ARGUMENTRELATIONS1

Samewords,differentstructures:anfMRIinvestigationofargumentrelationsandthe1angulargyrus23Abstract45InfMRI,increasedactivationforcombinatorialsyntacticandsemanticprocessingistypically6observedinasetoflefthemispherebrainareas:theangulargyrus(AG),theanteriortemporal7lobe(ATL),theposteriorsuperiortemporalsulcus(pSTS),andtheinferiorfrontalgyrus(IFG).8RecentworkhassuggestedthatsemanticcombinationissupportedbytheATLandtheAG,9withadivisionoflaborinwhichAGisinvolvedinthematic/eventrepresentationsandATLis10involvedinencodingconceptualfeaturesofentitiesand/ormoregeneralformsofsemantic11combination.ThecurrentfMRIstudywasdesignedtorefinehypothesesabouttheangular12gyrusprocessesinquestion.Inparticular,weaskwhethertheAGsupportsthecomputationof13argumentstructure(alinguisticnotionthatdependsonaverbtakingotherphrasesas14arguments)orthecomputationofeventconceptsmorebroadly.Todistinguishthese15possibilitiesweusedanovel,lexically-matchedcontrast:nounphrases(thefrightenedboy)and16verbphrases(VP)(frightenedtheboy),whereVPscontainedargumentstructure,denotingan17eventandassigningathematicroletoitsargument,andNPsdidnot,denotingonlya18semanticallyenrichedentity.Resultsshowedthatwhilemanyregionsshowedincreased19activityforNPsandVPsrelativetounstructuredwordlists(AG,ATL,pSTS,anteriorIFG),20replicatingevidenceoftheirinvolvementincombinatorialprocessing,neitherAGorATL21showeddifferencesinactivationbetweentheVPandNPconditions.Theseresultssuggestthat22increasedAGactivitydoesnotreflectthecomputationofargumentstructureperse,butare23compatiblewithaviewinwhichtheAGrepresentseventinformationdenotedbywordssuch24asfrightenedindependentoftheirgrammaticalcontext.Bycontrast,pSTSandposteriorIFGdid25showincreasedactivationfortheVPsrelativetoNPs.Wesuggestthattheseeffectsmayreflect26differencesinsyntacticprocessingandworkingmemoryengagedbydifferentstructural27relations.28291.Introduction3031Accuratelycomputingtherelationsbetweenverbsandtheirarguments,oftenreferredtoas32'argumentstructure',isafundamentalcomponentoflanguagecomprehension.Forexample,in33thesentence“thefleetdestroyedtheplanet”,themeaningofthesentencecentersaroundthe34verb,apredicatewhichdenotesadestroyingevent.Knowledgeofthepropertiesofthatverb35andthegrammarofEnglishallowsalistenertodeterminethatthetwoarguments'thefleet'36and'theplanet'beardifferentrelationstotheevent:thefleetcarryingoutthedestroying,and37theplanetbeingdestroyed.Theseverb-argumentrelationsarecriticalforlistenerstobeableto38drawappropriateinferencesaboutthespeaker'sintendedmessage.3940Areverb-argumentrelationsdifferentinkindfromotherlinguisticrelations,suchasthe41modificationrelationbetweenanadjectiveandanouninaphraselike'thedestroyedplanet'?42Whilebothargumentstructureandmodificationrelationscontributetothemeaningofa43sentence,manylinguistictheorieshavesuggestedthattheyrelyondistinctmechanisms.44

ARGUMENTRELATIONS2

Furthermore,thereisevidencefromcognitiveneurosciencethattheprocessingofverb-1argumentrelationsmayrelyonspecializedneuralmachinery.Thegoalofthepresentstudywas2tofurtherinvestigatethepossibilityofneuralspecializationfordistinctlinguisticmechanisms3throughafunctionalmagneticresonanceimaging(fMRI)experimentthatdirectlycomparesthe4neuralresponsetolexicallymatchedverbphrases('destroyedtheplanet'),denotingverb-5argumentrelations,andnounphrases('thedestroyedplanet'),denotingmodificationrelations.671.1.Verb-argumentrelations:linguisticandconceptualdistinctions89Almostalltheoriesofnaturallanguagesemanticsassumethatverbsarepredicatesthatrequire10nominalargumentstoappearinawell-formedsentence(Frege,1892).Manytheories11additionallypositthatthecompositionoperationthatcombinesthemeaningofaverbandits12argumentisdistinctfromthecompositionoperationthatcombinesthemeaningofan13adjectivalmodifierwithanoun(e.g.Heim&Kratzer's1998contrastbetween'argument14saturation'and'predicatemodification').Anotherlinguisticdifferencebetweenverb-argument15relationsandadjective-nounrelationsistheirobligatoriness;incommonusage,'syntactic16arguments'ofaverbmustbepresentinthesentenceforittobewell-formed1,butadjectives17arealwaysoptional.Manytheoriesalsoassumethatverbsareencodedinthelexiconwitha18particularsetofargumentsthatarerequired:boththenumberofarguments(e.g.,intransitive19verbslike‘arrive’requireoneargument,transitiveverbslike‘buy’requiretwoarguments),as20wellasthespecific'thematicroles',ofthosearguments(suchas'agent','patient','instrument',21etc.(e.g.Bresnan,1981,Joshi&Schabes,1997,Sag&Wasow,1999).Critically,however,such22lexically-specifiedargumentstructuresarespecifictotheverbrepresentationsofthesewords;23inthepresentstudy,wewillcapitalizeonwordsthatcansometimesbeusedasadjectives24withoutthesecorrespondingargumentstructures.2526Independentofthelinguisticassumptions,however,verbphrasesalsotendtodenotedifferent27conceptsfromnounphrases.Forexample,GentnerandKurtz(2005)drawakeydistinction28betweenentityconcepts,whichrefertoobjectsandarecharacterizedbytheirproperties,and29relationalconcepts,whichrefertorelationsbetweenthings(suchasevents)andare30characterizedbytheirrelationalstructure.Mostverbphrasesdenoteeventsorstates,i.e.31

1Wenotethattheterm'argument'itselfissomewhatambiguousbetweenaconceptualandalinguisticsense.Intheabsenceoflanguage,humanscanconceptualizeevents—suchaswhenwatchingasilentvideooftheplanetbeingdestroyedbythefleet.Ifweconsiderconceptssuchas'eat'or'kill',wecandefine'conceptualarguments'asthoseparticipantswhichareinsomeintuitivebutunspecifiedsense'core'totheevent--forexample,acommonintuitionisthattheconceptdenotedbytheverb'eat'hastwocoreparticipants,somethingthatiseatingandsomethingbeingeaten.Somewhatcorrelatedbutinprincipledistinctfromconceptual'coreness'issyntacticobligatoriness—particularverbsinparticularlanguagesrequirecertainparticipantstobecontainedinthesentence,andthesearedefinedas'syntacticarguments'.Forexample,inEnglishtheverb'put'requiresthegoaloftheeventtobecontainedinthesentenceasaprepositionalphrase('Iputthekeysonthetable'vs.'*Iputthekeys')buttheverb'throw'doesnot('Ithrewtheballonthefloor'vs.'Ithrewtheball').Aclassicillustrationofhowconceptualandsyntacticargumentsmaybedistinctistheexampleof'stealing';arguablytheconceptofastealingeventhasthree'core'participants(thepersonstealing,thepersonbeingstolenfrom,andthethingbeingstolen),buttheEnglishverb'steal'onlyrequirestwoofthosetobecontainedinthesentence'Istoleacar'.Williams(2015)providesanextremelycleardiscussionoftheseissues.

ARGUMENTRELATIONS3

relationalconcepts,whilemostnounphrasesdenoteobjects/entities,i.e.entityconcepts.With1respecttothebrain,Chatterjeeandcolleagueshavesuggestedthatrelationalconceptsmight2besupportedbyregionsoriginallydevelopedforrelationalprocessingofvisualmotionand3actions(Kableetal.,2006;Wuetal.,2007),whileentityconceptsmightbepartially4representedwithsensory-functionalattributesencodedinotherregions.Recent5neurophysiologicalworkusingsinglewordshastentativelyindicateddifferentlocalizationof6activitybasedontheevent/entitydistinction,evenwhenthiscross-cutsthenoun/verb7distinction(Bednyetal.,2014;Lapinskayaetal.,2016).Therefore,computingverb-argument8relationscouldrequiredifferentneuralresourcesthanotherkindsoflinguisticrelationssuchas9adjective-nounrelationseitherbecauseofneuralspecializationforverb-argumentstructure10itselforbecauseofthistendencytodenotedifferentconceptualrepresentationsthatare11computedandrepresentedindifferentbrainregions.Thepresentexperimentwasdesignedto12deconflatethelinguistic/conceptualdistinctioninordertohelpresolvethisissue,particularly13withrespecttothebrainareathatiscommonlyimplicatedinverb-argumentstructure14processing:theangulargyrus.15161.2.Argumentstructureandthematicprocessingintheangulargyrus1718Neuroimagingandneuropsychologicalresearchacrossthelastseveraldecadeshasidentified19severallefthemispherebrainregionsthatarecriticalforlanguage,includingtheanterior20temporallobe(ATL),theposteriorsuperiortemporalsulcus(pSTS),theinferiorfrontalgyrus21(IFG),andtheangulargyrus(AG).Alloftheseregionshavebeenshowntodemonstrate22increasedactivityforsentencesvs.unstructuredlists,makingthemgoodcandidatesfor23involvementincombinatorialcomputations.Amongthese,itistheangulargyrusthathasmost24frequentlybeensuggestedtospecificallysupportcomputationsrelatedtoargumentstructure.2526Manyneuropsychologicalstudiesusingthevoxel-wiselesion-symptommappingtechnique27(VLSM)havefoundageneralassociationbetweendamagetoAG,ordegenerationofAG,with28sentencecomprehensiondeficits(Dronkersetal.,2004;Mesulametal.,2015;Pillayetal.,292017;Magnusdottiretal.,2013;Rogalskyetal.,2018).Morespecifictoargumentstructure,30Thothathirietal.(2012)conductedaVLSManalysislookingforregionsthatcorrelatedwith31errorsininterpretingsemanticallyreversiblesentenceslikeThemanwasservedbythewoman32(reversiblemeaningthatbothmenandwomenplausiblyserveandareservedbyeachother).33Reversiblesentencesrequiresubjectstocorrectlyassignthethematicrolesdenotedbythe34verb,assubjectscannotcorrectlyguessthemeaningofthesentencebasedsolelyonindividual35argumentsandtheeventsthattheseargumentstypicallyengagein.Theyfoundastrong36correlationbetweendamagetothelefttemporoparietaljunction(TPJ),includingAG,and37deficitsincorrectargumentassignment.Thothathirietal.suggestthatTPJmayplayageneral38roleinbindingrelationalinformation,andinsentenceprocessingmayspecificallysupportthe39bindingofentitiestotheirrolesinanevent.4041ComplementaryevidencethatAGisinvolvedinrepresentingthematicrelationsbetween42predicatesandargumentsforcomesfromaVLSMstudyofpicturenaming(Schwartzetal.,432011),whichfoundthatdamagetotheAGwasassociatedwiththematicsubstitutionerrors44

