ERIC ED598417: Selective Attention, Diffused Attention, and the Development of Categorization

2016·1.1 MB·English

by  ERIC

CognitivePsychology91(2016)24–62
ContentslistsavailableatScienceDirect
Cognitive Psychology
journal homepage: www.elsevier.com/locate/cogpsych
Selective attention, diffused attention, and the
development of categorization
Wei (Sophia) Denga,⇑, Vladimir M. Sloutskyb,⇑
aDepartmentofPsychology,UniversityofMacau,China.
bDepartmentofPsychology,TheOhioStateUniversity,UnitedStates
a r t i c l e i n f o a b s t r a c t
Articlehistory: Howdopeoplelearncategoriesandwhatchangeswithdevelop-
Accepted24September2016 ment?Thecurrentstudyattemptstoaddressthesequestionsby
focusingontheroleofattentioninthedevelopmentofcategoriza-
tion. In Experiment 1, participants (adults, 7-year-olds, and
Keywords: 4-year-olds) were trained with novel categories consisting of
Categorization
deterministic and probabilistic features, and their categorization
Attention
andmemoryforfeaturesweretested.InExperiment2,participants’
Learning
Cognitivedevelopment attention was directed to the deterministic feature, and in
Experiment 3 it was directed to the probabilistic features.
Attentional cueing affected categorization and memory in adults
and7-year-olds:theseparticipantsreliedonthecuedfeaturesin
theircategorizationandexhibitedbettermemoryofcuedthanof
non-cued features. In contrast, in 4-year-olds attentional cueing
affected only categorization, but not memory: these participants
exhibited equally good memory for both cued and non-cued
features.Furthermore,acrosstheexperiments,4-year-oldsremem-
berednon-cuedfeaturesbetterthanadults.Theseresultscoupled
withcomputationalsimulationsprovidenovelevidence(1)point-
ing to differences in category representation and mechanisms of
categorizationacrossdevelopment,(2)elucidatingtheroleofatten-
tioninthe development ofcategorization, and(3) suggesting an
importantdistinctionbetweenrepresentationanddecisionfactors
incategorizationearlyindevelopment.Theseissuesarediscussed
withrespecttotheoriesofcategorizationanditsdevelopment.
(cid:1)2016PublishedbyElsevierInc.
⇑
Correspondingauthorsat:DepartmentofPsychology,247PsychologyBldg.,1835NeilAvenue,TheOhioStateUniversity,
Columbus, OH 43210, United States (V.M. Sloutsky) and Department of Psychology, FSS, University of Macau, E21-3037,
AvenidadaUniversidade,Taipa,MacauS.A.R.,China(W.(Sophia)Deng).
E-mailaddresses:[email protected](W.(Sophia)Deng),[email protected](V.M.Sloutsky).
http://dx.doi.org/10.1016/j.cogpsych.2016.09.002
0010-0285/(cid:1)2016PublishedbyElsevierInc.
W.(Sophia)Deng,V.M.Sloutsky/CognitivePsychology91(2016)24–62 25
1.Introduction
Categorization,ortheabilitytoformequivalenceclassesofdiscriminableentities,isanessential
componentofhumancognition.Thisabilitytoassignsamenesstoappreciablydifferentstimuliwas
identifiedbyWilliamJames(1983/1890)asthekeypropertyofthemind.Categoriesenablerecogni-
tionand differentiationof objects,people,and events, helporganizingour existing knowledge,and
promote generalization of knowledge to new situations. For example, having observed that all of
thepreviouslyencounteredbirdshavebeenabletofly,onemightinferthatanewlyencounteredbird
can fly as well. There are a number of key (and relatively uncontroversial) findings pertaining to
categorization.
