Conceptual Representations of Action in the Lateral ...

Conceptual Representations of Action in the LateralTemporal Cortex

Joseph W Kable Irene P Kan Ashley WilsonSharon L Thompson-Schill and Anjan Chatterjee

Abstract

amp Retrieval of conceptual information from action picturescauses greater activation than from object pictures bilaterallyin human motion areas (MTMST) and nearby temporal re-gions By contrast retrieval of conceptual information fromaction words causes greater activation in left middle and su-perior temporal gyri anterior and dorsal to the MTMST Weperformed two fMRI experiments to replicate and extend thesefindings regarding action words In the first experimentsubjects performed conceptual judgments of action and objectwords under conditions that stressed visual semantic informa-tion Under these conditions action words again activatedposterior temporal regions close to but not identical with theMTMST In the second experiment we included conceptual

judgments of manipulable object words in addition tojudgments of action and animal words Both action and mani-pulable object judgments caused greater activity than animaljudgments in the posterior middle temporal gyrus Both ofthese experiments support the hypothesis that middletemporal gyrus activation is related to accessing conceptualinformation about motion attributes rather than alternativeaccounts on the basis of lexical or grammatical factors Fur-thermore these experiments provide additional support forthe notion of a concrete to abstract gradient of motion repre-sentations with the lateral occipito-temporal cortex extendinganterior and dorsal from the MTMST towards the peri-sylviancortex amp

INTRODUCTION

A large body of work within cognitive neuroscience hasexamined how the brain mediates conceptual knowl-edge of the world Starting from the observation thatknowledge of specific categories of objects can beselectively impaired by brain damage a range of com-peting theories have been proposed for the neuralmediation of conceptual knowledge One class of theo-ries argues that category-specific deficits do not resultfrom category-specific organization per se but ratherbecause different kinds of information are more relevantto concepts in different conceptual categories (Gainotti2000 Saffran amp Schwartz 1994 Warrington amp Shallice1984) According to this view conceptual knowledgeis mediated by distributed neural networks partiallydissociable according to the kind of information repre-sented such as particular sensory or motor attributesVariations within this general class of theories havebeen elaborated to explain how such conceptual repre-sentations interact with perceptual or language systemsto allow for category-specific deficits restricted to lan-guage or perception (Damasio Tranel GrabowskiAdolphs amp Damasio 2004 Simmons amp Barsalou 2003

Pulvermuller 1999) A second class of theories proposesthat category-specific deficits are evidence of category-specific organization and that the brain contains do-main-specific knowledge systems shaped by evolutionand specialized for particular categories of knowledge(Caramazza amp Mahon 2003) Finally a third class oftheories also argues that conceptual knowledge is me-diated by distributed neural networks but these net-works are not organized along sensorimotor lines (Tyleramp Moss 2001 Caramazza Hillis Rapp amp Romani 1990)Rather category-specific deficits arise because of thestructure of knowledge in different conceptual catego-ries (intercorrelations vs uniqueness of semantic fea-tures) and how such structure breaks down withdamage to a unitary conceptual system

However most of the tests pitting these competingtheories have focused on knowledge of different cate-gories of objects Comparatively less attention has beengiven to what might be considered a broader distinctionbetween actions and objects In language verbs com-monly represent actions and nouns commonly repre-sent objects Brain damage can independently impairconceptual knowledge of actions and objects (DanieleGiustolisi Silveri Colosimo amp Gainotti 1994 MiceliSilveri Nocentini amp Caramazza 1988 McCarthy ampWarrington 1985) and a few functional imaging studieshave found distinct brain areas that are engaged to aUniversity of Pennsylvania

D 2005 Massachusetts Institute of Technology Journal of Cognitive Neuroscience 1712 pp 1855ndash1870

greater degree by actions or objects (Tranel MartinDamasio Grabowski amp Hichwa in press Kable Lease-Spellmeyer amp Chatterjee 2002 Damasio GrabowskiTranel Ponto et al 2001 Perani Cappa et al 1999Martin Haxby Lalonde Wiggs amp Ungerleider 1995) Anunderstanding of the neural substrates of actions com-pared to different categories of objects should provideimportant evidence for evaluating theories of concep-tual knowledge systems as well as a more complete pic-ture of the neural mediation of conceptual knowledge

We directly investigated the neural substrates of ac-tion knowledge in a previous fMRI study (Kable Lease-Spellmeyer et al 2002) Subjects saw three pictures orthree words simultaneously one at the top of the screenand two at the bottom and they were asked to deter-mine which of the two concepts presented at thebottom was most similar to the one presented at thetop For example subjects would decide that lsquolsquodiggingrsquorsquowas more similar to lsquolsquoshovelingrsquorsquo than lsquolsquolisteningrsquorsquo or thatlsquolsquodoctorrsquorsquo was more similar to lsquolsquocoprsquorsquo than lsquolsquozucchinirsquorsquoSubjects performed the same task once with picturesand once with words When subjects performed the taskwith pictures there was greater activity for actionscompared to objects bilaterally in human motion areaMTMST and in nearby areas of the middle temporalgyrus When subjects performed the task with wordsthere was greater activity for actions compared toobjects in the left middle and superior temporal gyrianterior and dorsal to the activation for pictures Actionwords unlike action pictures did not preferentially ac-tivate area MTMST

These results suggest that parts of the lateral occipito-temporal cortex are particularly important for me-diating conceptual knowledge of actions In additionconsidering the involvement of the MTMST the senso-rimotor theory of conceptual knowledge provides themost natural explanation for these findings Within thisframework we argued that the lateral occipito-temporalcortex likely mediates different representations of mo-tion which are presumably more relevant for actionconcepts compared to object concepts Furthermorewe suggested that these motion representations are or-ganized on a concrete to abstract gradient with moreconcrete representations preferentially accessed frompictures located more posteriorly near the MTMSTand more abstract representations preferentially ac-cessed from words located more anteriorly near theperi-sylvian cortex

However there are alternative possible explanationsfor our findings concerning action words especiallygiven the small overlap between areas showing greateractivation for action words and action pictures Actionwords may have failed to activate area MTMST be-cause the judgments for the particular stimuli used didnot stress detailed motion information and not be-cause of any intrinsic material specificity Furthermoreactivation differences in the lateral temporal cortex

may have been due to lexical differences between theaction and object words rather than the differences inconceptual category In particular all of the actionwords were verbs whereas all of the object words werenouns so perhaps grammatical processing accounts forthese activation differences Although other investiga-tors have failed to detect significant differences betweenverbs and nouns in the lateral temporal cortex (TylerBright Fletcher amp Stamatakis 2004 Tyler Stamatakiset al 2003 Tyler Russell Fadili amp Moss 2001) thisfailure highlights the additional possibility that our pre-vious findings with action words may have been idio-syncratic to the particular stimuli or subjects

We designed two fMRI experiments to test thesealternative hypotheses Both experiments test the hy-pothesis that activity for words in the lateral temporalcortex is driven by semantic differences as we previ-ously argued In the first experiment we designed newword triads for the conceptual judgment task thatfocused subjectsrsquo attention on a more constrained setof visual attributes (see Figure 1) In our previous studybecause the distractor item on each trial was oftendistantly related to the other two items (lsquolsquolisteningrsquorsquocompared to lsquolsquodiggingrsquorsquo and lsquolsquoshovelingrsquorsquo) subjects couldhave made their judgments with only minimal access tovisual attributes of different actions (unlike lsquolsquolisteningrsquorsquolsquolsquodiggingrsquorsquo and lsquolsquoshovelingrsquorsquo involve motion) We de-signed new stimuli in which all three words were closelyrelated to increase the importance of visual attributes inmaking the judgment For example subjects would judgethat lsquolsquoskippingrsquorsquo is more similar to lsquolsquobouncingrsquorsquo thanlsquolsquorollingrsquorsquo as both skipping and bouncing have similarmanners of motion or they would judge that lsquolsquozebrarsquorsquo ismore similar to lsquolsquohorsersquorsquo than lsquolsquocowrsquorsquo as both zebras andhorses have similar forms If activation in the lateraltemporal cortex is associated with accessing semanticinformation about actions then we would expect toreplicate our original finding of greater activation foraction words compared to object words Furthermore ifour hypothesis about a gradient of action representa-tions is correct we would expect the activation for actionwords to be closer to the MTMST in the current exper-iment because these stimuli emphasized visual infor-mation about actions at a greater level of detail than inour previous study

In the second experiment we again used the concep-tual judgment task but with three different categoriesof words actions animals and manipulable objectsAlthough the words for both animals and manipulableobjects are nouns manipulable object concepts morethan animal concepts rely on information about move-ment (Damasio Tranel et al 2004 Saffran Coslett ampKeener 2003 Buxbaum amp Saffran 2002 McRae de Sa ampSeidenberg 1997) Unlike animals most manipulableobjects are strongly associated with a constrained setof actions So if the previously observed activation dif-ferences between action and object words were driven

1856 Journal of Cognitive Neuroscience Volume 17 Number 12

by visual attributes associated with concepts then bothaction and manipulable object words should causegreater activation than animals in the lateral temporalcortex However if these activation differences weredriven by linguistic differences evoked by grammaticalclass then action words should cause greater activitythan both animal and manipulable object words in thelateral temporal cortex

RESULTS

Experiment 1

Behavioral Results

In the scanner subjects were more accurate on objectjudgments compared to action judgments [t(7) = 29p = 02] and they were faster on object judgmentscompared to action judgments [t(7) = 51 p = 001](see Table 1)