ARGUMENTRELATIONS4

(e.g.,substitutingwormforapple)whiledamagetotheATLwasassociatedwithtaxonomic1substitutionerrors(e.g.,substitutingpearforapple)(Schwartzetal.,2011).Similarly,another2lesionstudyusingeye-trackingfoundthatpatientswithdamagetoposteriortemporal-parietal3cortexincludingtheAGhadreducedanddelayedactivationtothematiccompetitorpictures4(e.g.lookingatapigwhenthetargetisbarn)comparedtobothhealthycontrolsandpatients5withamoreanteriorlesiondistribution(Mirman&Graziano,2012).67Additionalsuggestiveevidencecomesfromneuroimagingstudies.Severalstudiesthat8examinedprocessingofdifferentlevelsofsentencestructure(wordlists,phrases,andfull9sentences)foundthatAGappearedtoselectivelyactivateforfullsentences,incontrasttoATL10whichshowedagradedpatterninwhichactivityincreasedsteadilywithphrasesize(Pallieret11al.,2011;Matchinetal.2017).ThisisconsistentwiththeviewthatactivityinAGisprimarily12drivenbythethematicrelationsdenotedinasentencewhiletheATLplaysamoregeneralrole13insemanticrelationshipsthatholdinphrasesofallsizes.Someneuroimagingstudieshave14foundamorespecificrelationbetweenthematicprocessingandactivationintheAG.For15instance,Boylanetal.(2017)foundevidencethatbothleftandrightAGshowedlarger16differencesfrombaselineforrelationalcompounds(e.g.,woodstove,thematicrelationship17betweenwords)thanattributivecompounds(e.g.bullettrain,featurerelationshipbetweenthe18words).Bycontrast,theATLshowedsimilaractivationforthetwotypesofcompoundsbutwith19alaterpeakforthethematictype.Finally,anexperimentbyKalenineetal.(2009)required20participantstomakepicture-matchingjudgmentsaboutthematicortaxonomicrelations;the21AGshowedincreasedactivityforjudgmentsaboutthematicrelationsrelativetotaxonomic22relations.2324Asdiscussedabove,itcouldbethecasethattheapparentneuralspecializationforverb-25argumentrelationsintheAGisbecauseofspecializedlinguisticprocessingroutinesor26specializedconceptualprocessingroutines.Whileitisnotclearfromthepriorliteraturewhich27aredrivingtheeffectsreviewedabove,anumberofstudieshavereportedAGeffectswith28morefine-grainedmanipulationsoflinguisticstructure,whichmightbetakentoimplicate29linguisticprocessing.First,severalneuroimagingstudieshavefoundanassociationbetweenthe30presenceorcomplexityofaverb’sargumentstructureandactivationintheleftAG.Thompson31etal.(2007;2010)andMeltzer-Asscheretal.(2015)conductedsingle-wordlexicaldecision32experimentsinfMRIandshowedthatactivityintheAGandneighboringtemporalcortex33increasedwiththenumberofargumentsrequiredbyaverb:verbswiththreearguments(e.g.,34put)activatedtheseregionsmorethanverbswithtwoarguments(e.g.,chase),whichactivated35theseregionsmorethanverbswithoneargument(e.g.,sleep).Basedonthisandotherimaging36andpatientwork,theyproposethatinferiorfrontalregionsareinvolvedinselectingthe37appropriatesyntacticframeforaverb,andthatposteriorregions(includingAGandposterior38MTG)areinvolvedinintegratinglexicalmaterialthatsatisfiestheargumentstructure39requirements.AsimilarexperimentbyMeltzer-Asscheretal.(2012)foundthatverbswith40alternatingtransitivity(e.g.,breakcanoccurwithoneargument,asinthestatuebroke,orwith41twoarguments,asinthechildbrokethestatue)activatedleftAGmorethanverbsthatsolely42occurwithoneargument(e.g.,bark),furtherincreasingtheassociationoftheAGandverbal43argumentstructureprocessing.Finally,anfMRIstudybyBoylanetal.(2015)foundsimilarityin44

ARGUMENTRELATIONS5

activitypatternsintheleftAGfortwo-wordphrasessharingthesameverb,butnotforphrases1sharingthesameargumentbutwithaverbvs.apreposition,suggestingthattherelational2informationspecificallydenotedbytheverbdrivesactivityinAGratherthanconceptual3featuresofthenounorargumentstructureingeneral,whichwouldbethesamewhether4encodedbyaverborbyapreposition.56SuchevidenceraisesthepossibilitythattheassociationbetweentheAGandevent/relational7semanticsisactuallydrivenbyspecializedlinguisticoperationsforverb-argumentstructure.8However,mostofthesestudieshavecompareddifferentverbstoeachother,whichraisesthe9possibilitythattheseeffectsaredrivenbyconceptual-semanticdifferencesamongtheseverbs10(forinstance,thedegreetowhichagivenverbisassociatedwitheventsemantics)ratherthan11linguisticargumentstructurerelationsthemselves.12131.3.Generalsemanticprocessingandtheangulargyrus1415OtherevidencesuggeststhattheAGmaybeinvolvedinaverygeneralformofsemantic16processing.Forinstance,severalstudieshaveshownthattheAGshowsincreasedactivationfor17realnounsrelativetopseudowords(Binderetal.,2003;2005;Bonneretal.,2013),and18increasedactivationforconcretenounsrelativetoabstractnouns(Binderetal.,2005).Given19thatisolatednounspresumablydonotdenoteeventsorhaveargumentstructure,theseresults20speakagainstarolefortheAGineithergeneralrelationalconceptprocessingorverbargument21structure(attheveryleast,theysuggestatleastsomerolefornon-relationalsemantics).22Relatedly,whilePillayetal.(2017)foundanassociationbetweenAGdamageandsentence23comprehension,theyalsofoundanassociationbetweenAGdamageandpicturenaming;the24factthatbotheffectshighlightedAGsuggestthatthisregionisinvolvedinageneralsemantic25processunderlyingbothtasks.2627Additionally,Priceetal.(2015,2016)useanadjective-nounparadigmtoprobeAG,where28participantsneededtojudgewhethersequenceswerereadilycombinable(plaidjacket)ornot29(mosspony,turnipchapel,fastblueberry).Theyfindmoreactivityforthecongruousitemsin30AG,thatpatientswithdamagetoAGwerelessaccurateinthejudgment,andthatsubjectswith31“positive”tDCStoAGwerefasterintheirjudgments.Thedirectionoftheimagingresultsis32slightlysurprisinggiventhatEEG/MEGstudiesofsemanticincongruityfindeitherincreased33responsesforincongruity(e.g.Kutas&Hillyard1980,Molinaroetal.2012)orlittledifference34betweenconditions(Lauetal.2016),butthismaybeafunctionofthejudgmenttask,whichis35rareintheEEG/MEGliterature.Priceetal.suggestedthatAGplaysacriticalroleingeneral36semanticcombination;thisroledoesnotappeartobestraightforwardlyattributableto37relational/eventsemanticsorlinguisticargumentstructuregiventhattheirmaterialswere38simpletwowordmodificationswithoutverbs.AtleastoneMEGstudyofadjective-noun39combinationalsoreportsincreasedactivityforAG(Bemis&Pylkkanen,2012),althoughthis40findingisinconsistentacrosssimilarstudies(e.g.Bemis&Pylkkanen,2011;Westerlund&41Pylkkanen,2014).MorerecentlyWilliamsetal.(2017)associateAGwithrelationalitymore42generallybasedonastudycomparingnounphrasesdesignedtobehighorlowinrelationality43(director’schildvs.director’schair).Similarly,Lewisetal.(2015)reportevidenceforprocessing44

ARGUMENTRELATIONS6

ofbothtaxonomicandthematicrelationsinAG,andthusarguethatitplaysabroaderrolein1semanticcombination.23Tosumup,whilemuchpriorworkspecificallyassociatestheleftAGwithprocessingrelational4informationthathastodowitheitherverbargumentstructureorconceptualrelations5betweenentitiesandevents,otherpriorworkchallengesthisconclusionbyassociatingthis6regionwithmoregeneralsemanticorconceptualcombination.781.3.Thepresentstudy910ThepresentstudywasdesignedtoinvestigatethehypothesisthattheAGisinvolvedin11linguisticprocessingofverb-argumentstructurespecificallyasopposedtoamoregeneral12semanticfunction.Weusedasimpleparadigmthatcontraststheresponsetolexically-matched13verbphrases(VPs)andnounphrases(NPs)infMRI(seeFigure1forexamples).Weassumed14thatVPs,suchasfrightenedthechild,thoughnotacompletesentence,denoteeventsof15frightening,e.g.theargumentthechildreceivesanexperiencerthematicroleassignedbythe16verbfrighten.WecomparedVPssuchasthesetoNPsformedbytakingtheverbandswitching17itsorderwiththedeterminer.Thiscreatedphrasessuchasthefrightenedchild,inwhichthe18verbfrightenisusedasaparticipleadjective,modifyingthenounchild.Thus,VPsandNPsin19thisstudyinvolvedtheexactsamewords,denotingverysimilarconceptualrepresentations,but20theVPscontainargumentstructureandeventrepresentations,whiletheNPs(withoutverbsto21assignthematicroles)didnot2.Inaddition,NPsinvolvedmodificationofthenounbyan22adjective,whiletheVPsdidnot.Contrastingnounphraseslikethefrightenedchildwithverb23phraseslikefrightenedthechildallowsusaneleganttestofthehypothesisthattheAGis24especiallyengagedinprocessingargumentstructure.AccordingtohypothesesinwhichAGis25involvedinprocessingrelationalinformationspecifictoverbs,increasedactivitymightbe26predictedinthisregionfortheverbphraseconditionoverthenounphrasecondition,even27thoughthewordsineachphraseareidentical.Weincludedanunstructuredwordlistusingthe28samesetofwordsasinthephrasestimuliasacontroltoensurethatbrainareasshowinga29differencebetweenVPandNPwouldalsoactivateforbasiccombinatorialprocessing.3031AlternativehypothesesaboutthefunctionoftheleftAGmightpredictequivalentactivityinthe32verbphraseandnounphraseconditions.IftheAGisinvolvedinprocessingsemanticrelations33ofallkinds,itmightrespondequivalentlytoverb-argumentfulfillmentandadjectival34modification.Anotherpossibilityisthatthisregionisinvolvedinrepresentingbasiclexical35semanticsforsomesubclassofwords,andthattheseincluderootsthataretypicallyusedto36refertodescribeevents,suchasfrighten.Accordingtothishypothesis,eveniffrightenisused37asanadjectiveinsteadofaverb,itwouldactivatethesamelexicalsemanticrepresentationin38theAG.39401.4.Argumentstructure:syntaxandsemantics41

2AlthoughBresnan(1978)proposedthattheadjectivalformderivedfromatransformationofaverbalstructure,mostmodernsyntacticanalysesdonotholdthisview.

