The Science of Word Recognition.docx

The Science of Word Recognition.docx

《The Science of Word Recognition.docx》由会员分享，可在线阅读，更多相关《The Science of Word Recognition.docx（22页珍藏版）》请在冰豆网上搜索。

TheScienceofWordRecognition

TheScienceofWordRecognition

Introduction

Evidencefromthelast20yearsofworkincognitivepsychologyindicatethatweusetheletterswithinawordtorecognizeaword.ManytypographersandothertextenthusiastsI’vemetinsistthatwordsarerecognizedbytheoutlinemadearoundthewordshape.Somehaveusedthetermboumaasasynonymforwordshape,thoughIwasunfamiliarwiththeterm.ThetermboumaappearsinPaulSaenger’s1997bookSpaceBetweenWords:

TheOriginsofSilentReading.ThereIlearnedtomychagrinthatwerecognizewordsfromtheirwordshapeandthat“Modernpsychologistscallthisimagethe‘Boumashape.’”

Thispaperiswrittenfromtheperspectiveofareadingpsychologist.Thedatafromdozensofexperimentsallcomefrompeerreviewedjournalswheretheexperimentsarewellspecifiedsothatanyonecouldreproducetheexperimentandexpecttoachievethesameresult.Thispaperwasoriginallypresentedasatalkatthe ATypI conferenceinVancouverinSeptember,2003.

Thegoalofthispaperistoreviewthehistoryofwhypsychologistsmovedfromawordshapemodelofwordrecognitiontoaletterrecognitionmodel,andtohelpotherstocometothesameconclusion.Thispaperwillcovermanytopicsinrelativelyfewpages.AlongthewayIwillpresentexperimentsandmodelsthatIcouldn’thopetocovercompletelywithoutboringthereader.Ifyouwantmoredetailsonanexperiment,allofthereferencesareattheendofthepaperaswellassuggestedreadingsforthoseinterestedinmoreinformationonsometopics.Mostpapersarewidelyavailableatacademiclibraries.

Iwillstartbydescribingthreemajorcategoriesofwordrecognitionmodels:

thewordshapemodel,andserialandparallelmodelsofletterrecognition.Iwillpresentrepresentativedatathatwasusedasevidencetosupporteachmodel.Afteralltheevidencehasbeenpresented,Iwillevaluatethemodelsintermsoftheirabilitytosupportthedata.AndfinallyIwilldescribesomerecentdevelopmentsinwordrecognitionandamoredetailedmodelthatiscurrentlypopularamongpsychologists.

Model#1:

WordShape

Thewordrecognitionmodelthatsayswordsarerecognizedascompleteunitsistheoldestmodelinthepsychologicalliterature,andislikelymucholderthanthepsychologicalliterature.Thegeneralideaisthatweseewordsasacompletepatternsratherthanthesumofletterparts.Someclaimthattheinformationusedtorecognizeawordisthepatternofascending,descending,andneutralcharacters.Anotherformulationistousetheenvelopecreatedbytheoutlineoftheword.Thewordpatternsarerecognizabletousasanimagebecausewehaveseeneachofthepatternsmanytimesbefore.JamesCattell（1886）wasthefirstpsychologisttoproposethisasamodelofwordrecognition.Cattellisrecognizedasaninfluentialfounderofthefieldofpsycholinguistics,whichincludesthescientificstudyofreading.

Figure1:

Wordshaperecognitionusingthepatternofascending,descending,andneutralcharacters

characters

Figure2:

Wordshaperecognitionusingtheenvelopearoundtheword

Cattellsupportedthewordshapemodelbecauseitprovidedthebestexplanationoftheavailableexperimentalevidence.CattellhaddiscoveredafascinatingeffectthattodaywecalltheWordSuperiorityEffect.Hepresentedletterandwordstimulitosubjectsforaverybriefperiodoftime（5-10ms）,andfoundthatsubjectsweremoreaccurateatrecognizingthewordsthantheletters.Heconcludedthatsubjectsweremoreaccurateatrecognizingwordsinashortperiodoftimebecausewholewordsaretheunitsthatwerecognize.

Cattell’sstudywassloppybymodernstandards,butthesameeffectwasreplicatedin1969byReicher.Hepresentedstringsofletters – halfthetimerealwords,halfthetimenot – forbriefperiods.Thesubjectswereaskedifoneoftwoletterswerecontainedinthestring,forexample D or K.Reicherfoundthatsubjectsweremoreaccurateatrecognizing D whenitwasinthecontextof WORD thanwheninthecontextof ORWD.Thissupportsthewordshapemodelbecausethewordallowsthesubjecttoquicklyrecognizethefamiliarshape.Oncetheshapehasbeenrecognized,thenthesubjectcandeducethepresenceofthecorrectletterlongafterthestimuluspresentation.

