The nature of lexical representation in visual word recognition

The Nature of Lexical Representation in Visual Word Recognition
OxfordHandbooksOnline
TheNatureofLexicalRepresentationinVisualWordRecognition
MarcusTaft
TheOxfordHandbookofReading(Forthcoming)
EditedbyAlexanderPollatsekandRebeccaTreiman
OnlinePublicationDate: Oct
2014
Subject: Psychology,CognitivePsychology
DOI: 10.1093/oxfordhb/9780199324576.013.6
AbstractandKeywords
Thischapterexploreshowinformationaboutwordsisrepresentedforthepurposesofrecognizingthosewords
whenreading.Adescriptionisfirstgivenofthevariousarchitecturesthathavebeenproposedtoframeour
understandingoflexicalprocessing,withanemphasisonthewaytheyportraylexicalrepresentation.The
importanceofmorphologicalstructuretothenatureoflexicalrepresentationishighlighted,andattentionis
directedtospecificmodelsthatattempttocapturethatstructure.Themodelthatformsthemajorfocusofthe
chapter,theAUSTRALmodel,isonewhereidentificationofaletterstringisbasedoninformationassociatedwith
anabstractlevelthatmediatesbetweenformandfunction,namely,alemmalevel.Theincorporationofalemma
levelintothelexicalprocessingsystemprovidesalocusformorphologicalstructure.Itcapturesaleveloflexical
representationthatnotonlyunderliesbothvisualandspokenwordrecognitionbutisalsocompatiblewithmodels
ofwordproduction.
Keywords:hierarchicalactivation,lemmas,lexicalrepresentation,modelsoflexicalprocessing,morphologicalprocessing,orthographicprocessing
Introduction
ItiseasyforaliterateEnglishspeakertorecognizethatcatisawordwhilelatisnot.Onlytheformeris
representedinthespeaker’slong-termmemoryforwords(i.e.,thementallexicon),andaccesstothis
representationallowsthespeakertodifferentiatebetweenrealwordsandpronounceablestringsoflettersthatare
nonwords(alsocalledpseudowords,e.g.,frink).Suchalexicaldecisionjudgmentthereforeprovidesawindowinto
thenatureofthelexicalrepresentationsthatareaccessedwhenweread,and,becauseofthis,hasbeenadopted
widelyasalaboratory-basedtasktoexploretheissueoflexicalrepresentationandretrieval.
Participantsinthelexicaldecisiontaskaretypicallypresentedarandomlymixedseriesofwordsandnonwords
andaskedtopressa“yes”or“no”buttonasquicklybutasaccuratelyaspossibleinresponsetowhetherthe
presentedletterstringisawordornot.Reactiontimesanderrorratesaremeasured.Reactiontimes(RTs)and
errorratesaremeasured.TheRTforaworditem(usuallyaveragingaround500ms)reflectsthetimeittakesto
accesstherelevantlexicalinformation,aswellastodecidethatsuchinformationissufficienttoindicatethatthe
letterstringisaword.TheRTforanonword(usuallylongerthanforaworditem)reflectstheamountoflexical
informationaccessedonthebasisoftheletterstringandthetimeittakestodecidethatthisisinsufficientfora
wordresponsetobemade.Differenttypesofitemarecomparedbyincluding15ormoreexamplesofeach.The
twoconditionsmakingupeachtypevaryonthefactorofinterestwhilebeingmatchedcloselyonasmany
importantfactorsaspossible(e.g.,frequencyofoccurrenceinthelanguageinthecaseofrealwords).Inthecase
ofprimingresearch,thesametargetiscomparedwhenprecededbydifferentprimesandtheimpactofdifferent
relationshipsbetweentheprimeandtargetcanthereforebemeasured.
Linguistsarelikelytointerpretthenotionoflexicalrepresentationintermsofthelinguisticinformationassociated
Page 1 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2014. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: University of New South Wales; date: 07 October 2014
The Nature of Lexical Representation in Visual Word Recognition
withaword(i.e.,itssemanticandsyntacticfunctions).Forthecognitivepsychologistwhoisinterestedinhowwe
read,however,anunderstandingofthenatureofthefunctionalinformationitselfislessimportantthanthemental
representationsthatprovideaccesstothatinformation.Forthisreason,theissuethatwillbeaddressedinthis
chapteristhenatureoftherepresentationthatallowsidentificationoftheword.Couchedintermsofthelexical
decisiontask,thequestioniswhatrepresentationisaccessedthatallowsaletterstringtobeidentifiedasa
particularword(ortoalternativelybeclassifiedasanonword).Itisthisrepresentationthatconstitutesthegateway
throughwhichtheincomingletterstringcanbeassociatedwithitsfunctionalinterpretationduringreading.
ModelsofLexicalProcessing
EarlyViews
Thenotionoflexicalaccessinvisualwordrecognitionwasfirstexploredinthelate1960sandearly1970s,with
twodistinctapproachesbeingadopted.
LexicalSearch
Forster(1976)outlinedamodeloflexicalprocessingwhereallinformationaboutaword(semantic,syntactic,
phonological,andorthographic)isstoredinamasterfilethatisaccessedviaaserialsearchthroughmodalityspecificperipheralaccessfiles.Theorthographicaccessfilethatisusedinreadingisalistofwordsinorderof
theirfrequencyofoccurrence,althoughdividedintosmallersizedbinsaccordingtoform-basedcharacteristics.A
visuallypresentedwordmightthereforeberecognizedwhenfoundtomatchwithanentryintheorthographic
accessfile.However,TaftandForster(1975,1976)arguedthatapolymorphemicword(suchasrevive,
henchman)whosestemisnotafree-standingwordisrecognizedwhenthatstem(e.g.,vive,hench)isaccessedin
theorthographicaccessfile,withinformationaboutthewholewordbeingsubsequentlyextractedfromthemaster
fileentry.Assuch,theaccessfileincludesnonwords(e.g.,vive,hench),whichmeansthatwordrecognition
requiresaccesstothemasterfilebecausethatisthelocusofinformationnecessarytodiscriminaterealwords
fromnonwords.
LexicalActivation
Analternativeideathatwordsareaccessedinlexicalmemoryviaaparallelactivationsystemwasincorporatedby
Morton(1969,1970)inhislogogenmodel.Thelexicalentryforeachwordisseenasaninformation-collecting
device(i.e.,alogogen)whoseactivationlevelincreasesinresponsetorelevantfeaturesbeingcontainedinthe
stimulus.Onceenoughevidenceaccumulatesinoneofthelogogensforitsthresholdtobereached,the
correspondingwordbecomesavailableforrecognitionandtherelevantfunctionalinformationassociatedwiththat
wordcanbeaccessed.Thereareseparatesetsoforthographicandphonologicallogogensthatareactivated
dependingonthemodalityoftheinput.Inaddition,Mortonarguedfortheexistenceoforthographicand
phonologicaloutputlogogensthatareusedforwritingandspeaking,respectively,althoughitisunclearhow
parallelincrementalactivationwouldworkwithinanoutputsystem.
Lexicalrepresentationsinthelogogenmodelarethewordscorrespondingtoeachlogogen.However,onfinding
thatinflectedwordsprimerecognitionoftheirstem(e.g.,carsprimescar,whilecarddoesnot),MurrellandMorton
(1974)concludedthatlogogensactuallycorrespondtomorphemes.Sincethelogogenmodeldoesnotspecify
howpolymorphemicwordsmightberepresented,itsnotionoflexicalrepresentationisrathervague.
MoreRecentApproaches
Theideaofserialsearchthroughlexicalmemoryhasbecomelargelyoutdated,withmostinvestigatorsnow
adoptinganaccountthatincorporatesaparallelactivationmechanism.Eventhemainproponentoflexicalsearch
hasrecentlyproposedamodelthatcombinesparallelactivationwithserialprocessing(Forster,2012).Therefore,
thenatureoflexicalrepresentationwillnowbeconsideredinthelightoftheactivationframeworksthatare
currentlymostinfluential.
Interactive-Activation
Page 2 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2014. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: University of New South Wales; date: 07 October 2014
The Nature of Lexical Representation in Visual Word Recognition
Theinteractive-activation(IA)model,asfirstoutlinedbyMcClellandandRumelhart(1981)andRumelhartand
McClelland(1982),elaboratesuponthenotionofparallelactivation.WordsarerepresentedintheIAmodelinmuch
thesamewayasinthelogogenaccount.However,thereisalsoalayerofactivationunitscorrespondingto
individuallettersthatfeedtheiractivationtothewordlevel,andalayerofactivationunitscorrespondingtovisual
featuresthatfeedtheiractivationtotheletterlevel.Thusactivationpassesupfromfeaturestoletterstowords,
withactivatedunitsinhibitingcompetitorsatthesamelevel.Asaunitincreasesinactivation,itfeedsactivation
backdownthehierarchysothatthelowerlevelunitswhoseactivationhasbeenmostproductiveatthehigher
levelwillbestrengthened.Fromthisinteractionofactivationthroughoutthesystem,asinglewordunitwill
eventuallyreachathresholdthatallowstheletterstringtoberecognizedasthatword.
