Reading Ability- Lexical Quality to Comprehension

SCIENTIFIC STUDIES OF READING,11(4),357–383

Copyright ©2007, Lawrence Erlbaum Associates, Inc.

Reading Ability: Lexical Quality

to Comprehension

Charles Perfetti

University of Pittsburgh

Thelexicalqualityhypothesis(LQH)claimsthatvariationinthequalityofwordrep-resentationshasconsequencesforreadingskill,includingcomprehension.Highlexi-calqualityincludeswell-specifiedandpartlyredundantrepresentationsofform(or-thographyandphonology)andflexiblerepresentationsofmeaning,allowingforrapidandreliablemeaningretrieval.Low-qualityrepresentationsleadtospecificword-relatedproblemsincomprehension.Sixlinesofresearchonadultreadersdem-onstratesomeoftheimplicationsoftheLQH.First,large-scalecorrelationalresultsshowthegeneralinterdependenceofcomprehensionandlexicalskillwhileidentify-ingdisassociationsthatallowfocusoncomprehension-specificskill.Second,word-levelsemanticprocessingstudiesshowcomprehensionskilldifferencesinthetimecourseofform-meaningconfusions.Studiesofrarevocabularylearningusingevent-relatedpotentials(ERPs)showthat,third,skilledcomprehenderslearnnewwordsmoreeffectivelyandshowstrongerERPindicatorsformemoryofthewordlearningeventand,fourth,suggestskilldifferencesinthestabilityoforthographicrepresentations.Fifth,ERPmarkersshowcomprehensionskilldifferencesinmean-ingprocessingofordinarywords.Finally,intextreading,ERPresultsdemonstratemomentarydifficultiesforlow-skillcomprehendersinintegratingawordwiththepriortext.Thestudiesprovideevidencethatword-levelknowledgehasconsequencesfor word meaning processes in comprehension.

Inreading,thesingularrecurringcognitiveactivityistheidentificationofwords.Fromthisfollowtwoother,relatedobservationsaboutreading:Comprehensiondependsonsuccessfulwordreading.Skilldifferencesincomprehensioncanarisefrom skill differences in word reading.

Thesesimpleobservationsformthecoreofatheoryofcomprehensionskillpublishedover20yearsago(Perfetti,1985).Verbalefficiencytheoryclaimedthat

CorrespondenceshouldbesenttoCharlesPerfetti,LRDC,UniversityofPittsburgh,Pittsburgh,PA15260. E-mail: Perfetti@pitt.edu

358PERFETTI

wordidentification,therapidretrievalofaword’sphonologyandmeaning,wasalimitingfactorincomprehension.Ireferredtothesecognitiveeventsofwordiden-tificationas“retrievals”becausetheyoperatedoninformationaboutawordstoredinareader’sorthographicallyaddressablememory.Butattheheartofwordidenti-ficationwerethephonologicalproceduresthatallowedaword(oranonword)tobedecoded,whetherornotmeaningwasalsoretrieved.Thetheoryassumedtheabilitytodecodenonwordswasthehallmarkofbasicalphabeticreadingskill.Infact,phonologywasimportantenoughinthisaccountthatithadredundantpartici-pation.Phonologywasbothstoredaspartoftheword(andthusretrievedduringidentification)andgeneratedbyconnectionsamongsubwordunitsthatwerepartoftheword.ThisconceptualizationwasexplicitintheRestrictedInteractiveModel,whichfocusedonthedevelopmentwithexperienceofspecificandredun-dant sublexical components suggested in Perfetti (1992).

Inthetheory,thelinkfromword-levelreadingtocomprehensionwasthroughtheassumptionthatcomprehensionincludedhigherlevelprocessesthatrequiredcognitiveresources(workingmemory),forexample,integrativeprocesses,infer-ences,syntacticrepairs.Wordidentification,andcertainlythesublexicalpro-cessesthatproduceit,werecandidatesforlow-resourceorautomaticprocesses(LaBerge&Samuels,1974)thatcouldpreserveprocessingresourcesforhigherlevelcomprehension.Automatic,resource-cheapword-levelprocesses—verbalefficiency—wereassumedtosupportcomprehension.Childrenwhohavethisefficiencywouldbeabletoachievehighlevelsofcomprehension,andchildrenwithinefficientword-levelprocesseswouldhaveproblemswithcomprehension.Theresearchshowingcorrelationsbetweenchildren’sdecodingskillandcom-prehensionwasconsistentwiththisaccount.Furthermore,thereisnoreasontosupposethatthisrelationshipdisappearsforolderreaders(Shankweileretal.,1999).

Thisgeneralaccountcontinuestoseemcorrecttome.However,Ithinkitsem-phasisoncompletelygeneralprocesses—decoding,phonologicalprocesses,re-trieval,memory,automaticity—althoughtheoreticallyconsistent,seemedtoleaveknowledgeoutofthepicture.Skilledreadingwasaboutefficientprocessingmech-anismsandlessskilledreadingwasaboutthesesamemechanismsexecutedineffi-ciently.Thisdescriptionseemedtopredictthatbecomingfasteratwordidentifica-tionleadstobettercomprehension.Inefficientreaderscanindeedbecomemoreefficient(Breznitz&Share,1992),andimprovingindividualwordreadingspeedmayincreasefluency(Martin-Chang&Levy,2005)and,undersomecircum-stances,comprehension(Breznitz&Share,1992;Tan&Nicholson,1997).How-ever,increasingdecodingspeedbyitselfhasnotalwaysincreasedcomprehension(Fleisher,Jenkins,&Pany,1979;Perfetti,1985).Overall,althoughthehypothesisthattrainingword-readingspeedraisescomprehensionhassomeresearchsupport,itisnottheprimarypracticalimplicationofthegeneralideathatcomprehensiondepends on efficient word reading.

LEXICAL QUALITY359

Efficiencyisnotthesameasspeed.Efficiencyisaratioofoutcometoeffort,withtimeasaproxyforeffort.Soalthoughprocessingdescriptionsmakeacoher-entframeworkforefficiency,theyleaveoutthebasicnatureandsourceofthewordreadingoutcomesonwhichefficiencydepends.Theseoutcomesarewordidenti-tiesthatmomentarilyrepresentformandmeaningcomponentsthatarethebasicelementsofcomprehension.Onthisdescription,thethingtounderstandisnotspeedbutrathertheabilitytoretrievewordidentitiesthatprovidethemeaningsthereaderneedsinagivencontext.Thissourceofthisabilityistheknowledgeareaderhas about words, specific lexical representations.