Imaging Results

Lateral temporal cortex We used a region-of-interest(ROI) approach to test for activity differences in thelateral temporal cortex First anatomical ROIs weredefined individually in each subject for the superiortemporal middle temporal and inferior temporal gyrusThen within each ROI we averaged the fMRI time seriesacross all voxels that showed significant activity for themain effect of conceptual judgments compared to thebaseline condition (see Figure 2) For each averagedtime series we calculated a measure of the difference inactivity between the action and object conditions and

then tested whether this difference was significantlydifferent from zero across subjects for each ROI (arandom effects test see Methods for details of allanalyses) Because our previous findings for actionwords were limited to the left hemisphere and most ofthe activity for the main effect was in the left hemi-sphere we focused our analyses on left hemisphereROIs Across subjects activity was consistently greaterfor action judgments in the left middle temporal gyrus[t(7) = 40 p = 005] and in the left superior temporalgyrus [t(7) = 41 p = 005] Activity did not differbetween the conditions in the left inferior temporalgyrus [t(7) = 03 p = 76] (see Figure 3 and Table 2)

Table 1 Behavioral Performance (Mean plusmn Standard Error) ofSubjects in the fMRI Experiments for Different Task Conditions

Accuracy () Reaction Time (msec)

Experiment 1

Actions 87 plusmn 2 1956 plusmn 101

Objects 92 plusmn 1 1802 plusmn 94

Baseline 98 plusmn 1 1201 plusmn 60

Experiment 2


Manipulable Objects 77 plusmn 3 1984 plusmn 101

Animals 81 plusmn 3 2047 plusmn 144


Figure 1 Schematic of

the experimental paradigm

Subjects were presented with

18-sec blocks of conceptualjudgments interspersed

with baseline blocks

Kable et al 1857

MTMST For each subject we also defined an ROI forthe MTMST based on a localizer scan Because therewas little activity for the main effect in the MTMST weaveraged across all ROI voxels in both hemispheres

Across subjects there was no consistent difference be-tween action judgments and object judgments [t(7) =09 p = 39] This pattern did not change if we consid-ered only voxels in the left hemisphere

Figure 2 Example ROIs are shown The main effect for the conceptual judgment task is displayed for three different subjects in Experiment 2

with some ROIs for these subjects highlighted on the different views ROIs were defined in each subject as all voxels significant for the main effect

contrast within a specified anatomical area Activity in these ROIs across both experiments is displayed in Figures 3 and 5

Figure 3 Displayed are averaged time series (and standard error) for lateral temporal ROIs in Experiments 1 and 2 Time series were averaged

across all blocks for each subject and then across all subjects for each experiment Solid line depicts the average dotted lines show average plus or

minus standard error (calculated across subjects)


Comparison of peak activity The ROI analyses rep-licated our previous findings for action words in thelateral temporal cortex however these analyses mightobscure a shift in activation within an ROI across studiesIn order to look for such a shift we calculated the loca-tion of the peak difference between action and objectjudgments in the left lateral temporal cortex Unthresh-olded t-maps for the actions versus objects contrastwere created for each subject and normalized into MNIspace The location of the voxel in the lateral occipito-temporal cortex demonstrating the largest differencebetween actions and objects was identified for eachsubject and the average location of this peak wascomputed across subjects For the eight subjects in thisexperiment the mean location (and standard error)of the peak difference between actions and objectswithin the left lateral occipito-temporal cortex was(55 plusmn 3 60 plusmn 2 5 plusmn 4) We performed a similaranalysis for our previous study (Kable Lease-Spellmeyeret al 2002) For the six subjects in the previous studythe mean location of the voxel with the largest differ-ence between actions and objects in the word condi-tion was (56 plusmn 3 26 plusmn 13 0 plusmn 4) whereas the meanlocation of the voxel with the largest difference be-tween actions and objects in the picture condition was(50 plusmn 3 78 plusmn 3 3 plusmn 2) (see Figure 4) Thedifferences in the anteriorndashposterior location of thepeak voxel were statistically significant for each com-

parison [pictures and words from the previous studyt(10) = 40 p = 0025 pictures from the previousstudy and words from current experiment t(12) = 49p = 0004 words from the previous study and currentexperiment t(12) = 30 p = 01] None of theinferiorndashsuperior or medialndashlateral differences in peaklocation was significant

Experiment 2

Behavioral Results

Subjects were more accurate on action judgments thaneither animal or manipulable object judgments [actionsvs animals t(8) = 29 p = 02 actions vs manipulableobjects t(8) = 29 p = 02] Reaction times did notdiffer between the three conceptual conditions (seeTable 1)

Imaging Results

Lateral temporal cortex We used a similar ROI analysisas in the first experiment to look for differences betweenthe three conditions in the left lateral temporal cortexIn the left superior temporal gyrus activity for actionjudgments was significantly greater than activity foranimal judgments [actions vs animals t(8) = 31p = 02] Activity for manipulable object judgments

Table 2 Functional MRI Results

Experiment 1 Experiment 2

Actions Objects

Actions Animals

Actions Manipulable Objects

Manipulable Objects Animals

Lateral Temporal ROIs

Superior Temporal Gyrus 164 plusmn 040 073 plusmn 024 052 plusmn 051 021 plusmn 038

Middle Temporal Gyrus 214 plusmn 053 129 plusmn 045 059 plusmn 064 070 plusmn 026

Inferior Temporal Gyrus 021 plusmn 066 078 plusmn 038 168 plusmn 056 090 plusmn 031

Other ROIs

Lateral Orbital Gyrus 108 plusmn 044 030 plusmn 027 045 plusmn 035 015 plusmn 028

Superior Frontal Gyrus 138 plusmn 046 101 plusmn 031 034 plusmn 039 067 plusmn 035

Premotor Cortex 125 plusmn 054 004 plusmn 031 069 plusmn 049 065 plusmn 031

Motor Cortex 127 plusmn 045 035 plusmn 036 024 plusmn 036 059 plusmn 019

Inferior Parietal Cortex 075 plusmn 043 088 plusmn 031 108 plusmn 043 020 plusmn 027

Fusiform Gyrus 041 plusmn 093 029 plusmn 030 102 plusmn 033 074 plusmn 032

Middle Occipital Gyrus 116 plusmn 047 060 plusmn 066 085 plusmn 053 026 plusmn 062

Retrosplenial Cortex 205 plusmn 057 177 plusmn 055 122 plusmn 042 055 plusmn 028

Effect sizes are given as mean plusmn standard error for contrasts between conceptual conditions within each ROI for Experiments 1 and 2 All effectsizes are from left hemisphere ROIs Effect sizes that are consistently different from zero across subjects ( p lt 05) are in bold

Kable et al 1859

was intermediate between the two and not significantlydifferent from either In the left middle temporal gyrusboth action judgments and manipulable object judg-ments caused significantly greater activity than animaljudgments [actions vs animals t(8) = 29 p = 02 manip-ulable objects vs animals t(8) = 27 p = 03] In the leftinferior temporal gyrus activity for manipulable objectjudgments was significantly greater than both actionjudgments and animal judgments [actions vs manipula-ble objects t(8) = 30 p = 02 manipulable objects vsanimals t(8) = 29 p = 02] (see Figure 3 and Table 2)

Comparison of peak activity In the same manner asthe first experiment we calculated the mean location(and standard error) of the peak difference within theleft lateral occipito-temporal cortex between actions andanimals (56 plusmn 3 54 plusmn 6 1 plusmn 4) and betweenmanipulable objects and animals (53 plusmn 2 56 plusmn 76 plusmn 4) The location of these two peaks was notsignificantly different However both peaks were moreanterior than the peak for action pictures in the pre-vious study [action words and pictures from the previ-ous study t(13) = 29 p = 01 manipulable objectwords and pictures from the previous study t(13) = 23p = 04] and more posterior than the peak for actionwords in the previous study [action words and wordsfrom the previous study t(13) = 22 p = 049 ma-nipulable object words and words from the previousstudy t(13) = 22 p = 049] (see Figure 4)

Other Left Hemisphere Regions

Although the main focus of our investigation was onthe pattern of activity in the lateral occipito-temporal

cortex we also looked for activation differences outsideof this region To have similar sensitivity we wanted touse the same ROI approach In order to identify poten-tial ROIs we first performed an exploratory whole-brainanalysis in normalized space on the data from the firstexperiment Because the purpose of the exploratoryanalysis was to generate hypotheses we used a liberalheight threshold of p lt 01 (uncorrected) and extentthreshold of 5 voxels At this threshold there were eightclusters with greater activity for actions seven were inthe lateral temporal cortex and one was near the borderof the ventral premotor and motor cortex (MNI coor-dinates of peak voxel 525 75 15) There were also10 clusters with greater activity for objects 3 in theanterior portions of the superior frontal gyrus (75525 10 15 2625 40 and 30 15 35) 1 in the lateralorbital gyrus (225 2625 20) 2 in the fusiform gyrus(3375 4875 35 and 225 45 25) 1 in middleoccipital gyrus (45 7875 15) 1 in the inferiorparietal cortex (3375 8625 35) and 2 extendingfrom the posterior cingulate into the visual cortex(375 6375 0 and 375 75 10) Based on thesefindings we decided to define additional anatomicalROIs in each subject in eight left hemisphere regionslateral orbital gyrus premotor cortex motor cortexsuperior frontal gyrus fusiform gyrus inferior parietalcortex middle occipital gyrus and retrosplenial cortex(see Methods for details on the anatomical criteria forthese ROIs) For each of these new ROIs we performedthe same analysis as the one used in the lateral temporalROIs