ThewordunitsintheIAmodelarelexicalrepresentationsinthesensethataletterstringcanbeidentifiedonce
onesuchunitreachesitsrecognitionthreshold.However,theterm“lexicalrepresentation”neednotbetakeninits
literalsenseof“thementalportrayalofacompleteword,”butcanbemorebroadlydefinedas“thestored
informationthroughwhichawordcanberecognized.”Giventhatthesublexicalunits(i.e.,featuresandletters)are
integraltotheactivationofwordunitsintheIAmodel,itcanbearguedthattheyarealsoapartofthelexical
representation.Thuswhenconsideringthenatureoflexicalrepresentation,thebreadthofitsdefinitionneedstobe
clear.Tofurtherourunderstandingofvisualwordrecognitionwewillprimarilybeinterestedinthebroadest
definition,becausetheactofreadinginvolvesthewholeprocedureofgettingfromtheletterstringtoits
identificationasaparticularword.
ParallelDistributedProcessing
LiketheIAapproach,theparalleldistributedprocessing(PDP)modelisaconnectionistframework.However,while
theformerrepresentswordsaslocalist(i.e.,specified)lexicalunits,thelattercaptureslexicalinformationwithin
patternsofactivationdistributedacrosssetsof“hidden”unitsthatmediateamongtheorthographic,phonological,
andsemanticlevelsofrepresentation(e.g.,Harm&Seidenberg,2004;Plaut,McClelland,Seidenberg,&Patterson,
1996;Seidenberg&McClelland,1989).Assuch,thereisnoexplicitrepresentationofthewholeword(i.e.,no
lexicalentryperse),onlyapatternofconnectionweightsthathasbeensettleduponthroughrepeatedexperience
withtheletterstring.SincerepresentationsinthePDPmodelarecommontomorethanonewordandalso
participateintheprocessingofnonwords,itcanbesaidthatlexicalrepresentationsdonotexist(cf.Seidenberg&
McClelland,1989,p.560).
However,inageneralsense,theremustbesometypeoflexicalrepresentationwithinthePDPsystemgiventhatit
ispossibletodiscriminatewordsfromnonwords.Lexicaldecisionjudgmentsaremadebycomparingthe
orthographicinputwiththeorthographicoutputthatisgeneratedbythehiddenunitsonthebasisofthatinput.This
providesameasureoforthographicfamiliaritybecausethemorethesystemencountersaletterstring,themore
accuratelytheweightswithinthehiddenunitswillsettleonapatternthatreflectstheorthographicformofthat
letterstring.Sinceanonwordhasneverbeenpreviouslyencountered,thepatternofactivationgeneratedinthe
hiddenunitswillbealessaccuratereflectionoftheorthographicinput,and,ifthismatchbetweeninputandoutput
fallsbelowsomecriteriallevel,anonwordclassificationcanbemade.Withinsuchanaccount,then,itisthesettled
patternofactivationwithinthehiddenunitsthatisequivalenttoalexicalrepresentation.Whiletherepresentation
ofawordmaybedistributedacrossanumberofunitsthatoverlapwiththedistributionofunitsrepresentingother
wordsinthevocabulary,thepatternofweightedactivationisneverthelessuniqueandthereforefunctionsasa
lexicalrepresentation.
WithinthePDPframework,apatternofconnectionweightsbecomesmorestablethemoreanorthographicformis
systematicallyassociatedwithanoutput,eitherphonologicalorsemantic.So,systematicsublexicalrelationships
thatexistbetweenprintandsoundarecapturedwithinthehiddenunitsthatmediatebetweentheorthographicand
phonologicallevels(e.g.,thefactthat‹EE›istypicallypronounced/i:/),whilesystematicsublexicalrelationships
thatexistbetweenprintandmeaningarecapturedwithinthehiddenunitsthatmediatebetweentheorthographic
andsemanticlevels.Infact,theonlysublexicalsystematicitybetweenprintandmeaningisatthelevelofthe
morphemeand,assuch,thehiddenunitsmediatingbetweenorthographyandsemanticsmustreflectmorphemic
ratherthansubmorphemicinformation(e.g.,Gonnerman,Seidenberg,&Andersen,2007;Rueckl&Raveh,1999).
Giventhesimilarlackofsubmorphemicsystematicitybetweenphonologyandsemantics,itisparsimoniousto
amalgamatethehiddenunitsthatmediatebetweenorthographyandmeaningwiththosethatmediatebetween
phonologyandmeaning,andthatiswhatGonnermanetal.(2007)propose.
Page 3 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2014. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: University of New South Wales; date: 07 October 2014
The Nature of Lexical Representation in Visual Word Recognition
ThePDPapproach,regardlessofitsspecificcomputationalimplementation,isimpressiveinthewayitsimulates
knowndatabycapturingthestatisticalrelationshipsthatexistbetweentheorthographic,phonological,and
semanticdomains.However,themodelcanbeseenmoreasafacsimileofhumanreadingperformancethan
providinganunderstandingofit.ThePDPexplanationforhowwereadisessentiallythatthepresentedletterstring
setsupapatternofneuralactivitythatcorrespondstoapatternthatwaspreviouslyacquiredinresponsetothat
letterstring.Suchanaccountisthereforenotveryinstructiveand,accordingly,thePDPapproachhasnotproven
verysuccessfulovertheyearsingeneratingnewresearchintothereadingprocess.Forthisreason,manyhave
soughtamorerevealingapproachtothequestionoflexicalrepresentationandtheprocessesinvolvedinreading
words.ThistypicallyinvolvestheadoptionoftheIAframeworkwherethelocalistdescriptiongivesaclearer
pictureofwhattheunitsthatareinvolvedintheprocessingofthewordrepresent.Adescriptionofonesuch
approachfollows.
TheAUSTRALModel
ThemodeltobeoutlinedhereingreaterdetailwasintroducedinTaft(1991)anddevelopedfurtherinTaft(2006).
ItadoptstheIAframework,butreplacesthewordlevelwithalevelofrepresentationthatprovidesalinkbetween
function(semantic,syntactic,pragmatic,etc.)andformregardlessofthemodalityofinputwhich,assuch,
incorporatesinformationaboutmorphemicstructure.Taft(2006)referstothisasthelemmalevelfollowingthelead
ofBaayen,Dijkstra,andSchreuder(1997)whoadoptedthenotionfromtheproductionliterature(cf.Kempen&
Huijbers,1983;Roelofs,1992).Theinclusionofsuchalevelofrepresentationmakesitthelocalistequivalentof
thePDPmodelofGonnermanetal.(2007)inasmuchasmeaningislinkedtobothorthographicandphonological
formviathesamesetofunitsthatcapturemorphemicinformation.
Clicktoviewlarger
Figure1 :ExampleofhowcatisrepresentedintheAUSTRALmodel.
Figure1depictsaversionofthismodel,whichwillbereferredtoasAUSTRALforreasonstobegivenshortly.The
figureillustratesthewayinwhichthewordcatisbothrecognizedandpronouncedwhenvisuallypresented,and
depictsthelexicalrepresentationinitsbroadestsense.Thatis,itdescribestherepresentationsrequiredtoaccess
thewordduringreading.Representationalunitsattheformlevelaresublexicallybased,whileunitsatthefunction
levelrepresentcomponentialsemanticandsyntacticfeaturesandanyotherinformationrelevanttotheword.A
lemmacanbeseenthenastheunitthatbringsthesecomponentstogether.So,thelemmafor“cat”representsthe
unionofthegraphemesc,a,andt(inthatorder)andthephonemes/k/,/æ/,and/t/withthenotionofacountable
concretenounthathasthesemanticfeaturesrelevanttotheconceptof“cat.”
Althoughneutralwithregardtothewayinwhichthefunctionalinformationmightberepresented,theAUSTRAL
modelspecifiesthenatureofthesublexicalformunits.Inparticular,therehasbeenstrongevidencethat
Page 4 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2014. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: University of New South Wales; date: 07 October 2014
The Nature of Lexical Representation in Visual Word Recognition
consonantalonsetsaretreatedseparatelyfromtherestofthesyllablewhenreadingmonosyllabicEnglishwords
(e.g.,Andrews&Scarratt,1998;Taraban&McClelland,1987;Treiman&Chafetz,1987).Therestofthesyllableis
referredtoasthebodyoftheword(ororthographicrime),andiscomposedofavowelplusconsonantalcodaif
thereisone(e.g.,stristheonsetofstreet,andeetisitsbody,comprisingvoweleeandcodat).Thusthe
sublexicalunitsoftheAUSTRALmodelformahierarchywherebygraphemeunitsactivatebodyunits(e.g.,the
graphemesAandTactivatethebodyAT),andthelemmaisthenactivatedthroughthecombinationoftheonset
andbody.Itisthis“activationusingstructurallytieredrepresentationsandlemmas”thatcharacterizesthemodel
andcreatestheacronymAUSTRAL.1
Byincorporatingstructurallytieredorthographicrepresentationslinkedtophonologicalunitsofacorresponding
structure,theAUSTRALmodelembodieslexicalandsublexicalinformationinadifferentwayfromdual-route
modelsofreadingaloudsuchasDRC(e.g.,Coltheart,Rastle,Perry,Langdon,&Ziegler,2001)andCDP++(e.g.,
Perry,Ziegler,&Zorzi,2010),wherelexicalandsublexicalinformationareprocessedthroughdifferentpathways.