LEXICAL QUALITY

Underlyingefficientprocessesareknowledgecomponents;knowledgeaboutwordforms(grammaticalclass,spellingsandpronunciations)andmeanings.Addeffectivepractice(readingexperience)oftheseknowledgecomponents,andtheresultisefficiency:therapid,low-resourceretrievalofawordidentity.Lexicalquality(LQ)referstotheextenttowhichthereader’sknowledgeofagivenwordrepresentstheword’sformandmeaningconstituentsandknowledgeofwordusethatcombinesmeaningwithpragmaticfeatures.Thusthevocabularyofagivenlanguageincludes,foragivenreader,wordsofwidelyvaryingLQ,fromrarewordsneverencounteredtofrequentlyencounteredandwell-knownwords.Like-wise,individualreadersdifferintheaverageLQoftheirwords.Thisreadervari-abilityisnotjustaboutthesizeofvocabulary,althoughitincludesthis;itisabouttherepresentationofwords,thestableandlessstableknowledgethereaderhasabout the word’s form and meaning.

Ofcourse,thequestionbecomeswhatis“quality,”awordthatcouldevokesus-picionwithoutsomedefinition.Qualityistheextenttowhichamentalrepresenta-tionofawordspecifiesitsformandmeaningcomponentsinawaythatisbothpre-ciseandflexible.Theprecisionisneededbecause“prettyandpetty”and“knightandnight”arenotthesame.Theflexibilityisneededbecausethemeaningsof“roamingcharge”and“afeechargedbyamobilephoneserviceforcallsinitiatedorreceivedoutsideacontractedservicearea”arethesame.Bothprecisionandflexibilityareneededtounderstandandpronouncerecordin“Youneedarecordofthetransaction”and“Theycan’trecordtheconversation.”Thesesimpleexamplesarejustthetipoftheicebergofform-meaningcomplexities.LQprovidesameansforsafepassagethroughthem.Earlierchapters(Perfetti&Hart,2001,2002)con-tain additional examples and theoretical discussions of LQ.

OnewaytobecomemorespecificaboutLQistoidentifythefeaturesthatwehy-pothesizetodistinguishhigherqualityfromlowerqualityrepresentations.Table1doesthis.Itidentifiesfivefeaturesoflexicalrepresentationthatdistinguishhighandlowqualityandshowsthree(theremaybemore)hypothesizedconsequencesof

360PERFETTI

TABLE 1

Properties and Consequences of Lexical Quality

Representational Properties

of Lexicon

Orthography

PhonologyHigh QualityFully specified; letters areconstantsRedundant word-specific

phonology and context-

sensitive grapheme-phoneme

phonology

All grammatical classes of the

word represented; morpho-

syntactic inflections

represented

More generalized, less

context-bound; fuller range

of meaning dimensions to

discriminate among words in

same semantic field.

Orthographic, phonological, and

semantic constituents are

tightly boundLow QualityNot fully specified; some lettersare variablesLess stable because of variableword-specific phonology and/or grapheme-phonemephonologyIncomplete range of form classuses; less stablemorpho-syntaxMore context bound; fewerrelevant meaning dimensionsto discriminate among relatedwordsOrthographic, phonological, andsemantic constituents are lesstightly boundGrammarMeaningConstituent binding

Possibleprocessingconsequences

duringreading

StabilityHigher; word identity is reliably

retrieved from an

orthographic or phonological

input

SynchronicityWord identity constituents are

activated and retrieved in

synchrony as a word identity

Meaning integrationHigher; word identities

available for building

comprehensionLower; word identity issometimes not retrieved froman orthographic orphonological inputWord constituents may beactivated and retrievedasynchronously; (e.g. laboreddecoding; activation ofincorrect meanings frompartial input)Lower; comprehensionprocesses that operate overword identities at risk

thesequalityfeaturesforreadingprocesses.Therepresentationalfeaturesarethefourconstituentsofwordidentity—orthography,phonology,morpho-syntax,andmeaning.Thefifthrepresentationfeatureisconstituentbinding,thedegreetowhichthefirstfourfeaturesareboundtogether(especiallythefirstthree;thegrammaticalfeaturemightbeconsideredtobeimplementedbyagrammaticalprocessthatoper-atesonthelexeme).Bindingsareconnectionsthatsecurecoherenceamongthecon-stituents,theorthographic,phonological,andsemanticrepresentations,whichto-getheraretheword’sidentity.Thebindingfeatureisnotindependentbutrathera

LEXICAL QUALITY361

consequenceoftheorthographic,phonologicalandsemanticconstituentsbecomingwellspecifiedinassociationwithanotherconstituent.

TheconsequencesofhighqualityinsublexicalandlexicalknowledgearealsoshowninTable1.Thesearehypothesizedconsequencesthataresubjecttoempiri-caltesting,andsomehaveatleastindirectevidence.Forexample,thehypothesisthatlowLQcanleadtotheasynchronousactivationofwordconstituentsisconsis-tentwithresultsofBreznitzandMisra(2003),whofoundthatERPindicatorsoforthographyandphonologyforlow-skillreadingweremoreasynchronousthanthoseofskilledreaders.Moregenerally,readingwordsincontextisaffectedbyLQ,providingtheimportanthypothesizedlinkbetweenLQandcomprehension(Perfetti&Hart,2001).ThegeneraldescriptionofthislinkisthatlocalprocessesofintegratingwordmeaningswithinandacrosssentenceboundariesareaffectedbytheLQofwordsthatareidentifiedaspartofthecomprehensionprocess.Onefi-nalobservationconcernsthedifferencebetweenaspokenandwrittenword.Myfocusisonreading,andforthatorthographyispartofLQ.However,alexicalanal-ysiscanbeappliedtojustspokenlanguagewithafocusonphonologicalrepresen-tations and meaning.

