Neural correlates of semantic and morphological processing of Hebrew nouns and verbs.

Abstract

Neuropsychological evidence regarding grammatical category suggests that deficits affecting verbs tend to localize differently from those affecting nouns, but previous functional imaging studies on healthy subjects fail to show consistent results that correspond to the clinical dissociation. In the current imaging study, we addressed this issue by manipulating not only the grammatical category but also the processing mode, using auditory presentation of Hebrew words. Subjects were presented with verbs and nouns and were instructed to make either a semantic decision ("Does the word belong to a given semantic category?") or a morphological decision ("Is the word inflected in plural?"). The results showed different patterns of activation across distinct regions of interest. With respect to grammatical category effects, we found increased activation for verbs in the posterior portion of the left superior temporal sulcus, left dorsal premotor area, and posterior inferior frontal gyrus. In each of these regions, the effect was sensitive to task. None of the ROIs showed noun advantage. With respect to task effects, we found a semantic advantage in left anterior inferior frontal gyrus, as well as in left posterior middle temporal gyrus. The results suggest that cerebral verb-noun dissociation is a result of localized and subtle processes that take place in a set of left frontal and temporal regions, and that the cognitive and neural processes involved in analyzing grammatical category depend on the lexical characteristics of the stimuli, as well as on task requirements. The discrepancy between functional imaging and patient data is also discussed.

Full text

Page 1

NEURAL CORRELATES OF SEMANTIC AND MORPHOLOGICAL
PROCESSING OF HEBREW NOUNS AND VERBS

Dafna Palti1,4*, Michal Ben Shachar3, Talma Hendler1,2,4 and Uri Hadar1

1Department of Psychology, Tel Aviv University, Tel Aviv, Israel
2Faculty of medicine, Tel Aviv University, Tel Aviv, Israel
3Department of Psychology, Stanford University, Stanford, California
4 Wohl Institute for Advanced Imaging, Sourasky Medical Center, Tel Aviv, Israel.


Short title: Neural correlates of processing nouns and verbs


Correspondence to:
Dafna Palti
Department of Psychology
Tel Aviv University
Tel Aviv, 69978
Israel

Tel: +972-3-697-3953
Fax: +972-3-697-3080
paltida@post.tau.ac.il








Keywords: fMRI, verbs, nouns, neurolinguistics, language processing, Hebrew

Page 2

2

Abstract: Neuropsychological evidence regarding grammatical category suggests
that deficits affecting verbs tend to localize differently from those affecting nouns,
but previous functional imaging studies on healthy subjects fail to show consistent
results that correspond to the clinical dissociation. In the current imaging study we
addressed this issue by manipulating not only the grammatical category but also the
processing mode, using auditory presentation of Hebrew words. Subjects were
presented with verbs and nouns, and were instructed to make either a semantic
decision (“Does the word belong to a given semantic category?”), or a morphological
decision (“Is the word inflected in plural?”). The results showed different patterns of
activation across distinct regions of interest. With respect to grammatical category
effects, we found increased activation for verbs in the posterior portion of the left
superior temporal sulcus, left dorsal premotor area and posterior inferior frontal
gyrus. In each of these regions the effect was sensitive to task. None of the ROIs
showed noun advantage. With respect to task effects, we found a semantic advantage
in left anterior inferior frontal gyrus, as well as in left posterior middle temporal
gyrus. The results suggest that cerebral verb-noun dissociation is a result of localized
and subtle processes that take place in a set of left frontal and temporal regions, and
that the cognitive and neural processes involved in analyzing grammatical category
depend on the lexical characteristics of the stimuli, as well as on task requirements.
The discrepancy between functional imaging and patient data is also discussed.


.