Because the data from first experiment were used inthe exploratory whole-brain analysis the ROI analyses

Figure 4 Plotted is the

location of the voxel showing

the peak difference within the

lateral occipito-temporalcortex for each subject in the

present study and a previous

one (Kable Lease-Spellmeyeret al 2002) Peak voxels are

shown for five contrasts ac-

tion words minus object words

in judgments with unrelateddistractors (6 n = 6 previous

study) action pictures minus

object pictures (5 n = 6

previous study) action wordsminus object words in

judgments with related

distractors ( n = 8Experiment 1) action words

minus animal words in


distractors (amp n = 9Experiment 2) and

manipulable object words minus animal words in judgments with related distractors (~ n = 9 Experiment 2) Locations are projected in the

medialndashlateral dimension onto a sagittal slice at x = 51 The most posterior peaks appear outside the brain for this projection because their true

location is medial to this slice


for these data are not an independent test However theexploratory analysis did not use a corrected significancethreshold so these ROI analyses do serve to confirmwhether apparent differences in the exploratory analy-sis are significant More importantly the ROI analyses forthe second experiment are independent from the hy-pothesis-generating whole-brain analysis Thereforeconsistent results across the two experiments wouldprovide inferentially sound evidence for activation differ-ences in an ROI

In the first experiment there was greater activityfor action judgments in the left motor cortex [t(7) =28 p = 03] and a trend in the same direction thatmissed significance in the left premotor cortex [t(7) =23 p = 055] The left retrosplenial cortex showedthe opposite pattern with consistently greater ac-tivity for object judgments compared to action judg-ments across subjects [t(7) = 36 p = 009] Inaddition increased activity for object judgments reachedsignificance in the left superior frontal gyrus [t(7) =30 p = 02] the left orbital gyrus [t(7) = 25p = 04] and the left middle occipital gyrus [t(7) =25 p = 04] In the left fusiform and inferior pari-etal ROIs we detected no significant differences be-tween the action and object conditions (see Figure 5and Table 2)

In the second experiment these differences were onlyreplicated in the retrosplenial and superior frontal ROIswhich again showed greater activity for objects com-pared to actions In the left retrosplenial cortex therewas greater activity for both animal and manipulableobject judgments compared to action judgments [ac-tions vs animals t(8) = 32 p = 01 actions vsmanipulable objects t(8) = 29 p = 02] In the leftsuperior frontal gyrus only animal judgments showedgreater activity than action judgments [actions vs ani-mals t(8) = 32 p = 01] In addition to these tworeplications there was greater activity for manipulableobject judgments compared to animal judgments in theleft motor cortex [manipulable objects vs animals t(8) =31 p = 01] and a trend in the same direction inthe left premotor cortex [manipulable objects vs ani-mals t(8) = 21 p = 07] However action judgmentswere not significantly different from either animal ormanipulable object judgments in the left motor andpremotor cortex We also detected no significant differ-ences in the lateral orbital or middle occipital ROIs Inthe left fusiform gyrus manipulable object judgmentsshowed greater activity than action judgments andtrended towards greater activity than animal judgments[actions vs manipulable objects t(8) = 31 p = 01manipulable objects vs animals t(8) = 23 p = 052]In the left inferior parietal cortex activity was greaterfor both manipulable objects and animals compared toactions [actions vs animals t(8) = 28 p = 02 actionsvs manipulable objects t(8) = 25 p = 04] (seeFigure 5 and Table 2)

Right Hemisphere Regions

As explained above we focused our analyses on lefthemisphere ROIs because our previous findings (andcurrent hypotheses) concerned the left hemisphere Inaddition because our stimuli were words we did notexpect much activity in the right hemisphere a prioriIndeed there was much less activity for the concep-tual judgment task in right hemisphere areasmdashfor ex-ample the number of voxels significant for the maineffect in the right middle temporal gyrus was 20 thenumber of voxels in the left middle temporal gyrusWhen we used the same approach to define the cor-responding right hemisphere ROIs to those used abovethere were only a few significant findings across thetwo experiments In the first experiment there wasgreater activity for action words compared to objectwords in the right middle temporal gyrus [t(7) = 24p = 048] The right middle temporal gyrus showeda similar trend (greater activity for actions comparedto animals) in the second experiment that did not reachsignificance [t(8) = 18 p = 10] Also in the secondexperiment there was greater activity in the right fusi-form gyrus for tools compared to actions [t(8) = 27p = 025] and for animals compared to actions [t(8) =29 p = 02]

DISCUSSION

This study continues our investigations of the role ofthe lateral temporal cortex in mediating conceptualknowledge of action In the first experiment wereplicated our previous finding that conceptual judg-ments of action words activate left posterior temporalregions close to but not identical with area MTMSTunder revised conditions where these judgmentsstressed detailed distinctions between visual motionfeatures In the second experiment we extendedthese results by demonstrating that judgments of bothaction and manipulable object words caused greateractivity than animal judgments in the middle temporalgyrus Thus semantic differences rather than lexicalor sublexical factors seem to drive activation in themiddle temporal gyrus Finally across both experi-ments we found greater activity for object judgmentsin two unanticipated regions the retrosplenial cortexand the superior frontal gyrus We start by discussing ourfindings on the role of the lateral occipito-temporalcortex in verbally mediated conceptual knowledge be-fore briefly touching on our findings regarding otherregions

Lateral Occipito-temporal Cortex

In our first experiment we tested the whether lateraloccipito-temporal cortex would demonstrate greater

Kable et al 1861

activity for action words than for object words in a taskthat stressed detailed distinctions between visual motionfeatures Under these conditions action words elicitedgreater activity than object words in the left middle andsuperior temporal gyri Action words however did notactivate the MTMST more than object words Theseresults replicate our previous findings that action wordscause greater activity in left-lateralized posterior tempo-ral regions but not within the MTMST (Kable Lease-Spellmeyer et al 2002) They are also consistent withour suggestion for a gradient of motion representationswithin the lateral occipito-temporal cortex such thatareas processing verbal descriptors of motion activateregions close to but not identical with areas thatmediate motion perception

In our second experiment we extended these find-ings by including two categories of object words animalsand manipulable objects Again action judgments eli-cited greater activity than animal judgments in the

middle and superior temporal gyri on the left Manipu-lable object judgments caused greater activation thananimal judgments in the middle temporal gyrus andgreater activation than both action and animal judg-ments in the inferior temporal gyrus This pattern of re-sults supports the argument that semantic factors ratherthan grammatical class drives activation within themiddle temporal gyrus

Thus the results across both experiments supportthe hypothesis that middle temporal gyrus activity is as-sociated with semantic processing Because activity inthe middle temporal gyrus cuts across grammaticalcategory boundaries these findings are more consistentwith theories that conceptual knowledge is mediatedby a distributed network partially organized accordingto sensorimotor attributes rather than theories positingconceptual organization along strict category lines In-deed our neural data show a similar pattern to onedemonstrated in recent computational models of fea-

Figure 5 Displayed are averaged time series (and standard error) for other ROIs showing consistent effects in Experiments 1 and 2 Time series

were averaged across all blocks for each subject and then across all subjects for each experiment Solid line depicts the average dotted lines showaverage plus or minus standard error (calculated across subjects)


ture-based semantic networks where the clusters ofconceptual similar items in self-organizing maps crossgrammatical category boundaries (Vinson amp Vigliocco2002) Within such a feature-based network the middletemporal gyrus may be particularly involved in thesemantic processing of motion information whichwould play a central role in conceptual representationsof actions and manipulable objects

A range of previous findings are consistent with thesesuggestions about the role of the middle temporalgyrus In a set of studies similar to ours Tyler and col-leagues found greater activity in the lateral temporalcortex for judgments about biological actions comparedto animal judgments (Tyler Stamatakis et al 2003) butnot for all verb judgments compared to all noun judg-ments (a category that included manipulable objects)(Tyler Stamatakis et al 2003 Tyler Russell et al2001) The lateral temporal cortex also demonstratesgreater activity when subjects generate action wordscompared to color words (Martin et al 1995) Further-more damage in this area is associated with impair-ments on tests of conceptual knowledge for actions(Tranel Kemmerer Adolphs Damasio amp Damasio2003) In addition activity in this region is found in avariety of tasks for tool words compared to animal words(Chao Haxby amp Martin 1999 Perani Schnur et al1999 Thompson-Schill Aguirre DrsquoEsposito amp Farah1999 Cappa Perani Schnur Tettamanti amp Fazio1998) and damage in this area is associated with bothimpaired knowledge and impaired naming of tool pic-tures (Tranel Damasio amp Damasio 1997 DamasioGrabowski Tranel Hichwa amp Damasio 1996) Few pre-vious studies however have directly compared actionsand manipulable objects Thus our finding that actionand manipulable object judgments cause distinct acti-vations in the temporal cortex which overlap in themiddle temporal gyrus provides important new evi-dence concerning the role of the lateral temporal cortexin conceptual processing

Interpretation of the distinct areas of activity foractions and manipulable objects is less clear One pos-sibility is that these two categories are associated withdiverging motion featuresmdashaction words may morestrongly evoke articulated biological motion whereasmanipulable objects may more strongly evoke unarticu-lated nonbiological motion Consistent with this ideaBeauchamp Lee Haxby and Martin (2002 2003) direct-ly compared pictures of moving tools with movingbodies and found greater activation for moving bodiesin the superior temporal sulcus and greater activationfor moving tools in the inferior temporal sulcus TranelMartin et al (in press) found a similar overall pattern inthe lateral temporal cortex for naming of actions andtools with activation for actions more dorsal and toolsmore ventral Our current results with words follow thissame dorsalndashventral pattern However we cannot ruleout possible alternative explanations In particular ac-

tivity in the superior temporal gyrus may be based onlexical or sublexical factors such as word length orgrammatical class especially considering the proximityof this activity to areas of the supramarginal gyrus im-plicated in phonological processing