Thesecondkeypieceofexperimentaldatatosupportthewordshapemodelisthatlowercasetextisreadfasterthanuppercasetext.Woodworth（1938）wasthefirsttoreportthisfindinginhisinfluentialtextbook ExperimentalPsychology.ThisfindinghasbeenconfirmedmorerecentlybySmith（1969）andFisher（1975）.Participantswereaskedtoreadcomparablepassagesoftext,halfcompletelyinuppercasetextandhalfpresentedinstandardlowercasetext.Ineachstudy,participantsreadreliablyfasterwiththelowercasetextbya5-10%speeddifference.Thissupportsthewordshapemodelbecauselowercasetextenablesuniquepatternsofascending,descending,andneutralcharacters.Whentextispresentedinalluppercase,alllettershavethesametextsizeandthusaremoredifficultandslowertoread.

Thepatternsoferrorsthataremissedwhileproofreadingtextprovidethethirdkeypieceofexperimentalevidencetosupportthewordshapemodel.Subjectswereaskedtocarefullyreadpassagesoftextforcomprehensionandatthesametimemarkanymisspellingtheyfoundinthepassage.Thepassagehadbeencarefullydesignedtohaveanequalnumberoftwokindsofmisspellings:

misspellingsthatareconsistentwithwordshape,andmisspellingsthatareinconsistentwithwordshape.Amisspellingthatisconsistentwithwordshapeisonethatcontainsthesamepatternsofascenders,descenders,andneutralcharacters,whileamisspellingthatisinconsistentwithwordshapechangesthepatternofascenders,descenders,andneutralcharacters.If test isthecorrectlyspelledword, tesf wouldbeanexampleofamisspellingconsistentwithwordshapeand tesc wouldbeanexampleofamisspellinginconsistentwithwordshape.Thewordshapemodelwouldpredictthatconsistentwordshapeswouldbecaughtlessoftenthananinconsistentwordshapebecausewordsaremoreconfusableiftheyhavethesameshape.Haber&Schindler（1981）andMonk&Hulme（1983）foundthatmisspellingsconsistentwithwordshapeweretwiceaslikelytobemissedasmisspellingsinconsistentwithwordshape.

Targetword:

 test

Errorrate

Consistentwordshape （tesf）

13%

Inconsistentwordshape （tesc）

7%

Figure3:

Misspellingsthatareconsistentwithwordshapearemissedmoreoften

Thefourthpieceofevidencesupportingthewordshapemodelisthatitisdifficulttoreadtextinalternatingcase.AlTeRnAtInGcaseiswherethelettersofawordchangefromuppercasetolowercasemultipletimeswithinaword.Thewordshapemodelpredictsthatthisisdifficultbecauseitgivesapatternofascending,descending,andneutralcharactersthatisdifferentthanexistsinawordinitsnaturalalllowercaseform.Alternatingcasehasbeenshowntobemoredifficultthaneitherlowercaseoruppercasetextinavarietyofstudies.Smith（1969）showedthatitslowedthereadingspeedofapassageoftext,Mason（1978）showedthatthetimetonameawordwasslowed,Pollatsek,Well,&Schindler（1975）showedthatsame-differencematchingwashindered,andMeyer&Gutschera（1975）showedthatcategorydecisiontimesweredecreased.

Model#2:

SerialLetterRecognition

Theshortestlivedmodelofwordrecognitionisthatwordsarereadletter-by-letterseriallyfromlefttoright.Gough（1972）proposedthismodelbecauseitwaseasytounderstand,andfarmoretestablethanthewordshapemodelofreading.Inessence,recognizingawordinthementallexiconwasanalogoustolookingupawordinadictionary.Youstartoffbyfindingthefirstletter,thanthesecond,andsoonuntilyourecognizetheword.

ThismodelisconsistentwithSperling’s（1963）findingthatletterscanberecognizedatarateof10-20msperletter.Sperlingshowedparticipantsstringsofrandomlettersforbriefperiodsoftime,askingifaparticularletterwascontainedinthestring.Hefoundthatifparticipantsweregiven10msperletter,theycouldsuccessfullycompletethetask.Forexample,ifthetargetletterwasinthefourthpositionandthestringwaspresentedfor30ms,theparticipantcouldn’tcompletethetasksuccessfully,butifstringwaspresentedfor40ms,theycouldcompletethetasksuccessfully.Goughnotedthatarateof10msperletterwouldbeconsistentwithatypicalreadingrateof300wpm.

Theserialletterrecognitionmodelisalsoabletosuccessfullypredictthatshorterwordsarerecognizedfasterthanlongerwords.Itisaveryrobustfindingthatwordrecognitiontakesmoretimewithlongerwords.Ittakesmoretimetorecognizea5-letterwordthana4-letterword,and6-letterwordstakemoretimetorecognizethan5-letterwords.Theserialletterrecognitionmodelpredictsthatthisshouldhappen,whileawordshapemodeldoesnotmakethisprediction.Infact,thewordshapemodelshouldexpectlongerwordswithmoreuniquepatternstobeeasiertorecognizethanshorterwords.

TheserialletterrecognitionmodelfailsbecauseitcannotexplaintheWordSuperiorityEffect.TheWordSuperiorityEffectshowedthatreadersarebetterabletoidentifylettersinthecontextofawordthaninisolation,whiletheserialletterrecognitionmodelwouldexpectthataletterinthethirdpositioninawordshouldtakethreetimesaslongtorecognizeasaletterinisolation.

Model#3:

ParallelLetterRecognition

Themodelthatmostpsychologistscurrentlyacceptasmostaccurateistheparallellett