First, at least a rudimentary ability to form categories appears in early infancy (Eimas & Quinn,
1994; Oakes, Madole, & Cohen, 1991) and is manifested in a variety of species (Lazareva,
Freiburger, & Wasserman, 2004; Smith et al., 2012). And second, there is evidence of remarkable
development in the ability to form and represent categories (e.g., Huang-Pollock, Maddox, &
Karalunas, 2011; Kloos & Sloutsky, 2008; Minda, Desroches, & Church, 2008; Quinn & Johnson,
2000;Rabi,Miles,&Minda,2015;Rabi&Minda,2014;Smith,1989;Younger&Cohen,1986;seealso
Quinn,2011;Sloutsky,2010,forreviews).Itishardlycontroversialthatadultscanacquireexceedingly
abstract(oftennon-perceptual)categories,whereasthereislittleevidencethatinfantsorevenyoung
childrencanacquirecategoriesofsimilarlevelsofabstraction.Althoughmanyagreethatcategoriza-
tiondoesdevelop,thereislessagreementastowhatchangeswithdevelopmentandwhy.
Accordingtosomeexplanations,thedevelopmentisdrivenbyacquisitionofdomain-specific(or
even concept-specific) knowledge (e.g., Carey, 1999; Inagaki & Hatano, 2002; Keil, 1992; Keil &
Batterman,1984).Accordingtoother(moredomain-general)explanations,thedevelopmentisdriven
bychangesinbasiccognitiveprocesses,suchasselectiveattention,workingmemory,andcognitive
control (see Fisher, Godwin, & Matlen, 2015; Rabi & Minda, 2014; Sloutsky, 2010; Sloutsky, Deng,
Fisher,&Kloos,2015;Smith,1989).Selectiveattentionisaparticularlypromisingcandidatebecause,
asdiscussedindetailbelow,it(a)hasbeenidentifiedasanimportantcomponentofadultcategory
learningand(b)clearlyundergoesdevelopment.
Althoughthesepossibilitiesprovideradicallydifferentdevelopmentalaccounts,theydonothave
tobemutuallyexclusive.Perhapstheformeraccountexplainsdevelopmentalchangesinhowfamiliar
categoriesareinterconnectedandused(seeFisher,Godwin,Matlen,&Unger,2015),whereasthelat-
terexplainsdevelopmentalchangesinacquisitionandrepresentationofnovelcategories.
The goal of current research is to better understand developmental changes in how novel cate-
gories are learned and represented. We propose that changes in selective attention may drive this
development, derive specific hypotheses from this general proposal, and test these hypotheses in
thereportedexperiments.Intheremainderofthissection,wefirstreviewtheroleofselectiveatten-
tionincategorylearningandcategoryrepresentation.Wethendiscussthedevelopmentofselective
attentionanditsimplicationforcategorylearning.
1.1.Selectiveattention,categorylearning,andcategoryrepresentation
SincethepioneeringresearchoncategorylearningbyShepard,Hovland,andJenkins(1961),selec-
tive attention1 has been an important component of models of categorization and category learning.
Exemplarmodels(Hampton,1995;Medin&Schaffer,1978;Nosofsky,1986),prototypemodels(Smith
& Minda, 1998), clustering models (Love, Medin, & Gureckis, 2004), and dual process models (Ashby,
Alfonso-Reese, Turken, & Waldron, 1998) all include some form of selective attention as a factor
1 Notethatintheattentionliterature(e.g.,Egeth&Yantis,1997;Pashler,Johnston,&Ruthruff,2001;Posner&Petersen,1990)
selective attention has been conceptualized as either involuntary, bottom-up, and stimulus-driven (when it is captured
automaticallybyahighlysalientstimulus)orasvoluntary,top-down,andgoal-driven(whenthegoalistofindaredobjectina
pileofthingsofdifferentcolors).Inthecategorizationliterature,selectiveattentionhasbeenconceptualizedinthelattersense.For
thepurposeofconsistency,inthispaperwewillusetheconceptualizationadoptedinthecategorizationliterature.However,we
willdiscusslimitationsofthisconceptualizationinSection6.
26 W.(Sophia)Deng,V.M.Sloutsky/CognitivePsychology91(2016)24–62
determiningtheinfluence(orweight)ofstimulusdimensionsoncategorization.Accordingtosomeof
thesemodels,aslearningprogresses,theseweightschange,withchangesoccurringgraduallyinassocia-
tivemodelsandabruptlyinrule-basedmodels(seeRehder&Hoffman,2005foradiscussion).