In the first experiment we confirmed that activationfor action words encompassed the posterolateral tem-poral cortex but not the human MTMST In contrastprevious experiments with pictures have found greateractivation for action pictures compared to object pic-tures that extends more posteriorly and includes theMTMST (Kable Lease-Spellmeyer et al 2002 DamasioGrabowski Tranel Ponto et al 2001 Peigneux et al2000) Furthermore Martin et al (1995) observed asimilar anterior shift for words compared to pictures ina verb generation task These anteriorndashposterior differ-ences between words and pictures could be evidenceof a gradient of motion information represented in theoccipito-temporal cortex with areas closer to the MTMST representing more concrete visual information andareas closer to the middle temporal gyrus representingmore abstract propositional information Consistentwith this proposal the peak of greater activation foraction (and manipulable object) words in the presenttwo experiments was posterior to the peak of greateractivation for action words in our previous study Be-cause the current task used related distractors tofocus subjectsrsquo attention on the type of motion in-volved in different actions subjects in the currentstudy may have recruited comparatively more con-crete visual information to perform the task Further-more given this posterior shift we would predict thatthere would be a greater overlap between the areasactivated by action words and action pictures in theconceptual judgment task with related distractorscompared to the little overlap we observed previouslywith unrelated distractors (Kable Lease-Spellmeyeret al 2002) as the related distractor task constrainssubjects to use similar kinds of semantic informationwith both formats

In both experiments the difference between action(or manipulable object) and animal words in the mid-dle temporal gyrus was smaller than the differencebetween any of the conceptual conditions and the per-ceptual baseline As can be seen in Figure 3 there wasactivation in the lateral temporal cortex for the animaljudgments compared to the perceptual baseline taskHowever interpreting this activation or any activationreferenced to the baseline is difficult because the per-ceptual baseline task was not intended to be a pre-cise control for the conceptual judgment task (seeMethods) Thus much of the difference between ani-mal judgments and the baseline perceptual matchingtask may be explained by nonspecific factors such asattention or task difficulty That is not to say thatknowledge of animals might not be informed by motionattributes Rather our claim is that motion attributes

Kable et al 1863

make a relatively greater contribution to action andmanipulable object knowledge compared to animalknowledge

Other Regions

Left Frontal Cortex

In the first experiment activity in the left motor cortexwas consistently greater for action judgments comparedto object judgments In the second experiment we didnot replicate this effect in the left motor cortex al-though we did find greater activity for manipulableobject judgments compared to animal judgments Wealso failed to find differences in motor or premotorareas between action and object judgments in our pre-vious study This pattern of results provides weak sup-port for the idea that motor and premotor areas areimportant for conceptual representations of action

In contrast to our results many previous studies havefound an association between action-naming difficultiesand left frontal damage (Hillis Tuffiash Wityk amp Barker2002 Tranel Adolphs Damasio amp Damasio 2001Daniele et al 1994 Damasio amp Tranel 1993) or in-creased activity in the left frontal cortex for tool namingcompared to animal naming (Chao amp Martin 2000Grabowski Damasio amp Damasio 1998 Grafton FadigaArbib amp Rizzolatti 1997 Perani Cappa Bettinardi et al1995) However these findings could be explained byprocesses other than action semantics that have beenassociated with the left frontal cortex For example theleft frontal cortex has been associated with the mor-phological processing of verbs (Tyler Bright et al 2004Shapiro amp Caramazza 2003) which could explain action-naming deficits Differences in naming associated withthe left frontal cortex could also be due to more gen-eral processes such as selection from semantic memory(Thompson-Schill 2003 Thompson-Schill DrsquoEspositoamp Kan 1999 Thompson-Schill Swick et al 1998Thompson-Schill DrsquoEsposito Aguirre amp Farah 1997)that might be particularly stressed for certain catego-ries of stimuli Alternatively the left frontal cortex mayparticipate in representations of motor programs asso-ciated with different concepts (Hauk Johnsrude ampPulvermuller 2004 Kellenbach Brett amp Patterson2003 Kraut Moo Segal amp Hart 2002 MecklingerGruenewald Besson Magnie amp Von Cramon 2002Rizzolatti Fogassi amp Gallese 2001) but these motorattributes may play a less central role than sensoryones in the representation of action concepts gener-ally or at least for the specific stimuli we used Inaddition these different processes may explain effectsin different parts of the left frontal cortex Along theselines we recently found a dissociation between activ-ity during naming in the left ventral premotor cortexwhich was associated with manipulability and in theleft inferior frontal gyrus which was associated with se-

lection demands (Kan Kable van Scoyoc Chatterjee ampThompson-Schill in press)

Ventral Occipito-temporal Cortex

If the lateral occipito-temporal cortex plays an impor-tant role in action representations then one mightexpect the ventral occipito-temporal cortex to play asimilar role in object representations Consistent withthis notion we found greater activation for manipula-ble objects compared to actions in inferior temporal andfusiform areas However we only found greater activa-tion for animal judgments compared to actions in theright fusiform in the second experiment In both thefirst experiment and a previous study we failed to findgreater activation for animal judgments in ventral occi-pito-temporal areas One explanation for these incon-sistent findings may be that action concepts cannot beaccessed independently of the actors or objects involvedin the action Consequently one would predict thatobject-specific areas in the ventral occipito-temporalcortex would not respond differentially depending onwhether the object was participating in an action Thiswas exactly the pattern reported in a recent experimentby Beauchamp et al (2002) Ventral occipito-temporalareas showed object-specific responses but these re-sponses were not modulated by whether the objectwas moving whereas lateral occipito-temporal areasshowed greater responses for moving compared to sta-tionary stimuli

Retrosplenial Cortex

Across both experiments we found consistently greateractivity for object words in two unanticipated regionsone of which was the left retrosplenial cortex In thesecond experiment left retrosplenial activity was greaterfor both animals and manipulable objects compared toactions The retrosplenial cortex has previously beenassociated with episodic memory processing Damage tothe retrosplenial cortex is associated with memory def-icits (Valenstein et al 1987) and damage to the leftretrosplenial cortex is associated with verbal memorydeficits (McDonald Crosson Valenstein amp Bowers2001) In addition functional imaging studies find retro-splenial activation associated with episodic encoding(Shallice et al 1994) In both experiments of the cur-rent study the object words were more concrete thanthe action words and concrete words are better remem-bered than abstract words Thus one possible interpre-tation of our results is that left retrosplenial activity isrelated to encoding words into episodic memory aprocess unrelated to performing the conceptual judg-ment task Consistent with this explanation subjectsperforming the same conceptual judgment tasks usedin the two current experiments remember more object


words from the task than action words in a free recallparadigm (data not shown)

Superior Frontal Cortex

The superior frontal cortex was the second unantici-pated region showing greater activity for objects com-pared to actions In the second experiment the superiorfrontal cortex showed greater activity only for animalscompared to actions Previously activity was found inportions of the anterior superior frontal gyrus duringa lexical decision task as well as during other tasksthought to involve accessing word meaning (Binderet al 2003) A nearby region of the superior frontalgyrus shows greater activity during naming of uniquepersons (Damasio Tranel et al 2004) or judgmentsabout persons (Mitchell Heatherton amp Macrae 2002)Given this pattern of results the anterior superiorfrontal cortex might be involved in conceptual repre-sentations of animate objects

Conclusion

Accessing knowledge of actions and manipulable ob-jects causes distinct activations in the posterior lateraltemporal cortex which overlap in the middle temporalgyrus These findings are compatible with the generalview that conceptual knowledge is instantiated by dis-tributed neural regions partially organized along sen-sorimotor lines Specifically these data support theidea that the lateral temporal cortex is important in me-diating representations of motion features central toaction concepts Furthermore we suggest that differentmotion representations are organized along a concreteto abstract gradient which extends from the occipito-temporal junction into the middle temporal gyrus andperi-sylvian cortices

METHODS

Subjects

Eight right-handed individuals participated in the firstexperiment (4 women 4 men mean age = 23 years) Aninth subject was excluded from the study because ofexcessive head movement in the scanner (gt2 mm acrossthe scanning session) Nine right-handed individualsparticipated in the second experiment (5 women4 men mean age = 21 years) Given that sex dif-ferences have previously been reported during lan-guage tasks (Grabowski Damasio Eichhorn amp Tranel2003) we tried to balance the number of men andwomen in each experiment All subjects were recruitedfrom the university community and all spoke onlyEnglish before school age None had a history ofneurologic or psychiatric symptoms All subjects gaveinformed consent in accordance with the procedures

of the Institutional Review Board of the University ofPennsylvania

Behavioral Tasks

In both experiments subjects performed a conceptualsimilarity judgment task with words during scanning(see Figure 1) In each trial subjects were presentedwith three words one at the top of the screen and twoat the bottom Subjects pressed a button (right or lefthand) to indicate which of the two words at the bottomwas more closely related in meaning to the word at thetop In baseline trials subjects were presented withthree false font strings and they indicated which ofthe two at the bottom of the screen was the same asthe one on top