ARGUMENTRELATIONS7

1TheminimalVPsandNPsweconstructedforthisexperimentwerespecificallydesignedtotest2whethertheAGprocessesverbargumentstructure.However,wenotethatourmaterialsalso3containafundamentalsyntacticdistinction:ahead-complementconfigurationforVPs,andan4adjunctionconfigurationforNPs,withtheadjectivemodifieradjoinedtothenoun.This5structuraldistinctionmightinduceincreasedsyntacticprocessingforVPs,giventhathead-6complementconfigurationshavebeenarguedtorequiremoresyntacticlabelingthan7adjunctionstructures(Hornstein,2009).Also,wenotethatverbsnotonlydenotesemantic8argumentstructurebutalsohavesyntacticselectionrequirementsforarguments(i.e.,syntactic9subcategorization).Thisimpliesthat(most)contrastsofsemanticverbargumentstructureare10confoundedwithsyntacticselection.AnfMRIstudybyShetreetetal.(2006)attempedtode-11correlatesyntacticsubcategorizationandsemanticargumentstructurebytestingHebrewverbs12(insentencecontext)thathadtwosyntacticselectionoptionsbutonlyonethematicoption.13Thisstudyfoundincreasedactivityfortheseverbs(relativetothosethathadonlyonesyntactic14selectionoption)inbothpSTSandsuperior/posteriorIFG;thisstudydidnotreportactivityin15AG,althoughthepSTSclusterwasveryposteriorandpotentiallyrelatedtoAGeffects.Other16fMRIstudiesofargumentstructurealsoreportactivationinposteriortemporallobe(clearly17anteriortoAG)andIFGforargumentstructurecomplexity(Shetreetetal.,2010;Ben-Shachar18etal.,2003).ThissuggeststhatAGmaybeinvolvedinsemanticargumentstructure,whilepSTS19andIFGmaybeinvolvedinsyntacticstructure.Thisisalsoconsistentwiththemodeloflexical-20semanticprocessingproposedbyLauetal.(2008),wherelexicalaccessoccursintheposterior21temporallobe,withsemanticcombinatoricsandintegrationwithcontextinATLandAG.2223Overall,thereisevidencethatthesyntacticcategorydifferencebetweenverbandnounmight24affecttheactivationpatternsfortheVPsandNPs,eventhoughtheunderlyinglexical-semantic25rootwordisthesameinbothstructures.Forthisreason,weselectedregionsofinterest(ROIs)26forbrainareasimplicatedinbothsemanticandsyntacticprocessing,testingwhetheranyof27theseregionsshoweddifferencesbetweenNPsandVPs,andsignificantdifferencesforthis28comparisonmustbeevaluatedonbothasemanticandasyntacticlevel.2930

31Figure1.Examplestimuliforeachcondition,alongwithanillustrationoftheircorresponding32structures.LEFT:thelistconditioncontainedunstructuredsequencesofwordsderivedfromthe33phrases,notinvolvingcombinatorialsyntaxorsemantics.MIDDLE:theNPcondition,inwhich34theparticipleadjective(e.g.frightened)adjoinstothenounphraseandmodifiesitsmeaning.35RIGHT:theVPcondition,inwhichtheverbistheheadofthephrasewiththenounphraseasa36

ARGUMENTRELATIONS8

syntacticallyselectedcomplement.Thearrowindicatesthattheverbassignsanexperiencer1thematicroletothenounphraseaspartoftheargumentstructuredenotedbytheverb.232.Materialsandmethods452.1.Subjects6720right-handed,nativespeakersofEnglish(12female)wererecruitedforthisstudyandwere8paid$25anhourfortheirparticipation.Oneadditionalsubjectwasexcludedfromanalysesfor9fallingasleepearlyinthescanningsession.Consentwasacquiredfromeachsubjectbeforethe10studybegan,andallprocedureswereapprovedbytheInstitutionalReviewBoardofthe11UniversityofMaryland.12132.2.Stimuli1415Themainexperimenthadthreeconditions:nounphrases(NP),verbphrases(VP),and16unstructuredwordlists(LIST).Asinglestimulusineachconditionconsistedofasequenceof17threewords.IntheNPcondition,thesequenceconsistedofadeterminerfollowedbya18participleadjectivefollowedbyanoun(e.g.,thefrightenedchild;ascrubbedcountertop;our19woundedofficer).IntheVPcondition,thesequenceconsistedofaverbfollowedbya20determinerfollowedbyanoun(e.g.,frightenedthechild;scrubbedacountertop;woundedour21officer).IntheLISTcondition,thesequenceconsistedofthreedeterminers,threeverbs,or22threenouns(e.g.,frightenedscrubbedwounded;atheour;childcountertopofficer).Thesame23setofwordswasusedtogeneratestimuliforallconditions,withthreedifferentexperimental24listssuchthattherewasnorepetitionofitemsforeachsubject;itemswerecounterbalanced25acrosssubjects.Thethree-wordsequencesintheVPandNPconditionswereidenticalexcept26fororderandanoccasionalchangefrom“an”to“a”andviceversaasnecessary.2728Togeneratethesephrases,wefirstcreatedalargesetofthree-wordsequencesthatwere29ratedfornaturalnessbytheauthors,andweselectedonlythosestimulithatwereratedas30consistentlynaturalbyalltheauthorstobeexperimentalitemsforthemainexperiment.We31ensuredthatboththeNPandVPversionofeachstimuluswasnaturalandcoherent.We32selected270wordsequencesandthenselected90forwhichwecreatedprobestimulithat33wereusedforthesubjects’task,detailedbelow.Theseprobestimuliwereratedandselected34usingthesamecriteriaasthemainitems.3536Moststudiesofstructuredlinguisticmaterialshaveusedrapidevent-relateddesigns,whichare37mostefficientforestimatingthehemodynamicresponseassociatedwitheachcondition.38However,rapidevent-relateddesignshavetheconsequencethattheorderofconditionsis39random,sothatsubjectstypicallydonotknowtheconditionofeachtrialasitbegins.We40wantedtoensurethatsubjectswereconfidentabouttheidentityofeachconditionasthey41wereprocessingtheprobes(e.g.,thattheywouldnottrytoimposestructureaccidentallyon42theLISTcondition).Wethereforeusedablockdesign,consistingofsixthree-wordstimuliina43rowfromthesamecondition,withalabelindicatingwhethertheupcomingblockwas44

ARGUMENTRELATIONS9

comprisedofNPorVPtrials(“PHRASE”)orlisttrials(“LIST”).Weusedthesetofwordsfrom1eachVP/NPblocktocreatethecorrespondingLISTblock.LISTsequenceswerecreatedby2groupingtogetherthedeterminers,nouns,andverbsacrosstheVP/NPblocksintotheirown3three-wordsequences.Thisprocedurecounterbalancedthelexicalmaterialusedineachblock4acrosstheconditionswhileremovinglinguisticstructurefromtheLISTcondition.Theorderof5sequenceswithineachblockwasrandomizedforeachcondition.Stimulusblockswere6counterbalancedacrossconditions;thesamesetofsequenceswasusedinthecorresponding7blockacrosstheNP,VP,andLISTconditions.892.3.Taskandprocedure1011Thesubjects’taskwastomakeasemanticsynonymy/similarityjudgmenttoprobesthatwere12presentedtwiceperblock,oneprobeafterthefirstthreesequencesandoneprobeafterthe13lastthreesequences.Theprobewasamodifiedversionofoneoftheprecedingsequences;one14wordintheprobewasreplacedwithadifferentword,theothertwowordsbeingidentical.The15replacedwordwasalwayseitherthenounortheverb.Thewordwasreplacedwithonethat16waseithersemanticallysimilar(roughlysynonymous)ordissimilartoit.E.g.“thetiredman”->17“theexhaustedman”(similar)or->“thefrightenedman”(dissimilar).Thesubjectswere18instructedtopressabuttontoindicatewhethertheprobe,evaluatedasanentirephrase,was19semanticallysimilartooneoftheprecedingitemsornot.Thesequencethatwasprobedwas20randomlyselected,witharoughlyequalmixoffirst,second,andthirdsequencesbeingprobed.21Similaranddissimilarprobesoccurredwithequalfrequency.TheprobesfortheLISTitemswere22somewhatdifferentfromtheNPandVPitems;fortheseitemswedevelopedprobesusingthe23sameprocedure,butthesincetheLISTsequenceswereunstructured,thesemanticjudgment24wasrestrictedtothereplacedword.E.g.“fanstreatybystander”->“fansagreementbystander”25(similar)or“defendantcorpsephone”->“defendantequipmentphone”(dissimilar).Thisledto26asubstantialincreaseindifficultyperformingthetask,whichweaddressinthediscussion.27Subjectsmadetheirresponsesusingtwobuttonboxes,oneineachhand;mappingof28similar/dissimilartohandwascounterbalancedacrosssubjects.2930Weexplainedtosubjectsthenatureofthestimuliandtaskandtrainedthemonapracticerun31beforetheyenteredthescanner.Thispracticerunwasidenticaltotheexperimentrunsexcept32forbeingoutsidethescannerwithmaterialsnotusedinthemainexperiment,anditwasthe33sameforeverysubject.Itconsistedofthreesix-sequencebocksofeachconditioninrandom34order.Followingthispracticesession,subjectswereplacedinthescanner.Beforebeginningthe35mainexperiment,theyweregivenoneblockofeachconditionformthispracticesessiontore-36familiarizethemwiththetask.3738Themainexperimentconsistedof15blocksfromeachcondition,dividedinto5experimental39runsof3blocksfromeachcondition.Inaddition,therewere3blocksofblankscreen(rest)40presentedineveryexperimentalrunlastingtoestimatethebaseline;everyrunendedonarest41blockinordertoallowthehemodynamicresponsetoreturntobaseline.Wegavesubjectsa42shortbreakin-betweenruns.Theexperimentlastedabout1hourand15minutes.4344

ARGUMENTRELATIONS10

WepresentedtheexperimentvisuallyusingthePsychToolBoxMatlabpackage(Brainard,1997;1Pelli,1997;Kleineretal,2007).Thescreenwasblackandthefontwaswhite.Eachtrialblock2beganwithafixationcrosspresentedforarandomjitteredduration,either1.5,2.5,or3.53seconds(withuniformfrequencyamongjitterdurations)(1350-3350mson,150msblank),and4thenalabelthatsaideither“PHRASE”(VPanNPconditions)or“LIST”(LISTcondition)which5appearedonthescreenfor2s(1850mson,150msblank).Wepresentedthesecuessothat6subjectswouldbepreparedforthatconditionandadapttheirexpectationsappropriately.The7conditioncuewasfollowedby1soffixation(850mson,150msblank),andthenthefirst8sequenceoftheblockstarted.Allwordswerepresentedwithrapidserialvisualpresentation,9witheachitemappearingfor600ms(450mson,150msblank).Aftereachsequencewas10fixationfor1.5seconds(1350mson,150msblank).Afterthreesequencesandfixationperiods,11theprobeingreenfontappearedonthescreenfor2.85s(2700mson,150msblank),followed12bythefixation(1.5seconds)precedingthenextthree-wordsequence.Weinstructedthe13subjectstomaketheirresponsetotheprobeatanypointbeforethefixationcrossreappeared.14Eachblocklastedatotalof29.5seconds,notcountingthevariable(jittered)fixationthat15precededtheonsetofeachblock.TheRESTperiodsconsistedofaninitialnon-jitteredfixationof161.5seconds(1350mson,150msblank),thelabelfor2seconds(1850secondson,150msoff),17ablankscreenfor13seconds,followedbyanumericalcountdownfrom5-1,5secondstotalin18duration(850mson,150msblankpernumber),thatallowedsubjectstogetreadyforthenext19block.Therestblocklastedatotalof20seconds,notincludingthe1.5secondinitialfixation.A20schematicoftheexperimentaldesignispresentedinFigure2.2122