WORD PROCESSING, WORD LEARNING,

AND COMPREHENSION STUDIES

WiththisbackgroundonthegeneralnatureoftheLQhypothesis(LQH)anditslinkstotheprocess-orientedaccountofverbalefficiency,IturnnowtoareviewofsomestudiesofreadingthatbearontheLQH.Althoughthesearestudiesofadultreading, I believe their conclusions apply also to children’s reading.

TheStructureofLexicalandComprehensionSkillinReading

Lexicalknowledgeandcomprehensionshouldbeassociated,andtheyare.Posi-tivecorrelationsbetweenword-processingmeasuresofvariouskindsandreadingcomprehensionassessmentsarewellestablishedinbothchildren(Perfetti,1985)andadults(Haenggi&Perfetti,1994).Moreinterestingisthefactthatthiscorrela-tionisgenerallyinthemoderaterange,leavingplentyofroomfordisassociationbetweenthetwo.We’vemaintainedalargedatabaseofcollegestudentsforwhomwetakevariousreadingandreading-relatedmeasures.Ingeneral,weareinter-estedinknowingtheassociatesofreadingcomprehensionskillandintheout-comesofexperimentsthatrelatedspecificreadingprocessestothesemeasures.PerfettiandHart(2002)reportedsomeresultsoffactoranalysisonasampleof445individualsfromthisdatabase.Theyconcludedthatskilledreaders’knowledgeofspelling,phonology,anddecodingcouldberepresentedreasonablywellbyasin-glewordformfactorwithasecondfactorreflectingmeaningandcomprehension.However,lessskilledreaders,inadditiontoameaningfactor,requiredtwoform

362PERFETTI

factors—oneloadedwithmorephonologicaltasksandtheotherwithmoreortho-graphic tasks, suggesting less coherence of word identities.

Sincethentwodissertationshaveassessedlargesamplesfromthisdatabase,basedonpartlydifferenttasks,onebyHart(2005)andalateronebyLandi(2005).BothHartandLandiwereinterestedinthedisassociationofword-levelskillfromcomprehensionskill,asassessedbytheNelson–Dennycomprehen-siontest.Hartanalyzedthescoresof792students,concludingthat,amongthosewhocouldbemostconfidentlyclassified,18%werebelowthemedianincom-prehensionbutatorabovethemedianonlexicalmeasures;64%showedthemoretypicalassociationpatternoflexicalandcomprehensionscoresbothhighorbothlow.Inherexperimentswithacarefullydefinedsubsetofthissample,Hartaskedwhethercertainaspectsoflearninganartificiallanguagemightde-pendmoreonfirst-languagelexicalknowledgecomparedwithfirst-languagecomprehensionskill.Herresultsshowedacomplexpattern,withbothcompre-hensionandlexicalknowledgepredictingvariousmeasuresofperformancewiththenovellanguageduringlearningandinpostlearningtransfer.However,lexicalknowledge,morethancomprehension,predictedthelearningofthisartificiallanguage,includingitsnovelorthographyanditsdecodingmappings.Italsopredictedresistancefrominterferencebyhomophonesthatwereplantedinthenewlanguage,replicatingaresultreportedinPerfettiandHart(2001)forEng-lish.

Landi’ndi’sfactoranalysisoffivetasksyieldedacomprehensioncomponentandalexicalcomponent,whichshethenusedtoweightnormalizeindividualparticipantscoresoneachtest.Figure1showsascatterplotofthesescores.Inthisnormalizedanalysis,23%werebelowthemedianonthecomprehensioncomponentbutabovethemedianonthelexicalcomponent;thereversepattern,highcomprehensionbutlowlexicalcomponents,wasobservedfor9%ofthesample.Thisasymmetryisthepatternonewouldex-pect(butmightnotfindbasedonmediansplitsoftestscores).Lexicalknowledgeisnotsufficientforcomprehension,sothelow-comprehension/highlexicalpatternismoreprominentthanthehigh-comprehension/low-lexicalpattern.Theapproxi-mately20%(18%inHart’ssample,23%inLandi’ssample)ofcollegestudentswhosecomprehensionlevelsundershootthelevelexpectedbylexicalskillsidenti-fiesagroupforwhomhypothesesaboutothersourcesofcomprehensionproblemscan be meaningfully tested.

Theideathatlexicalprocessesarenotsufficientforcomprehensionshouldnotbecontroversial,althoughacarelessreadingofverbalefficiencytheorymighthaveledsometobelievethatthetheoryassumedtherewasnothingtocomprehen-sionbeyondefficientwordreading.Observationstothecontraryhavelongbeenintheliterature,andatleastafewofthemseemtohavecontrolledadequatelyforword-levelskills.StudiesbyOakhillandcolleagues(Cain&Oakhill,1999;

LEXICAL QUALITY

363

FIGURE1Scatterplotofnormalizedcomponentscoresfromprincipalcomponentsanalysisofreadingtestscoresfromasampleof799collegestudents,basedonLandi’s(2005)disserta-tion.Inthisprocedure,normalizedZscoresforeachtestforeachparticipantweremultipliedbythefactorscoreforthattestdeterminedfromthePrincipleComponentsAnalysis.Thus,theplotisafactor-weightedcompositeoffivetests(decoding,spelling,vocabulary,comprehension,and the Author Recognition Test) that weighted differentially on the two components.

Oakhill,Cain,&Bryant,2003)seemtoshowthatsomechildrenhavetroubledrawinginferencesduringcomprehension,despitehavinggooddecodingskills.Moregenerally,comprehensionproblemscanarisefromgenerallanguagecom-prehensionproblemsevenwhenworddecodingappearstobeadequate(Stodhard&Hulme,1996).(Forreviews,seeNation,2006,andPerfetti,Landi,&Oakhill,2005).Anothercandidateforcomprehensionproblems,closertotheideaofLQ,isthehypothesisthatchildrenwithadequatedecodingandphonological-levelskillscanhaveword-levelsemanticproblemsthataffectcomprehension.Thissemanticdeficithypothesis(Nation&Snowling,1998,1999),whichhassomeevidenceinstudiesofchildren,allowsforseveralinterpretationsofsemanticdeficits,includ-ingaproblemwithsemanticcategories.Fornow,theLandiandHartstudiesex-tendtotheadultpopulationtheobservationthatdecodingisnotsufficientforcom-prehension.