Therearethreeimportantsourcesofevidenceofhowselectiveattentionaffectscategorylearning
and under what conditions: (1) selective attention has consequences in the form of attention opti-
mizationandlearnedinattention,andtheseconsequencesareoftenobservedafteradultslearncate-
gories; (2) attention optimization is specific to learning categories of particular structures; (3)
attentionoptimizationisspecifictolearningcategoriesundercertaintaskconditions.
1.1.1.Consequencesofselectiveattention:attentionoptimizationandlearnedinattention
Ifcategorylearningisaccompaniedbyattendingselectivelytofeaturesdistinguishingtheto-be-
learnedcategories(i.e.,diagnosticfeatures),suchselectivitymayresultinanumberoftestableconse-
quences (see Hoffman & Rehder, 2010, for a review). The most important of these consequences is
shifting attention to the diagnostic features (i.e., attention optimization) and learning to ignore less
diagnosticorirrelevantfeatures(i.e.,learnedinattention).
Consider a situationin which one learns two categories, such as squirrels versus chipmunks.As
learningprogresses,thelearner’sattentionmayshifttostripes(whichisadiagnosticfeature),which
would indicate attention optimization. At the same time, the learner would attend less to or even
ignore the tail, which is a non-diagnostic feature. This phenomenon of learned inattention to non-
diagnosticfeatureswouldtranspireif,afterlearningthecategoriesofsquirrelsandchipmunks,alear-
nerembarksonanewcategorizationtask–differentiatingbetweensquirrelsand hamsters.Inthis
case,thetailthatwasnon-diagnosticforpreviouslearningbecomesdiagnosticforcurrentlearning.
Importantly,priorhistoryofignoringthetailwouldmakeitmoredifficulttoshiftattentiontothetail
thanitwouldhavebeenwithoutlearningofthefirstsetofcategories.
Toexaminethisissue,HoffmanandRehder(2010)presentedadultswithamulti-phasecategory
learning task, such that dimensions that were diagnostic in phase 1 of category learning became
non-diagnosticinphase2,whereasdimensionsthatwerenon-diagnosticinphase1becamediagnos-
ticinphase2.Usingacombinationofbehavioralandeyetrackingmethodologies,theauthorsfound
thatadultlearnersoptimizedattentioninphase1byshiftingittothecategory-relevant(ordiagnos-
tic)dimensionandexhibitedlearnedinattentioninphase2.Thesefindingssuggestthatinthecourse
ofcategorylearning,adultstendtoattendselectively,tryingtoextractthemostdiagnostic(orrule)
dimension(s).2Attentionoptimizationaccompanyingcategorylearninginadultshasbeenalsoreported
inaneye-trackingstudybyBlair,Watson,andMeier(2009).
1.1.2.Attentionisallocateddifferentlytocategoriesofdifferentstructures
Intheirseminalstudyofcategorylearning,Shepardetal.(1961)identifiedanumberofcategory
structures that elicited different learning profiles in human adults. For example, type I category is
theeasiestandthemostbasiccategorystructure:categorizationdecisioncanbemadeonthebasis
ofasingledimension(e.g.,blueitemsvs.reditems).Incontrast,typeVIcategoryisthemostdifficult
one,asnodimensionortheircombinationsupportscategorization:theassignmentofeachitemtoa
categoryshouldbelearnedbyroteandmemorized.Therefore,whereasitisadaptivetoattendselec-
tivelywhenlearningtypeIcategory,itishardlyusefulwhenlearningtypeVIcategory.Toexamine
thisissue, RehderandHoffman(2005)recordedadultparticipants’ eyemovementsduringlearning
ofcategoriesofdifferentstructures.Theirresultsindicatedthatparticipantsexaminedprimarilythe
diagnosticdimensioninthecaseoftypeIcategory,thuspresentingevidenceofattentionoptimiza-
tion.However,itcouldbearguedthatattentionoptimizationisaconsequenceofanylearning,notjust
selective attention to diagnostic features. This possibility was rejected when researchers examined
attentionallocationinthecaseoftypeVIcategory,whennoneofthedimensionswasdiagnostic.Upon
observingthatparticipantsexaminedalldimensions,theresearchersconcludedthatparticipantsallo-
2 Notethatinmostresearchdiscussedhereparticipantslearnedvisualcategories.However,whencategorieswerepresentedas
listsoffeaturesthatparticipantsread,lessselectivity/optimizationwasobserved(e.g.,Bott,Hoffman,&Murphy,2007).