Page 3

3
INTRODUCTION
During the past two decades, ample evidence in the neuropsychological
literature has shown that the retrieval of verbs and nouns can be selectively damaged
(to mention a few: Caramazza and Hillis, 1991; Damasio and Tranel, 1993; Hillis and
Caramazza, 1995; McCarthy and Warrington, 1985; Miceli et al., 1988; Miceli et al.,
1984; Saffran et al., 1989; Silveri and Di Betta, 1997; Zingeser and Berndt, 1988;
Zingeser and Berndt, 1990). Commonly, impairment in the retrieval of verbs relative
to nouns has been associated with damage in the left frontal lobe, whereas the
opposite impairment has been ascribed to lesions in the left temporal lobe or temporo-
parietal regions. However, exceptions to this, as well as conflicting and inconsistent
results, challenged the association of the retrieval of different grammatical categories
with different brain areas (Corina et al., 2005; Luzzatti et al., 2002).
Theoretically, the processing origin of verb-noun dissociation is subject to
further controversy (for a recent review, see Crepaldi et al., 2005). Lexical accounts
suggest that this is rooted in either a lexical-syntactic level (Berndt et al., 2002;
Berndt et al., 1997) or, more specifically, in a lexico-morphological level (Shapiro
and Caramazza, 2003a; Shapiro et al., 2001; Shapiro et al., 2000). By contrast,
semantic accounts suggest that the dissociation originates in conceptual (Damasio and
Tranel, 1993; McCarthy and Warrington, 1985), or featural (Bird et al., 2000; Bird et
al., 2003) differences.
Recently, functional brain-imaging was used to investigate the neuronal
representation of verbs and nouns in healthy brains. This did not yield consistent
results and failed to confirm the neuropsychological observations. Specifically, while
most studies revealed large networks of brain regions that were activated during
lexical processing, no clear association of cortical regions with grammatical category

Page 4

4
was observed. Early PET studies found greater activation for verbs than for nouns in
a large network of regions, including the temporal, parietal and prefrontal regions, as
well as the supplementary motor area (SMA), whereas the opposite contrast elicited
more activation in the right prefrontal cortex (Warburton et al., 1996). A more recent
PET study by Perani et al. (1999) employed visual lexical decision for both verbs and
nouns in Italian. Here the verbs(+) nouns(-) contrast yielded activation in the left
dorsolateral prefrontal, left superior parietal, left middle temporal and occipital
cortices. No regions were activated more by nouns than by verbs.
Additional studies in English (Tyler et al., 2004; Tyler et al., 2003) and Chinese
(Li et al., 2004) failed to replicate the results of Perani et al. (1999) and did not find
any neuronal correlates of grammatical category differences. A subsequent study by
Tyler et al. (2004) found that inflected verbs evoked significantly greater activations
than inflected nouns only in left IFG, in support of the claim that differences in
processing verbs and nouns were rooted in morphological rather than semantic
properties. However, none of the above studies identified brain regions that were
activated more by nouns than by verbs. Shapiro et al. (2005) ascribed the latter failure
to the inability of the tasks to tap semantics-independent, grammatical knowledge. In
their PET study, Shapiro et al. (2005) used a morphological task in which subjects
produced singular and plural forms of written nouns (N), verbs (V), pseudo-nouns
and pseudo-verbs. By conjuncting real and pseudo- word contrasts, they aimed to
neutralize semantics and reveal activations that appertain to grammatical differences
as such. For the V>N contrast, the conjunction yielded activation in the left superior
frontal gyrus, the left anterior temporal gyrus, the cerebellum and thalamus. For N>V,
the conjunction also yielded activations, mainly in the right superior temporal gyrus,
left fusiform, left precentral gyrus and cerebellum. This result supports the idea that

Page 5

5
some aspects of processing nouns were functionally distinct from those of processing
verbs. However, the cerebral locations that showed this in Shapiro et al (2005)
differed from those implicated in noun impairments following brain damage.
Markedly, neither of the conjunction contrasts in the latter study activated the main
language regions on the left IFG and left STG (Broca’s and Wernicke’s area
respectively). It may well be that the choice of a task was crucial here, as one could
infer from Shapiro and Carmazza (2003b), Shapiro et al. (2005) and Tyler et al.
(2004). However, the effect of task on the V-N dissociation has never been tested in a
direct comparison within a single experiment.
The present fMRI study was set out to test the processing mode effect directly,
by comparing the grammatical category effect under two different tasks: semantic and
morphological, within the same experiment. We employed comprehension rather than
production tasks in order to avoid in-scanner motion artifacts associated with overt
articulation. In the semantic task, subjects judged the association of Hebrew verbs and
nouns with a given semantic category. In the morphological task, subjects had to
judge the number inflection of these stimuli (number inflection for both verbs and
nouns is specified in Hebrew by a morphological suffix). Applying the latter task in a
morphologically rich language like Hebrew enabled us to test the role of inflectional
morphology in the dissociation, as was suggested by Tyler et al. (2004). The tasks
were identical for verbs and nouns so that grammatical category was an implicit
property of the stimuli. To neutralize imageability effects (Bird et al., 2000; Bird et
al., 2003), we used only imageable words from both categories. Behavioral data on
other semantic features of the words (e.g., association with body movement) were
collected separately. The experimental blocks were carefully arranged to fit both
tasks. This allowed us to swap the tasks relative to stimuli in half of the subjects for