Studies of Form–Meaning Confusions

OnewaytostudytheeffectsofLQoncomprehensionistoexperimentallycreatethreatstoquality.ThisisthebasisforexperimentsreportedinPerfettiandHart(2001),whichIonlybrieflyreviewhere.Thekeyideaisthatforawordlikewails,

364PERFETTI

thequalityofitsidentity,whichentailsitsspelling,pronunciation,andmeaning,isthreatenedbytheexistenceofwhales,whichsharesitspronunciation.Sohowdoreadersofhigherandlevelskillhandlethiskindofthreat?Ourhypothesiswasthatbetterreaders,definedbycomprehensionassessment,havehigherqualityrepre-sentations,sogivenwail,theyshouldretrievewailassociates,notwhaleassoci-ates,evenifthereismomentaryactivationofboththepresentedwordanditsho-mophone (Gernsbacher & Faust, 1991).

InthemeaningtaskreportedinPerfettiandHart(2001),readersdecidedwhethertwowordspresentedinsuccessionweresemanticallyrelated.Sometrialscontainedhomophonesofwordsthatwouldhavebeenrelated,aswhenwailswasfollowedbydolphins.Skilledcomprehendersshowedfaster-meaningdecisionsforbothcontrolpairsandhomophones,andtheyshowedlesshomophoneconfu-sionwhenthepresented(confusable)formwastheoneofhigherfrequency(whales–criesratherthanwails–dolphins).However,theydidshowconfusionswhenpresentedwithaformoflowfrequency,andthiseffectoccurredatshorterla-tenciesthanitdidforlessskilledcomprehenders.Sonight–armordidnotproduceconfusion(intheformoflongerdecisiontimes)forskilledcomprehenders,al-thoughitdidforlessskilledcomprehenders.However,knight–eveningproducedinterferenceforskilledcomprehenders,andthisemergedveryrapidly,ataStimu-lusOnsetAsynchronyof150msec,whereasforlessskilledcomprehendersthein-terferenceeffectdidnotemergeuntil450msec.Thus,skilledcomprehendersshowlessinterferencebasedonform,andwhentheyshowinterference,itoccurswithin150msofexposuretothehomophone,suggestinganearlyactivationofword phonology.

Thisfrequencyeffectinhomophoneconfusionsappearstobedependentonwordexperience.InastudyreportedinHartandPerfetti(inpress),readerswereprovidedwithexperienceonthememberofahomophonepairthatwasjudgedtobelessfamiliar,soastomakeitmorefamiliarthanitsmate.Forexample,inthepairnight–knight,knightwasratedlowerinfamiliarity,soitwastheonepartici-pantsexperiencedintraining;similarly,inhair–hare,harewasratedlower,soitwasthetrainingword.Theresultofthetraining,whichconsistedofvisualexpo-surestothewordassociatedwithmeaning,wasthereversalofthefrequencyeffectinhomophoneinterference.Intermsoftheexamples,trainingonknightcausedse-manticdecisionsonknight–eveningtoproducelessinterferencethansemanticde-cisionsonnight–armor.Soform-basedconfusionsdependontherelativefrequen-ciesofcompetingforms,whichinturndependonreadingexperience.Ahighlyfrequentformisrelativelyprotectedfrominterference,becauseitretrievesitsmeaningandpronunciationrapidlyasastable,uniquewordidentity.Aninfrequentformislessprotectedbecauseitismorelikelytoretrieveanunstableidentitybasedonsharedphonologywiththemorefrequentform.Theapplicationofthistoread-ingskillisthatLQdependsonexperiencewithwords.Askilledcomprehenderhas

LEXICAL QUALITY365

hadmoreexperiencewithagivenwordthanhasalessskilledreader,andthishasimportant implications.1

Therelationshipbetweenwordfrequencyestimatesandvariousword-process-ingtasksisnonlinear,generallylogarithmic.Inlowfrequencyranges,agivendif-ferenceinfrequencybetweentwowordsmayhavealargeeffectonmeasuresofspeedofprocessing;inhighfrequencyranges,thatsamefrequencydifferencehasasmallereffect.Skilldifferencesinword-readingexperimentsareusuallygreaterforlow-frequencywordsthanhigh-frequencywords.Thismayreflecttheimpor-tanceofsomeminimumnumberofexposuresforawordtobeidentifiedwithloweffort.Ifweassumethatagivenwordhasbeenreadmorefrequentlybyaskilledreaderthanalessskilledreader,thenitfollowsthattheskilldifferencesweobserveinprocessingthatwordreflectthisfrequencyofexperiencedifference.However,thedifferenceinexposuresseemstohaveaneffectonlyforlow-frequencywords,consistentwithassumptionthatitisthelow-frequencyrangewhereincrementsinfrequencyaremostimportant.Althoughrecencyeffectscanbedisguisedasfre-quencyeffects,thisdoesnotmatterfortheskillconclusion.Morereadingleads,statistically,tomorefrequentandmorerecentencountersandbothmayhavethisnonlineareffectonword-readingefficiency.Althoughthisstatisticalperspectiveisimportant,itdoesnotmeanthatallexperienceswithwordsareequal.Infact,inthenextsection,Ireviewastudythatsuggeststhatskilledcomprehendersmakebetteruse of their experiences with words they are trying to learn.

Learning the Meanings of New Words

GiventheimplicationthatLQisacquiredthrougheffectiveexperiencewithwords,wemightbeabletoobservetheacquisitionofLQduringwordlearning.Further-more,bycomparingthelearningofreaderswhodifferincomprehensionskill,wecanexaminethelinkbetweenLQandcomprehensioninasituationthatcontrolsthewordexperiences.Inthisresearch,wehaveusedbothbehavioralandERPmeasures.

Perfetti,Wlotko,andHart(2005)taughtthemeaningsofveryrarewordstoun-dergraduatesandthentestedtheeffectsofthislearninginasimplemeaningjudg-menttaskwhilerecordingelectroencephalograms(EEGs).Examplesofthewordstaughtincludethefollowing:gloaming,flexion,clement,ibex,agog,bastion,tiglon,andquisling.Inapretestlexicaldecisiontask,ourrarewordswerejudgedtoberealwordsonlyatarateof8%onaverage.Toassurethatthewordswewould1GernsbacherandFaust(1991;alsoGernsbacher,1990)explainedlessskilledreaders’problemsinmeaningprocessingasduetoproblemsinsuppressingirrelevantmeaningsthatareactivatedbyaword.Differencesbetweentheirmechanism-basedaccountandtheknowledge-basedaccountoftheLQHarediscussed further in Perfetti and Hart (2001).