LikethePDPmodel,AUSTRALpronouncesnonwordsviathesameunitsthroughwhichwordsarenamed,except
thatnonwordslackthesupportthatwordsreceivefromthelemmalevel.Taft(1991,2006)discussesthewayin
whichAUSTRALreadilyexplainswhyregularwords(e.g.,pink)arenamedwithashorterlatencythanirregular
words(e.g.,pint)andwhyregularwordswithabodythatisinconsistentlypronounced(e.g.,hint)alsoshow
slowernaminglatencies.However,thefocusofthischapterisnotonthegenerationofsoundfromprintbutonthe
identificationofvisuallypresentedwords.Sodiscussionwillnowberestrictedtothewayinwhichavisually
presentedletterstringisrepresentedforthepurposesofrecognitionduringreading.
Informationaboutthewholewordisfoundatthelemmalevel,whichpotentiallymakesitthelocusoflexical
representationinitsnarrowestsense.Thatis,wordscanpotentiallybediscriminatedfromnonwordsatthelemma
levelbecauseonlywordshavedevelopedaunitlinkingformwithfunction.However,thisisnotentirelytrue
because,aswillbeexplainedlater,therearerealwordsthatmightnotberepresentedbyalemmaandthereare
alsononwordsthatmightbe.Toelucidate,weneedtoconsiderhowmorphologicalstructureisrepresentedinthe
lexicalprocessingsystem(seealsoHyönä,thisvolume).Aswillbeseen,thewayinwhichmorphemicallycomplex
wordsareprocessediscentraltoourunderstandingoflexicalrepresentation.
MorphologicalProcessing
Amorphemeisusuallydefinedasthesmallestunitofformassociatedwithasemanticorsyntacticfunction.Given
thatalemmaisaunitthatencapsulatestheassociationbetweenformandfunction,itfollowsthatthelemmalevel
capturesmorphemicstructureanddrivestheprocessingofmorphemicallycomplexwords(e.g.,cats,unfriendly,
daydream).Howaresuchpolymorphemicwordsrecognized?
ObligatoryDecompositionintheAUSTRALModel
Inaccordwithitsnotionthattheformlevelrepresentscomponentsofthewholeword,theAUSTRALmodelhasall
polymorphemicletterstringsbeingdecomposedintotheirapparentmorphemesforrecognitiontotakeplace(e.g.,
un,friend,andly).ThisideaofobligatorydecompositionhasbeenaroundsinceTaftandForster(1975),andhas
beensupportedinmorerecenttimesbyresearchusingthemaskedprimingparadigm.AsoverviewedbyRastle
andDavis(2008),manyexperimentshaveshownnotonlythatthemaskedpresentationofatransparentlyderived
word(e.g.,hunter)facilitatessubsequentrecognitionofitsstem(hunt),butthatthesameistrueofa
pseudoderivedword(e.g.,corner-corn).Becausenofacilitationisreportedwhentheprimedoesnotincludea
putativesuffix(e.g.,turnip-turn),itisconcludedthatapseudoderivedwordisblindlydecomposedintoitsapparent
morphemes(e.g.,cornander).Onlyatalaterstageisthisanalysisoverturnedsothatthepseudoderivedwordis
correctlytreatedasamonomorphemicword.Howthenisapolymorphemicwordrecognizedaftertheform
representationsofitscomponentmorphemesareaccessed?Taft(2003,2004)andTaftandNguyen-Hoan(2010)
arguedthatthelemmasforeachofthecomponentmorphemesareactivatedviatheirformrepresentationsand
thentherearetwopossiblewaysinwhichthewholewordmightberecognized.
First,ifthefunctionofthepolymorphemicwordisentirelytransparentwithrespecttothefunctionofitscomponent
morphemes,asistypicallythecaseforregularlyinflectedwords(e.g.,cats,jumped,eating),thewholewordcan
berecognizedpurelyonthebasisofthefunctionalinformationassociatedwitheachmorphemelemma.Thestem
Page 5 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2014. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: University of New South Wales; date: 07 October 2014
The Nature of Lexical Representation in Visual Word Recognition
andtheaffixeachhavetheirownlemma.Forexample,onceitisknownwhatacatisandthatthesuffixscan
denotethepluralofacountablenoun,everythingthatisknownaboutthewordcatscanbedetermined.So,asina
printeddictionary,thereisnoneedforawhole-wordrepresentationtoexistinthementallexicon,becauseitwould
beredundant.Thesearethetypesofrealwordsreferredtoearlierthatwouldnotberepresentedatthelemma
levelinthemodel,butthatcannonethelessberecognizedonthebasisoffunctionalinformation.
Second,andincontrasttowordswhosemorphologicalcompositionisentirelytransparent,apolymorphemicword
thathasanysemanticorgrammaticalfunctionthatcannotbedeterminedonthebasisofitscomponent
morphemesmustberepresentedbyawhole-wordlemmatoprovidealinktothatidiosyncraticinformation.For
example,thereneedstobealemmaforfriendlytoassociateitwiththefunctionalknowledgethatitmeansmore
thanjust“characteristicofafriend.”Similarly,thepresentparticipleofmeet(asinIammeetingherforthefirst
time)isentirelyunderstandablefromthecombinationofthefunctionsassociatedwiththestemandaffixlemmas
and,hence,doesnotrequirealemma,buttheredoesneedtobeawhole-wordlemmaformeetingwhenusedasa
gerund(asinwe’llholdameetingtonight)inordertounderstandthatitspecificallymeans“anassemblyof
peopleforthepurposesofdiscussion.”Taft(2003,2004)andTaftandNguyen-Hoan(2010)proposethatsucha
whole-wordlemmaisactivatedviathelemmasforitscomponentmorphemes,creatingahierarchyoflemmasfrom
monomorphemictopolymorphemic.
SupportfortheexistenceofsuchahierarchyoflemmasispresentedbyTaftandNguyen-Hoan(2010)froma
maskedprimingexperimentwithambiguoustargets(e.g.,stickmeaningeither“atwig”or“adhere”).Theresults
showedthatwhenaskedtoprovidethemeaningofthetarget,participantswerebiasedbythemeaningsuggested
byaprimethatwasanaffixedversionofthetarget.Forexample,moreparticipantsgavethe“adhere”meaningof
stick(asopposedtothe“twig”meaning)whenprecededbythemaskedprimestickythanwhenprecededbyan
unrelatedword.Moreover,therewasnosuchbiaswhentheprimewasonlysemanticallyrelatedtothatmeaning
withoutbeingamorphologicalvariantofthetarget(e.g.,thewordglue).Thislackofabiastowardthe“adhere”
meaningofstickwhengluewastheprimeindicatesthattheobservedbiastothe“adhere”meaningwhensticky
wastheprimecouldnothavearisensolelyatthesemanticlevel.Neithercoulditslocusbetheformlevelbecause,
logically,thetwoversionsofahomographarenotdifferentiatedatthatlevel,beingidenticalinform.Itwas
thereforeconcludedthatthelocusofmeaningbiasmusthavebeenalevelthatmediatesbetweenformand
semantics,namely,thelemmalevel.So,thelemmaforstickyisactivatedviathelemmaforonlyoneversionof
stick(i.e.,the“adhere”version)and,whenstickyispresentedastheprime,thatversionremainsactivewhenthe
targetarrives,hencebiasingtheresponse.
BoundMorphemes
Ifmorphemesarerepresentedatthelemmalevelbecausetheycapturethecorrelationbetweenformandfunction,
thisshouldbeequallytruewhetherthemorphemeisfreeorbound,thatis,whetherornotitcanstandasawordin
itsownright.Affixesarethetypicalboundmorphemes(e.g,un,y,ing),butsomestemsarealsobound.For
example,vengecannotstandonitsownasaword,yetitoccursinrevenge,avenge,vengeful,andvengeance,
whichclearlyhaveoverlappingmeanings.Therefore,itisargued(seeTaft,2003)thatvengedevelopsalemmato
capturethisform-meaningcorrelation,throughwhichthewhole-wordlemmasforrevenge,avenge,andsoonare
activated.