Page 6

6
counterbalancing purposes. We performed an ROI analysis to assess the effect of
grammatical category in the classical language regions, compared to other regions of
the cortical language network.

MATERIALS AND METHODS
Subjects
Fourteen healthy volunteers (5 males and 9 females), aged 21-50 (mean age:
30.5), with no psychiatric or neurological history, participated in the study. All
subjects gave written informed consent. The Tel-Aviv Sourasky Medical Center and
Tel Aviv University ethics committees approved the experimental protocol. All
subjects were native speakers of Hebrew, and Hebrew was their sole mother tongue.
They were all right handed as assessed by the Edinburgh handedness inventory
(Oldfield, 1971).
Experimental design
The experiment included four word conditions in a 2x2 factorial block design,
created by manipulating grammatical category (Verbs/Nouns) and task
(Semantic/Morphological). Word stimuli consisted of single words in Hebrew that
were presented auditorily. In order to control for task-stimuli interactions, all blocks
were designed for both tasks. Consequently, by swapping the tasks relative to stimuli,
we could create two versions that were counterbalanced between subjects (See figure
1). The tasks consisted of two “yes/no” decisions: (1) Semantic, in which the subject
had to decide whether the word was related to a given category (“food and drinks” in
version 1 and “agriculture” in version 2. During each version, the category remained
constant for all blocks requiring a semantic decision). (2) Morphological, in which the
subject had to decide whether the word was inflected in plural.

Page 7

7
In addition to the four experimental conditions, we included an auditory control
condition which consisted of Hebrew words played backwards. This condition
created a baseline activation of low level speech perception.
Insert figure 1 here
Materials
A brief description of Hebrew grammatical system with respect to inflectional
morphology of verbs and nouns is given in appendix 1. The crucial grammatical
difference here is that, for some inflections, morphological properties fully identify
the verb forms. Consequently, unlike in English, the recognition of grammatical
category is unambiguous (even with regard to 'inflected' non-words). Only
unambiguous verbal inflections were used in the study.
Preliminary tests. Stimuli were selected on the basis of preliminary ratings on
264 non abstract verbs and nouns made by 37 subjects who did not participate in the
imaging experiment. In these, subjects were asked to rate for each word the perceived
familiarity (ranging from 1: “I hardly come across the word” to 5: “I frequently come
across the word"), degree of imageability (ranging from 1:"unimaginable" to 5:"very
easily imaginable") and semantic associations with certain categories (ranging from1:
“not related at all” to 5: “very much related”). All verbs were transitive with one or
two complements and none of the nouns had an argument structure.
Selection of stimuli. Based on the preliminary ratings, 48 verbs and 48 nouns
were selected for the imaging experiment, according to the following criteria: (1)
Imageability rate above 3.4. (2) Mean rank of semantic association to the given
semantic category above 3.5 for the words that matched a “yes” response, and no
more than 1 for the words that matched a “no” response. (3) Verbs were of the three
active grammatical patterns (‘Pa'al’, ‘Pi'el’, ‘Hif'il’, see appendix 1). (4) All nouns