366

PERFETTI

trainandthentestwereunknown,theto-be-learnedwordswereindividuallytai-loredforeachparticipantaccordingtothepretest.Followingsimpleassociationinstruction(50mininwhichtherarewordswerepairedwithbriefdefinition-likeparaphrases),thetrainedwords,untrainedrarewords,andfamiliarwordswerepresentedformeaningjudgments.Inthemeaningjudgmenttask,thefirstwordap-pearedfor1secandthendisappeared,replacedbythesecondwordthatwasre-latedinmeaningon50%oftrials.Forexample,gloamingfollowedbytwilight,shouldgetayesresponse.EEGswererecordedcontinuouslyduringthesejudg-ments,soweobtainedERPindicatorsassociatedwithviewingthefirstword(gloaming) and its meaning mate (twilight).

Figure2showsthebehavioralresultsobtainedduringtheposttrainingmeaningjudgmentstask.Thethingtonoticeisthatskilledcomprehenderswerecorrectsig-nificantlymoreoftenthanlessskilledcomprehendersinmeaningjudgmentsmadetotherarewordsthatwetaughtthembutnottoeitheruntrainedrarewordsorfa-miliarwords(alsonottrained).Thelackofaskilldifferenceforuntrainedrarewordsshowsthatoverallknowledgeofrarewordswasnotdifferentacrossthetwogroups.Instead,theconclusionisthatthehighercomprehendersactuallylearnedthe new words better.

TheresultsoftheERPanalysis,whichareshowninFigure3forthegroupofskilledcomprehenders,addtothispicture.Plottedarethegrandaveragewave-formsatoneelectrode(thecentralreferenceelectrode)foreachconditionformeaning-relatedtrials.Theconditionsshowasimilarpatternforthefirst200msecorso,reflectingvisualorthographicprocessessharedbyallwords.Thefirstpointofseparationoccursat200msec,wheretrainedwordsseparatefrombothun-trainedandfamiliarwordsinanegativegoingshift.Thisreflectsanearly“notice”FIGURE2Comprehensionskilldifferencesinwordmeaningjudgmentsfollowinglearningofrarewords.Skilledcomprehendersshowedhigheraccuracyinjudgingmeaningrelationsfortrainedrarewordsbutnotforuntrainedrarewordsorfamiliar(known)words.BasedonPerfetti, Wlotko, and Hart (2005).

LEXICAL QUALITY

367

FIGURE3Anevent-relatedpotential(ERP)recordforskilledcomprehendersduringameaningjudgment.Thegrandaveragewaveformisshownforthereferenceelectrode(Cz).Theonsetofthefirstwordisexemplifiedforthetrainedrarewordgloaming.Theonsetofthesecondwordisexemplifiedfortherelatedwordtwilight.ThethreecurvesrepresentERPrecordsfortrainedrarewords(darkestline),untrainedfamiliar(known)words(intermediatedarkness),anduntrainedrarewords(lightestline).Twosignificanteffectsoftrainingarevisible,at200msecandabout550msec,thelatterrepresentingword-levelepisodicmemoryforthetrainedword.About400msecaftertheonsetofarelatedword,areductionintheN400isobservedfortrainedandfamiliarwords.Forthetrainedwords,lessskilledcomprehenders(notshown)showweakerepisodiceffectsat550msecandweakerN400meaningeffectsforthesecondrelatedword. See Perfetti, Wlotko, & Hart (2005)for fuller skill comparisons.

ofwordsthathadbeenrecentlyviewedintraining,basedonpreidentitysublexicalpatterns.Asecondseparationataround550msec,whichdoesmarkwordidentity,furtherdistinguishestrainedwordsfromtheothertwoclasses,nowinapositivegoingshift.Thisshiftisthesameinkeyrespects(distributionandpolarity)astheP600thatisobservedinmemorystudieswhenapreviouslyvieweditemispre-sented(Curran,1999;Rugg,1995).ThisERPshiftmarksrecognitionoftheepi-sodicmemorylaiddownbythetrainingevent:Ineffect,thebrainrespondstothiswordasfamiliarbecausethewordhasbeenpartoftheprevioushour’straining.Awordthathasgaineditsfamiliaritythroughexperiencespriortotheexperiment(thefamiliarwords)showsnosucheffect.Theimplicationofthisisthatwehaveidentifiedamarkeroffamiliarity-basedlearningthatisexpressedwhenareaderviews a word.

368PERFETTI

Sonowthequestioniswhetherthisword-levelepisodicmemoryeffectisob-servedequallyinourskilledandlessskilledcomprehenders.Theanswerisno.LessskilledreadersshowedthesameFigure3patternofERPshiftsduringthemeaningjudgments.However,thekeymarkerofepisodicmemoryat550msec(theP600trainingeffect)wassignificantlyreducedinamplitudeforthelessskilledcomprehenders.Thus,onaverage,awordthathadbeenlearnedjustpriortotheexperimentmadelessofanimpressiononthelessskilledreaders.Itisinterest-ingthatthefirsteffectoftraining,the200-msecnegativityfortrainedwords,wasnotdifferentforthetwoskillgroups.Ifourinterpretationofthesetwocomponentsiscorrect—thatthe200msecisbasedonsublexicalfamiliaritywhereasthe550-mseceffectisbasedonlexicalidentity—thenweconcludethatalllearnersre-spondtothedistinctiveletterpatternofarecentlytrainedword,buttheydifferinrecognizingthewordepisode—forexample,thisisthatwordgloamingthatIjustexperienced a few minutes ago.