Withnonwordsexistingatthelemmalevel(i.e.,whentheyareboundmorphemes),itcannotbethecasethat
lexicaldecisionresponsesaremadepurelyonthebasisoftherebeingalemmacorrespondingtothepresented
letterstring.Whileclassifyingaboundstemasanonwordiscertainlydifficult(e.g.,Taft,1994;Taft&Forster,
1975),itisneverthelesspossibletodoso,andtheAUSTRALmodelneedstoexplainhow.Thesimplestexplanation
isthatthereisinformationlinkedtothebound-stemlemmathatstipulatesthatitcannotbeusedasawordinits
ownright;informationthatwouldbeparticularlyimportantwhenitcomestoproduction.Itmightbethecasethat
thisinformationtakestheformofafurtherlevelthatrepresentslexicalconcepts,ashasbeenproposedinrelation
tospeechproduction(e.g.,Levelt,Roelofs,&Meyer,1999).Thatis,onlyfree-standingwords,bethey
monomorphemicorpolymorphemic,correspondtoholisticconcepts.Boundmorphemesdonotandcan,therefore,
berejectedaswordsonthatbasis.
Anotherpossibilityisthatlemmasvaryinsomewayasafunctionoftheform-meaningcorrelationthatthey
capture.Asystematicrelationshipbetweenformandfunctionislikelytobemostobviouswhentheformretainsits
Page 6 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2014. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: University of New South Wales; date: 07 October 2014
The Nature of Lexical Representation in Visual Word Recognition
meaningwithinavarietyofcontexts,becausetheconstancyoftherelationshipcontrastswiththevariabilityofthe
linguisticinformationsurroundingit.Assuch,arealwordwillhaveastrongercorrelationthanwillabound
morpheme.Althoughaboundmorphememightrecurinseveraldifferentcontexts(i.e.,withdifferentaffixes
attached),thesewillbefewerthanthenumberofcontextsinwhicharealwordcanrecur(i.e.,allthesentencesin
whichthatwordisencountered).Therefore,theremightbeathresholdofcorrelationabovewhichtheletterstring
isclassifiedasaword.Ofcourse,ifsuchanargumentweretobepursued,themechanismbywhichalemmais
abletovaryonthebasisofform-meaningcorrelationwouldneedgreaterspecification.
MorphologicalDecompositioninOtherModels
Whileotherwordrecognitionmodelsthatfocusonmorphologicalprocessingalsoincorporatethenotionofa
lemmalevel,theydifferfromAUSTRALinanumberofways.
SchreuderandBaayen(1995)
AsinAUSTRAL,SchreuderandBaayen(1995)proposetheseparationofaccessrepresentationsfromlemmas
(which,priortoBaayenetal.,1997,werelabeledas“concepts”;seealsoTaft,1991).However,accordingto
SchreuderandBaayen(1995),on-linedecompositiononlyoccursattheearlieststagesofacquiringanew
polymorphemicword.Otherwise,polymorphemicwordsareidentifiedthroughawhole-wordaccessrepresentation
thatactivateseitheralemmacorrespondingtothewholepolymorphemicwordorlemmascorrespondingtoits
componentmorphemes,dependingonhowtransparentlyrelatedthosemorphemesaretothewholeword.
Therefore,apartfromnewlyexperiencedpolymorphemicwords,theonlydecompositionthatoccursinthemodel
ofSchreuderandBaayen(1995)isatthelemmalevelafterwhole-wordaccess,andonlyforsomewords.
However,suchanotionofpostlexicalactivationofconstituentmorphemes(asalsoproposedbyBurani&
Caramazza,1987,andGiraudo&Grainger,2000,2001)failstoexplainthepseudoderivedmaskedpriming(e.g.,
cornerprimingcorn)thatwasdescribedearlier(seeRastle&Davis,2008).Thatis,accordingtothisaccount,atno
pointinitsrecognitioniscornereverdecomposedintocornander,becauseitswhole-wordaccessrepresentation
onlyactivatesawhole-wordlemma.Therefore,thereisnoreasonfortheprocessingofcornertoinfluencethe
processingofcornotherthanthroughorthographicoverlap,inwhichcaseturnipshouldequallyprimeturn.For
thisreason,othermodelshaveincorporatedearlymorphologicaldecompositionbasedpurelyonforminorderto
tapintoaso-calledmorpho-orthographiclevelofrepresentation.
Diependaele,Sandra,andGrainger(2009)
ThemodelproposedbyDiependaele,Sandra,andGrainger(2009)hastwolevelsbasedonform;themorphoorthographiclevelwherehunterisrepresentedbyhuntander,andthelexicalformlevelwhereallwordsare
represented,includinghunter,hunt,andcorner.Lexicalformrepresentationsforpolymorphemicwordsare
activatedbothviathedecompositionalpathwaythatismediatedbytherelevantmorpho-orthographicunits(e.g.,
huntander),anddirectlyfromtheletterstringwithoutmediation.Itisthroughtheformerpathwaythatcornerwill
primecorn,sincethelexicalformofcornwillbepreactivatedviaitsmorpho-orthographicunitthatisinadvertently
activatedwhencornerisblindlydecomposed.Thedifferencebetweentheprocessingofapseudoderivedandtruly
derivedwordisthatthelexicalformofsuchwords(e.g.,hunter,corner)receivesactivationfromthemorphoorthographiclevelwhenithasatruestem(e.g.,hunt),butnotwhenithasapseudostem(e.g.,corn).2 Links
betweenthemorpho-orthographiclevelandlexicalformlevelarisefromfeedbackfromahighermorpho-semantic
levelwherewordsarerepresentedasmorphemeunits(suchthatthelexicalformunitforhunteractivatesthe
morpho-semanticunitsforhuntander,butthelexicalformunitforcornerdoesnotactivatesemanticunitsforcorn
ander).Assuch,themorpho-semanticunitsfunctionasmorpheme-basedlemmas,withinformationaboutthe
wholepolymorphemicwordonlybeingfoundatthelexicalformlevel.Therefore,themodelincorporatesbotha
prelexicaldecompositionpathway(i.e.,basedonsublexicalinformation)andapostlexicaldecompositionpathway
(i.e.,basedonlexicallystoredinformation),withlexicaldecisioncenteringontheexistenceofanintermediate
lexicalformrepresentation.However,ifawordcanberecognizedthroughdirectaccesstothewholelexicalform,
whatisthepurposeofprelexicaldecomposition?Ifitsomehowmakesaccesstothelexicalformofacomplexword
easierthanwhole-wordaccess,whatthenisthepurposeofpostlexicaldecomposition?
Crepaldi,Rastle,Coltheart,andNickels(2010)
Page 7 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2014. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: University of New South Wales; date: 07 October 2014
The Nature of Lexical Representation in Visual Word Recognition
Clicktoviewlarger
Figure2 :TherepresentationofderivedandpseudoderivedwordsaccordingtoCrepaldi,Rastle,Coltheart,
andNickels(2010).
Diependaeleetal.(2009)differentiatetheearlyprocessingoftrulyderivedandpseudoderivedwordsinorderto
capturetheapparentfactthatmaskedprimingisstrongerfortheformerthanthelatter.Incontrast,Crepaldi,
Rastle,Coltheart,andNickels(2010)maintainthatthemagnitudeofmaskedprimingisnotsignificantlydifferentfor
trulyderivedandpseudoderivedwords,thoughonlytheformergenerateprimingwhenunmaskedand,therefore,
thetwotypesofwordsareonlydistinguishedatalatesemanticstage.AccordingtoCrepaldietal.(2010),morphoorthographicrepresentationsareactivatedviaobligatorydecompositionandthese,inturn,combinetoactivatethe
formrepresentationforthewholeword(atalevelreferredtoastheorthographiclexicon).Whilesucha
decompositionalpathwayalsofeaturesintheDiependaeleetal.(2009)account,Crepaldietal.(2010)specifythat
activationofthewhole-wordformismediatedbythecomponentmorphemesregardlessofsemantictransparency.
AsseeninFigure2,themorpho-orthographicrepresentationsCORNandERactivatetheorthographiclexicalunit
forCORNERinexactlythesamewaythatthemorpho-orthographicrepresentationsHUNTandERactivatethe
orthographiclexicalunitforHUNTER.
AccordingtoCrepaldietal.(2010),activationfromtheorthographiclexiconpassestoalemmalevelandthenon
tothesemanticsystem.However,unlikeAUSTRAL,derivationallyrelatedwordsdonotsharealemma.Thismeans
thattherelationshipbetweenthelemmasforcornerandcornisexactlythesameasthatforhunterandhunt.Itis
onlyinthesemanticsystemthatthetwotypesofwordsdiffer,becausethelatterhaveoverlappingsemantic
featuresandtheformerdonot.ThepurposeofthelemmalevelintheCrepaldietal.(2010)accountissolelyto
capturetherelationshipbetweeninflectionallyrelatedwords,wheresuchwordssharealemmaregardlessof
whethertheinflectionisregular(e.g.,catsandcat)orirregular(e.g.,fallandfell).