Page 8

8
were of masculine grammatical gender (see appendix 1). (5) No verb-noun
homophones were included. Verbs were inflected in the past tense, 3rd person
masculine form, half in singular (e.g., ‘mazag’ – he poured) and half in plural (e.g.,
‘pizru’- they distributed). Similarly, half of the nouns were inflected in singular (e.g.,
‘mazleg’ – a fork) and the other half in plural (e.g., ‘patishim’ – hammers). In this
way, all other inflections except for the number suffixes of the words were constant
throughout the experiment. In order to further evaluate the selected words for relevant
semantic dimensions, 14 additional subjects judged their visual clarity and the extent
to which they associated with body movement (also on 1-5 scales). The results of
these assessments are presented in figures 2A and 2B. The visual clarity scale (figure
2A) established that both verbs and nouns were associated with clear visual shapes
(none of the words was rated as having no visual shape). Yet, nouns had greater
clarity ratings than verbs, which difference coheres with previous reports regarding
imageability differences between verbs and nouns (Bird et al., 2003). This seems to
hold even when only highly imageable words from both categories are compared.
Results of the movement assessment showed that all verbs, but not all nouns, were
associated with body movement, and that 90% of the verbs were rated as “always or
usually associated with body movement” as opposed to only 4% of the nouns (figure
2B).
Insert figure 2 here
Timing parameters
Stimuli were presented in a standard block design. Each of the 5 conditions repeated
in four different blocks, 6 words in each block. The resulting 20 blocks were ordered
randomly within the experiment. Among them, there were 16 experimental blocks, 8
of verbs and 8 of nouns. Within each block, both singular and plural forms were

Page 9

9
included and the number of syllables, as well as imageability and familiarity ratings,
were matched as closely as possible (see appendix 2). Blocks lasted 10 seconds, with
an ISI of 1650 msec(1). In each block, 2 or 3 words matched a "yes" response. Word
blocks were separated by silence periods of 7.5 seconds. Each block was preceded by
a pre-recorded spoken instruction that determined the task during the block, and ended
with an auditory end-of-block cue.
Procedure
Each experimental session lasted approximately 1-1.5 hours and comprised of
both anatomical and functional scans. Anatomical scanning included the acquisition
of structural images in axial plane (see also ‘data acquisition’ below) and a 3D-SPGR
scan. In functional scans, subjects were requested to press a ‘yes’/'no' response button
after each word with their left pointer finger, according to the instructions of each
block. During the control condition (reversed words), subjects were asked to press the
button at the end of each stimulus. Each functional run lasted 7 minutes and 15
seconds. Responses were recorded online. No further information regarding type of
stimuli was given to the subjects. Subjects performed a short practice prior to entering
the magnet.
Data acquisition
MRI scans were conducted in a whole-body 1.5 Tesla, Signa Horizon, LX, 8.25
GE scanner, located at the Whol Institute for Advanced Imaging in Tel-Aviv
Sourasky Medical Center. Anatomical images were acquired using a standard T1-
weighted SE pulse sequence (voxel size: 0.8 mmX1.5mmX5mm). Fourteen slices, 5

1 The rate of the stimuli was unsynchronized with the TR, in order to distribute the timing of

data acquisition throughout the peristimulus time, as recommended by
Price CJ, Veltman DJ, Ashburner J, Josephs O, Friston KJ. (1999): The critical relationship between
the timing of stimulus presentation and data acquisition in blocked designs with fMRI. Neuroimage
10(1):36-44..

Page 10

10
mm thick with 1 mm gap, were selected, based on a mid-sagital slice covering most
of the cerebrum (excluding the dorsal and ventral tips). In addition, a 3D spoiled
gradient echo (SPGR) sequence with high resolution was acquired for each subject, in
order to allow volume statistical analyses. Functional MRI protocols included T2*-
weighted EPI images (at the same locations as the T1-weighted anatomical images).
A total of 174 volumes were acquired in each functional run, with FOV of 24 cm2 and
matrix size of 80x80, TR=2500ms, TE=55, flip angle=90°.
Data analysis
fMRI data were processed using BrainVoyager 4.1 software package
(http://www.brainvoyager.com, Goebel et al., 1998). The first six scans of the time
series were discarded. Functional images were superimposed on 2D (T1 weighted)
anatomical images, and projected on the 3D data sets through trilinear interpolation.
The complete data set was transformed into Talairach space (Talairach, 1988). Pre-
processing of functional scans included head motion estimation, as well as high pass
and low pass temporal frequency filtering. Volume statistical parametric maps were
calculated separately for each subject using a general linear model (Friston et al.,
1995), by contrasting all word conditions (both verbs and nouns) with the control
(reversed words) condition. In this way we identified brain regions that were
activated during the high-level processing of words, without imposing our
preliminary hypothesis on the data. Time courses of activated clusters within the
predefined ROIs (see below) were collected using voxel-number criterion: The exact
threshold of each ROI was set to the point of 150 activated voxels in pIFG and aIFG,
and 500 voxels in other regions of interest (see Regions of Interest below). To
account for the hemodynamic response delay, a lag of 2.5-5 seconds was inserted.
Shifts were determined individually (per ROI and per subject), in a manner that