23Figure2.Schematicofstimuluspresentation.MIDDLE:illustrationofthesequenceofblocks24withinarun.TOP:timingofstimuliwithinagivenblock.BOTTOM:timingofstimulifortherest25blocks.26272.4.Languagelocalizer2829Beforethemainexperimentwepresentedsubjectswithashortlanguagelocalizerruninorder30toderiveindependentROIsforanalysisofthemainexperiment.Thisrunconsistedofsentences31takenfromRogalsky&Hickok(2009)aswellasunstructuredwordliststhatwerecreatedby32

ARGUMENTRELATIONS11

randomizingtheorderofwordsineachsentence.Sentencesandlistswerepresentedin1randomorderbutwithalabeltoindicatewhethertheupcomingstimuluswasasentenceor2list.Forsentencetrials,thesubjects’taskwastodetectanoccasionalsentencethatwas3semanticallyanomalous(weremovedthesetrialsfromouranalysis).Forthewordlist4condition,thesubjects’taskwastodeterminewhetheraprobewordwaspresentintheword5listornot.Westartedwithasetof40sentences,andscrambledtheorderofwordstocreate6thelistconditions.Wethencreatedtwostimulusliststhatincludedwordliststhatdidnot7overlapwithsentences,andcounterbalancedacrosssubjects.Wepresentedstimuliusingrapid8serialvisualpresentation.Theconditionlabelappearedonthescreenfor1.3s(1050mson,9250msblank),theneachwordwaspresentedfor600ms(450mson,150msblank),followed10byfixationfor900ms(750mson,150msblank).Theprobewaspresentedonthescreenin11greenfontfor1700ms(1550mson,150msblank).Weinstructedsubjectstomaketheir12responseswhiletheprobewasstillonthescreen.Followingeachtrialwasarandomfixation13foraminimumdurationof900ms,variablefrom.9secondsto15seconds,optimizedto14estimatethehemodynamicresponseusingFreeSurfer’sOPTSEQprogram15(https://surfer.nmr.mgh.harvard.edu/optseq/).Thelocalizerrunlastedabout10minutes.16172.5.fMRIdatacollectionandanalysis1819MRimageswereobtainedinaSiemensTRIO3Tscanner(SiemensMedicalSystems)usinga32-20channelheadcoil.Wefirstcollectedahigh-resolutionT1-weightedanatomicalimageinthe21axialplane(voxeldimension:0.45mmX0.45mmX0.9mm).Wethencollectedatotalof120322T2*-weightedEPIvolumesover6runs(1localizerrunof298volumes,5experimentalrunsof23181volumesapiece).Eachvolumecontained36obliqueslicesorientedapproximately2024degreesclockwiserelativetotheAC-PCaxis(TR=2s,TE=25ms,flipangle=90°,in-plane25resolution=3mmx3mm,slicethickness=3mmwith0.3mmgap).Thefirstfourvolumesof26eachrunwerecollectedbeforestimuluspresentationanddiscardedtocontrolforT1saturation27effects.Slice-timingcorrection,motioncorrection,andspatialsmoothingwereperformedusing28AFNIsoftware(Cox,1996;http://afni.nimh.nih.gov/afni).Motioncorrectionwasachievedby29usinga6-parameterrigid-bodytransformation,witheachfunctionalvolumeineachrunfirst30alignedtoasinglevolumeinthatrun.Functionalvolumeswerealignedtotheanatomical31image,andsubsequentlyalignedtoTalairachspace(TalairachandTournoux,1988).Functional32imageswereresampledto3mmisotropicvoxels,andspatiallysmoothedusingaGaussian33kernelof6mmFWHM. 3435First-levelanalyseswereperformedoneachindividual’sdatausingAFNI’s3dDeconvolve36function.Theregressionanalysiswasperformedtofindparameterestimatesthatbest37explainedvariabilityinthedata.Eachpredictorvariablerepresentingthetimecourseof38stimuluspresentationwasenteredintoaconvolutionanalysisusingacanonicalhemodynamic39responsefunction(AFNI’sBLOCKparameter).Forthelocalizerexperiment,wemodeledthe40threeconditions(list,goodsentence,anomaloussentence).Forthemainexperiment,we41includedaregressorforeachcondition,VP,NP,andLIST,modelingthedurationofthethree42consecutivephraseswithineachhalfoftheblockseparately.Weaddedoneregressorforthe43conditionlabelsthatprecededeachlanguageblockincludingtheprecedingandfollowing44

ARGUMENTRELATIONS12

fixationcross.Weaddedanotherregressorfortheconditionlabelsthatprecededtherest1block,includingtheprecedingfixationperiod.Wealsomodeledthe5secondcountdown2periodattheendoftherestblock,andmodeledthetwotaskportionsofeachlanguageblock,3includingthefollowingfixationcrossforthefirsttaskportion.Thesixmotionparameterswere4includedasregressorsofnointerest. 56Toperformthegroup-levelanalysis,wefirstidentifiedROIsfromthelanguagelocalizertask.7Weenteredtheparameterestimatesforthesentenceandlistconditionsforeachsubjectinto8AFNI’s3dANOVA2functionandidentifiedROIsusingthefollowingcontrasts:sentences>rest,9andsentences>wordlists.Weusedavoxel-wisethresholdofp<0.005(one-tailed)andasmall10volumecorrectionof20voxelsinordertoidentifysignificantclusters.Thisthresholdresultedin11acorrespondingFalseDiscoveryRateqvalue<0.05forbothofthesecomparisons,indicating12thattheseresultspassedacorrectionformultiplecomparisons.Weusedthesamevoxel-wise13thresholdtoensurethattheclusterswereofapproximatelythesamesizeforbothcontrasts.14WethenselectedthefollowinglefthemisphereROIsfromthesentence>listcontrast:anterior15temporallobe(ATL),inferiorangulargyrus(AG),andinferiorfrontalgyrus(IFGanterior),16straddlingtheparstriangularisandparsorbitalis.Fromthesentence>restcontrast,we17selectedthepSTSROIandthepotionoftheIFGthatcenteredontheparsopercularis(IFG18opercularis).Wethenaveragedtheestimatedpercentsignalchangevaluesforeachcondition19ofthemainexperiment(LIST,VP,NP)withineachROIandranpairwisecomparisonsamongall20theconditions.2122ToensurethatourROIanalyseswerenotmissingsignificanteffectsinthemainexperiment,we23alsoperformedwholebrainanalysesgroupanalysesbyenteringtheparameterestimatesforall24conditionsandsubjectsintoAFNI’s3dANOVA2function.WeperformedthecontrastsofVP>25LIST,NP>LIST,VP>NP,andNP>VP,usingaliberalvoxel-wisethresholdofp<0.01(two-tailed)26andaclustersizecorrectionof20voxels.Weusedthisliberalanduncorrectedthresholdin27ordertoensurethatnoeffectsthatmightcompromiseourROIanalyseswerelefthidden,and28asinformationforfutureresearcherswhomightseektoperformasimilarexperimental29manipulation.30313.Results32333.1.Behavioralperformance3435Forallbehavioralanalyses,weperformedallthreeplannedpairwisecomparisonsamong36conditionsusingtwo-tailedt-testsandaBonferronicorrectionformultiplecomparisons37(adjustedalphaofp<0.0167).Subjects’accuracyonthebehavioraltaskisshowninFigure2,38left.SubjectsperformedsignificantlybetterontheNPconditionrelativetotheListcondition:39t(19)=5.529,p<0.001(two-tailed),significantlybetterontheVPconditionrelativetotheList40condition:t(19)=4.704,p<0.001(two-tailed),withnosignificantdifferencebetweentheNP41andVPconditions:t(19)=1.064,p=0.301(two-tailed).TheseresultsindicatethattheList42conditionwassubstantiallyharderthanthetwophraseconditions.Thiseffectislikelyinlarge43partduetothemoreextremeworkingmemoryburdenintheListcondition.Overall,these44

ARGUMENTRELATIONS13

resultsindicatethatsubjectswereattentivetothetask,giventhefactthatperformanceon1boththeNPandVPconditionswashigherthanad’of1.5(byconvention,ad’of1is2consideredgoodperformance),andtheListconditionwascloseto1.34ResponsetimesareshowninFigure3,right.SubjectsperformedsignificantlyfasterontheNP5conditionrelativetotheListcondition:t(19)=6.145,p<0.001(two-tailed),andsignificantly6fasterontheNPconditionrelativetotheVPcondition:t(19)=2.984,p=0.008(two-tailed).The7VPconditiontrendedtowardsbeingfasterrelativetotheListcondition,butthiseffectdidnot8survivemultiplecomparisonscorrection:t(19)=2.392,p=0.027(two-tailed).910

11Figure3.BehavioralperformanceduringthefMRIexperiment.Errorbarsreflectonestandard12errorofthemeanwithsubjecteffectsremoved(Cousineau,2005).LEFT:sensitivity.RIGHT:13reactiontime.Seetextfordetailsofstatisticalanalysis.*=significanteffect,~=trendtowards14significance.15163.2.fMRI:Languagelocalizer1718Thecontrastofsentencesvs.restinthelocalizerrevealedeffectsinbilateraloccipito-temporal19cortex,theleftpSTS,andalargeclusterintheleftprecentralgyrusandinferiorfrontalgyrus,20withsomeadditionalsmallclustersintherighthemispherefrontalandparietallobes.The21contrastofsentencesvslistsrevealedsignificantclustersintheleftATL,rightATL,leftAG,left22IFG(parsorbitalis/triangularis),aswellasseveralmedialareas,includingtheanteriorcingulate,23ventromedialprefrontalcortex,leftprecuneus,leftamygdala,andleftparahippocampalgyrus.24Asmentionedabove,weselectedthefollowingsignificantlefthemisphereclustersfromthese25analyses.Fromthesentence>listcontrast:anteriortemporallobe(ATL),inferiorangulargyrus26(AG),andinferiorfrontalgyrus(IFGanterior)(straddlingtheparstriangularisandthepars27orbitalis).Fromthesentence>restcontrast:thepSTSandtheIFG(parsopercularis).2829

ARGUMENTRELATIONS14

1Figure4.Whole-brainactivationmapsforthelocalizerexperiment.TOP:sentence>rest.2BOTTOM:sentence>list.ClustersselectedtoserveasROIsforthemainexperimentareshown3surroundedbyawhitecirclewiththeircorrespondinganatomicallabels.Activationsare4displayedonaninflatedtemplatebraininTalairachspace(Talairach&Tournoux,1988).56Region Hemisphere x y z Clustersize

(voxels)Sentence>Rest Inferiorfrontalgyrus/precentralgyrus

Left -45 7 33 670

IFG(parsopercularis/triangularis)peak

Left -44 13 26

IFG(parstriangularis/orbitalis)peak

Left -50 32 4

ARGUMENTRELATIONS15

Precentralgyruspeak

Left -49 -8 50

Occipito-temporalcortex

Left -32 -72 -9 653

Inferioroccipitalgyruspeak

Left -26 -85 -5

Fusiformgyruspeak Left -39 -46 -13 Occipito-temporalcortex

Right 31 -77 -8 515

Inferioroccipitalgyruspeak

Right 25 -86 -4

Fusiformgyruspeak Right 41 -60 -10 Fusiformgyruspeak Right 36 -40 -15 Inferiorfrontalgyrus/precentralgyrus