IrregularlyInflectedWordsandWhole-WordFormRepresentation
ThemajormotivationforCrepaldietal.(2010)toincludeanorthographiclexiconintheirmodelistocapturethe
factthatrealinflectedwords(e.g.,jumped)canbedistinguishedfromnonwordscomposedofarealstemandaffix
(e.g.,falled,sheeps).Thelatterhavenorepresentationintheorthographiclexicon,whilerealinflectedwordsdo.
Irregularlyinflectedwords(e.g.,fell,taught,sheep,teeth)arenotdecomposedattheformlevel,butactivatethe
lemmafortheirstem,andthisistrueinAUSTRALaswell(seealsoAllen&Badecker,2002).Presentationofeither
fellorfallwilldirectlyactivatethelemmaforfall.AccordingtoAUSTRAL,theformerwillalsoactivatethelemma
correspondingtothepasttensejustasthesuffixedwould(seeTaft,2003),becauseotherwisetherewouldbeno
waytodistinguishfellfromfall.Withwordsbeingidentifiedonthebasisofinformationassociatedwiththelemmas
intheAUSTRALmodel,whatstopsfalledfrombeingrecognizedasaword,giventhatitwillactivatelemmasthat
canbecombinedonthebasisoffunctionalinformation(e.g.,fallisaverbthatcantakethepasttense)?
ThewaytheAUSTRALmodelcanhandlethisisbysimplyhavingthelemmaforthestembeingexplicitlyassociated
withinformationstipulatingthattheworddoesnotfollowregularinflectionalpatterns.Thatis,a“yes”response
couldultimatelybeavoidediffalledwerepresentedbecause,afterdecomposition,informationassociatedwiththe
lemmaforfallwouldspecifythatitspasttenseisactuallyfellor,moregenerally,thattheregularinflectionedisnot
appropriateforthisword.So,againsttheclaimofCrepaldietal.(2010),itispossiblefortheAUSTRALmodelto
distinguishallwordsandnonwordswithouttheneedforaform-basedlexiconthatincludesallpossiblewords,
Page 8 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2014. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: University of New South Wales; date: 07 October 2014
The Nature of Lexical Representation in Visual Word Recognition
includinginflectedones.
However,thereisafurtherissueinrelationtoirregularlyinflectedwordsthathasimplicationsforthenatureof
form-basedrepresentations.InthedescriptionoftheAUSTRALmodelaspresentedhereandinTaft(1991,2006),
thesubsyllabicunitsofonsetandbodyaredepictedasthehighestlevelofformrepresentation.Atothertimes,
though,themodelhasbeenpresentedwithwhole-wordformrepresentations,atleastwhenthewordsare
monosyllabic(e.g.,Taft,2003,2004;Taft&Nguyen-Hoan,2010).Inotherwords,whetherornotthehighestlevel
offormrepresentationcorrespondstothewholewordhasbeenanopen-endedaspectofthemodel.However,the
proposedaccountofirregularwordprocessingseemstonecessitatetheexistenceofawhole-wordform
representation,oratleastawhole-syllablerepresentation.Theonlywayinwhichanirregularlyinflectedwordcan
activatealemmaassociatedwithitsrelevantsyntacticfunction(e.g.,thepast-tenselemmawhenfellispresented
ortheplurallemmawhenteethispresented)isifthereisawhole-wordformrepresentationthatcanbelinkedto
thatlemma.Itisthewhole-wordformFELLthatisassociatedwiththepast-tenselemma,ratherthaneitheritsonset
ForitsbodyELL.IfitwerethebodyELLthatactivatedthepast-tenselemma,thiswouldhappennotonlywhenfell
waspresented,butwhenanyotherwordthatcontainsellwaspresented,suchasyellorspell.Obviously,itwould
beinappropriatetoactivateinformationaboutthepasttensewhenthewordisnotactuallyapast-tenseverb.
Therefore,inalocalistmodelsuchasAUSTRALitseemsnecessarytoallowforwhole-wordformrepresentations,
evenifnotforallwords(i.e.,notforregularlyaffixedwords,orpossiblyevenpolysyllabicwords;seesection
“RepresentationofPolysyllabicWords”).
GradedEffectsofMorphologicalRelatedness
Thereisafurtheraspectofmorphologicalstructurethatisinformativewithregardtolexicalrepresentation.The
transparencyofwhetherthereisaderivationalrelationshipbetweenanaffixedwordanditsstemisnotanall-ornonepropertyofthosewords.Itisclearthathunterandhuntarederivationallyrelated(i.e.,ahunteris“someone
whohunts”),whilecornerandcornarenot.However,manycasesshowapartialrelationship,suchasarcherand
archwheretheshapeofthelatteriscapturedinthebowusedbytheformer,orheartyandheartwherea
metaphoricalsenseofthelatterseemstobecontainedintheformer(asitalsoisinheartfeltandwholehearted).
Speakersaresensitivetosuchagradationinderivationaltransparencybothwhenaskedtoratesemantic
relatednessandinthedegreeoffacilitationoflexicaldecisionresponseswhenthestemisprecededbyan
unmaskedversionofthecomplexwordthatcontainsit(e.g.,Gonnermanetal.,2007).Howissuchacontinuumof
transparencytobeincorporatedintomodelsoflexicalprocessing?
Gonnermanetal.(2007)arguethatthetransparencycontinuumisexpectedwithinaPDPmodelbecausehidden
unitsstatisticallycapturetherelationshipbetweenformandmeaning,andthemoretransparentlyrelatedtwowords
areinbothformandmeaning,thegreatertheoverlapintheirpatternofactivationwithinthosehiddenunits.This
overlapprovidesthebasisforagradationinpriminginlinewithderivationaltransparency.Accordingtosuchan
account,thereisnoneedforamorpho-orthographicstageinwhichlettercombinationsthatcorrespondtoaffixes
areblindlystripped,becausepatternsofactivationcorrespondingtodifferentmorphologicalstructuresareentirely
encapsulatedwithinthehiddenunits.Gonnermanetal.(2007)supportsuchaclaimbydrawingonanunpublished
maskedprimingstudybyGonnermanandPlaut(2000)wherepseudoaffixedwordsfailedtoprimetheir
pseudostems(e.g.,cornernotprimingcorn).However,morerecentresearchhasclearlyshownmaskedprimingfor
alllevelsoftransparencyincludingpseudoderivations(e.g.,Marslen-Wilson,Bozic,&Randall,2008;Rastle&
Davis,2008),anoutcomethatseemsincompatiblewiththePDPaccountasitstands(thoughseeRueckl&Aicher,
2008,forspeculationastohowitmightcomeabout).Themaskedprimingobservedforpseudoderivedwords
seemsmostreadilyexplainedbytheexistenceofastageofdecompositionthatisblindtosemanticfactors,
namely,amorpho-orthographicstage.
Canagradedeffectoftransparencyinratingsandunmaskedprimingbehandledwithinamodelthatalsoincludes
morpho-orthographicprocessing?Certainly,theCrepaldietal.(2010)accounthasnoproblemincorporatingthe
ideaofgradedfeedbackfromthesemanticleveldependingontherelationshipbetweenthecomplexwordandits
stem.Suchfeedbackistheonlysourceofdifferentiationbetweentransparentlyandopaquelyderivedwords,and
comesintoplaywhenthecomplexwordismorefullyprocessed(asintheunmaskedprimingparadigm).
Page 9 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2014. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: University of New South Wales; date: 07 October 2014
The Nature of Lexical Representation in Visual Word Recognition
Clicktoviewlarger
Figure3 :TheAUSTRALrepresentationofderivedandpseudoderivedwords.
TheideaofgradedeffectsoftransparencyismoreofanissuefortheAUSTRALaccount.AsdepictedinFigure3,
TaftandNguyen-Hoan(2010)supposethatatransparentlyderivedword(e.g.,hunter)isactivatedthroughthe
lemmaforitsstem(hunt),whereasapseudoderivedword(e.g.,corner)anditspseudostem(corn)competewith
eachotheratthelemmalevel.Thegreatertheopportunityfortheinhibitoryimpactofcompetitiontocomeinto
play,thegreateritscounteractingimpactonthemorpho-orthographicfacilitationthatoccursattheformlevel.
Therefore,underunmaskedconditions,pseudoderivedwordswillshownofacilitation,unliketransparentwords
whereprimingisgeneratedatboththeformandlemmalevels.Withsuchaqualitativedifferencebetweenthe
representationsofthetwotypesofwords,though,howcanpartiallyrelatedwords(e.g.,archerandarch)showa
gradedeffectofprimingintheunmaskedprimingparadigm?