Page 11

11
maximized the correlation between the time course and the “All-Words” predictor.
After shifting, the data were transformed into scores of percent signal change (PSC),
using the average values of the silent blocks as a baseline. For each ROI, the PSC
values of all data points in each condition were inserted into a within subject
ANOVA, in which grammatical category, task and repetition were used as within-
subject factors, while the version was used as a between subject factor. In order to
examine lateralization effects, a second ANOVA was carried out on the activations
from pIFG and pSTS in the left and right hemispheres, with hemisphere, grammatical
category and task as within subject variables.
Regions of interest
Six regions of interest were defined in the left hemisphere: three anterior (in the
frontal lobe) and three posterior (in the temporal and parietal lobes). The three
anterior regions were: 1. posterior Inferior Frontal Gyrus (pIFG; BA 44, pars
opercularis), defined in each subject separately by the common markers suggested in
the literature (Tomaiuolo et al., 1999), selected for its documented role in syntactic
processing (Bookheimer, 2002; Heim et al., 2003) 2. anterior IFG (aIFG; BA 45-7),
selected for its well documented involvement with aspects of semantic processing
(Bookheimer, 2002), and 3. Premotor, on the dorsal part of the pre-central sulcus and
gyrus (BA 4-6), chosen in order to check the relation of motor schemata represented
in this region to the processing of verbs (Grezes and Decety, 2001). The three
posterior regions were: 4. posterior Superior Temporal Sulcus (pSTS; BA 39)
considered a sub region within a more wide-ranging “Wernicke’s area”. It was chosen
on the basis of extensive evidence showing its involvement in speech perception
(Wise et al., 2001); 5. posterior middle temporal gyrus (pMTG; BA 37), in the
posterior part of the mid-inferior temporal gyrus, located ventrally to Wernicke’s

Page 12

12
area, chosen for its extensive involvement in semantic processing (Friederici et al.,
2000) and 6. IPS (Intraparietal sulcus), in the dorsal part of the parietal lobe, which is
known to be involved in general, language related processes (Cabeza and Nyberg,
2000) and was chosen as a control region.
On the right hemisphere, two additional ROIs were defined by homology with
left pIFG and left pSTS, in order to examine aspects of language processing that are
related to hemispheric dominance. Brain activations were measured within these
regions separately, as well as in comparison with the activation in the left hemisphere.
Mean Talairach coordinates of the activations in each ROI are given in table I.
Insert table I here

Reaction- time measurements
Reaction-Time measurements were collected outside of the magnet. Thirteen
psychology students aged 19-26 participated in this experiment for course credits.
None of them participated in the imaging experiment. Subjects were tested using the
same experimental protocol that was used in the scanner. Reaction times were
measured with homemade software installed on an IBM PC. A repeated-measures
ANOVA was performed on the reaction times, using a 2 (grammatical category:
verbs or nouns) X 2 (task: semantic or morphological) design. The version was added
as a between-subject independent variable, to control for stimuli and task interaction.

Page 13

13
RESULTS
Reaction- time measurements
Factorial ANOVA on the behavioral data revealed a main effect of grammatical
category: The reactions to verbs (mean RT= 655ms, SD=191ms) were slower than to
nouns (mean RT = 553ms, SD=173ms; F(1,11)=83.62; p<.00001). A main effect of
task was also documented, in which the semantic task evoked slower responses (mean
RT= 645ms, SD=203ms) than the morphological task (mean RT = 562ms,
SD=155ms), (F(1,11)=31.11; p<.0002). The interaction of category and task was also
significant (F(1,11)=6.22; p<.0298): the task effect was slightly greater in verbs than
in nouns.