Right 42 6 28 110

pSTS Left -53 -39 5 71Medialfrontalgyrus Left -5 0 57 71aSTS Left -54 -4 -5 33Inferiorparietal Right 28 -58 35 31Precentralgyrus Right 54 -6 44 25 Sentences>Lists Superiormedialgyrus Left/right -4 40 47 399IFGorbitalis Left ?Angulargyrus Left -46 -66 25 127aSTS Left -54 -4 -9 120aSTS Right 49 11 -15 83Ventromedialprefrontalcortex

Left/Right 1 45 -8 76

IFG(parsorbitalis) Right 42 28 -7 60Parahippocampalgyrus Left -27 -32 -11 50Precuneus Left -6 -52 21 45Amygdala Right 25 -4 -6 20Table1.Coordinatesofsignificantclustersfromthelanguagelocalizeranalyses.Individualvoxel1thresholdp<.005(one-tailed),clustersizethresholdof20voxels.AllvoxelspassedanFDR2correctionthresholdformultiplecomparisonsofq<0.05.Allcoordinatesarecenterofmass3(unlessnotedaslocalpeaks)reportedinTalairachspace(Talairach&Tournoux,1988).453.3.fMRI:Mainexperiment673.3.1.ROIanalyses89Region NPvs.List VPvs.List VPvs.NP

ARGUMENTRELATIONS16

AG t(19)=3.451*p=0.003

t(19)=2.279~p=0.034

t(19)=-0.326p=0.748

pSTS t(19)=1.883~p=0.075

t(19)=4.054*p<0.001

t(19)=2.790*p=0.012

ATL t(19)=3.052*p=0.007

t(19)=3.277*p=0.004

t(19)=0.466p=0.646

IFG(anterior) t(19)=3.283*p=0.004

t(19)=4.232*p<0.001

t(19)=1.817~p=0.085

IFG(opercularis) t(19)=-1.259p=0.235

t(19)=-0.449p=0.658

t(19)=2.750*p=0.013

Table2.StatisticalresultsoftheROIanalyses.*indicatesasignificanteffectwhencorrectedfor1multiplecomparisons,~indicatesatrendtowardssignificance.Allcomparisonsweretwo-2tailed.Significanceiscorrectedformultiplecomparisonswithafamily-wisealphathresholdofp3<.05(eachregionisdefinedasaseparatefamily),usingaBonferronicorrectionwithan4individualpthresholdofp<.0167.56

7Figure5.ROIanalyses.ROIsaredisplayedonaninflatedtemplatebraininTalairachspace8(Talairach&Tournoux,1988).Errorbarsreflectonestandarderrorofthemeanwithsubject9effectsremoved(Cousineau,2005).DetailsofteststatisticsarereportedinTable2.*=10significanteffect.~=trendtowardssignificance.11123.3.2.Whole-brainanalyses1314Weperformedwholebrainanalysesforthefollowingcontrasts:VP>list,NP>list,andVP–NP,15initiallyusingavoxel-wisethresholdofp<0.01(two-tailed)withaminimumclustersizeof2016voxels.Weperformedthisexploratoryanalysisbecausewewantedtoensurethatany17

ARGUMENTRELATIONS17

differencesweobtainedinourROIanalyseswereinterpretableinlightofthewhole-brain1contrasts.Forinstance,ifthecontrastbetweentheseconditionsresultedinoverallgreater2activityforoneconditionacrossmanybrainareas,thedifferencesbetweenNPandVPwe3obtainedinthepSTSandIFGROIswouldbelessmeaningful.45TheNPconditionshowedincreasedactivityrelativetotheListconditioninbilateral6ventromedialprefrontalcortex,biasedtothelefthemisphere(Figure6).TheVPcondition7showedincreasedactivityrelativetotheListconditioninatypicallefthemispherelanguage8network:thepSTS,theATL,andtheIFG(parstriangularis/orbitalis)(Figure7).Inthe9exploratoryanalysisofVPvs.NP,VPshowedincreasedactivityinpSTS,IFG(pars10opercularis/triangularis,anddorsalpremotorcortex,asimilarbutnotidenticalpatterntothe11contrastofVP>List.NPshowedincreasedactivityinabilateralnetworkthatroughly12correspondedtothedefaultnetwork:dorsalAG,ventromedialprefrontalcortex,and13precuneus(Figure8).1415

16Figure6.WholebrainanalysesforthecontrastofNP>list.Activationsaredisplayedonan17inflatedtemplatebraininTalairachspace(Talairach&Tournoux,1988).1819

ARGUMENTRELATIONS18

1Figure7.WholebrainanalysesforthecontrastsofVP>list.Activationsaredisplayedonan2inflatedtemplatebraininTalairachspace(Talairach&Tournoux,1988).34

5Figure8.WholebrainanalysesforthecontrastofVPvs.NP.Activationsaredisplayedonan6inflatedtemplatebraininTalairachspace(Talairach&Tournoux,1988).78Region Hemisphere x y z Clustersize

(voxels)VP>List IFG/aSTS/Anteriorcingulate

Left-35 15 -6 709

IFG(parsorbitalis)peak

Left-37 25 -9

Hippocampus Left -22 -16 -8

ARGUMENTRELATIONS19

aSTS Left -53 -4 -6 Anteriorcingulate

Left-1 38 -6

pSTS Left -48 -39 7 209Insula/Inferiorparietal

Left-32 -19 22 85

Superiormedialgyrus

Left-9 52 25 71

AG Left -42 -61 21 54Parahippocampalgyrus

Left-31 -32 -13 37

Temporalpole Right 48 21 -20 26Cerebellum Right 21 -75 -31 25Heschl’sgyrus Right 41 -17 4 23Caudate Left -13 9 25 23 NP>List Anteriorcingulate Left/Right -2 42 10 995IFG(parsorbitalis)

Left -35 22 -12 189

AG Left -43 -58 21 112aSTS Left -54 -2 -8 94Precuneus Left -7 -56 21 84Hippocampus Left -22 -13 -9 50Temporalpole Right 46 19 -17 45Insula Left -31 -17 18 44pSTS Left -46 -33 1 33aSTS Right 53 1 -10 30IFG(parsorbitalis)

Right29 12 -13 20

Table3.Coordinatesofsignificantclustersfromthewhole-brainanalysesofthemain1experimentforVP>ListandNP>List.Individualvoxelthresholdp<.01(two-tailed),cluster2sizethresholdof20voxels(uncorrectedformultiplecomparisons).Allcoordinatesarecenterof3massreportedinTalairachspace(Talairach&Tournoux,1988).456Region Hemisphere x y z Clustersize

(voxels)VP>NP Cerebellum Right 22 -56 -39 65IFG(parstriangularis)

Left -50 21 23 58

Cerebellum Right 17 -64 -22 41

ARGUMENTRELATIONS20

Precentralgyrus Left -45 -2 49 29pSTS Left -52 -43 5 26IFG(parsorbitalis)

Left -46 26 5 20

NP>VP Anteriorcingulate

Left/Right 7 51 11 168

Superiorfrontalgyrus

Right 26 19 43 45

Precuneus Right 3 -71 30 30Angulargyrus Right 50 -56 29 29Angulargyrus Right 38 -67 24 22Table4.Coordinatesofclustersfromtheexploratorywhole-brainanalysesofthemain1experimentforVP>NPandNP>VP.Voxel-wisep<0.01(two-tailed),clustersizethreshold202voxels.AllcoordinatesarecenterofmassreportedinTalairachspace(Talairach&Tournoux,31988).454.Discussion67ThecurrentfMRIstudywasdesignedtoinvestigatetheextenttowhichlefthemispherebrain8regionsthatshowincreasedactivityforstructuredlanguageinput(pSTS,ATL,AGandIFG)are9specificallyinvolvedincomputingverbargumentvs.modificationrelations.Replicatingprior10work(Pallieretal.,2011;Fedorenkoetal.,2012;Matchinetal.,2017),wefoundthatallofthe11lefthemispherelanguage-relatedregionswetestedexceptfortheparsopercularisshowed12increasedactivityforlexically-matchedverbphrases(VPs)andnounphrases(NPs)relativeto13listsoratrendtowardsthiseffect.However,onlypSTSandposteriorIFGshowedincreased14activityforVPsrelativetoNPs(therewasalsoanon-significantbutsuggestivetrendforthis15effectinanteriorIFG).Notably,althoughAGhasoftenbeenassociatedwiththematic16processinginthepreviousliterature,nodifferencesintheresponsetoVPsandNPswere17observedhere.Inwhatfollows,wediscusstheseresultsinturn.18194.1.TheroleoftheAGinprocessingargumentstructureandeventsemantics2021PreviousworkhasdemonstratedthatAGandsurroundingtemporalcortexissensitivetoverbal22argumentstructure,showingincreasedactivityinresponsetoverbsthatrequiremore23arguments(Thompsonetal.,2007;2010;Meltzer-Asscheretal.,2012;2015).Thecurrent24resultssuggestthattheseeffectsdonotdirectlyreflectpredicate-argumentrelationsinthe25logicalform,asweobservednodifferenceintheAGresponsebetweenVPsthatcontained26predicate-argumentrelationsandNPsthatdidnot.However,thefactthatbothtypesof27phraseselicitedalargerAGresponsethanunstructuredwordlistsisconsistentwithprevious28workimplicatingtheAGincombinatorialand/orconceptual-semanticprocessing(Priceetal.,292016;Boylanetal.,2015;Binderetal.,2009;Pallieretal.,2011;Williamsetal.,2017).One30