Recognizinga“gloaming”word-episodemayinvolveretrievingitsmeaning,butitmaynot.Themoredirecttestofameaningprocessisintheresponsetothesecondword,twilight.Iftheparticipanthaslearnedthemeaningofgloaming,thenthesecondword,twilight,whichiscloselyrelatedinmeaning,shouldproduceareducedN400.TheN400isasignatureforsemanticcongruence,alargenegativegoingshiftwhenawordisincongruentwithitsprecedingcontext.Whenawordiscongruentwithitsprecedingcontext,theN400isreduced.ThisN400reductioniswhathappenedwhengloamingwasfollowedbytwilight—providedthemeaningofgloamingwaslearned.AsshowninFigure3,anN400appearedfortheun-trainedwords,reflectingthefactthatifaparticipanthadnotexperiencedgloamingintraining,therewasnoparticularcongruenceprovidedbythewordtwilight.However,ifgloaminghadbeenlearned,thentheN400shouldbereduced,andthisreduction, as well as one for familiar words, is visible in Figure 3.

Onceagain,theERPeffectwasdifferentforskilledandlessskilledcom-prehenders.AlthoughbothgroupsshowedareductionoftheN400whenthesec-ondwordwasrelatedinmeaningtothefirstword,thisN400forlessskilledcomprehenderswassignificantlylessreducedfortrainedwordscomparedwiththereduction for skilled readers.

Wemustconcludethatafterlessthan50minspentlearningthemeaningsof60rarewordswhosemeaningswereunknownpriortothestudy,skilledcom-prehendersmademoreeffectiveuseofthelearningperiod.EspeciallyinterestingisthefactthatthisdifferencewasseeninERPsrecordedduringthebriefperiodinwhichthewordwasbeingviewed,priortotheappearanceofasecondword.TheN400onthesecondwordreflectedthestrongerlearningofthemeaningbyskilledcomprehenders;theepisodicmarkeronthefirstwordindicatedastrongerassocia-tion between the word and its training.

Onemightarguewiththeinterpretationthatthesearelearningeffectsasopposedtosubtleexperienceeffects.Althoughwechosewordsthatwereindividually

LEXICAL QUALITY369

tailoredtobeunknownforagivenparticipant,perhapsthatprocedureunderesti-matedveryslightfamiliaritydifferencesthatfavoredthemoreskilledandmoreex-periencedreaders.Thisseemsunlikely.Foronething,wedidnotfinddifferencesintheuntrainedwords,whichwerefromthesamepopulationofpretestedrarewords.Ifskilledreadershadsomeunmeasuredfamiliaritywiththetrainedwordspriortothestudy,thentheyshouldhavethesameunmeasuredfamiliarityfortheuntrainedwords.ButthebehavioralandERPresultsbothsaytheydidnot.Further-more,theinterpretationthatwehaveadifferenceintheabilitytolearnwordmean-ingsisconsistentwiththeresultsofHart’s(2005)dissertation.Readershighincomprehensionskillshowedbettermeaningdecisionperformancethanlow-skillcomprehenders on words learned in her artificial language.

Acquiring Lexical Form Stability

AsindicatedinTable1,oneofthefeaturesofLQisastablelexicalrepresentation.Stabilityoccursaslettersandphonemeconstituentsbecomespecifiableascon-stants(fullyspecified)ratherthanvariablesinthewordrepresentation.High-qual-ityrepresentationsarefullyspecified(Perfetti,1992).Inastudythatusedrare-wordlearningtoexamineformstabilityduringlearning(Yang&Perfetti,2006),skilledandlessskilledcomprehenderslearnedthemeaningsof42rarewordsoverfourmini–training“sessions”inthecourseofasingleday.Aftereachsession,par-ticipantsalsomadelexicaldecisions,choosingwhetheragivenletterstringwasthecorrectformofawordtheywerelearning.Thefoilsvariedsystematicallyintheirorthographicandphonologicaloverlapwiththecorrectform.Thismanipula-tionallowedtestsofthelearners’formstability.Iflearnersacquirewell-specifiedrepresentationsofthewordtheywerelearning,similarformsshouldproducelessinterference.

Table2illustrateswhatthelearnerswereupagainst.Inthemidstoftryingtolearnthemeaningofthenew,veryrareword,hebetude,learnersencountereitheroneofthespellingsshowninTable2orthecorrectformofthewordhebetude,de-cidingwhetherthepresentedformisthecorrectform.2Orthographicallysimilarfoils(twoleftcolumnsofTable2)hadhighspellingoverlapwiththetarget,andsomeofthesealsohadveryhighphonemicoverlap.Controlfoilssharedonlyaninitialletterwiththetarget.Anunusualfeatureofthisstudyistheuseofmultisyllabicwords,whichrequiredsomeflexibilityincreatingfoils.AsTable2shows,theorthographicfoilsforhebetudeallsharebothgraphemesandphonemesforthefirstofthethreesyllables.(Thismeanssubstantialphonemicoverlapevenamonglowoverlapfoils.)Highphonemicoverlapinthiscasewascarriedby

2AccordingtoWorldWideWords(Quinion,July2007),anAugust2001columninTheWashingtonPostobservedthat“ToomanyAmericansslouchtowardaterminalfunkofhebetudeandsloth.”Itwasalso the “word of the day” on on January 24, 2004.

370PERFETTI

TABLE 2

Foils for the Rare WordHebetude

High Orthographic OverlapHigher PhonologicalOverlap

hebitude

hebatude

hebutude

hebotudeLower PhonologicalOverlaphebetidehebetadehebetedehebetode

Control:First Letter Onlyhodilaneharisadehigiforehudufiseidenticalfirstandthirdsyllables(leftcolumn).Thefourhighoverlapfoilsdifferedonlyonthevowelletterofthesecondunstressedsyllable.Becausethesyllableisunstressed,vowelscanmigratetowardaminimalvowel.Onecouldconceivablypronounce all the second syllable vowels as unstresseduh.

Ofinterestareresultsofbothformandmeaninglearning.Themeaningpartofthestudyinvolvedviewingthewordandhearingaspokendefinition,whichcouldberepeatedatthelearner’soption,followedbyatestofmeaninginwhichthewordwasspoken.Weexplainedtoparticipantsthatlearningthemeaningsofrarewordslikehebetudewasthegoal.Eachofthefoursessionsofmeaninglearningwasfol-lowed by the form task (lexical decision).