The stimuli from the first experiment consisted of40 action word triads and 40 object word triads All threewords in a triad referred to actions or objects from thesame general category Action categories included man-ners of motion paths of motion and movements of dif-ferent body parts Object categories were restricted toanimals fruits and vegetables trees flowers and foodThe stimuli for the second experiment consisted of32 action word triads 32 animal word triads and32 manipulable object word triads Manipulable objectsincluded the categories of tools kitchen utensils clean-ing implements office implements and weapons Be-cause the second experiment was partially designedto test the effect of grammatical category words thatcould refer to both a tool and the use of that tool (eglsquolsquohammerrsquorsquo) were not used in either the action ormanipulable object categories Given that English wordscan frequently be used as both nouns and verbs itwould have been impossible to exclude all nounndashverbhomonyms However the context of comparing threerelated words as well as the consistent ending of actionwords served to constrain subjectsrsquo interpretations ofwords as denoting objects or actions

Because in all cases the three words in a triad werefrom the same category they tended to share severalsemantic attributes and the choice was usually distin-guished on the basis of one salient feature Using threewords from the same category also reduced the impor-tance of syntactic information in making judgmentsabout action words Although the syntactic frames ofa verb constrains verb meaning this information bestdistinguishes different categories of actions (FisherGleitman amp Gleitman 1991) For example motion verbslike lsquolsquowalkrsquorsquo and lsquolsquoslidersquorsquo can both occur in sentenceswith one noun argument and a prepositional phrasewhereas cognition verbs like lsquolsquothinkrsquorsquo and lsquolsquobelieversquorsquo canboth occur in sentences with one noun argument anda sentential complement

Triads were selected from a larger initial set on thebasis of pilot testing so that response agreement (theproportion of subjects choosing a particular response)

Kable et al 1865

and reaction time were similar across conditions Wetested 11 pilot subjects for the first experiment (9women 2 men mean age = 27 years) and 12 pilotsubjects for the second experiment (5 women 7 menmean age = 19 years) Equal numbers of left and rightresponses were chosen in each condition No triadswere repeated although individual words could be re-peated once Because many English words can refer toactions or objects all action words were presented inthe present participle form ending in lsquolsquo-ingrsquorsquo Althoughthis consistent ending might have created a potentialconfound (by making action words longer for exam-ple) we felt it was necessary to ensure that subjects un-derstood the words as clearly referring to actionsBaseline trials were created by replacing one of thechoice words with the top word and changing all threewords to strings in Wingdings font

We validated these tasks in a second pilot studywith 24 right-handed subjects (20 women 4 men meanage = 21 years) Twelve subjects performed the con-ceptual similarity judgment task from the first experi-ment and 12 performed the task from the secondexperiment In addition because eye movements cancause differences in neural activity (Barton et al 1996)we recorded eye position using an iView RED-II cam-era system (Sensomotoric Instruments Boston MAwwwsmide) to verify that eye movements did notsystematically differ across conditions For each con-dition we measured subjectsrsquo accuracy reaction timeand number of fixations Fixations an index of theamount of eye movements were defined as gaze po-sition remaining with a restricted area (radius lessthan approximately 2 mm) for greater than 100 msecThese data are reported in Table 3 There were nodifferences across conditions in reaction time or num-ber of fixations

In evaluating subjects we consider a response correctif it was the one chosen by the majority of pilot subjectsIn this context subjects could make an lsquolsquoincorrectrsquorsquo

choice for valid reasons based on a semantic propertythat choice shares with the target For our purposes itwas only important that subjects engaged in the taskand accessed semantic information in making theirjudgment which seemed likely given their 90 accu-racies in each condition

We decided to choose stimuli on the basis of reactiontime rather than on various lexical variables Values ofthese different variables are provided in Table 4 Weprioritized reaction time because it could clearly have aneffect on neural activity as the fMR response is sensitiveto the length as well as amount of neural activity Asalready mentioned action words were longer becauseof the morphological ending and verbs are systemati-cally rated lower in concreteness and imageability Pre-vious investigators have noted this confounding ofconceptual categories with lexical variables and arguedthat matching can result in idiosyncratic stimulus sets(Tranel Adolphs et al 2001) However we are sensi-tive to the possibility that these variables may affectactivation which forms part of the motivation for testingmultiple categories in the second experiment Becausethe manipulable object words are similar to animals inlength concreteness and imageability activation forboth actions and manipulable objects rules out thesepotential confounding explanations

In the first experiment subjects also participated ina scan to functionally localize the MTMST In this scansubjects passively viewed alternating 16-sec blocks ofradially moving rings and stationary rings The radiallymoving rings alternated between inward and outwardmotion every 2 sec

Stimulus Presentation

Stimuli were presented to the subjects using a high-resolution fiber-optic presentation system carrying in-formation from an external computer to goggles attachedto the head coil (Avotec Stuart FL wwwavotecorg)Timing of stimulus presentation was synchronized withthe scanner so that a new stimulus was always pre-sented at the beginning of a repetition Subject responseswere transmitted from the scanner room using a custom-designed fiber-optic motor response-recording device(Current Design Concepts Philadelphia PA) Stimuluspresentation and recording of responses were con-trolled by Psyscope software (psyscopepsycmuedu)for the conceptual tasks and Pixx software (psychologyconcordiacadepartmentCVLabCVLabhtml) for the vi-sual motion scan

MRI Acquisition

BOLD-sensitive T2-weighted fMRI data were acquiredon a 15-T GE Signa scanner using a gradient-echo echo-planar pulse sequence (TR = 2000 msec TEeff =

Table 3 Results (Mean plusmn Standard Error) of Pilot Study forStimuli Used in Experiments 1 and 2

TotalFixations

Accuracy()

Reaction Time(msec)

Experiment 1

Actions 846 plusmn 036 93 plusmn 1 1710 plusmn 77

Objects 835 plusmn 044 92 plusmn 1 1702 plusmn 73

Experiment 2


Manipulable Objects 788 plusmn 022 84 plusmn 2 1984 plusmn 78

Animals 779 plusmn 021 88 plusmn 2 1913 plusmn 80


50 msec flip angle = 908) The scanner was equippedwith a standard clinical quadrature radio-frequencyhead coil and a prototype gradient system (GESR230)capable of ultrafast imaging Data were acquired intwenty-one 5-mm axial slices covering the entire cortexbut omitting lower portions of the cerebellum andbrainstem The resolution within each slice was 375 375 mm (64 64 matrix within a 24-cm field ofview) Phase maps were acquired to correct for distor-tion in the echo-planar images Head motion was min-imized using foam padding Subjects performed no taskduring the first 20 sec of each scan as steady-statemagnetization was achieved Subjects in the first exper-iment completed two 6-min 20-sec scans while per-forming the conceptual judgment task for a total of360 observations per voxel per subject Subjects in thesecond experiment completed two 7-min 32-sec scanswhile performing the conceptual judgment task for atotal of 432 observations per voxel per subject For eachsubject high-resolution T1-weighted axial and sagittalscans were also acquired using a spin-echo sequence(TR = 500 msec TE = 8 msec flip angle = 908)

Data Analysis

Statistical Tests

Data processing was performed off-line using soft-ware developed at the University of Pennsylvania(wwwvoxboorg) After reconstructing images fromthe raw data data were sinc-interpolated in time to cor-rect for staggered slice acquisition realigned to the firstimage acquired for each subject using a six-parametermotion-correction algorithm and thresholded to ex-clude extra-parenchymal voxels from subsequent ana-lyses Within each subject a voxelwise analysis wasperformed using a version of the modified general linearmodel Included in this model were covariates modelingdifferent task conditions (eg lsquolsquoaction judgmentsrsquorsquo) a

subject-specific estimate of the intrinsic temporal auto-correlation and sine and cosine regressors for frequen-cies below those of the task and for frequencies in theelevated range of the noise spectrum Task covariateswere boxcar waveforms convolved with an estimate ofthe BOLD hemodynamic transfer function empiricallyderived from the motor cortex in a large group ofsubjects (Aguirre Zarahn amp DrsquoEsposito 1998) Toimprove specificity data were also smoothed in timewith the hemodynamic transfer function These meth-ods have been empirically demonstrated to controlthe false-positive rate (Aguirre Zarahn amp DrsquoEsposito1997 Zarahn Aguirre amp DrsquoEsposito 1997)

In the first experiment reaction times for our twoconditions of interest differed for the subjects partici-pating in the fMRI experiment even though they did notin our pilot testing Because reaction time differencesalone could account for activation differences we alsoincluded in our statistical model regressors that covariedout effects of reaction time differences (Kable Kimbergamp Chatterjee 2004) Using individual subject reactiontimes for each trial we modeled effects of reaction timefor the conceptual task (collapsed across action andobject conditions) using orthogonal polynomial covari-ates (orthogonalized with respect to the main effect ofthe conceptual task) (Buchel Holmes Rees amp Friston1998) Because reaction times were longer for actionjudgments than for object judgments covarying forreaction time reduces the amount of activation foractions (and increases the amount for objects) in thedirect contrast between these conditions

We tested hypotheses about activity differences be-tween conditions using an ROI approach After definingROIs we performed a random-effects group analysis asfollows Initially voxels were identified within eachsubject where activity was significant for the main effectof conceptual judgment compared to baseline condi-tions ( p lt 05 Bonferroni-corrected for multiple com-parisons within an ROI) Unlike in cognitive subtraction

Table 4 Values of Psycholinguistic Variables for the Words Used in Each Condition

Letter Length Syllable Length Log Frequency Familiarity Concreteness Imageability