ARGUMENTRELATIONS21

possibleexplanationforthepresentpatternofresultsisthattheleftAGplaysageneralrolein1semanticcombination,regardlessofwhetherthesemanticrepresentationsare2thematic/relational,taxonomic,orotherwiseinnature.Thiswouldbeconsistentwiththe3severalpriorstudiesthathavereportedanassociationbetweenleftAGandgeneralconceptual4combinationusingmaterialsthatdon’tincludeverb-argumentcombination(Priceetal.,2015;52016;Bemis&Pylkkanen,2012;2013;Williamsetal.,2017).67However,ageneralroleinsemanticcombinationdoesnotaccountforthefactsthatledtothis8proposedstudy–thelargevolumeofdataassociatingtheAGmorespecificallywith9event/relationalsemanticsandverbargumentstructure.Rather,theentirebodyofdata10incorporatingbothpresentandpastresultsseemsmoreconsistentwithamorespecificrolefor11AGincomputingand/oraccessingconceptualrelationsbetweeneventsandparticipants—in12otherwords,placingthespecificitynotinrepresentinglinguisticverb/predicate-argument13relations,butconceptualevent-participantrelations.Totakeourownexperimentasan14illustration,inthenounphrase“thefrightenedboy”,noverbisrepresentedinthesyntaxofthe15phrase,andnopredicate-argumentrelationsinthelogicalform.However,therootfrighten16denotesanevent,anditisconceivablethatthelexicalaccessroutinethatinvolvesretrieving17themorphemesthatcomposetheadjectivefrightenedmayautomaticallyengenderactivation18ofthisstoredeventrepresentation.Onthisaccountthen,theleftAGmaybeinvolvedin19processingtheeventinformationassociatedwithparticularwordsregardlessofwhetherthey20areusedinasyntacticcontextwhoselogicalformdenotesaneventornot.2122HerewestepbackandreconsiderthepreviousstudiesthatfoundanassociationbetweenAG23activityandargumentstructure.StudiesthathavefoundincreasedactivityinAGforthe24complexityofaverb’sargumentstructurehavemostlyusedsinglewordpresentationanda25lexicaldecisiontask(Thompsonetal.,2007;2010;Meltzer-Asscheretal.,2012;2015).Itmay26bethatthesesubjectsattemptedtoaccessthesemanticinformationassociatedwithwordsin27ordertomakethelexicalitydecision,includingeventsemantics.Underthepresenthypothesis,28verbswithmorecomplexargumentstructuresmayinducetheconstructionand/oraccessin29memoryofmorecomplexeventrepresentations,increasingactivationintheAG.3031ThishypothesiscanalsopotentiallyexplainwhytheAGactivatesforgeneralsemantictasks32(Binderetal.,2009).Forinstance,theactivationofeventscanonicallyassociatedwithparticular33words(ortheimplicitcomputationofsuchevents)regardlessofovertargumentstructure34couldexplainincreasedactivityintheAGforrealnounsrelativetopseudowordsinlexical35decisiontasks(Binderetal.,2003;2005;Bonneretal.,2013).Thisexplanationalsoextendsto36studiesfindingthatmeaningfulsemanticcombinationsactivatetheleftAGmorethanless37meaningfulones(Priceetal.,2015;2016).Forinstance,thefMRIstudybyPriceetal.(2015)38foundthatmeaningfulcombinationslikeplaidjacketactivatedtheleftAGmorethanless39meaningfulonessuchasmosspony.Plaidjacketsareassociatedwithmanyevent40representations,suchasalumberjackcuttingdownatreeoracocktailpartyinthe1960s.By41contrast,mossponiesarenotassociatedwithanyeventrepresentations.However,wenote42thatthisexplanationdoesnotaccountforsomestudiesthatobservesimilaractivityinAGfor43

ARGUMENTRELATIONS22

bothtaxonomicandthematicrelations(Lewisetal.,2015;Sachsetal.,2008).Thismeansthat1theseresultsshouldbereplicatedandexploredfurther.23Thishypothesisisalsoconsistentwiththelesiondata,whichshowageneralassociation4betweenleftAGdamageordegenerationandsentencecomprehensiondeficits(Dronkerset5al.,2004;Thothathirietal.,2011;Mesulametal.,2015;Pillayetal.,2017;Magnusdottiretal.,62013).ThefindingthatpicturenamingdeficitsareassociatedwithAGdamage(Pillayetal.,72017)isconsistentwithideathatsubjectsfindthewordassociatedwithapictureinpartby8identifyingthethematicinformationassociatedwithapictureandthenidentifyingwords9associatedwiththoseeventrepresentations.Finally,thefindingbySchwartzetal.(2011)that10thematicerrorswereassociatedwithAGdamage(andnottaxonomicerrors)ismuchbetter11explainedbyaroleforAGineventsemanticsthaningeneralsemanticprocessing.Similar12resultswereobtainedintheeyetrackingstudybyMirman&Graziano(2012)–patientswith13posteriorlesions(includingAG)showeddifferentprocessingofthematicrelationsthancontrol14patients,butnodifferencefortaxonomicrelations;thereversepatternwasfoundforpatients15withanteriorlesions(overlappingwithATL).Webelieveaparticularlyimportantgoalfor16researchonthematicprocessinginAGistoreplicateandextendthefindingsofadistinction17betweendamagetoATLandAGbeforemakingfirmconclusionsbasedonthesestudies.1819Overall,whileinourstudytheleftAGwasnotmoreengagedforVPs(withargumentstructure)20relativetoNPs(withoutargumentstructure),theresultsarereconcilablewitharoleforthis21brainregioninprocessingeventinformation,whetheritisaccesstostoredevent22representationsortheimplicitcomputationofevents.Ourresultshelpnarrowdownthe23functionalcontributionofthisregion,suggestingthatitdoesnotsimplyrespondtothe24presenceorabsenceofovertargumentstructuredenotedbyaverb,butrathermightbe25involvedinprocessingevent-levelrepresentationsmoregenerally.2627Thisresultthenraisesafurtherquestion:howdoesthebrainprocesslinguisticargument28structure?Suchaquestiongoesbeyondthescopeofthepresentinvestigation.However,the29failuretoprovidegoodevidenceforaplausibleneuralcorrelateoflinguisticargumentrelations30highlightstheneedtodevelopandtestmoredetailedmodelsoflinguisticandsemantic31processinginthebraintopursuethisandotherquestionsregardingtheneurobiologyof32language.33344.2.pSTS:lexical-syntacticprocessing3536ThepSTSandanteriorIFGshowedincreasedactivityforVPsrelativetoNPs,andincreased37activationforphrasesrelativetolistsinthemainexperiment.Inprinciple,thisactivationcould38reflectsemanticargumentstructure.However,thepSTShasnotbeenstronglyassociatedwith39semanticprocessing,butratherlexicalandsyntacticprocesses(Pallieretal.,2011;Wilsonetal.,402010;Bornkessel-Schlesewsky&Schlesewsky,2013;Matchinetal.,2017),suggestingthat41effectsintheSTSarenotduetosemanticargumentstructure.4243

ARGUMENTRELATIONS23

WesuggestthattheincreasedactivationforVPsrelativetoNPsinpSTSinthecurrentstudyis1duetotheactivationofsyntacticsubcategorizationframesstoredinthisregion.Allofourverbs2weretransitive,requiringanounphrasecomplement;suchinformationispresumablyaccessed3incomputingthesyntacticstructureoftheVPsinthisstudy.However,theadjectivalversionsof4thesewordsdonotrequireanycomplements;itmaybethatwhenthesewordsareclearly5adjectives(inNPcontext),theydonotactivatethesubcategorizationinformationstoredonthe6correspondingverbs.Previousresearchhasassociatedtheposteriortemporallobewithlexical7access(seeLauetal.,2008andHickok&Poeppel,2007forreview),consistentwiththeidea8thatlexicalizedsubcategorizationinformationisstoredinthepSTS,exhibitingincreased9activationforverbswithsubcategorizationrelativetoadjectiveswithoutsuch10subcategorization.1112Asnotedintheintroduction,semanticargumentstructureandsyntacticsubcategorizationare13usuallyconfounded.OnepriorfMRIstudybyShetreetetal.(2006)de-confoundedthese14variablesthroughthepropertyofambiguity,whichhasbeenshowntomodulateneuralactivity15intheposteriortemporallobe(Roddetal.,2005;Snijdersetal.,2008).Theyexaminedverbs16thatareambiguouswithrespecttothesyntacticframesthattheycanappearin,suchas17discover,whichcantakeanouncomplement(discoveredthetreasure)orasentence18complement(discoveredthattheearthisround).Shetreetetal.arguethatwhileinsomecases19thesyntacticambiguityisassociatedwiththematicambiguity(fordiscover,thecomplementis20interpretedaseitherathemeoraproposition),inothercasesbothsyntacticframesconveythe21samethematicrelations(forexample,inHebrewtastethesoupvs.tastefromthesoup,the22complementisalwaysinterpretedasatheme).Theyobservedincreasedactivityinposterior23temporallobe(aswellasIFG)forverbswithmorethanonesyntacticframe,butthisactivity24wasunaffectedbythenumberofthematicoptions,suggestingthattheseactivationsweredue25tosyntacticsubcategorizationratherthanargumentstructure.Ifthishypothesisiscorrect,the26increasedactivationforVPsrelativetoNPsinourstudymightbeduetotheactivationof27syntacticsubcategorizationinformationstoredwithalexicalitem.However,giventhatour28studywasnotdesignedspecificallytodiscriminateamongthesealternatives,wecannotbe29confidentinthisinterpretation,andmoreresearchisneededtodeterminewhether30subcategorizationisspecificallytiedtopSTSactivity.3132Finally,anadditionalpossibilityisthatincreasedactivityforVPsinpSTS(andposteriorIFG)33actuallyreflectsafrequencyeffectratherthanalinguisticone.Althoughthelexicalitemswere34controlledacrossconditionswithrespecttotherootwordandovertmorphology,itmaybe35thatthefrequencyofusagewithrespecttosyntacticcategorywasdifferentacrosstheNP36(adjectiveusage)andVP(verbusage)conditions.PreviousfMRIresearchhasoccasionally37observedthatsomelefthemisphereregionsshowsincreasedactivityforlowfrequencywords38relativetohighfrequencywords(Fiebachetal.,2002;Kronbichleretal.,2003;deZubicarayet39al.,2005;Hauketal.,2008),althoughthepresenceofthiseffectdependsontaskandsyntactic40context(Kelleretal.,2001;Carreirasetal.,2006).Futureresearchcouldinvestigatetheeffect41ofsuchcontextualfrequencyeffectsonbrainactivationinlanguageareas.42434.3.AnteriorIFG44

ARGUMENTRELATIONS24

1TheanteriorIFG(parsorbitalis/triangularis)showedanon-significanttrendtowardsincreased2activationforVPsrelativetoNPs,andincreasedactivationforbothkindsofphrasesrelativeto3lists.Theincreasedactivationforstructuredphrasesrelativetolistsisconsistentwithprevious4researchidentifyingeffectsoflinguisticstructureandsemantics(Badre&Wagner,2007;Binder5etal.,2009;Pallieretal.,2011;Gouchaetal.,2015;Rogalskyetal.,2015;Matchinetal.,2017).6WehesitatetointerpretthenonsignificanttrendtowardsadifferencebetweenVPsandNPs,7giventhatwedidnotexpectanargumentstructureeffectinthisregion;moredatawouldbe8neededtodetermineifthiseffectisindeedreliableandinformativeastothefunctionalroleof9theanteriorIFG.10114.3.Theparsopercularis–verbalworkingmemory1213TheparsopercularisROIshowedasignificantincreasedresponsetoVPsrelativetoNPs.Given14thestrongassociationofthisregionwithsyntacticprocessingintheliterature(Ben-Shacharet15al.,2003;Snijdersetal.,2008;Goucha&Friederici,2015;Zaccarellaetal.,2017),itistempting16tointerpretthisresultassyntacticstructurebuilding.However,alargepriorliteraturehas17attributedthefunctionofthemid-posteriorIFGinsentenceprocessinginsteadtocognitive18controland/orverbalworkingmemoryresources(Thompson-Schilletal.,1997;Novicketal.,192005;Rogalsky&Hickok,2011;Matchinetal.,2017).Consistentwiththis,unlikethemore20anteriorportionoftheIFGandthepSTS,activityinthisregionwasactuallynumericallyhigher21forthelistconditionrelativetotheNPcondition,andalmostnumericallyequaltotheVP22condition.Giventhatthelistconditiondidnotinvolvecombinatorialprocessing,andour23blockeddesignlikelypreventedsubjectsfrommistakenlybuildingstructureinthesestimuli,24theseeffectswouldbehardtoattributetosyntacticprocessing.2526Ontheotherhand,thesomewhatelevatedactivationforlistsrelativetoNPsinthisregionfits27witharoleforthisregioninverbalworkingmemoryand/orcognitivecontrol.Giventhat28unstructuredwordlistsdonotallowforefficientchunkingtoreduceworkingmemoryload,29whilesentencesfacilitateworkingmemorythroughsyntacticandsemanticchunking,theword30listsinourstudygreatlyincreaseddemandsonworkingmemoryresourcesand/orexecutive31functionresources.Thisisconsistentbothwiththemuchlowerperformanceonthewordlist32conditionrelativetothephraseconditions,aswellastheinformalreportsofoursubjectsthat33wordrecallinthelistconditionwasquitehard.Theparsopercularishasbeenpreviously34implicatedinverbalworkingmemorytasks,consistentwiththisexplanation(Hickoketal.,352003;Buchsbaumetal.,2011).Itisalsorelevanttonotethatthereducedthresholdwhole36brainanalysisofVP>NPrevealedaneffectinthedorsalprecentralgyrus,whichisalsostrongly37implicatedinverbalworkingmemorytasks(Buchsbaumetal.,2011).Conversely,itis38somewhatlessclearwhyverbalworkingmemoryorcognitivecontroldemandswouldbe39greaterinVPsrelativetoNPs,asbothconditionshadhierarchicalstructurethatlikelyfacilitated40demandsonverbalworkingmemoryresources.However,onecouldspeculatethatthe41conceptualdifferencesbetweenobjectsandeventsresultindifferencesintheresources42requiredtorepresenttheminworkingmemory,anditisworthnotingthatreactiontimeswere43significantlyslowerfortheVPscomparedtotheNPs.44