W.(Sophia)Deng,V.M.Sloutsky/CognitivePsychology91(2016)24–62 27
cateattentiondifferently,dependingonthestructureoftheto-be-learnedcategory:selectivelywhen
therearediagnosticfeaturesanddiffuselywhentherearenodiagnosticfeatures.
1.1.3.Attentionisallocateddifferentlyacrossdifferentcategorizationtasks
Multipletasks canbeused toelicitcategorylearning. Thetasks mostfrequentlyusedin thelab
studiesareclassificationlearningandinferencelearning.Intheformertask,participantslearnacate-
gorybypredictingthelabelofagivenitemonthebasisofpresentedfeatures:oneachtrial,apartic-
ipant is presented with an item and has to predict whether the item is labeled A or B. In inference
learningparticipantshavetoinferamissingfeatureonthebasisofcategorylabelandotherpresented
features.Oneachtrialanitemispresentedandlabeled,butoneofthefeaturesisnotrevealedtothe
participant.Aparticipanthastopredictwhetherthenon-revealedfeaturecomesfromfeaturesofcat-
egoryAorcategoryB.Thereisevidencethatclassificationandinferencelearningarenotequivalent
foradultsandthesetasksmayresultindifferentrepresentationsinadults(seeMarkman&Ross,2003;
Yamauchi&Markman,1998,forextensivearguments).
If classification and inference tasks result in different representations, perhaps adults allocate
attention differently in these tasks. This issue was addressed in the Hoffman and Rehder’s (2010)
study reviewed above. Recall that these authors presented adult participants with a multi-phase
category-learning task and recorded participants’ eye movements during learning. Hoffman and
Rehder(2010)foundevidencethatclassificationlearning,butnotinferencelearning,resultedinopti-
mizedattentioninphase1andlearnedinattentioninphase2.Theauthorsconcludedthereforethat
whereasselectiveattentiontranspiresinsomecategorylearningtasks,itdoesnottranspireinother
categorylearningtasks:incontrasttoclassificationlearnerswhoattendselectively,tryingtoextract
thediagnosticdimension,inferencelearnersattenddiffusely,tryingtolearnmultipledimensionsand
thewaystheyinterrelate.
Thesefindingsalsosuggestthatclassificationandinferencelearningleadtodifferencesinalloca-
tionofattentionandsubsequentlytodifferencesincategoryrepresentation.Inclassificationlearning,
adultsarelikelytoextractthemostdiagnostic(orrule)feature,whereasininferencelearningtheyare
morelikelytoextractwithin-categorysimilarity.Notethat,aswediscussbelow,differencesbetween
classificationandinferencelearningdonottranspireuntil6-to-7yearsofage,withyoungerchildren
exhibitingsimilarperformanceinbothtypesoftasks(Deng&Sloutsky,2013,2015a).Thesefindings
suggestthatyoungchildrenattendsimilarlyinbothtypesoftasksandsubsequentlyformsimilarrep-
resentationsacrossthesetasks.
Therefore, the reviewed evidence suggests that depending on the task and category structure,
adultsattendeitherselectivelyordiffuselyandformrepresentationsthatreflecttheirpatternofatten-
tion.Incontrast,youngchildrentendtodistributeattentionandformsimilarrepresentationsacross
differentconditions.Thesefindingsaretheoreticallyconsequentialforunderstandingofmaturecate-
gorylearningandofdevelopmentalchangesincategorizationandcategorylearning.