Experiment 1

Actions 78 plusmn 01 (120) 21 plusmn 003 (120) 112 plusmn 006 (108) 529 plusmn 6 (69) 487 plusmn 10 (64) 513 plusmn 8 (68)

Objects 55 plusmn 02 (120) 17 plusmn 01 (120) 088 plusmn 006 (86) 509 plusmn 6 (82) 602 plusmn 3 (82) 593 plusmn 4 (82)

Experiment 2


Manipulable Objects 56 plusmn 02 (96) 16 plusmn 01 (96) 097 plusmn 008 (67) 505 plusmn 8 (70) 579 plusmn 6 (61) 563 plusmn 6 (61)

Animals 58 plusmn 02 (96) 18 plusmn 01 (96) 070 plusmn 007 (61) 476 plusmn 9 (64) 608 plusmn 3 (58) 595 plusmn 5 (58)

Values are from the MRC Psycholinguistic Database and are reported as mean plusmn standard error Numbers in parentheses are the number ofwords for which that variable was listed in the database

Kable et al 1867

paradigms the baseline condition was not designed tobe closely matched to the conditions of interest Ratherthe main effect comparison simply reduced the numberof voxels within which specific hypotheses about differ-ential activity for the contrasts of interest were testedThe fMRI time series was averaged for all voxels signif-icant for the main effect within a defined ROI in eachsubject Next a measure of the effect size for the or-thogonal contrast between conditions of interest (egactions minus objects in the first experiment or actionsminus animals in the second experiment) was extractedfrom the spatially averaged ROI time series in eachsubject We used t values as a measure of effect sizerather than percent signal change because the residualerror term in the denominator of the t value mosteffectively corrects the effect size for scaling effectsdue to differences in overall MR signal intensity acrossscanning sessions (Postle Zarahn amp DrsquoEsposito 2000)Finally paired t tests addressed whether this effectwas consistently greater for one condition or the otherin that ROI across subjects This analysis gains sensi-tivity by reducing the effect size in each ROI to onevalue and avoiding the need to correct for multiplecomparisons across voxels although it loses spatialresolution by averaging across all active voxels withinan ROI

We performed an additional analysis to compare theresults from this study to those of our previous studyIn particular a shift in the location of activation withinthe lateral occipito-temporal cortex between the twostudies might not be revealed by our ROI analyses Inorder to look for such a shift we derived unthresh-olded t-maps for the direct contrast of actions andobjects in each subject These t-maps were then nor-malized into the standard coordinate space of the MNIbrain using a 12-parameter affine transformation withnonlinear deformations Normalization parameters wereinitially calculated using each subjectrsquos high-resolutionanatomical scan After normalization the maps werespatially smoothed with a kernel with full-width half-maximum of 75 mm We found the location of thevoxel with the peak difference between actions andobjects in the lateral occipito-temporal cortex in eachsubject and computed the average location in MNIcoordinates of this peak across subjects We performeda similar analysis on the data from our previous studyand used t tests to compare the location of the peakdifference in this study to the location of the peakdifferences (for both words and pictures) in our previ-ous study

We also used the normalized unthresholded t-mapsfrom the first experiment in a hypothesis-generatingwhole-brain analysis Specifically we performed a ran-dom-effects analysis at each voxel by testing whetherthe difference between actions and object was signifi-cantly different from zero across subjects Becausethis analysis was exploratory we thresholded the results

at a height threshold of p lt 01 and an extent thresholdof 5 voxels

Definition of ROIs

The locations of anatomical ROIs were determinedwith reference to a combined MRIanatomical atlasAnatomical ROIs were traced on each subjectrsquos high-resolution T1-weighted image by one author andchecked by a second author The lateral orbital ROIincluded the lateral orbital gyrus inferior to the infe-rior frontal ROI The lateral premotor ROI correspondedto Brodmannrsquos area 6 laterally including the cortexsurrounding the precentral sulcus below the superiorfrontal gyrus The motor ROI included the precentralgyrus excluding any areas in the premotor ROI Thesuperior frontal ROI included the anterior half of thesuperior frontal gyrus The fusiform ROI included boththe occipital and temporal portions of this gyrus Theinferior middle and superior temporal ROIs includedthe posterior two-thirds of these gyri The inferiorparietal ROI included the angular and supramarginalgyri corresponding to Brodmannrsquos areas 39 and 40The middle occipital ROI included all of the middleoccipital gyrus The retrosplenial ROI included portionsof the posterior cingulate cortex immediately surround-ing the splenium

For the first experiment the MTMST ROI wasfunctionally identified in each subject in a separate lo-calizer scan In accordance with previous studies wedefined the visual motion cortex on the basis of thecontrast between moving rings and stationary rings Weincluded in the MTMST all significant voxels contigu-ous with the peak activation on the lateral occipitalsurface This region appears to be the human homo-logue of motion-sensitive cortical areas in the macaqueincluding the MT and MST (Tootell et al 1995 Watsonet al 1993) Within this group of subjects both thesize of the MTMST (20 voxelshemisphere) and itslocation (within a sulcus near the junction of the in-ferior temporal sulcus and the ascending limb of theinferior temporal sulcus) were consistent with previousstudies (Dumoulin et al 2000)

Acknowledgments

This research was supported by an NSF graduate fellowship toJ K and NIH grant R01 DC004817 to A C We thank ScottGreenwald for running some of the pilot experiments andGeoff Aguirre John Detre and Dan Kimberg for discus-sions about issues involved in covarying for reaction time infMRI analyses

Reprint requests should be sent to Anjan Chatterjee Depart-ment of Neurology University of Pennsylvania 3 West GatesBuilding 3400 Spruce Street Philadelphia PA 19104-4283 orvia e-mail anjanmailmedupennedu


The data reported in this experiment have been depositedwith the fMRI Data Center archive (wwwfmridcorg) Theaccession number is 2-2005-118X8

REFERENCES

Aguirre G K Zarahn E amp DrsquoEsposito M (1997) Empiricalanalyses of BOLD fMRI statistics II Spatially smoothed datacollected under null-hypothesis and experimentalconditions Neuroimage 5 199ndash212

Aguirre G K Zarahn E amp DrsquoEsposito M (1998) Thevariability of human BOLD hemodynamic responsesNeuroimage 8 360ndash369

Barton J J Simpson T Kiriakopoulos E Stewart CCrawley A Guthrie B Wood M amp Mikulis D (1996)Functional MRI of lateral occipitotemporal cortex duringpursuit and motion perception Annals of Neurology 40387ndash398

Beauchamp M S Lee K E Haxby J V amp Martin A (2002)Parallel visual motion processing streams formanipulable objects and human movements Neuron34 149ndash159

Beauchamp M S Lee K E Haxby J V amp Martin A (2003)FMRI responses to video and point-light displays of movinghumans and manipulable objects Journal of CognitiveNeuroscience 15 991ndash1001

Binder J R McKiernan K A Parsons M E Westbury C FPossing E T Kaufman J N amp Buchanan L (2003) Neuralcorrelates of lexical access during visual word recognitionJournal of Cognitive Neuroscience 15 372ndash393

Buchel C Holmes A P Rees G amp Friston K J (1998)Characterizing stimulusndashresponse functions using nonlinearregressors in parametric fMRI experiments Neuroimage 8140ndash148

Buxbaum L J amp Saffran E M (2002) Knowledge of objectmanipulation and object function Dissociations in apraxicand nonapraxic subjects Brain and Language 82179ndash199

Cappa S F Perani D Schnur T Tettamanti M amp Fazio F(1998) The effects of semantic category and knowledge typeon lexicalndashsemantic access A PET study Neuroimage 8350ndash359

Caramazza A Hillis A E Rapp B C amp Romani C (1990)The multiple semantics hypothesis Multiple confusionsCognitive Neuropsychology 7 161ndash189

Caramazza A amp Mahon B Z (2003) The organization ofconceptual knowledge The evidence from category-specificsemantic deficits Trends in Cognitive Sciences 7354ndash361

Chao L L Haxby J V amp Martin A (1999) Attribute-basedneural substrates in temporal cortex for perceiving andknowing about objects Nature Neuroscience 2 913ndash919

Chao L L amp Martin A (2000) Representation of manipulableman-made objects in the dorsal stream Neuroimage 12478ndash484

Damasio A R amp Tranel D (1993) Nouns and verbs areretrieved with differently distributed neural systemsProceedings of the National Academy of Sciences USA 904957ndash4960

Damasio H Grabowski T J Tranel D Hichwa R D ampDamasio A R (1996) A neural basis for lexical retrievalNature 380 499ndash505

Damasio H Grabowski T J Tranel D Ponto L LHichwa R D amp Damasio A R (2001) Neural correlatesof naming actions and of naming spatial relationsNeuroimage 13 1053ndash1064

Damasio H Tranel D Grabowski T Adolphs R amp

Damasio A (2004) Neural systems behind word andconcept retrieval Cognition 92 179ndash229

Daniele A Giustolisi L Silveri M C Colosimo C ampGainotti G (1994) Evidence for a possible neuroanatomicalbasis for lexical processing of nouns and verbsNeuropsychologia 32 1325ndash1341

Dumoulin S O Bittar R G Kabani N J Baker C L JrLe Goualher G Bruce Pike G amp Evans A C (2000)A new anatomical landmark for reliable identification ofhuman area V5MT A quantitative analysis of sulcalpatterning Cerebral Cortex 10 454ndash463

Fisher C Gleitman H amp Gleitman L R (1991) On thesemantic content of subcategorization frames CognitivePsychology 23 331ndash392