OnewaytocapturethiswithintheAUSTRALframeworkwouldbetoproposethatlemmasthatareactivatedviathe
sameformrepresentationarealwayslinked,butwithweightingsthatrangefromzerothroughtoastronglypositive
value,dependingonfeedbackbasedontheirsemanticoverlap.So,thelemmasforcornandcornerarelinked,but
withaminimalweighting.Whencornerispresentedastheprime,thelemmaforcornwillbeactivated,butwillsend
negligibleactivationtothelemmaforcornerwhichisactivateddirectlyfromtheformlevel.Inturn,thelemmafor
cornerwillprovidenegligiblesupporttothelemmaforcorn,whicheffectivelyplacesthetwolemmasincompetition
witheachother.Incontrast,thelemmaforhuntwillsendpositiveactivationtothelemmaforhunterandvice
versa,aswillthelemmasforarchandarcher,buttoalesserdegree.Whentheprimeisunmasked,therewillbe
sufficientopportunityfortheimpactoftherelationshipbetweenthetwolemmastomodulatetheeffectofpriming.In
thisway,thegradedeffectofpartialtransparencyintheunmaskedprimingtaskcanbehandledwithinthe
AUSTRALmodel.
WithreferencetoFigure3,theinclusionofaweightedlinkbetweenthelemmasforacomplexwordandthosefor
itscomponentmorphemeswouldmeanthatthereisanadditionallinkbetweenthelemmaforcornandthatfor
corner.However,beingsoweak,itwouldloseouttothepathwaythatlinksthecornerlemmadirectlytotheform
unitsCORNandER.Ifconceptualizedinthisway—namely,asacompetitionbetweenadirectform-to-lemmalink
andamediatedlemmalink—thesameshouldholdforallcomplexwords,includingtrulyderivedwords.Thatis,not
onlyisthelemmaforhunterlinkedtothelemmaforhunt,butitalsoreceivesdirectactivationfromtheformunits
forHUNTandER.Inthiscase,however,thepositivelinkagebetweenlemmasmakesitamoreeffectivepathway
thanthedirectlinkfromtheformleveland,therefore,recognitionofthetransparentlyderivedwordwilltypicallybe
basedonthemediatedpathway.
Theideaofcompetingpathwayshasbeenpreviouslyproposedinrelationtomorphologicalprocessing(e.g.,
Baayenetal.,1997;Bertram,Schreuder,&Baayen,2000;Colé,Beauvillain,&Segui,1989;Diependaeleetal.,
2009;Niswander,Pollatsek,&Rayner,2000).Inthoseaccounts,though,thetwopossiblepathwayscorrespondto
decompositionversuswhole-wordaccess.Incontrast,form-baseddecompositionalwaysoccursinAUSTRAL.Itis
onlyafterdecompositionthataccesstothewhole-wordrepresentationmightfollowtwocompetingpathways:one
directfromtheform-basedcomponentsandoneviathelemmascorrespondingtothosecomponents.Thus
flexibilitycanbeintroducedintotheAUSTRALmodel,withtheoverlapinsemanticfeaturesdeterminingwhich
postdecompositionalpathwayisstronger.
RelationshipBetweenReceptionandProduction
Themodelsdescribedinrelationtothereadingprocessaddressthelexicalmechanismsinvolvedingettingfrom
Page 10 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2014. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: University of New South Wales; date: 07 October 2014
The Nature of Lexical Representation in Visual Word Recognition
printtomeaning,andsimilarmechanismsarelikelytounderlietherecognitionofspokenwordsoncephonetic
variationistakenintoaccount.Itwouldalsomakesensetoproposethatthesamelexicalmechanismsareinvolved
inspeechproduction,althoughrunninginthereversedirection.Infact,AUSTRALowesmuchtothemodelsof
speechproductionproposedbyDell(1986)andbyLeveltetal.(1999).TheDell(1989)modelhasalevelof
morphemerepresentation(equivalenttothelowestlevelofthelemmahierarchyinAUSTRAL)abovewhichisa
wordlevelthatincludesderivationallycomplexwords(potentiallyequivalenttothehigherlevelofthelemma
hierarchyinAUSTRAL),butdoesnotincludeinflectedwordswhichareinsteadactivatedthroughthefunctionallydeterminedcombinationofthestemandaffix,asinAUSTRAL.Furthermore,thehighestlevelofformrepresentation
intheDellmodelcomprisessyllableswhich,inturn,activatetheircomponentonsetsandrimespriortoactivating
theirphonemes.ThisisthesameastheformleveloftheAUSTRALmodelinreverse,albeitphonologicallybased
ratherthanorthographic.
TheexistenceofalemmalevelisamajorfeatureoftheWEAVER++modelofLeveltetal.(1999),butthereisalso
aseparatelevelthatrepresentslexicalconcepts.Theproductionofawordbeginswithactivationofa
semantically-basedconceptwhich,inturn,activatesalemma.Thelemmaisseenasalinktosyntacticinformation
abouttheword.InreceptionmodelssuchasAUSTRAL,littleemphasishasbeenplacedonthisdistinctionbetween
semanticandsyntacticinformation.Thereasonforthisisthatwhenpassingfromtheformleveltothefunction
level(asisthecaseinwordrecognition),thelemmaprovidesdirectlinkstobothsyntaxandmeaningandtheir
orderofactivationisimmaterial.Assuch,thedistinctionbetweenthelocusoflexicalsemanticsandlexicalsyntax
maybecriticalforspeechproduction,buthaslittleimpactonlexicalprocessinginreading.
FutureDirections
ImpactofDerivationalTransparencyinMaskedPriming
Theequivalenceofmaskedprimingfortrulyderivedandpseudoderivedwordswouldseemtobecriticalforthe
Crepaldietal.(2010)model,sincethetwotypesofwordsaretreatedinexactlythesamewaythroughoutmuchof
theirprocessing.However,variableresultshavebeenobservedwithregardtothisissue.AlthoughRastleand
Davis(2008)concludedfromtheiroverviewofsuchresearchthatthereisnodifferenceinprimingbetween
transparentlyderivedandpseudoderivedwords,astatisticalmeta-analysisindicatesthatprimingmightbegreater
fortheformerthanforthelatter(cf.Feldman,O’Connor,&MoscosodelPradoMartín,2009;Taft&Nguyen-Hoan,
2010;butseeDavis&Rastle,2010).Ifthisisindeedthecase,itwouldbehardtomaintaintheCrepaldietal.
(2010)account.Therefore,itisimportanttoestablishwhetheragenuinedifferencebetweentransparentand
pseudoderivedprimingcanbefoundundermaskedconditions.TheAUSTRALaccountisflexiblewithregardtothis
questionbecausewhetherornotprimingisgreaterfortransparentlyderivedthanpseudoderivedwordswill
dependonwhethercompetitioninthelattercasehastheopportunitytocomeintoplay.
DualPathwaysfromFormtoLemma
AccordingtothemodificationtotheAUSTRALaccountwherebygradedeffectsoftransparencyarisefromthe
differentialuseofmediatedanddirectlinks,thereshouldalsobegradedeffectsintheimpactofstemfrequencyon
complexwordrecognition.Iflemmaactivationisinfluencedbywordfrequency,recognitionofaderivedwordwill
beinfluencednotonlybyitsownfrequencyofoccurrencebutalsobythefrequencyofitsstem,ifrecognitionis
mediatedthroughthelemmaforthatstem.Theimpactofstemfrequencyhasbeenwelldocumented(e.g.,Baayen
etal.,1997;Bertram,etal.,2000;Coléetal.,1989;Niswanderetal.,2000;Taft,1979b,2004;Taft&Ardasinski,
2006),soitwouldmakesensetoexaminefurtherwhetherthestrengthofthiseffectvariesasafunctionof
transparency.Itshoulddosoifsemantictransparencydetermineswhichofthecompetingpathwaysucceeds,
becauseonlylemmamediationwillbeinfluencedbystemfrequency.TheCrepaldietal.(2010)modelwould
expectstemfrequencyeffectsregardlessoftransparencybecauseaccesstothewhole-wordrepresentationisthe
sameforalltypesofwords.
RepresentationofPolysyllabicWords
ItcanbeseenfromFigure3thatAUSTRALbreaksdownmonomorphemicwordslikecornerintosubunitsattheform
Page 11 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2014. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: University of New South Wales; date: 07 October 2014
The Nature of Lexical Representation in Visual Word Recognition
level.Whileitispossiblethatthisonlyhappenswhenthestructureofthewordismorphologicallycomplexinits
appearance,itismoreparsimonioustoassumethatallpolysyllabicwordsaresimilarlybrokendown,evenif,unlike
CORNandER,eachoftheform-basedsubunitsisnotassociatedwithitsownlemma.Forexample,thelemmafor
walrusmightbeactivateddirectlyfromorthographicunitsrepresentingthemeaninglesssyllablesWALandRUS
(seee.g.,Taft&Krebs-Lazendic,2013).Wheredoesthesyllableboundaryfallincasesthatarepotentially
ambiguousinthisregard?Forexample,sermoncouldbeorthographicallybrokendownintoserandmonin
correspondencetothewayitispronounced,or,alternatively,theinformativenessofitsfirstsubunitmightbe
increasedbybreakingitdownintosermandon,maximizingthecodaofthefirstsyllable.Thelatteranalysishas
beenproposedbyTaft(1979a,1987),TaftandKougious(2004),andTaftandKrebs-Lazendic(2013),withthe
maximizedfirstsyllable(e.g.,serm)beingreferredtoastheBasicOrthographicSyllabicStructure(orBOSS);
thoughsuchanideahasbynomeansfoundwidesupport(seee.g.,Katz&Baldasare,1983;Lima&Pollatsek,
1983;Perry,2013).
