The role of polysemy in masked semantic and translation prim(15)

M.Finkbeineretal./JournalofMemoryandLanguage51(2004)1–2215

controltrials.If,though,theprimestimulusisprimingthetargetbyactivatingasemanticsenseincommonbetweentheprimeandtarget,participantsshouldbesigni cantlyfasterontheexperimentaltrials.Designandprocedure

TheexperimentaldesignandprocedurewasidenticaltothatofExperiment5,exceptthatweonlytestedinonedirection.Becausethesourceofthetaskdi erencebetweenlexicaldecisionandsemanticcategorizationintheone-to-manydirectionisthecriticalquestionthatthisexperimentaddresses,itwasonlynecessarytotestinthisdirection.Asigni cantpriminge ectintheone-to-manydirectionwhenusingexemplarsascontrolprimeswouldallowustoruleoutthepossibilityofadecisioncon ict.

Resultsanddiscussion

Thesametrimmingproceduresusedinthepreviousexperimentswereusedagainhere.Contrarytothepredictionsofthedecisioncon ictaccount,semanticprimingintheone-to-manydirectionwasnota ectedbythepresenceofcongruent,butsemanticallydissim-ilar,exemplarsasprimes.Meanresponsetimefortar-gets(e.g.,‘‘HAND’’)precededbyexperimentalprimes(e.g.,‘‘wrist’’)was463ms;andmeanresponsetimeforthesametargetsprecededbycontrolprimes(e.g.,‘‘kidney’’)was475ms.Thispriminge ectwasfoundtobereliableinboththesubjectanditemsanalyses(F1ð1;22Þ¼5:79,P¼:02;F2ð1;40Þ¼6:25,P¼:01).These ndingscon rmthosethathavebeenreportedonatleasttwooccasionsnow(Bueno&Frenck-Mes-tre,2002;Forsteretal.,2003).Thatis,maskedpriminge ectsforexemplarsinsemanticcategorizationaresimilarinmagnituderegardlessofwhetherthecontrolconditionincludescategorycongruentitemsornot.Assuch,wecanconcludethatthetaskdi erencebetweenlexicaldecisionandsemanticcategorizationintheone-to-manydirection(andbyextension,intheL2–L1direction)isnotattributabletoadecisioncon ictinsemanticcategorization.Rather,these ndingssuggestthatthetaskdi erenceisduetothefactthatthese-manticcategorizationtaskservestofocushowdeci-sionsaremade,sothatonlythesemanticsensethatisrelevanttothecategoryistakenintoaccount.Taking‘‘hand’’asanexample,onedoesnotneedtotakeintoconsiderationthemultiplesensesthat‘‘hand’’mayhaveinordertodecidewhetheritisabodypartornot.Essentially,wearguethatthesemanticcategorizationtaskturnsmany-sensetargetsintoone-sensetargetsby‘‘ ltering’’outcategory-irrelevantsensesfromthede-cisionmakingprocess,andthatthisenhancesthee -cacyoftheone-senseprime.ThisissuewillbeconsideredagainintheGeneraldiscussion.

Generaldiscussion

Theresearchhereestablishesseveral ndingsthat,whentakentogether,presentanimportantchallengetomodelsofbilinguallexicalrepresentationandprocess-ing.These ndingsare:(1)asymmetricaltranslationpriminginlexicaldecision,(2)symmetricaltranslationpriminginsemanticcategorization,(3)robustwithin-L2maskedpriminge ects,and(4)thesamepatternof ndingsforwithin-languagewordpairsthathavethesamerepresentationalasymmetryassumedbetweentranslationequivalents(seeTable2).Aswehaveseen,modelsofbilinguallexicalprocessingthatproposelim-itingfactorssuchasweakL2form–meaningconnectionstrengths(e.g.,TheRevisedHierarchicalModel),orthattheL2lexiconmaybestoredinaseparatememorymodulefromtheL1lexicon(asintheseparatememorysystemsaccount),cannotaccountforthese ndings.Inwhatfollows,wediscussinmoredetailhowtheSenseModelprovidesaparsimoniousexplanationforeachofthese ndingsinturn.

MaskedtranslationprimingasymmetryinlexicaldecisionTheaccountoftheprimingasymmetrygivenbytheSenseModelfollowsdirectlyfromtheassumptionsthatitmakeswithrespecttohowlexicalsemanticinforma-tionisrepresented.AswementionedinthediscussionofExperiment3,theSenseModel,muchlikethedistrib-utedconceptualfeaturemodel(DCFM)(deGroot,1992;deGroot,Dannenburg,&vanHell,1994;Kroll&deGroot,1997),holdsthatlexical(form-level)repre-sentationsmapontodistributedlexicalsemanticrepre-sentations,asopposedtolocalistconceptualrepresentations.UnliketheDCFM,though,acentralassumptionoftheSenseModelisthatsemanticfeaturesarebundledintosemanticsenseswithindistinctlexicalsemanticrepresentations.Followingfromthis,theSenseModelassumesthatmaskedtranslation(andsemantic)primingisattributednotonlytotheoverlappingse-manticfeaturesbetweenprimeandtarget,butalsototheratioofprimedtounprimedsenses.Whenthese-manticfeaturesincommonbetweenprimeandtargetconstitutealargeproportionofthetargetÕslexicalse-manticrepresentation,primingshouldbeobserved.Ontheotherhand,whenthesemanticfeaturesincommonbetweenprimeandtargetconstituteonlyasmallpro-portionofthetargetÕslexicalsemanticrepresentation,theamountofprimingmaybetoosmalltoa ectdeci-sionmakinginanysigni cantway.

Whyshouldweassumethatlexicaldecisionissen-sitivetotheproportionofsemanticsensesactivatedforagiventarget?SomesupportforthisassumptioncomesfromrecentworkbyRoddetal.(2002).Roddetal.(2002)reportthatthecontroversialclaimthatthereisanambiguityadvantage(i.e.,thatwordswithmany