ThemeaningresultsareshowninFigure4.Skilledcomprehenderslearnedmorethanlessskilledcomprehendersfromthefirstsession,andtheiradvantageremainedconstantoverthefoursessions.ThisresultconvergeswiththatofFIGURE4Increaseinrarewordlearningoverfourlearning“sessions”(allwithinoneday).Skilledcomprehenderslearnedmoreduringthefirstsessionandmaintainedthisslightadvan-tage.

LEXICAL QUALITY

371

Perfetti,Wlotko,&Hart(2005)infindingthatskilledcomprehenderslearnthemeanings of new words more effectively than less skilled comprehenders.

Formlearning,whichisshowninFigure5,wasfairlygoodoverall,withfalsealarmstofoilsrangingaround10%.Nevertheless,skilledcomprehendersweremoreaccurateatrejectingsimilarfoilsandselectingthecorrectform.Afteroneses-sionoflearning,lessskilledcomprehenderschoseafoilon20%oftrials.Figure5doesnotdistinguishamongfoiltypes,butthereweredifferences:Thefoilswithhighphonemicoverlapattractedthehighestpercentageoffalsealarms,andfoilswithlessphonemicoverlapbuthighorthographicoverlapwerenext.Controlfoils,whichsharedonlytheinitialletterwiththetarget,attractedfewfalsealarms.However,therewasnointeractionofthetypeoffoilwithreadergroup.Bothskilledandlessskilledcomprehendersmademoreerrorstofoilswhenbothorthographicandphone-micoverlapwashigh.Thus,theconclusionisthatskilledcomprehenderslearnnotonlynewwordmeaningsmoreeffectivelybutalsonewformsmoreeffectively.LQimpliescoherencebetweenformandmeaningcomponents,andskillinreadingisas-sociatedwithhigherLQrightfromthebeginningoflearning.

Theideaofstabilityismorespecificthaneffectiveformlearning.Itimpliesthattheword’srepresentationcomestocomprisespellingandpronunciationpatternsthatareidenticalonsuccessiveobservations.OnewaytoexaminestabilityinthisFIGURE5Formaccuracy(lexicaldecisions)overfourmeaninglearningsessions.Hitratesforrealrarewords(e.g.,hebetude)comparedwithfalsealarmstosimilarforms(hebitude,hebetade, etc.). Less skilled comprehenders show less word-form accuracy.

372

PERFETTI

senseisshowninFigure6,whichplotsperformanceonsuccessivelexicalchoicetrials.Theperformanceplottedisconditionalforsessionsbeyondthefirst:Givenacorrectchoiceinonetestsession,theprobabilityofbeingcorrectthenexttime.Thuschoosinghebetudeatonetestbutthenhebitudeonthenextwouldcontributenegativelytothismeasure,whichisoneofstability—choosingthesameformthenexttime.Figure6showsagaintheslightlybetterperformanceofskilledcompre-hendersrightfromthebeginning,thatis,fromthefirstsession.Stability(condi-tionalprobabilitiesofcorrectresponses)showsasmallbutconsistentdifferenceonthenexttwosessions,andthedifferencebecomeslargestonthefinalsession,whenskilledcomprehendersproduced96%stabilitycomparedwith86%forlessskilled comprehenders.

Aninformalcharacterizationofthisresultisthatthelessskilledgroup,onaver-age,showsasmallbutnoticeableinstabilityevenafterfoursessionsoflearning.This is evidence in favor of the stability implication of LQ.

Less Skilled Comprehenders’Semantic Processing

of Ordinary Words

Thetwoprecedingsectionshaveconcludedthatproblemsinlearningwordmean-ingsandinlearningwordformsbothassociatewithcomprehensionskill.IntheFIGURE6Anindicatorofformstabilityduringwordlearning.Theconditionalprobabilityofacorrectdecisiononrarewordsgivenacorrectresponseontheprecedingtrialoversucces-sivelexicaldecisionsessions.Lessskilledcomprehendersshowslightlylesswordformstabil-ity by this measure.

LEXICAL QUALITY373

StudiesofForm–MeaningConfusionssection,weconcludedthatproblemsinmeaningprocessesthatwereassociatedwithformweremorelikelyforlessskilledreaders.Inthissectionweconsidersemanticprocessingdifferencesforordinarywords already known to the reader.

Thehypothesisthatsemanticdeficitsmayexplaincomprehensionproblemshasbeendevelopedandtestedforchildren(Nation&Snowling,1998,1999)ndi(2005;Landi&Perfetti,2007)extendedtherangeofthishypothesistoincludeadults,linkingbehavioralandERPmeasurestotheassessmentsofspecificlexicalandcomprehensionskill.Ineffect,Landi’sexperimentstargetedthegroupofread-ersofFigure1whoareinthelowerrightquadrant—belowthediagonalandtotherightofthex-axismidpoint:readersofhighlexicalskillbutlowcomprehensionskill.

Landitestedthehypothesisthatsuchreadersarelessabletoeffectivelyusese-manticcategoryinformation.Participantsmademeaningdecisionsforwordpairsthatwererelatedassociativelyandcategoricallyorcategoricallyonly.Forexam-ple,brother–sister,dog–cat,pillow–sleepwerepairsbothassociativelyandcate-goricallyrelated;green–pink,banana–tomato,kite–ndi’shypothesiswasbasedonNationandSnowling’s(1999)conclusionthatchildren’scomprehensionproblemsreflectedfailurestorepresentsemanticcategoryrelations.Thus,lessskilledcomprehenderswouldbecompara-bletoskilledcomprehendersinsemanticprocesseswhenassociativerelationscouldfacilitatedetectionofsemanticrelations,buttheywouldbelesssuccessfulthanskilledcomprehenderswhentheyhadtorelyonlyoncategoricalrelations.Toprovidemoreinformativecomparisons,participantsalsomadesemanticjudg-mentsonpairsofpictures(controllingforwordreading)andhomophonedeci-sions,forexample,boar–bore,chants–chance,tacks–tax(assessingphonologicalprocessing).