Ifallpolysyllabicwordsarerepresentedattheformlevelassubunitsthatcorrespondtothemaximalcoda
analysis,awordliketurnipwillberepresentedatthatlevelasTURNandIP.Assuch,itmightbeexpectedthat
turnipwillfacilitateresponsestoturninthemaskedprimingparadigmdespitethefactthatipdoesnothavethe
appearanceofamorpheme.Thefactthatsuchanorthographicconditionhasnotshownmaskedprimingeffectsin
previousstudiesisaddressedbyTaftandNguyen-Hoan(2010).Theypointoutthattheitemsusedinthatcondition
haveactuallybeenamixtureofcaseswherethetargetistheBOSS(asinturnip-turn,brothel-broth)andwhereit
isnot(asinfreeze-free,shunt-shun).Therefore,ithasnotyetbeenshownthatmaskedprimingisabsentwhenthe
targetisspecificallytheBOSSoftheprime(asinturnip-turn)andthisissomethingthatfutureresearchcould
pursue.
Conclusions
Thepurposeofthischapterhasbeentoexplorehowmodelsofvisualwordrecognitionenvisagethewayinwhich
wordsarerepresentedinlexicalmemory.Asaworkingdefinition,lexicalrepresentationwastakentoembraceall
informationrequiredtoestablishthatapresentedletterstringcorrespondstoaknownword.Ifsuchawordhasa
representationthatanonwordcannothave,thenaccesstothisrepresentationshouldbesufficienttorecognize
theletterstringasawordand,indeed,asthatparticularword.Forthisreason,anumberofmodels(e.g.,Crepaldi
etal.,2010;Diependaeleetal.,2009;Schreuder&Baayen,1995)incorporateform-basedrepresentationsfor
everyknownword,nomatterwhattypeofworditis,andthisprovidesthebasisfordiscriminatingrealwordsfrom
nonwords.Nevertheless,insomeaccountswhenthewhole-wordrepresentationcorrespondstoapolymorphemic
word,itisaccessedviaformrepresentationsforitscomponentmorphemes(e.g.,Crepaldietal.,2010),oratleast
itcanbe(e.g.,Diependaeleetal.,2009).
Incontrast,theAUSTRALmodelhighlightedinthischapterhasaformlevelthatonlyrepresentssinglesyllables,
whetherthesecreateawholeword(e.g.,thehuntofhunterorthecornofcorner)ornot(e.g.,er).Therecognition
ofaletterstringasaparticularwordthereforetakesplaceatalevelbeyondthatofform.Thelemmalevelmediates
betweenformandfunctionandbringstogetherthesyllablesrepresentedattheformlevel.Moreover,lemmasare
hierarchicallystructuredwherebyaderivationallycomplexword(e.g.,hunter)hasitsownlemmaactivatedviathe
lemmasforitscomponentmorphemes(huntander).Thelemmalevelthereforeprovidestheprimarylocusof
lexicalrepresentation,thoughwithcertaincaveatsattached.
Onecaveatisthatanyaffixedwordwhosemeaningisentirelypredictablefromitscomponents(e.g.,jumped,cats)
isnotactuallyrepresentedatthelemmalevel,butratherisrecognizablethroughthecombinationofhigherlevel
functionalinformationassociatedwiththelemmaforeachofitsmorphemes(i.e.,semantic,syntactic,and/or
idiosyncraticfeaturesofthemorpheme).Whenawordactivatessuchcombinablefunctions,buthasan
exceptionalform(e.g.,fell,teeth),informationaboutitsdistinctivecharacteristicscanalsobefoundatthehigher
functionallevel.Thispreventsaregularizationoftheirregularword(e.g,falled,tooths)frombeingacceptedasa
word.
Second,thelemmalevelincludesrepresentationsofboundmorphemes(e.g.,venge,er)which,bydefinition,are
notwordsintheirownright.Therefore,thereneedstobesomewayofdiscriminatingsuchmorphemesfromreal
words.Suggestionsforachievingthiscanbegivenintermsofspecificinformationstoredatthehigherfunctional
Page 12 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2014. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: University of New South Wales; date: 07 October 2014
The Nature of Lexical Representation in Visual Word Recognition
level,theexistenceofaconceptlevel,ordifferingdegreesofform-functioncorrelationcapturedatthelemma
level.
Weseethenthatthelexicalrepresentationsinvolvedinreadingcanbeconceptualizedinanumberofdifferent
waysrangingfromorthographicunitscorrespondingtothewholewordthroughtopatternsofactivationwithina
distributedsetofunitsthatmediatebetweenformandfunction.Theaccountfavoredinthischapter,AUSTRAL,is
onewhereinformationaboutawordisassociatedwithalocalistunitthatmediatesbetweenformandfunction(i.e.,
alemma),whilebeingactivatedthroughsublexicalunitsattheformlevel.Suchanaccountprovidesaconcrete
frameworkforhelpingunderstandtheprocessesinvolvedinreadingalltypesofwords,bothintermsofretrieving
theirmeaningandingeneratingtheirpronunciation.Anunderstandingoftherecognitionofpolymorphemicwords
isthusnotjustanichetopicwithinthedomainoflexicalprocessingbuthasimportantimplicationsforthe
conceptualizationofthewholelexicalprocessingsystem.
References
Allen,M.,&Badecker,W.(2002).Inflectionalregularity:Probingthenatureoflexicalrepresentationinacrossmodalprimingtask.JournalofMemoryandLanguage,44,705–722.
Andrews,S.,&Scarratt,D.R.(1998).Ruleandanalogymechanismsinreadingnonwords:HoughDoupeapelrede
gnewwirds?JournalofExperimentalPsychology:HumanPerceptionandPerformance,24,1052–1086.
Baayen,R.H.,Dijkstra,T.,&Schreuder,R.(1997).SingularsandpluralsinDutch:Evidenceforaparalleldual
routemodel.JournalofMemoryandLanguage,37,94–117.
Bertram,R.,Schreuder,R,&Baayen,R.H.(2000).Thebalanceofstorageandcomputationinmorphological
processing:Theroleofwordformationtype,affixalhomophony,andproductivity.JournalofExperimental
Psychology:Learning,Memory,andCognition,26,489–511.
Burani,C.,&Caramazza,A.(1987).Representationandprocessingofderivedwords.LanguageandCognitive
Processes,2,217–227.
Colé,P.,Beauvillain,C.,&Segui,J.(1989).Ontherepresentationandprocessingofprefixedandsuffixedderived
words:Adifferentialfrequencyeffect.JournalofMemoryandLanguage,28,1–13.
Coltheart,M.,Rastle,K.,Perry,C.,Langdon,R.&Ziegler,J.(2001).DRC:ADualRouteCascadedmodelofvisual
wordrecognitionandreadingaloud.PsychologicalReview,108,204–256.
Crepaldi,D.,Rastle,K.,Coltheart,M.,&Nickels,L.(2010).“Fell”primes“fall,”butdoes“bell”prime“ball”?Masked
primingwithirregularly-inflectedprimes.JournalofMemoryandLanguage,63,83–99.
Davis,M.H.,&Rastle,K.(2010).Formandmeaninginearlymorphologicalprocessing:CommentonFeldman,
O’Connor,andMoscosodelPradoMartin(2009).PsychonomicBulletin&Review,17,749–755.
Dell,G.S.(1986).Aspreadingactivationtheoryofretrievalinsentenceproduction.PsychologicalReview,93,
283–321.
Diependaele,K.,Sandra,D.,&Grainger,J.(2009)Semantictransparencyandmaskedmorphologicalpriming:The
caseofprefixedwords.Memory&Cognition,37,895–908.
Feldman,L.B.,O’Connor,P.A.,&MoscosodelPradoMartín,F.(2009).Earlymorphologicalprocessingismorphosemanticandnotsimplymorpho-orthographic:Anexceptiontoform-then-meaningaccountsofwordrecognition.
PsychonomicBulletin&Review,16,684–691.
Forster,K.I.(1976).Accessingthementallexicon.InE.C.J.Walker&R.J.Wales(Eds.),Newapproachesto
languagemechanisms(pp.257–287).Amsterdam:North-Holland.
Forster,K.I.(2012).Aparallelactivationmodelwithasequentialtwist.InJ.S.Adelman(Ed.),VisualWord
RecognitionVolume1:ModelsandMethods,OrthographyandPhonology(pp.52–69).Hove,UK:Psychology
Page 13 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2014. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: University of New South Wales; date: 07 October 2014
The Nature of Lexical Representation in Visual Word Recognition
Press.
