FREQUENCY-EFFECTS-IN-LANGUAGE-PROCESSING

SSLA,24,143–188DOI:10.1017.S0272263102002024FREQUENCYEFFECTSINLANGUAGEPROCESSINGAReviewwithImplicationsforTheoriesofImplicitandExplicitLanguageAcquisitionNickC.EllisUniversityofWales,BangorThisarticleshowshowlanguageprocessingisintimatelytunedtoinputfrequency.Examplesaregivenoffrequencyeffectsintheprocessingofphonology,phonotactics,reading,spelling,lexis,mor-phosyntax,formulaiclanguage,languagecomprehension,grammat-icality,sentenceproduction,andsyntax.TheimplicationsoftheseeffectsfortherepresentationsanddevelopmentalsequenceofSLAarediscussed.Usage-basedtheoriesholdthattheacquisitionoflan-guageisexemplarbased.Itisthepiecemeallearningofmanythou-sandsofconstructionsandthefrequency-biasedabstractionofregularitieswithinthem.Determinantsofpatternproductivityincludethepowerlawofpractice,cuecompetitionandconstraintsatisfac-tion,connectionistlearning,andeffectsoftypeandtokenfrequency.Theregularitiesoflanguageemergefromexperienceascategoriesandprototypicalpatterns.Thetypicalrouteofemergenceofcon-structionsisfromformula,throughlow-scopepattern,toconstruction.Frequencyplaysalargepartinexplainingsociolinguisticvariationandlanguagechange.Learners’sensitivitytofrequencyinallthesedomainshasimplicationsfortheoriesofimplicitandexplicitlearningandtheirinteractions.Thereviewconcludesbyconsideringthehis-toryoffrequencyasanexplanatoryconceptintheoreticalandap-pliedlinguistics,its40yearsofexile,anditsnecessaryreinstatementasabridgingvariablethatbindsthedifferentschoolsoflanguageac-quisitionresearch.Addresscorrespondenceto:NickC.Ellis,SchoolofPsychology,UniversityofWales,Bangor,GwyneddLL572AS,UnitedKingdom;e-mail:[e-mail].2002CambridgeUniversityPress0272-2631/02$9.50143144NickC.Ellis“Perceptionisofdefiniteandprobablethings”(James,1890,p.82).Thereisalotmoretotheperceptionoflanguagethanmeetstheeyeorear.Aperceptisacomplexstateofconsciousnessinwhichantecedentsensationissupplementedbyconsequentideasthatarecloselycombinedtoitbyasso-ciation.Thecerebralconditionsoftheperceptionofthingsarethusthepathsofassociationirradiatingfromthem.Ifacertainsensationisstronglyassoci-atedwiththeattributesofacertainthing,thatthingisalmostsuretobeper-ceivedwhenwegetthatsensation.Wherethesensationisassociatedwithmorethanonereality,however,unconsciousprocessesweightheodds,andweperceivethemostprobablething:“allbrain-processesaresuchasgiverisetowhatwemaycallFIGUREDconsciousness”(James,1890,p.82,emphasisinoriginal).Accurateandfluentlanguageperception,then,restsonthecompre-henderhavingacquiredtheappropriatelyweightedrangeofassociationsforeachelementofthelanguageinput.Psycholinguisticandcognitivelinguistictheoriesoflanguageacquisitionholdthatalllinguisticunitsareabstractedfromlanguageuse.Intheseusage-basedperspectives,theacquisitionofgrammaristhepiecemeallearningofmanythousandsofconstructionsandthefrequency-biasedabstractionofreg-ularitieswithinthem.Languagelearningistheassociativelearningofrepre-sentationsthatreflecttheprobabilitiesofoccurrenceofform-functionmappings.Frequencyisthusakeydeterminantofacquisitionbecause“rules”oflanguage,atalllevelsofanalysis(fromphonology,throughsyntax,todis-course),arestructuralregularitiesthatemergefromlearners’lifetimeanalysisofthedistributionalcharacteristicsofthelanguageinput.Learnershavetofigurelanguageout.Thisreviewillustrates(a)howfrequencyunderpinsregularityeffectsintheacquisitionoforthographic,phonological,andmorphologicalform,and(b)thatlearningaccordstothepowerlawofpractice.Itshows,forexample,thatthereareeffectsofbigramfrequencyinvisualwordidentificationandofpho-notacticknowledgeinspeechsegmentation,effectsofspelling-to-soundcorre-spondencesinreading,andcohorteffectsinspoken-wordrecognition.Thereareeffectsofneighborsandtheproportionoffriends(itemsthatsharesur-face-patterncuesandhavethesameinterpretation)toenemies(itemsthatsharesurface-patterncuesbuthavedifferentinterpretations)inreadingandspelling,morphology,andspoken-wordrecognition.Athigherlevels,itcanbeshownthatlanguagecomprehensionisdeterminedbythelisteners’vastamountofstatisticalinformationaboutthebehavioroflexicalitemsintheirlanguageandthat,atleastforEnglish,verbsprovidesomeofthestrongestconstraintsontheresolutionofsyntacticambiguities.Comprehendersknowtherelativefrequencieswithwhichindividualverbsappearindifferenttenses,inactiveversuspassivestructuresandinintransitiveversustransitivestruc-tures,thetypicalkindsofsubjectsandobjectsthataverbtakes,andmanyothersuchfacts.Suchinformationisacquiredthroughexperiencewithinputthatexhibitsthesedistributionalproperties.ItisnotsomeidiosyncraticfactFrequencyEffectsinLanguageProcessingandAcquisition145inthelexiconisolatedfrom“core”grammaticalinformation;rather,itisrele-vantatallstagesoflexical,syntactic,anddiscoursecomprehension.Compre-henderstendtoperceivethemostprobablesyntacticandsemanticanalysesofanewutteranceonthebasisoffrequenciesofpreviouslyperceivedutter-anceanalyses.Languageuserstendtoproducethemostprobableutteranceforagivenmeaningonthebasisoffrequenciesofutterancerepresentations.Theseaccountsreadilycontribut