Imaging experiment
No effect of version was observed in any of the ROIs, hence the results are
presented across versions (i.e., across specific task-stimulus assignments). Statistical
parametric maps, obtained by contrasting all words conditions with the reversed-
words condition are presented in figure 3. Above threshold activation was shown in
all 14 subjects in left pIFG, aIFG ,premotor, left pSTS and IPS. Twelve of the 14
subjects showed above threshold activations in pMTG and right pIFG, and 10/14
subjects in right pSTS. As none of the regions revealed an interaction of task and
category, the results are presented as main and simple effects in each factor
separately.
Insert figure 3 here
Grammatical category effects
A main effect of grammatical category, showing more activation for verbs than
for nouns (V>N), was revealed in the left pIFG (F(1,12) = 9.69, p<0.01), premotor

Page 14

14
(F(1,12) = 8.95, p<0.05) and left pSTS (F(1,12) = 36.6, p<0.01), but not in aIFG,
pMTG, IPS, right pIFG and right pSTS, (see figure 4). No region with the reversed
pattern was found. A simple-effect analysis revealed a V>N effect in left pIFG during
the semantic task only (F(1,12) = 7.3, p<0.05), in the premotor area during the
morphological task only (F(1,12)= 8.2, p<0.05), and in left pSTS during both tasks
(semantic: F(1,12)= 21.5, p<0.01, morphological: F(1,12)= 26.8, p<0.01,), see figure
5.
Insert figures 4 and 5 here
Task effects
A main effect of task, showing more activation for the semantic than for the
morphological task, was revealed in aIFG (F(1,12) = 40.51, p<0.01), left pSTS
(F(1,12) = 11.4, p<0.01) and pMTG (F(1,10) = 10.0, p<0.01), but not in left pIFG,
premotor, IPS, right pIFG and right pSTS, (see figure 6). No region with the reversed
pattern was found. A simple-effect analysis in these ROIs revealed more activation
for the semantic task than for the morphological task in both the verb and the noun
conditions in all three ROIs (in aIFG - verbs: F(1,12)= 11.77, p<0.01, nouns:
F(1,12)= 28.5, p<0.01, in pSTS - verbs: F(1,12)= 9.05 p<0.05, nouns: F(1,12)= 5.2,
p<0.05, in pMTG - verbs: F(1,10)= 5.7 p<0.05, nouns: F(1,10)= 8.16 p<0.05).
Insert Figure 6 here
Hemispheric differences
In pIFG, an overall lateralization effect was observed (F(1,12)=5.3, p<0.05),
with greater activation on the left hemisphere than on the right hemisphere, across
conditions (Figure 7A). Across hemispheres, there was also a marginal effect of
grammatical category, whereby activation for verbs was greater than activation for

Page 15

15
nouns (F(1,12)=4.62, p=0.052, Figure 7C-upper panel ). No other effects were
revealed in pIFG.
In pSTS, there was no overall lateralization in activation; right and left pSTS
showed similar levels of activation (F(1,12)=0.3, p=NS). Across sides, there was a
main effect of grammatical category, in which activation for verbs was greater than
activation for nouns (F(1,12)=54.5, p<0.001). There was also an interaction between
hemisphere and grammatical category (F(1,12)=10.5, p<0.001), showing that the
grammatical category effect was larger in left pSTS than in right pSTS (figure 7B and
7C-lower panel). A task effect showing more activation in the semantic than in the
morphological task across hemispheres, was also observed (F(1,12)=8.55, p<0.05).
Insert figure 7 here

DISCUSSION
Our results suggest that verbs and nouns are processed differently in the brain.
Three of the ROIs - left pIFG, left pSTS and left premotor- were activated more by
verbs than by nouns during the semantic task, the morphological task or both (figures
4 and 5). This result is consistent with previous neuroimaging studies in which an
overall greater activation while processing verbs relative to nouns has been observed
in several fronto-temporal regions (Perani et al., 1999; Warburton et al., 1996). With
respect to the frontal ROIs, the results also correspond to the verb deficit observed in
patients following damage to the left frontal lobe. However, in contrast to the double
dissociation suggested by the neuropsychological evidence, but still in line with most
previous imaging studies (Perani et al., 1999; Tyler et al., 2003; Warburton et al.,
1996), none of our ROIs seem to be activated by nouns more than by verbs (figure 4).
Moreover, while all previous imaging studies used only reading as the input modality