Giraudo,H.,&Grainger,J.(2000).Primewordfrequencyinmaskedmorphologicalandorthographicpriming.
LanguageandCognitiveProcesses,15,421–444.
Giraudo,H.,&Grainger,J.(2001).Primingcomplexwords:Evidenceforsupralexicalrepresentationofmorphology.
PsychonomicBulletin&Review,8,127–131.
Gonnerman,L.,&Plaut,D.C.(2000).Semanticandmorphologicaleffectsinmaskedpriming(2000).Departmentof
Psychology.Paper350.http://repository.cmu.edu/psychology/350.
Gonnerman,L.M.,Seidenberg,M.S.,&Andersen,E.S.(2007).Gradedsemanticandphonologicalsimilarityeffects
inpriming:Evidenceforadistributedconnectionistapproachtomorphology.JournalofExperimentalPsychology:
General,136,323–345.
Harm,M.W.,&Seidenberg,M.S.(2004).Computingthemeaningsofwordsinreading:Cooperativedivisionof
laborbetweenvisualandphonologicalprocesses.PsychologicalReview,111,662–720.
Katz,L.,&Baldasare,J.(1983).Syllablecodinginprinted-wordrecognitionbychildrenandadults.Journalof
EducationalPsychology,75,245–256.
Kempen,G.,&Huijbers,P.(1983).Thelexicalizationprocessinsentenceproductionandnaming:Indirectelection
ofwords.Cognition,14,185–209.
Levelt,W.J.M.,Roelofs,A.,&Meyer,A.S.(1999).Atheoryoflexicalaccessinspeechproduction.Behavioraland
BrainSciences,22,1–38.
Lima,S.D.,&Pollatsek,A.(1983).Lexicalaccessviaanorthographiccode?TheBasicOrthographicSyllabic
(BOSS)reconsidered.JournalofVerbalLearningandVerbalBehavior,22,310–332.
Marslen-Wilson,W.,Bozic,M.,&Randall,B.(2008).Earlydecompositioninvisualwordrecognition:Dissociating
morphology,form,andmeaning.LanguageandCognitiveProcesses,23,394–421.
McClelland,J.L.,&Rumelhart,D.E.(1981).Aninteractiveactivationmodelofcontexteffectsinletterperception:
Part1.Anaccountofbasicfindings.PsychologicalReview,88,375–407.
Morton,J.(1969).Interactionofinformationinwordrecognition.PsychologicalReview,76,165–178.
Morton,J.(1970).Afunctionalmodelofmemory.InD.A.Norman(Ed.),Modelsofhumanmemory(pp.203–254).
NewYork:AcademicPress.
Murrell,G.A.,&Morton,J.(1974).Wordrecognitionandmorphemicstructure.JournalofExperimentalPsychology,
102,963–968.
Niswander,E.,Pollatsek,A.,&Rayner,K.(2000).Theprocessingofderivedandinflectedsuffixedwordsduring
reading.LanguageandCognitiveProcesses,15,389–420.
Perry,C.(2013).Graphemicparsingandthebasicorthographicsyllablestructure.LanguageandCognitive
Processes,28,355–376.
Perry,C.,Ziegler,J.C.,&Zorzi,M.(2010).Beyondsinglesyllables:Large-scalemodelingofreadingaloudwiththe
ConnectionistDualProcess(CDP++)model.CognitivePsychology,61,106–151.
Plaut,D.C.,McClelland,J.L.,Seidenberg,M.S.,&Patterson,K.(1996).Understandingnormalandimpairedword
reading:Computationalprinciplesinquasi-regulardomains.PsychologicalReview,103,56–115.
Rastle,K.,&Davis,M.H.(2008).Morphologicaldecompositionbasedontheanalysisoforthography.Language
andCognitiveProcesses,23,942–971.
Roelofs,A.(1992).Aspreading-activationtheoryoflemmaretrievalinspeaking.Cognition,42,107–142.
Page 14 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2014. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: University of New South Wales; date: 07 October 2014
The Nature of Lexical Representation in Visual Word Recognition
Rueckl,J.G.,&Aicher,K.A.(2008).AreCornerandBrothermorphologicallycomplex?Notinthelongterm.
LanguageandCognitiveProcesses,23,972–1001.
Rueckl,J.G.,&Raveh,M.(1999).Theinfluenceofmorphologicalregularitiesonthedynamicsofaconnectionist
network.BrainandLanguage,68,110–117
Rumelhart,D.E.,&McClelland,J.L.(1982).Aninteractiveactivationmodelofcontexteffectsinletterperception:
Part2.PsychologicalReview,89,60–94.
Schreuder,R.,&Baayen,R.H.(1995).Modelingmorphologicalprocessing.InL.B.Feldman(Ed.),Morphological
aspectsoflanguageprocessing(pp.131–154).Hillsdale,NJ:LawrenceErlbaumAssociates.
Seidenberg,M.S.,&McClelland,J.L.(1989).Adistributed,developmentalmodelofwordrecognitionandnaming.
PsychologicalReview,96,523–568.
Taft,M.(1979a).Lexicalaccessviaanorthographiccode:TheBasicOrthographicSyllabicStructure(BOSS).
JournalofVerbalLearningandVerbalBehavior,18,21–39.
Taft,M.(1979b).Recognitionofaffixedwordsandthewordfrequencyeffect.Memory&Cognition,7,263–272.
Taft,M.(1987).Morphographicprocessing.TheBOSSre-emerges.InM.Coltheart(Ed.),Attentionand
performance,XII(pp.265–279).London:LawrenceErlbaumAssociates.
Taft,M.(1991).Readingandthementallexicon.Hove,UK:LawrenceErlbaumAssociates.
Taft,M.(1994).Interactive-activationasaframeworkforunderstandingmorphologicalprocessing.Languageand
CognitiveProcesses,9,271–294.
Taft,M.(2003).Morphologicalrepresentationasacorrelationbetweenformandmeaning.InE.Assink,&D.Sandra
(Eds.),Readingcomplexwords(pp.113–137).Amsterdam:Kluwer.
Taft,M.(2004).Morphologicaldecompositionandthereversebasefrequencyeffect.QuarterlyJournalof
ExperimentalPsychology,57A,745–765.
Taft,M.(2006).Alocalist-cum-distributed(LCD)frameworkforlexicalprocessing.InS.M.Andrews,Frominkmarks
toideas:Currentissuesinlexicalprocessing(pp.76–94).Hove,UK:PsychologyPress.
Taft,M.,&Ardasinski,S.(2006).Obligatorydecompositioninreadingprefixedwords.TheMentalLexicon,1,183–
199.
Taft,M.,&Forster,K.I.(1975).Lexicalstorageandretrievalofprefixedwords.JournalofVerbalLearningand
VerbalBehavior,14,638–647.
Taft,M.,&Forster,K.I.(1976).Lexicalstorageandretrievalofpolymorphemicandpolysyllabicwords.Journalof
VerbalLearningandVerbalBehavior,15,607–620.
Taft,M.,&Kougious,P.(2004).Theprocessingofmorpheme-likeunitsinmonomorphemicwords.Brainand
Language,90,9–16
Taft,M.,&Krebs-Lazendic,L.(2013).Theroleoforthographicsyllablestructureinassigningletterstotheirposition
invisualwordrecognition.JournalofMemoryandLanguage,68,85–97.
Taft,M.,&Nguyen-Hoan,M.(2010).Astickystick:Thelocusofmorphologicalrepresentationinthelexicon.
LanguageandCognitiveProcesses,25,277–296.
Taraban,R.,&McClelland,J.L.(1987).Conspiracyeffectsinwordpronunciation.JournalofMemoryand
Language,26,608–631.
Treiman,R.,&Chafetz,J.(1987).Arethereonset-andrime-likeunitsinprintedwords?InM.Coltheart(Ed.),
AttentionandPerformance,XII(pp.281–298).London:LawrenceErlbaumAssociates.
Page 15 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2014. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: University of New South Wales; date: 07 October 2014
The Nature of Lexical Representation in Visual Word Recognition
Notes:
(1)InTaft(2006)themodelwasreferredtoaslocalist-cum-distributed(LCD).Thereasonforabandoningthisterm
isthatmostmodelsactuallyincludeacombinationoflocalistanddistributedcharacteristics,evenifonedominates
theother.Forexample,theinputunitsofPDPmodelsaretypicallylocalist,representingspecifiedlettersorletter
groupings,whiletheletterunitsoftheIAmodelcanbeseenasbeingdistributedinthesensethatmorethanone
wordisactivatedthroughthesamesetofunits.
(2 )Infact,Diependaeleetal.(2009)donotexplicitlystatethatthelexicalformcornerisnotactivatedthroughthe
morpho-orthographicunitcorn.However,thisseemsasensibleconclusionbasedonthedescriptiontheygive.
MarcusTaft
MarcusTaft,SchoolofPsychology,UniversityofNewSouthWales
Page 16 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2014. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: University of New South Wales; date: 07 October 2014