ARGUMENTRELATIONS25

1OnealternativeinterpretationoftheincreasedresponsetotheVPconditioninposteriorIFGis2thatitreflectstheactivationofembodiedsemanticrepresentationsofverbsand/oractionsin3premotorcortex.Someimagingstudieshaveshownthatverbsactivateinferiorfrontalregions4morestronglythannouns(seeViglioccoetal.,2011forareview),consistentwiththeideathat5actionmeaningsprimarilydenotedbyverbsareembodiedinregionsneighboringmotorcortex6(Pulvermuller,2005).However,thischaracterizationhasbeenchallengedboththeoreticallyand7empirically(e.g.deZubicarayetal.,2013),andwithrespecttothecurrentstudydoesnot8explainwhyunstructuredlistsofnouns,determiners,andverbswoulddriveactivationtonearly9thesameextent.Thus,thisexplanationofthecurrenteffectsseemslesslikely.10114.4.Conclusions1213Ourstudyintroducedanovel,fullylexically-matchedparadigmtotestwhetheractivityinthe14leftangulargyruscontributesspecificallytothecomputationofargumentstructureinlinguistic15input.ThelackofdifferenceweobservedintheresponsetoVPsandNPsinthisregionsuggests16thatitdoesnot.However,thispatternisconsistentwitharoleforthisregionintheprocessing17ofeventinformationassociatedwithagivenword,independentofitslinguisticcontext,and18futureresearchshouldaimtofurtherdiscriminatebetweenthematicandmoregeneral19semanticprocessinginthisregion.Inaddition,thesignificantincreaseinactivationweobtained20forVPsrelativetoNPsinthepSTSfurthersupportsaroleforthisregioninsyntacticprocessing,21generallydefined(Pallieretal.,2011;Matchinetal.,2017),whichmayberelatedtothe22processingofsyntacticselectionorsubcategorization.23245.Acknowledgements2526ThisresearchwassupportedbyinternalresearchfundsfromtheUniversityofMaryland.We27wouldliketothankAlexanderWilliamsandtheaudienceoftheCognitiveNeuroscienceSociety28annualmeetingin2017forhelpfuldiscussion.29306.References3132Badre,D.,&Wagner,A.D.(2007).Leftventrolateralprefrontalcortexandthecognitivecontrol33

ofmemory.Neuropsychologia,45(13),2883-2901.3435Bedny,M.,Dravida,S.,&Saxe,R.(2014).Shindigs,brunches,androdeos:Theneuralbasisof36

eventwords.Cognitive,Affective,&BehavioralNeuroscience,14(3),891-901.3738Bemis,D.K.,&Pylkkänen,L.(2011).Simplecomposition:Amagnetoencephalography39

investigationintothecomprehensionofminimallinguisticphrases.Journalof40Neuroscience,31(8),2801-2814.41

42Bemis,D.K.,&Pylkkänen,L.(2012).Basiclinguisticcompositionrecruitstheleftanterior43

ARGUMENTRELATIONS26

temporallobeandleftangulargyrusduringbothlisteningandreading.Cerebral1Cortex,23(8),1859-1873.2

3Ben-Shachar,M.,Hendler,T.,Kahn,I.,Ben-Bashat,D.,&Grodzinsky,Y.(2003).Theneural4

realityofsyntactictransformations:Evidencefromfunctionalmagneticresonance5imaging.PsychologicalScience,14(5),433-440.6

7Binder,J.R.,McKiernan,K.A.,Parsons,M.E.,Westbury,C.F.,Possing,E.T.,Kaufman,J.N.,&8

Buchanan,L.(2003).Neuralcorrelatesoflexicalaccessduringvisualwordrecognition.9JournalofCognitiveNeuroscience,15(3),372-393.10

11Binder,J.R.,Westbury,C.F.,McKiernan,K.A.,Possing,E.T.,&Medler,D.A.(2005).Distinct12

brainsystemsforprocessingconcreteandabstractconcepts.JournalofCognitive13Neuroscience,17(6),905-917.14

15Binder,J.R.,Desai,R.H.,Graves,W.W.,&Conant,L.L.(2009).Whereisthesemanticsystem?16

Acriticalreviewandmeta-analysisof120functionalneuroimagingstudies.Cerebral17Cortex,19(12),2767-2796.18

19Bonner,M.F.,Peelle,J.E.,Cook,P.A.,&Grossman,M.(2013).Heteromodalconceptual20

processingintheangulargyrus.Neuroimage,71,175-186.2122Bornkessel-Schlesewsky,I.,&Schlesewsky,M.(2013).Reconcilingtime,spaceandfunction:a23

newdorsal–ventralstreammodelofsentencecomprehension.BrainandLanguage,24125(1),60-76.25

26Boylan,C.,Trueswell,J.C.,&Thompson-Schill,S.L.(2015).Compositionalityandtheangular27

gyrus:Amulti-voxelsimilarityanalysisofthesemanticcompositionofnounsandverbs.28Neuropsychologia,78,130-141.29

30Boylan,C.,Trueswell,J.C.,&Thompson-Schill,S.L.(2017).Relationalvs.attributive31

interpretationofnominalcompoundsdifferentiallyengagesangulargyrusandanterior32temporallobe.BrainandLanguage,169,8-21.33

34Brainard,D.H.(1997).Thepsychophysicstoolbox.SpatialVision,10,433-436.3536Bresnan,J.(1978).Arealistictransformationalgrammar.InHalle,M.,Bresnan,J.&Miller,G.A.37

(eds.),LinguisticTheoryandPsychologicalReality.Cambridge,MA:TheMITPress.1-59.3839Bresnan,J.(1981).AnapproachtoUniversalGrammarandthementalrepresentationof40

language.Cognition,10(1-3),39-52.4142Buchsbaum,B.R.,Baldo,J.,Okada,K.,Berman,K.F.,Dronkers,N.,D’esposito,M.,&Hickok,G.43

ARGUMENTRELATIONS27

(2011).Conductionaphasia,sensory-motorintegration,andphonologicalshort-term1memory–anaggregateanalysisoflesionandfMRIdata.BrainandLanguage,119(3),2119-128.3

4Caplan,D.,Stanczak,L.,&Waters,G.(2008).Syntacticandthematicconstrainteffectsonblood5

oxygenationleveldependentsignalcorrelatesofcomprehensionofrelative6clauses.JournalofCognitiveNeuroscience,20(4),643-656.7

8Carreiras,M.,Mechelli,A.,&Price,C.J.(2006).Effectofwordandsyllablefrequencyon9

activationduringlexicaldecisionandreadingaloud.HumanBrainMapping,27(12),963-10972.11

12deZubicaray,G.I.,McMahon,K.L.,Eastburn,M.M.,Finnigan,S.,&Humphreys,M.S.(2005).13

fMRIevidenceofwordfrequencyandstrengtheffectsduringepisodicmemory14encoding.CognitiveBrainResearch,22(3),439-450.15

16deZubicaray,G.,Arciuli,J.,&McMahon,K.(2013).Puttingan“end”tothemotorcortex17

representationsofactionwords.JournalofCognitiveNeuroscience,25(11),1957-1974.1819Dronkers,N.F.,Wilkins,D.P.,VanValinJr,R.D.,Redfern,B.B.,&Jaeger,J.J.(2004).Lesion20

analysisofthebrainareasinvolvedinlanguagecomprehension.Cognition,92(1-2),145-21177.22

23Fedorenko,E.,Nieto-Castanon,A.,&Kanwisher,N.(2012).Lexicalandsyntacticrepresentations24

inthebrain:anfMRIinvestigationwithmulti-voxelpattern25analyses.Neuropsychologia,50(4),499-513.26

27Fiebach,C.J.,Friederici,A.D.,Müller,K.,&Cramon,D.Y.V.(2002).fMRIevidencefordual28

routestothementallexiconinvisualwordrecognition.JournalofCognitive29Neuroscience,14(1),11-23.30

31Frege,G.(1892).Übersinnundbedeutung.ZeitschriftfürPhilosophieundphilosophische32

Kritik,100(1),25-50.3334Goucha,T.,&Friederici,A.D.(2015).Thelanguageskeletonafterdissectingmeaning:a35

functionalsegregationwithinBroca'sArea.Neuroimage,114,294-302.3637Hauk,O.,Davis,M.H.,&Pulvermüller,F.(2008).Modulationofbrainactivitybymultiplelexical38

andwordformvariablesinvisualwordrecognition:AparametricfMRIstudy.39Neuroimage,42(3),1185-1195.40

41Hickok,G.,Buchsbaum,B.,Humphries,C.,&Muftuler,T.(2003).Auditory–motorinteraction42

revealedbyfMRI:speech,music,andworkingmemoryinareaSpt.JournalofCognitive43Neuroscience,15(5),673-682.44

ARGUMENTRELATIONS28

1Hickok,G.,&Poeppel,D.(2007).Thecorticalorganizationofspeechprocessing.Nature2ReviewsNeuroscience,8(5),393.3

4Hornstein,N.(2009).Atheoryofsyntax:Minimaloperationsanduniversalgrammar.5

CambridgeUniversityPress.67Joshi,A.K.,&Schabes,Y.(1997).Tree-adjoininggrammars.InHandbookofformallanguages8

(pp.69-123).Springer,Berlin,Heidelberg.910Kalénine,S.,Peyrin,C.,Pichat,C.,Segebarth,C.,Bonthoux,F.,&Baciu,M.(2009).Thesensory-11

motorspecificityoftaxonomicandthematicconceptualrelations:Abehavioraland12fMRIstudy.Neuroimage,44(3),1152-1162.13

