Effects of Categorization and Discrimination Train

EffectsofCategorizationandDiscriminationTrainingonAuditoryPerceptualSpaceFrankH.Guenthera,b,FatimaT.Husaina,MichaelA.Cohena,andBarbaraG.Shinn-CunninghamaJournaloftheAcousticalSocietyofAmerica(1999)Vol.106,pp.2900-2912aDepartmentofCognitiveandNeuralSystemsBostonUniversity677BeaconStreetBoston,MA,02215FaxNumber:[phone]email:[e-mail]ElectronicsMassachusettsInstituteofTechnologyCambridge,MA02139ABSTRACTPsychophysicalphenomenasuchascategoricalperceptionandtheperceptualmagneteffectindicatethatourauditoryperceptualspacesarewarpedforsomestimuli.Thispaperinvestigatestheeffectsoftwodif-ferentkindsoftrainingonauditoryperceptualspace.Itisfirstshownthatcategorizationtrainingusingnon-speechstimuli,inwhichsubjectslearntoidentifystimuliwithinaparticularfrequencyrangeasmem-bersofthesamecategory,canleadtoadecreaseinsensitivitytostimuliinthatcategory.Thisphenomenonisanexampleofacquiredsimilarityandapparentlyhasnotbeenpreviouslydemonstratedforacate-gory-relevantdimension.Discriminationtrainingwiththesamesetofstimuliwasshowntohavetheoppo-siteeffect:subjectsbecamemoresensitivetodifferencesinthestimulipresentedduringtraining.Furtherexperimentsinvestigatedsomeoftheconditionsthatarenecessarytogeneratetheacquiredsimilarityfoundinthefirstexperiment.Theresultsoftheseexperimentsareusedtoevaluatetwoneuralnetworkmodelsoftheperceptualmagneteffect.Thesemodels,incombinationwithourexperimentalresults,areusedtogenerateanexperimentallytestablepredictionconcerningchangesinthebrain’sauditorymapsunderdifferenttrainingconditions.PACSnumbers:43.71.An(modelsandtheoriesofspeechperception),43.66.Ba(modelsandtheoriesofauditoryprocesses).FrankGuentherissupportedinpartbytheAlfredP.SloanFoundationandtheNationalInstituteonDeaf-nessandotherCommunicationDisorders(NIDCDR2902852).FatimaHusainissupportedbyNIDCDR2902852.BarbaraShinn-CunninghamissupportedinpartbyAFOSRgrantF49620-98-1-0108.TheauthorswouldliketothankHarlanLane,DavidPisoni,JacobBeck,JamesBeale,PeterChiu,andKennethStevensforhelpfuldiscussionsregardingthiswork.Guenther,Husain,Cohen,andShinn-Cunningham(1999)21.IntroductionItiswell-knownthatourperceptualspacesforsomeauditorystimuli,suchasphonemes,arewarped.Thatis,theperceptualdistancebetweentwostimuli,asevidencedbyasubject’sabilitytodiscriminatethem,isnotalwaysastraightforwardfunctionoftheirdistanceasmeasuredalongphysicaldimensionssuchasfre-quencyortime1.Englishstopconsonants,forexample,havelongbeenknowntoexhibitcategoricalper-ception(seeJusczyk,1986;Liberman,1996;LiebermanandBlumstein,1988;andRepp,1984forreviews).Forexample,ifsubjectsarepresentedwithsyntheticspeechstimulicreatedbyvaryingthesec-ondformanttransitioninsmallstepsthrougharangecorrespondingtothephonemes/b/,/d/,and/g/,theyshowverypoordiscriminabilitywhentwostimulibothfallwithinoneofthecategoriesandverygooddis-criminabilityforstimulithatstraddlecategoryboundaries,eventhoughthestimuluspairsinthesetwocasesareequidistantinfrequencyspace(Liberman,Harris,Hoffman,andGriffith,1957;Eimas,1963).Otherexperimentshaveshownsimilarcategoricaleffectsforvoiceonsettime(VOT)distinctionsbetween/d/and/t/(Liberman,Harris,Kinney,andLane,1961)andbetween/b/and/p/(Liberman,Harris,Eimas,Lister,andBastian,1961).Similareffectshavealsobeenreportedforavarietyofnon-speechstimuli,includingmelodicmusicalintervals(BurnsandWard,1978),simplevisualshapes(Lane,1965;Goldstone,1994),andmorphedfacesalonganintriguing“JohnF.KennedytoBillClinton”continuum(BealeandKeil,1995).Researchershavealsoshown,relativelyrecently,thattheperceptualspaceforsomesyntheticvowelsandsemivowelsappearstobewarped(e.g.,Aaltonen,Eerola,Hellström,Uusipaikka,andLang,1997;Iverson,Diesch,Siebert,andKuhl,1994;IversonandKuhl,1994,1995;Kuhl,1991,1995;Kuhl,Williams,Lac-erda,Stevens,andLindblom,1992;SussmanandLauckner-Morano,1995).Kuhl(1991)referredtothiswarpingasa“perceptualmagneteffect,”thusdistinguishingitfromcategoricalperception.Roughlyspeaking,theeffectischaracterizedbyawarpingofperceptualspacesuchthatacousticpatternsnearpho-nemiccategoryprototypesareperceivedasclosertogetherthanequallyspacedacousticpatternsthatarefurtherawayfromphonemiccategoryprototypes.AccordingtotheKuhletal.account,themagneteffectdiffersfromcategoricalperceptioninthatitischaracterizedbydifferencesindiscriminabilityforprototyp-icalvs.nonprototypicalstimulithatfallwithinthesamephonemiccategory.Specifically,betterdiscrimi-nationisfoundnearnon-prototypicalmembersofacategorythannearprototypicalmembers.However,otherresearchershaveclaimedthatcategoricalperceptionandtheperceptualmagneteffectareessentiallythesame.Forexample,Lotto,Kluender,andHolt(1998)concludefromtheirstudythatthemagneteffect“maybenothingmorethanfurtherdemonstrationthatgeneraldiscriminabilityisgreaterforcross-categorystimuluspairsthanforwithin-categorypairs”(p.3648).Byeitheraccount,though,theperceptualspaceforvowelsandsemivowelsappearstobewarped,althoughapparentlynotasdramaticallyasforconso-nants.Itisverylikelythatsomeofthewarpingofauditoryspaceis“builtin”totheauditorynervoussystem.Evi-denceforthiscome