsky/infins.pdf.
Petrides M. 1995. Functional organization of the human frontal cortex
for mnemonic processing: evidence from neuroimaging studies. Ann
N Y Acad Sci. 769:85--96.
Poldrack R, Wagner A, Prull M, Desmond J, Glover G, Gabrieli JD. 1999.
Functional specialization for semantic and phonological processing
in the left interior prefrontal cortex. Neuroimage. 10:15--35.
Radford A. 1997. Syntax: a minimalist introduction. Cambridge (UK):
Cambridge University Press.
Rochette A. 1988. Semantic and syntactic aspects of Romance sentential
complementation [doctoral dissertation]. Cambridge (MA): MIT.
Rösler F, Friederici AD, Pütz P, Hahne A. 1993. Event-related brain
potentials while encountering semantic and syntactic constraint
violations. J Cogn Neurosci. 5:345--362.
Rowe JB, Toni I, Josephs O, Frackowiak RS, Passingham RE. 2000. The
prefrontal cortex: response selection or maintenance within working memory? Science. 288:1656--1660.
Rubin SS, Newhoff M, Peach RK, Shapiro LP. 1996. Electrophysiological
indices of lexical processing: the effects of verb complexity and age.
J Speech Lang Hear Res. 39:1071--1080.
Sakai KL, Hashimoto R, Homae F. 2001. Sentence processing in the
cerebral cortex. Neurosci Res. 39:1--10.
Schlosser MJ, Aoyagi N, Fulbright RK, Gore JC, McCarthy G. 1998.
Functional MRI studies of auditory comprehension. Hum Brain
Mapp. 6:1--13.
Schmauder AR. 1991. Argument structure frames. J Exp Psychol learn
Mem Cogn. 17:49--65.
Schumacher EH, Lauber E, Awh E, Jonides J, Smith EE, Koeppe RA. 1996.
PET evidence for an amodal verbal working memory system.
Neuroimage. 3:79--88.
Schmauder R, Kennison S, Clifton C. 1991. On the conditions necessary
for obtaining argument structure complexity effects. J Exp Psychol
learn Mem Cogn. 17:1188--1192.
Shapiro LP, Brookins B, Gordon B, Nagel N. 1991. Verb effects during
sentence processing. J Exp Psychol learn Mem Cogn. 17:983--996.
Shapiro LP, Gordon B, Hack N, Killackey J. 1993. Verb argument
processing in complex sentences in Broca and Wernicke’s aphasia.
Brain Lang. 45:423--447.
Shapiro LP, Levine BA. 1990. Verb processing during sentence comprehension in aphasia. Brain Lang. 38:21--47.
Shapiro LP, Zurif E, Grimshaw J. 1987. Sentence processing and the
mental representation of verbs. Cognition. 27:219--246.
Shapiro LP, Zurif E, Grimshaw J. 1989. Verb processing during sentence
comprehension: contextual impenetrability. J Psycholinguist Res.
18:223--243.
Shivde GS, Thompson-Schill SL. 2004. Dissociating semantic and
phonological maintenance using fMRI. Cogn Affect Behav Neurosci.
4:10--19.
Smith EE, Jonides J, Koeppe RA. 1996. Dissociating verbal and spatial
working memory using PET. Cereb Cortex. 6:11--20.
Stowell TA. 1981. Origins of phrase structure [doctoral dissertation].
Cambridge (MA): MIT.
Swinney D, Prather P. 1989. On the comprehension of lexical ambiguity
by young children: investigations into the development of mental
modularity. In: Gorfein DS, editor. Resolving semantic ambiguity.
New York: Springer Verlag. p. 225--238.
Tanenhaus MK, Boland J, Garnsey SM, Carlson GN. 1989. Lexical structure
in parsing long-distance dependencies. J Psycholinguist Res. 18:37--50.
Thompson CK, Bonakdarpour B, Blumenfeld H, Fix SC, Parrish TB,
Gitelman DR, Mesulam MM. 2005 May. Neural correlates of verb
argument structure processing. Paper Presented at the 35th Clinical
Aphasiology Conference, Sanibel Island (FL).
Thompson CK, Lange KL, Schneider SL, Shapiro LP. 1997. Agrammatic
and non-brain-damaged subjects’ verb and verb argument structure
production. Aphasiology. 11:473--490.
Thompson-Schill SL, D’Esposito M, Aguirre JK, Farah MJ. 1997. Role of
the left inferior prefrontal cortex in retrieval of semantic knowledge:
a reevaluation. Proc Natl Acad Sci USA. 9:14792--14797.
Thompson-Schill SL, D’Esposito M, Kan IP. 1999. Effects of repetition and
competition on prefrontal activity during word generation. Neuron.
23:513--522.
Thompson-Schill SL, Kan IP, Oliver RT. 2006. Neuroimaging of semantic
memory. In: Cabeza R, Kingstone A, editors. Handbook of functional
neuroimaging. 2nd ed. Cambridge (MA): MIT Press. p. 149--190.
Trueswell JC, Kim AE. 1998. How to prune a garden-path by nipping it in
the bud: fast priming of verb argument structure. J Mem Lang. 39:
102--123.
Trueswell JC, Tanenhaus MK, Kello C. 1993. Verb-specific constraints in
sentence processing: separating effects of lexical preference from
garden-paths. J Exp Psychol Learn Mem Cogn. 19:528--553.
van Valin RD. 2001. An introduction to syntax. Cambridge (UK):
Cambridge University Press.
Vigneau M, Beaucousin V, Herve H, Duffau PY, Crivello F, Houde O,
Mazoyer B, Tzourio-Mazoyer N. 2006. Meta-analyzing left hemisphere language areas: phonology, semantics, and sentence processing. Neuroimage. 30:1414--1432.
Waters GS, Caplan D, Rochon E. 1995. Processing capacity and sentence
comprehension in patients with Alzheimer’s disease. Cogn Neuropsychol. 12:1--30.
Worsley KJ, Friston KJ. 1995. Analysis of fMRI time-series revisited—
again. Neuroimage. 2:173--181.
Zhang JX, Feng CM, Fox PT, Gao JH, Tan LH. 2004. Is left inferior frontal
gyrus a general mechanism for selection. Neuroimage. 23:281--287.
Cerebral Cortex August 2007, V 17 N 8 1969