1.2.Diffusedattentionandearlycategorizationandcategorylearning
Ifadultsattendselectively,atleastundersomecategorystructureandlearningtaskconditions,itis
reasonabletoask:howdochildrenlearncategoriesundertheseconditions?Thequestionispoten-
tiallyinformativebecausechildrenyoungerthan5yearsofageoftenhavedifficultyfocusingonasin-
glerelevantdimension,whileignoringmultipledistractingdimensions(see,Hanania&Smith,2010;
Plude,Enns,&Brodeur,1994,forreviews).Notethatthesedifficultiestranspirewhennodimension
capturesattentionautomaticallyandtop-downattentionisrequired.Incontrast,whenasinglehighly
salientfeatureordimensioncapturesattentionautomatically,youngchildrenfocusonthisdimension
(Deng&Sloutsky,2012),andtheyoftenhavedifficultyignoringthisfeatureordimension(Napolitano
&Sloutsky,2004;Robinson&Sloutsky,2004;seealsoRobinson&Sloutsky,2007,forsimilartendency
ininfancy.).
There is much evidence supporting the idea of developmental differences in top-down selective
attention,witholderchildrenandadultsbeinggenerallybetterthanyoungerchildrenatselectively
attending to a single dimension. These findings transpire in a variety of tasks, including rule use
(e.g. Frye, Zelazo, & Palfai, 1995), discrimination learning (e.g. Kendler & Kendler, 1962), free
28 W.(Sophia)Deng,V.M.Sloutsky/CognitivePsychology91(2016)24–62
classification(Smith,1989;Smith&Kemler,1977),speededclassification(e.g.Smith&Kemler,1978),
andcategorylearning(Best,Yim,&Sloutsky,2013).
Forexample,inSmithandKemler(1977,seealsoSmith,1989)participantswerepresentedwith
triadsoftwo-dimensionalstimuli.Oneofthestimulimatchedthetargetonasingledimension,but
hadaverydifferentvalueontheseconddimension(e.g.,thetwocouldhavethesamecolor,butdif-
ferentshapes).Atthesametime,theotherstimuluswassimilartothetargetonbothdimension,with
neitherdimensionvaluebeingthesame(e.g.,thetwohadsomewhatsimilarcolorandshape).When
askedtoselecttwooutofthreeitemsthatwouldgotogether,5-year-oldsoptedfortheoverallsim-
ilarity,whereasolderchildrenpreferreddimensionalmatches.Thesefindingssuggestthatyoungchil-
dren tend to distribute attention across multiple dimensions, rather than focusing on a single
dimension.
Morerecently,Bestetal.(2013)presented6–8-month-oldinfantsandadultswithatwo-phasecat-
egorylearningtask,suchthatthedimensionsthatwererelevantinthefirstphasebecameirrelevant
inthesecondphase.Resultsindicatedthatalthoughbothgroupslearnedcategories,theirpatternsof
allocatingattentiondiffered.Adultsoptimizedattentiontocategoryrelevantdimensionsinphase1
andcontinuedattendingtothesedimensionsinphase2(whenthesedimensionswerenolongerrel-
evant).In contrast,infantsallocatedattention toall dimensionsin bothphases. Therefore,whereas
adults attended selectively when learning categories, infants attended diffusely. There is also some
evidence suggesting a lack of attention optimization in category learning of 4–5year-olds
(Robinson,Best,&Sloutsky,2011):evenwhenchildrenexhibitedrobustlearningofarule-basedcat-
egory,theyfailedtoexhibitevidenceofattentionoptimization.
Takentogether,thesestudiessupporttheideathatearlyindevelopmentchildrentendtodistribute
attentionacrossmultipledimensions,unlessasingledimensioncapturesattentionautomatically.This
patternofattentionallocationmayhaveimportantconsequencesforhowpeoplelearnandrepresent
categories and what they remember after learning. Current research examines these consequences
acrossdevelopment,withthegoalofbetterunderstandingofdevelopmentaldifferencesinthemech-
anismofcategorization.Weconsiderthesegeneralconsequencesandmorespecificpredictionsinthe
nexttwosections.