Thekeyresultscanbesummarizedasfollows:Accuracywasgenerallyhighforalltasksandnotdifferentbetweenthetwogroups.However,timesforcorrectdeci-sionswerefasterfortheskilledgroupthanthelessskilledgroupacrossalltasks.Thus,skilledcomprehendersshowedfastersemanticprocessingwhetherthestim-uliwerewordsorpicturesandfasterwordjudgmentswhetherthedecisionwasbasedonmeaningorpronunciation.Forwordmeanings,thedecisiontimestosemanticcategorypairswereslowerthantoassociativepairs.Thisconfirmstheas-sumptionthatdetectingcategoryrelationswouldrequiremoreprocessingthande-tectingrelationsbetweenassociations.Correctdecisionsbasedoncategoryrela-tionstookanaverageof68mseclongerthanthosebasedonassociativerelationsfortheskilledgroupand139mseclongerforthelessskilledgroup.Thesegroupdifferences,althoughinthedirectionimpliedbyLandi’sextensionofNationandSnowling’s (1999) hypothesis, were not reliable.

374PERFETTI

TABLE 3

Summary Skill Pattern in Landi ERP Results

Comprehension

Skill Differences in

Decision Times?

Yes

Yes

Yes

YesComprehensionSkill Differencesin ERP Measures?NoNoYes, but small(P200, N400)Yes, large (P200,N400)TaskPhonological decisionsSemantic picture decisionsSemantic word decisions—CategoricalSemantic word decisions—Associative

Note.Source: Landi and Perfetti (2007). ERP = event-related potential.

Whereasthedecisionstimesshowedverygeneralprocessing-speeddiffer-ences,theERPmeasuresshoweddifferencesthatwererestrictedtothesemanticworddecisions,andtheseresultspresentaslightlydifferentpicture.BothgroupsshowedareliablereductionintheN400forrelatedtrialsforbothcategoryandas-sociativepairs,althoughineachcasethereductionshownbyskilledcompre-henderswassomewhatlarger.Moreover,thehigh-skillgroup,butnotthelow-skillgroup,showedanadditionalN400reductionfortheassociativelyrelatedpairsrel-ativetothecategorypairs.Thus,contrarytotheexpectationthatlessskilledcomprehenderswouldhavespecificproblemswithcategoryrelations,theERPssuggestthatsemanticcategoryprocesseswerecomparableinthetwogroups,buttheERPsforassociativerelationswerenot.(Thetendencyforlessskilledcom-prehenderstotakerelativelylongerthanskilledcomprehendersoncategorydeci-sionsmayberelevant;howeversuchadifferencecouldbeinterpretedasanaddi-tionalcheckingprocesswellbeyondthemoreautomaticsemanticprocessthatisreflected in the N400.)3

Table3showsasummaryofskilldifferencesacrosstasks.Takentogether,theysuggestdetailedevidenceforsemantic-processingdifferencesbetweentheskilledandlessskilledcomprehenders.TheERPevidencesuggestedthatskilledcom-prehendershadastrongermeaningcongruenceresponsewhenwordswerecate-goricallyrelated,andthisresponsewasevenstrongerwhenthewordswerealsoas-sociativelyrelated.Thelessskilledgroupshowedacongruenceresponsealso,butthisresponsewasnotstrengthenedbyassociativerelations.Noticethatifallwehadwerebehavioraldata,ourconclusionwouldbedifferent,andperhapsmislead-ing.Wewouldconcludethatthereisnosemanticprocessingskilldifference3TheERPresultsalsoproducedevidenceforearlysemanticeffectsat200msecthatweremorecon-sistentacrosselectrodesitesandtrialconditionsforskilledcomprehenders.Theyalsoshowed200-mseceffectsforthephonologicaltask,comparinghomophonesversusnonhomophonesthatwerecomparable across the two groups.

LEXICAL QUALITY375

specifictolanguage,becauseofdifferencesinpicturedecisionspeed.Andwewouldhaveconcludedthatthesegroupsdifferedeitheronphonologicalprocesses(becauseofthephonologicaldecisionspeeddifferences)or,morelikely,onnon-specificprocessing-speeddifferences.TheERPdataaretellingusmoredirectlyaboutthebrain’sresponsetowords,andtheyseemtosaythattherearespecificskill differences related to word-meaning processes.

Lexical Processing During Text Comprehension

Myfinalexamplereturnstothekeylinkbetweenlexicalprocessesandcompre-hension.Thislinkismostdirectatthelevelofshortrunsoftext,asentenceortwo,whereonecanobservewordprocessing“on-line”aspartoftextreading.Thefocalissueistheprocessesthatintegratethewordcurrentlybeingreadwiththeongoingrepresentationofthetext.Thisintegrationisacentralconnectingeventbetweenword identification and text comprehension.

Theskillquestioniswhetherdifferencesingloballyassessedcomprehensionariseatthislocal(i.e.,onewordatatime)processing.Doskilledcomprehendersintegratewordsimmediatelyintothetext?Dolessskilledcomprehenders?TheLQH(andalsoverbalefficiencytheory)predictsword-textintegrationproblemsfor less skilled readers.

Onceagain,ourstudiesonthisquestionuseERPstoexaminetheN400seman-ticcongruenceindicator.TheN400hasbeenfoundtovarywithdemandsonsen-tenceandtext-levelintegration(vanBerkum,Hagoort,&Brown,1999),anditisthatfactthatweexploitedintwoparallelstudies,onewithskilledcomprehenders(Yang,Perfetti,&Schmalhofer,2007)andonewithlessskilledcomprehenders(Yang,Perfetti,&Schmalhofer,2005).Theapproachofthestudiesisillustratedinthe following text:

Afterbeingdroppedfromtheplane,thebombhitthegroundandexploded.Thewas ….

ERPswereacquiredaseachwordwasread.Theunderlining(notvisibletopartici-pants) marks the target word (explosion) for our analysis.

Whenthereadercomestoexplosionatthebeginningofthesecondsentence,itshouldberelativelyeasytointegratethewordwithanunderstandingofthetextbasedonthefirstsentence.Thatis,wecanassumethatthereader’stextmemoryincludestheevent-proposition[bombexploded],amongotherpropositionsfromthefirstsentence,andthewordexplosionisintegratedwiththisevent-proposition;thatis,itistakenascoreferentialwiththeproposition.Thisconditionistermed“explicit” because there are explicit coreferential phrases (exploded, explosion).Asecondcondition,theparaphrasedcondition,wasofspecialinterestfromasemanticprocessingpointofview.Insteadofthefirstsentencereferringto

本文来源：https://www.bwwdw.com/article/7gim.html