The role of polysemy in masked semantic and translation prim(10)

10M.Finkbeineretal./JournalofMemoryandLanguage51(2004)1–22

Jiang,1999;Jiang&Forster,2001;Keatlyetal.,1994;Sanchez-Casasetal.,1992;aswellasExperiment2ofthepresentstudy).

IfitiscorrecttoassumethatL2–L1primingdoesnotoccur(orisveryweak)inlexicaldecisionbecauseaninsu cientnumberofL1sensesarepreactivatedbytheL2prime,thentheremustbesomethingaboutthesemanticcategorizationtaskthatservestorestrictthenumberofsensesrecruitedwhenadecisionisgener-ated.Aswehavealreadyindicated,itisplausibletoarguethatthecategoryactsasacontextthatservestofocushowdecisionsaremadeinresponsetothetarget,sothatonlythesemanticsensethatisrelevanttothecategoryistakenintoaccount(e.g.,inordertodecidethat‘‘black’’isacolor,onedoesnotneedtotakeintoconsiderationthemultiplesensesthat‘‘black’’mayhave).Essentially,wearguethatthesemanticcatego-rizationtaskturnsmany-sensetargetsintoone-sensetargetsby‘‘ ltering’’outcategory-irrelevantsensesfromthedecisionmakingprocess.Assumingthattheprime–targetpairshavebeenproperlyselected,thissuggeststhattheL2primewillbemoree ectiveinprimingtheL1targetbecausetheonlysensethatisrelevantforthedecisionwillbetheverysensethathasbeenprimedbytheL2prime.

TheSenseModeldi ersfromothermodelsofbi-linguallexicalrepresentationandprocessinginthatitsexplanationofthetranslationprimingasymmetryisnotrestrictedtoapeculiarityofthebilinguallexicon(e.g.,weakconnectionstrengths).Thatis,theSenseModelpredictsthesamepatternofpriminge ects(andtaskdi erences)forwordpairswithinalanguagewhoserelationshipmeetstherepresentationalasym-metrycriteriaspeci edabove.Infact,therearereasonstothinkthatthestrongesttestoftheSenseModelwouldbetousewithin-languagewordpairsinamaskedprimingexperimentwithlexicaldecisionandsemanticcategorization.IftheassumptionsoftheSenseModelarecorrect,thenwordpairslike‘‘head–skull,’’whiche ectivelyinstantiatetheassumedrepre-sentationalasymmetrybetweentranslationequivalents,shouldproducethesameprimingasymmetryinlexicaldecisionbutnotsemanticcategorization.Thesepre-dictionsoftheSenseModelaretestedinthefollowingexperiments.

Experiment4—within-languageprimingasymmetryinlexicaldecision

Inthepresentexperiment,wetestthepredictionsoftheSenseModelinamaskedprimingtaskwithlexicaldecision.IftheassumptionsoftheSenseModelarecorrect,Englishmaterialsthate ectivelyrecreatetherepresentationalasymmetrythoughttoexistbe-tweentranslationequivalentsshouldproducepriming

inthemany-to-onedirection(e.g.,‘‘head–SKULL’’),butnotintheone-to-manydirection(‘‘skull–HEAD’’).Participants

Forty-fourundergraduatesattheUniversityofAri-zonaparticipatedforcoursecredit.Twenty-twooftheparticipantsweretestedinthemany-to-one(thoughttobeanalogoustoL1–L2)direction(Experiment4a)and22weretestedintheone-to-many(thoughttobeanal-ogoustoL2–L1)direction(Experiment4b).Allpartic-ipantswerenativespeakersofEnglish.Materials

Severaldi erentselectioncriteriawereemployed.First,itwasimperativetoselectwordpairsthatwerethoughttoshareacommonsense.Thiswasdonebyselectingsynonyms,orclosesynonyms.Second,itwasimportantthatthematerialssatis edthenumericaldiscrepancycriterionoftherepresentationalasymme-trybetweenthewords.Thiswasdonebyensuringthatonewordinthepairhadonlyonesense(e.g.,‘‘oar’’hasonlyonesense)andthattheotherwordinthepairhadseveral(e.g.,‘‘paddle’’has10senses).SensecountsweretakenfromWordNet1.6(Fellbaum,1998).One-sensewordswerepairedwithwordshavingelevensensesonaverage(seeAppendixAforacompletelistofwordpairsandsensecounts).Fortywordpairsintotalwerecreated.Anequalnumberofnon-wordswasgeneratedbytheARCNonwordDatabase(Rastle,Harrington,&Coltheart,2002)andinterleavedwiththeexperimentalitems.Additionally,40wordsthatwerematchedinlengthandfrequencywiththeex-perimentalprimeswereselectedtoserveascontrolprimes.

Designandprocedure

Experiments4aand4bwereidenticalindesignandprocedure.ParticipantsinbothexperimentswereaskedtodecidewhetherthetargetitemwasanEn-glishwordornot.InExperiment4a,theprimeshadmanysenses,whilethetargetshadjustone.ThiswasreversedinExperiment4b.Eachtrialconsistedofthefollowingsequenceofevents,eachonefollowingimmediatelyaftertheother:forwardmask(########)for500ms,primeinlowercaselettersfor41ms,andtargetinuppercaselettersfor500ms.TwolistswereconstructedforbothExperiments4aand4binordertocounterbalanceacrosstheprimefactor.TargetsthatwereprecededbytheirrelatedprimeonListAwereprecededbyacontrolprimeonListBandviceversa.ItemswerepresentedrandomlytoeachparticipantandRTswererecordedtothenearestmillisecond.