14Keller,T.A.,Carpenter,P.A.,&Just,M.A.(2001).Theneuralbasesofsentence15

comprehension:afMRIexaminationofsyntacticandlexicalprocessing.Cerebral16Cortex,11(3),223-237.17

18Kleiner,M.,Brainard,D.,Pelli,D.,Ingling,A.,Murray,R.,&Broussard,C.(2007).What'snewin19

Psychtoolbox-3.Perception,36(14),1.2021Kratzer,A.,&Heim,I.(1998).Semanticsingenerativegrammar(Vol.1185).Oxford:Blackwell.2223Kronbichler,M.,Hutzler,F.,Wimmer,H.,Mair,A.,Staffen,W.,&Ladurner,G.(2004).Thevisual24

wordformareaandthefrequencywithwhichwordsareencountered:evidencefroma25parametricfMRIstudy.Neuroimage,21(3),946-953.26

27Kutas,M.,&Hillyard,S.A.(1980).Readingsenselesssentences:Brainpotentialsreflect28

semanticincongruity.Science,207(4427),203-205.2930Lapinskaya,N.,Uzomah,U.,Bedny,M.,&Lau,E.(2016).Electrophysiologicalsignaturesof31

eventwords:Dissociatingsyntacticandsemanticcategoryeffectsinlexicalprocessing.32Neuropsychologia,93,151-157.33

34Lau,E.F.,Phillips,C.,&Poeppel,D.(2008).Acorticalnetworkforsemantics:(de)constructing35

theN400.NatureReviewsNeuroscience,9(12),920.3637Lau,E.,Namyst,A.,Fogel,A.,&Delgado,T.(2016).AdirectcomparisonofN400effectsof38

predictabilityandincongruityinadjective-nouncombination.Collabra:Psychology,2(1).3940Lewis,G.A.,Poeppel,D.,&Murphy,G.L.(2015).Theneuralbasesoftaxonomicandthematic41

conceptualrelations:AnMEGstudy.Neuropsychologia,68,176-189.4243Magnusdottir,S.,Fillmore,P.,DenOuden,D.B.,Hjaltason,H.,Rorden,C.,Kjartansson,O.,...&44

ARGUMENTRELATIONS29

Fridriksson,J.(2013).Damagetoleftanteriortemporalcortexpredictsimpairmentof1complexsyntacticprocessing:Alesion-symptommappingstudy.HumanBrain2Mapping,34(10),2715-2723.3

4Matchin,W.,Hammerly,C.,&Lau,E.(2017).TheroleoftheIFGandpSTSinsyntactic5

prediction:EvidencefromaparametricstudyofhierarchicalstructureinfMRI.Cortex,688,106-123.7

8Matchin,W.(2017).Aneuronalretuninghypothesisofsentence-specificityinBroca’s9

area.PsychonomicBulletin&Review.1011Meltzer-Asscher,A.,Schuchard,J.,denOuden,D.B.,&Thompson,C.K.(2013).Theneural12

substratesofcomplexargumentstructurerepresentations:Processing“alternating13transitivity”verbs.LanguageandCognitiveProcesses,28(8),1154-1168.14

15Meltzer-Asscher,A.,Mack,J.E.,Barbieri,E.,&Thompson,C.K.(2015).Howthebrainprocesses16

differentdimensionsofargumentstructurecomplexity:EvidencefromfMRI.Brainand17Language,142,65-75.18

19Mesulam,M.M.,Thompson,C.K.,Weintraub,S.,&Rogalski,E.J.(2015).TheWernicke20

conundrumandtheanatomyoflanguagecomprehensioninprimaryprogressive21aphasia.Brain,138(8),2423-2437.22

23Mirman,D.,&Graziano,K.M.(2012).Damagetotemporo-parietalcortexdecreasesincidental24

activationofthematicrelationsduringspokenword25comprehension.Neuropsychologia,50(8),1990-1997.26

27Molinaro,N.,Carreiras,M.,&Duñabeitia,J.A.(2012).Semanticcombinatorialprocessingof28

non-anomalousexpressions.Neuroimage,59(4),3488-3501.2930Novick,J.M.,Trueswell,J.C.,&Thompson-Schill,S.L.(2005).Cognitivecontrolandparsing:31

ReexaminingtheroleofBroca’sareainsentencecomprehension.Cognitive,Affective,&32BehavioralNeuroscience,5(3),263-281.33

34Pallier,C.,Devauchelle,A.D.,&Dehaene,S.(2011).Corticalrepresentationoftheconstituent35

structureofsentences.ProceedingsoftheNationalAcademyofSciences,201018711.3637Pelli,D.G.(1997).TheVideoToolboxsoftwareforvisualpsychophysics:Transformingnumbers38

intomovies.SpatialVision,10,437-442.3940Pillay,S.B.,Binder,J.R.,Humphries,C.,Gross,W.L.,&Book,D.S.(2017).Lesionlocalizationof41

speechcomprehensiondeficitsinchronicaphasia.Neurology,10-1212.4243Price,A.R.,Bonner,M.F.,Peelle,J.E.,&Grossman,M.(2015).Convergingevidenceforthe44

ARGUMENTRELATIONS30

neuroanatomicbasisofcombinatorialsemanticsintheangulargyrus.Journalof1Neuroscience,35(7),3276-3284.2

3Price,A.R.,Peelle,J.E.,Bonner,M.F.,Grossman,M.,&Hamilton,R.H.(2016).Causalevidence4

foramechanismofsemanticintegrationintheangulargyrusasrevealedbyhigh-5definitiontranscranialdirectcurrentstimulation.JournalofNeuroscience,36(13),3829-63838.7

8Pulvermüller,F.(2005).Brainmechanismslinkinglanguageandaction.NatureReviews9

Neuroscience,6(7),576.1011Rodd,J.M.,Davis,M.H.,&Johnsrude,I.S.(2005).Theneuralmechanismsofspeech12

comprehension:fMRIstudiesofsemanticambiguity.CerebralCortex,15(8),1261-1269.1314Rogalsky,C.,&Hickok,G.(2008).Selectiveattentiontosemanticandsyntacticfeatures15

modulatessentenceprocessingnetworksinanteriortemporalcortex.Cerebral16Cortex,19(4),786-796.17

18Rogalsky,C.,&Hickok,G.(2011).TheroleofBroca'sareainsentencecomprehension.Journal19

ofCognitiveNeuroscience,23(7),1664-1680.2021Rogalsky,C.,Almeida,D.,Sprouse,J.,&Hickok,G.(2015).Sentenceprocessingselectivityin22

Broca'sarea:evidentforstructurebutnotsyntacticmovement.Language,Cognition23andNeuroscience,30(10),1326-1338.24

25Rogalsky,C.,LaCroix,A.N.,Chen,K.H.,Anderson,S.W.,Damasio,H.,Love,T.,&Hickok,G.26

(2018).Theneurobiologyofagrammaticsentencecomprehension:alesionstudy.27JournalofCognitiveNeuroscience,30(2),234-255.28

29Sag,I.A.,&Wasow,T.(1999).Syntactictheory:Aformalintroduction(Vol.92).Stanford,CA:30

CenterfortheStudyofLanguageandInformation.3132Shetreet,E.,Palti,D.,Friedmann,N.,&Hadar,U.(2006).Corticalrepresentationofverb33

processinginsentencecomprehension:Numberofcomplements,subcategorization,34andthematicframes.CerebralCortex,17(8),1958-1969.35

36Shetreet,E.,Friedmann,N.,&Hadar,U.(2010).Theneuralcorrelatesoflinguisticdistinctions:37

Unaccusativeandunergativeverbs.JournalofCognitiveNeuroscience,22(10),2306-382315.39

40Schwartz,M.F.,Kimberg,D.Y.,Walker,G.M.,Brecher,A.,Faseyitan,O.K.,Dell,G.S.,...&41

Coslett,H.B.(2011).Neuroanatomicaldissociationfortaxonomicandthematic42knowledgeinthehumanbrain.ProceedingsoftheNationalAcademyofSciences,43201014935.44

ARGUMENTRELATIONS31

1Snijders,T.M.,Vosse,T.,Kempen,G.,VanBerkum,J.J.,Petersson,K.M.,&Hagoort,P.(2008).2

Retrievalandunificationofsyntacticstructureinsentencecomprehension:anfMRI3studyusingword-categoryambiguity.CerebralCortex,19(7),1493-1503.4

5Talairach,J.,&Tournoux,P.(1988).Co-planarstereotaxicatlasofthehumanbrain.NewYork,6

NY:ThiemeMedicalPublishers.78Thompson-Schill,S.L.,D’Esposito,M.,Aguirre,G.K.,&Farah,M.J.(1997).Roleofleftinferior9

prefrontalcortexinretrievalofsemanticknowledge:areevaluation.Proceedingsofthe10NationalAcademyofSciences,94(26),14792-14797.11

12Thompson,C.K.,&Meltzer-Asscher,A.(2014).Neurocognitivemechanismsofverbargument13

structureprocessing.StructuringtheArgument.Amsterdam:JohnBenjamins,141-168.1415Thompson,C.K.,Bonakdarpour,B.,Fix,S.C.,Blumenfeld,H.K.,Parrish,T.B.,Gitelman,D.R.,&16

Mesulam,M.M.(2007).Neuralcorrelatesofverbargumentstructureprocessing.17JournalofCognitiveNeuroscience,19(11),1753-1767.18

19Thompson,C.K.,Bonakdarpour,B.,&Fix,S.F.(2010).Neuralmechanismsofverbargument20

structureprocessinginagrammaticaphasicandhealthyage-matchedlisteners.Journal21ofCognitiveNeuroscience,22(9),1993-2011.22

23Thothathiri,M.,Kimberg,D.Y.,&Schwartz,M.F.(2012).Theneuralbasisofreversible24

sentencecomprehension:evidencefromvoxel-basedlesionsymptommappingin25aphasia.JournalofCognitiveNeuroscience,24(1),212-222.26

27Vigliocco,G.,Vinson,D.P.,Druks,J.,Barber,H.,&Cappa,S.F.(2011).Nounsandverbsinthe28

brain:areviewofbehavioural,electrophysiological,neuropsychologicalandimaging29studies.Neuroscience&BiobehavioralReviews,35(3),407-426.30

31Westerlund,M.,&Pylkkänen,L.(2014).Theroleoftheleftanteriortemporallobeinsemantic32

compositionvs.semanticmemory.Neuropsychologia,57,59-70.3334Williams,A.,Reddigari,S.,&Pylkkänen,L.(2017).Earlysensitivityofleftperisylviancortexto35

relationalityinnounsandverbs.Neuropsychologia,100,131-143.3637Wilson,S.M.,Dronkers,N.F.,Ogar,J.M.,Jang,J.,Growdon,M.E.,Agosta,F.,...&Gorno-38

Tempini,M.L.(2010).Neuralcorrelatesofsyntacticprocessinginthenonfluentvariant39ofprimaryprogressiveaphasia.JournalofNeuroscience,30(50),16845-16854.40

41Zaccarella,E.,Meyer,L.,Makuuchi,M.,&Friederici,A.D.(2017).Buildingbysyntax:theneural42

basisofminimallinguisticstructures.CerebralCortex,27(1),411-421.4344

ARGUMENTRELATIONS32

1