Gainotti G (2000) What the locus of brain lesion tells usabout the nature of the cognitive defect underlyingcategory-specific disorders A review Cortex 36 539ndash559

Grabowski T J Damasio H amp Damasio A R (1998)Premotor and prefrontal correlates of category-relatedlexical retrieval Neuroimage 7 232ndash243

Grabowski T J Damasio H Eichhorn G R amp Tranel D(2003) Effects of gender on blood flow correlates ofnaming concrete entities Neuroimage 20 940ndash954

Grafton S T Fadiga L Arbib M A amp Rizzolatti G (1997)Premotor cortex activation during observation and namingof familiar tools Neuroimage 6 231ndash236

Hauk O Johnsrude I amp Pulvermuller F (2004)Somatotopic representation of action words in humanmotor and premotor cortex Neuron 41 301ndash307

Hillis A E Tuffiash E Wityk R J amp Barker P B (2002)Regions of neural dysfunction associated with impairednaming of actions and objects in acute stroke CognitiveNeuropsychology 19 523ndash534

Kable J W Kimberg D Y amp Chatterjee A (2004) Dealingwith potential reaction time confounds in theinterpretation of fMRI results Paper presented at theCognitive Neuroscience Society Eleventh Annual MeetingSan Francisco CA

Kable J W Lease-Spellmeyer J amp Chatterjee A (2002)Neural substrates of action event knowledge Journal ofCognitive Neuroscience 14 795ndash805

Kan I P Kable J W van Scoyoc A Chatterjee A ampThompson-Schill S L (in press) Tools Dissociable effectsof motor experience and lexical competition Journal ofCognitive Neuroscience

Kellenbach M L Brett M amp Patterson K (2003) Actionsspeak louder than functions The importance ofmanipulability and action in tool representation Journalof Cognitive Neuroscience 15 30ndash46

Kraut M A Moo L R Segal J B amp Hart J Jr (2002)Neural activation during an explicit categorization taskCategory- or feature-specific effects Brain ResearchCognitive Brain Research 13 213ndash220

Martin A Haxby J V Lalonde F M Wiggs C L ampUngerleider L G (1995) Discrete cortical regionsassociated with knowledge of color and knowledge ofaction Science 270 102ndash105

McCarthy R amp Warrington E K (1985) Category specificityin an agrammatic patient The relative impairment ofverb retrieval and comprehension Neuropsychologia 23709ndash727

McDonald C R Crosson B Valenstein E amp Bowers D(2001) Verbal encoding deficits in a patient with a leftretrosplenial lesion Neurocase 7 407ndash417

McRae K de Sa V R amp Seidenberg M S (1997) On thenature and scope of featural representations of wordmeaning Journal of Experimental Psychology General126 99ndash130

Kable et al 1869

Mecklinger A Gruenewald C Besson M Magnie M N ampVon Cramon D Y (2002) Separable neuronal circuitries formanipulable and non-manipulable objects in workingmemory Cerebral Cortex 12 1115ndash1123

Miceli G Silveri M C Nocentini U amp Caramazza A (1988)Patterns of dissociation in comprehension and production ofnouns and verbs Aphasiology 2 351ndash358

Mitchell J P Heatherton T F amp Macrae C N (2002)Distinct neural systems subserve person and objectknowledge Proceedings of the National Academy ofSciences USA 99 15238ndash15243

Peigneux P Salmon E van der Linden M Garraux GAerts J Delfiore G Degueldre C Luxen A Orban Gamp Franck G (2000) The role of lateral occipitotemporaljunction and area MTV5 in the visual analysis ofupper-limb postures Neuroimage 11 644ndash655

Perani D Cappa S F Bettinardi V Bressi S Gorno-Tempini M Matarrese M amp Fazio F (1995) Differentneural systems for the recognition of animals and man-madetools NeuroReport 6 1637ndash1641

Perani D Cappa S F Schnur T Tettamanti M CollinaS Rosa M M amp Fazio F (1999) The neural correlatesof verb and noun processing A PET study Brain 1222337ndash2344

Perani D Schnur T Tettamanti M Gorno-Tempini MCappa S F amp Fazio F (1999) Word and picture matchingA PET study of semantic category effects Neuropsychologia37 293ndash306

Postle B R Zarahn E amp DrsquoEsposito M (2000) Usingevent-related fMRI to assess delay-period activity duringperformance of spatial and nonspatial working memorytasks Brain Research Brain Research Protocols 5 57ndash66

Pulvermuller F (1999) Words in the brainrsquos languageBehavioral and Brain Sciences 22 253ndash336

Rizzolatti G Fogassi L amp Gallese V (2001)Neurophysiological mechanisms underlying theunderstanding and imitation of action Nature ReviewsNeuroscience 2 661ndash670

Saffran E M Coslett H B amp Keener M T (2003)Differences in word associations to pictures and wordsNeuropsychologia 41 1541ndash1546

Saffran E M amp Schwartz M F (1994) Of cabbages andthings Semantic memory from a neuropsychologicalperspectivemdashA tutorial review In C Umilta amp M Moscovitch(Eds) Attention and performance XV Conscious andnonconscious information processing (pp 507ndash536)Cambridge MIT Press

Shallice T Fletcher P Frith C D Grasby P FrackowiakR S amp Dolan R J (1994) Brain regions associated withacquisition and retrieval of verbal episodic memory Nature368 633ndash635

Shapiro K amp Caramazza A (2003) The representation ofgrammatical categories in the brain Trends in CognitiveSciences 7 201ndash206

Simmons W K amp Barsalou L W (2003) The similarity-in-topography principle Reconciling theories of conceptualdeficits Cognitive Neuropsychology 20 451ndash486

Thompson-Schill S L (2003) Neuroimaging studies ofsemantic memory Inferring lsquolsquohowrsquorsquo from lsquolsquowherersquorsquoNeuropsychologia 41 280ndash292

Thompson-Schill S L Aguirre G K DrsquoEsposito M ampFarah M J (1999) A neural basis for category and modalityspecificity of semantic knowledge Neuropsychologia 37671ndash676

Thompson-Schill S L DrsquoEsposito M Aguirre G K ampFarah M J (1997) Role of left inferior prefrontal cortex inretrieval of semantic knowledge A reevaluationProceedings of the National Academy of Sciences USA94 14792ndash14797

Thompson-Schill S L DrsquoEsposito M amp Kan I P (1999)Effects of repetition and competition on activity in leftprefrontal cortex during word generation Neuron 23513ndash522

Thompson-Schill S L Swick D Farah M J DrsquoEsposito MKan I P amp Knight R T (1998) Verb generation in patientswith focal frontal lesions A neuropsychological test ofneuroimaging findings Proceedings of the NationalAcademy of Sciences USA 95 15855ndash15860

Tootell R B Reppas J B Kwong K K Malach R BornR T Brady T J Rosen B R amp Belliveau J W (1995)Functional analysis of human MT and related visual corticalareas using magnetic resonance imaging Journal ofNeuroscience 15 3215ndash3230

Tranel D Adolphs R Damasio H amp Damasio A R (2001) Aneural basis for the retrieval of words for actions CognitiveNeuropsychology 18 655ndash674

Tranel D Damasio H amp Damasio A R (1997) A neuralbasis for the retrieval of conceptual knowledgeNeuropsychologia 35 1319ndash1327

Tranel D Kemmerer D Adolphs R Damasio H ampDamasio A R (2003) Neural correlates of conceptualknowledge for actions Cognitive Neuropsychology 20409ndash432

Tranel D Martin C Damasio H Grabowski T J ampHichwa R (in press) Effects of nounndashverb homonymyon the neural correlates of naming concrete entities andactions Brain and Language

Tyler L K Bright P Fletcher P amp Stamatakis E A (2004)Neural processing of nouns and verbs The role ofinflectional morphology Neuropsychologia 42 512ndash523

Tyler L K amp Moss H E (2001) Towards a distributedaccount of conceptual knowledge Trends in CognitiveSciences 5 244ndash252

Tyler L K Russell R Fadili J amp Moss H E (2001) Theneural representation of nouns and verbs PET studiesBrain 124 1619ndash1634

Tyler L K Stamatakis E A Dick E Bright P Fletcher Pamp Moss H (2003) Objects and their actions Evidence fora neurally distributed semantic system Neuroimage 18542ndash557

Valenstein E Bowers D Verfaellie M Heilman K MDay A amp Watson R T (1987) Retrosplenial amnesiaBrain 110 1631ndash1646

Vinson D P amp Vigliocco G (2002) A semantic analysis ofgrammatic class impairments Semantic representations ofobject nouns action nouns and action verbs Journal ofNeurolinguistics 15 317ndash351

Warrington E K amp Shallice T (1984) Category specificsemantic impairments Brain 107 829ndash854

Watson J D Myers R Frackowiak R S Hajnal J V WoodsR P Mazziotta J C Shipp S amp Zeki S (1993) Area V5of the human brain Evidence from a combined study usingpositron emission tomography and magnetic resonanceimaging Cerebral Cortex 3 79ndash94

Zarahn E Aguirre G K amp DrsquoEsposito M (1997) Empiricalanalyses of BOLD fMRI statistics I Spatially unsmootheddata collected under null-hypothesis conditionsNeuroimage 5 179ndash197


greater degree by actions or objects (Tranel MartinDamasio Grabowski amp Hichwa in press Kable Lease-Spellmeyer amp Chatterjee 2002 Damasio GrabowskiTranel Ponto et al 2001 Perani Cappa et al 1999Martin Haxby Lalonde Wiggs amp Ungerleider 1995) Anunderstanding of the neural substrates of actions com-pared to different categories of objects should provideimportant evidence for evaluating theories of concep-tual knowledge systems as well as a more complete pic-ture of the neural mediation of conceptual knowledge

We directly investigated the neural substrates of ac-tion knowledge in a previous fMRI study (Kable Lease-Spellmeyer et al 2002) Subjects saw three pictures orthree words simultaneously one at the top of the screenand two at the bottom and they were asked to deter-mine which of the two concepts presented at thebottom was most similar to the one presented at thetop For example subjects would decide that lsquolsquodiggingrsquorsquowas more similar to lsquolsquoshovelingrsquorsquo than lsquolsquolisteningrsquorsquo or thatlsquolsquodoctorrsquorsquo was more similar to lsquolsquocoprsquorsquo than lsquolsquozucchinirsquorsquoSubjects performed the same task once with picturesand once with words When subjects performed the taskwith pictures there was greater activity for actionscompared to objects bilaterally in human motion areaMTMST and in nearby areas of the middle temporalgyrus When subjects performed the task with wordsthere was greater activity for actions compared toobjects in the left middle and superior temporal gyrianterior and dorsal to the activation for pictures Actionwords unlike action pictures did not preferentially ac-tivate area MTMST

These results suggest that parts of the lateral occipito-temporal cortex are particularly important for me-diating conceptual knowledge of actions In additionconsidering the involvement of the MTMST the senso-rimotor theory of conceptual knowledge provides themost natural explanation for these findings Within thisframework we argued that the lateral occipito-temporalcortex likely mediates different representations of mo-tion which are presumably more relevant for actionconcepts compared to object concepts Furthermorewe suggested that these motion representations are or-ganized on a concrete to abstract gradient with moreconcrete representations preferentially accessed frompictures located more posteriorly near the MTMSTand more abstract representations preferentially ac-cessed from words located more anteriorly near theperi-sylvian cortex

However there are alternative possible explanationsfor our findings concerning action words especiallygiven the small overlap between areas showing greateractivation for action words and action pictures Actionwords may have failed to activate area MTMST be-cause the judgments for the particular stimuli used didnot stress detailed motion information and not be-cause of any intrinsic material specificity Furthermoreactivation differences in the lateral temporal cortex

may have been due to lexical differences between theaction and object words rather than the differences inconceptual category In particular all of the actionwords were verbs whereas all of the object words werenouns so perhaps grammatical processing accounts forthese activation differences Although other investiga-tors have failed to detect significant differences betweenverbs and nouns in the lateral temporal cortex (TylerBright Fletcher amp Stamatakis 2004 Tyler Stamatakiset al 2003 Tyler Russell Fadili amp Moss 2001) thisfailure highlights the additional possibility that our pre-vious findings with action words may have been idio-syncratic to the particular stimuli or subjects

We designed two fMRI experiments to test thesealternative hypotheses Both experiments test the hy-pothesis that activity for words in the lateral temporalcortex is driven by semantic differences as we previ-ously argued In the first experiment we designed newword triads for the conceptual judgment task thatfocused subjectsrsquo attention on a more constrained setof visual attributes (see Figure 1) In our previous studybecause the distractor item on each trial was oftendistantly related to the other two items (lsquolsquolisteningrsquorsquocompared to lsquolsquodiggingrsquorsquo and lsquolsquoshovelingrsquorsquo) subjects couldhave made their judgments with only minimal access tovisual attributes of different actions (unlike lsquolsquolisteningrsquorsquolsquolsquodiggingrsquorsquo and lsquolsquoshovelingrsquorsquo involve motion) We de-signed new stimuli in which all three words were closelyrelated to increase the importance of visual attributes inmaking the judgment For example subjects would judgethat lsquolsquoskippingrsquorsquo is more similar to lsquolsquobouncingrsquorsquo thanlsquolsquorollingrsquorsquo as both skipping and bouncing have similarmanners of motion or they would judge that lsquolsquozebrarsquorsquo ismore similar to lsquolsquohorsersquorsquo than lsquolsquocowrsquorsquo as both zebras andhorses have similar forms If activation in the lateraltemporal cortex is associated with accessing semanticinformation about actions then we would expect toreplicate our original finding of greater activation foraction words compared to object words Furthermore ifour hypothesis about a gradient of action representa-tions is correct we would expect the activation for actionwords to be closer to the MTMST in the current exper-iment because these stimuli emphasized visual infor-mation about actions at a greater level of detail than inour previous study

In the second experiment we again used the concep-tual judgment task but with three different categoriesof words actions animals and manipulable objectsAlthough the words for both animals and manipulableobjects are nouns manipulable object concepts morethan animal concepts rely on information about move-ment (Damasio Tranel et al 2004 Saffran Coslett ampKeener 2003 Buxbaum amp Saffran 2002 McRae de Sa ampSeidenberg 1997) Unlike animals most manipulableobjects are strongly associated with a constrained setof actions So if the previously observed activation dif-ferences between action and object words were driven



RESULTS

Experiment 1

Behavioral Results


Imaging Results





Experiment 1




Experiment 2










Kable et al 1857












Experiment 2

Behavioral Results


Imaging Results




Actions Objects

Actions Animals







Other ROIs










Kable et al 1859

































DISCUSSION




Kable et al 1861














Kable et al 1863


Other Regions

Left Frontal Cortex












Conclusion


METHODS

Subjects



Behavioral Tasks





Kable et al 1865









MRI Acquisition



TotalFixations

Accuracy()

Reaction Time(msec)

Experiment 1



Experiment 2






Data Analysis

Statistical Tests







Experiment 1



Experiment 2





Kable et al 1867





Definition of ROIs



Acknowledgments





REFERENCES





































Kable et al 1869





































RESULTS

Experiment 1

Behavioral Results


Imaging Results





Experiment 1




Experiment 2










Kable et al 1857












Experiment 2

Behavioral Results


Imaging Results




Actions Objects

Actions Animals







Other ROIs










Kable et al 1859

































DISCUSSION




Kable et al 1861














Kable et al 1863


Other Regions

Left Frontal Cortex












Conclusion


METHODS

Subjects



Behavioral Tasks





Kable et al 1865









MRI Acquisition



TotalFixations

Accuracy()

Reaction Time(msec)

Experiment 1



Experiment 2






Data Analysis

Statistical Tests







Experiment 1



Experiment 2





Kable et al 1867





Definition of ROIs



Acknowledgments





REFERENCES





































Kable et al 1869















































Experiment 2

Behavioral Results


Imaging Results




Actions Objects

Actions Animals







Other ROIs










Kable et al 1859

































DISCUSSION




Kable et al 1861














Kable et al 1863


Other Regions

Left Frontal Cortex












Conclusion


METHODS

Subjects



Behavioral Tasks





Kable et al 1865









MRI Acquisition



TotalFixations

Accuracy()

Reaction Time(msec)

Experiment 1



Experiment 2






Data Analysis

Statistical Tests







Experiment 1



Experiment 2





Kable et al 1867





Definition of ROIs



Acknowledgments





REFERENCES





































Kable et al 1869




































































DISCUSSION




Kable et al 1861














Kable et al 1863


Other Regions

Left Frontal Cortex












Conclusion


METHODS

Subjects



Behavioral Tasks





Kable et al 1865









MRI Acquisition



TotalFixations

Accuracy()

Reaction Time(msec)

Experiment 1



Experiment 2






Data Analysis

Statistical Tests







Experiment 1



Experiment 2





Kable et al 1867





Definition of ROIs



Acknowledgments





REFERENCES





































Kable et al 1869

















































Kable et al 1863


Other Regions

Left Frontal Cortex












Conclusion


METHODS

Subjects



Behavioral Tasks





Kable et al 1865









MRI Acquisition



TotalFixations

Accuracy()

Reaction Time(msec)

Experiment 1



Experiment 2






Data Analysis

Statistical Tests







Experiment 1



Experiment 2





Kable et al 1867





Definition of ROIs



Acknowledgments





REFERENCES





































Kable et al 1869





































Other Regions

Left Frontal Cortex












Conclusion


METHODS

Subjects



Behavioral Tasks





Kable et al 1865









MRI Acquisition



TotalFixations

Accuracy()

Reaction Time(msec)

Experiment 1



Experiment 2






Data Analysis

Statistical Tests







Experiment 1



Experiment 2





Kable et al 1867





Definition of ROIs



Acknowledgments





REFERENCES





































Kable et al 1869







































Conclusion


METHODS

Subjects



Behavioral Tasks





Kable et al 1865









MRI Acquisition



TotalFixations

Accuracy()

Reaction Time(msec)

Experiment 1



Experiment 2






Data Analysis

Statistical Tests







Experiment 1



Experiment 2





Kable et al 1867





Definition of ROIs



Acknowledgments





REFERENCES





































Kable et al 1869












































MRI Acquisition



TotalFixations

Accuracy()

Reaction Time(msec)

Experiment 1



Experiment 2






Data Analysis

Statistical Tests







Experiment 1



Experiment 2





Kable et al 1867





Definition of ROIs



Acknowledgments





REFERENCES





































Kable et al 1869





































Data Analysis

Statistical Tests







Experiment 1



Experiment 2





Kable et al 1867





Definition of ROIs



Acknowledgments





REFERENCES





































Kable et al 1869








































Definition of ROIs



Acknowledgments





REFERENCES





































Kable et al 1869





































REFERENCES





































Kable et al 1869




































Conceptual Representations of Action in the Lateral ...

Documents

Transcript of Conceptual Representations of Action in the Lateral ...