Importantly,diffusedattentionseemstobemorethanjustalimitationinfocusingattentionearly
indevelopment,itcouldbealsoanimportantmechanism,sub-servingearlycategorylearning.Inone
study, Deng and Sloutsky (2015b) presented infants with a category-learning task. While category
learningwasestablishedusingatraditionalnoveltypreferenceprocedure,attentionduringcategory
learningwasexaminedbyrecordingparticipants’eyemovements.Resultsindicatedthatmoresuc-
cessful learning was accompanied by more distributed attention evidenced by a greater numberof
gazeshiftsacrossdifferentfeaturesofpresentedobjects.
1.3.Attentionandcategoryrepresentationacrossdevelopment
Ifyoungchildrenandadultsallocateattentiondifferentlyinthecourseofcategorylearning,they
arealsolikelytoformdifferentrepresentations:youngchildrenshouldrepresentallormostdimen-
sions,whereasadultsshouldrepresentprimarilycategory-relevantdimensions.Ifthisisthecase,then
categorization in adults should be accompanied by different representations across different situa-
tions(i.e.,dependingonasituation,theyshouldrepresentdifferentdiagnosticdimensions),whereas
categorizationinyoungchildrenshouldbeaccompaniedbysimilarrepresentations(i.e.,acrossthesit-
uations, both diagnostic and non-diagnostic dimensions should be represented). In a recently pub-
lished study, Deng and Sloutsky (2015a) trained 4-year-olds, 6-year-olds, and adults with either a
classificationtaskoraninferencetaskandtestedtheircategorizationperformanceandmemoryfor
items.Adultsand6-year-oldsexhibitedanasymmetry:theyreliedonasingledeterministicfeature
andformedrule-basedrepresentationsduringclassificationtraining,butnotduringinferencetrain-
ing.Incontrast,regardlessofthelearningregime,4-year-oldsreliedonmultipleprobabilisticfeatures
andformedsimilarity-basedrepresentations.Thesefindingssuggestthatwhereasolderchildrenand
adultsattendselectivelytoasubsetoffeaturesthatwereparticularlyusefulforagiventask,younger
childrentendedtoattenddiffuselyacrossthetasks.Thesedevelopmentaldifferencesinattentionallo-
cationduringcategorylearningmayhaveimportantconsequencesforwhatisrememberedaboutthe
W.(Sophia)Deng,V.M.Sloutsky/CognitivePsychology91(2016)24–62 29
categoryandwhatisnotandforhowthesememoriesdifferacrossdevelopment.Wediscussthese
consequencesinthenextsection.
1.4.Howtoinfercategoryrepresentation?
Attentionisimportantforcategoryrepresentationbecauseitaffectswhichinformationisencoded
intolong-termmemoryandwhichisnot(Chun,Golomb,&Turk-Browne,2011).Therefore,analysesof
memorydatafollowinglearningcanprovideusefulinformationaboutattentioninthecourseoflearn-
ingandaboutcategoryrepresentation.Specifically,ifadimensionisrememberedafterlearning,itwas
likelytobeattendedtointhecourseoflearning.Inaddition,ifadimensionisnotremembered(or
rememberedpoorly)itisunlikelytobepartofcategoryrepresentation.
Furthermore,categorizationandgeneralizationdata(i.e., howparticipantscategorizeitemscon-
sisting of old and new features) may provide further evidence about representations being formed
inthecourseoflearning.Forexample,ifagivenfeatureisnotusedincategorizationandgeneraliza-
tion,itisunlikelytobeapartofcategoryrepresentation,whereasifthefeatureisusedincategoriza-
tionandgeneralization,itislikelytobeapartofcategoryrepresentation.
Inadditiontothese relativelystraightforward cases,itisalsopossiblethatfeaturesthatare not
used in categorization and/or generalization are still remembered. Such cases may highlight differ-
encesbetweenrepresentationanddecisioncomponentsincategorization.Forexample,learningacat-
egory of red objects versus blue objects may be achieved by color being encoded (because it is an
attended dimension), and shape and texture not being encoded (because these are ignored dimen-
sions).Alternatively,itispossiblethatalldimensionsare encoded,butcategorizationdecisionsare
madeonlyonthebasisofthediagnosticdimension.Therefore,caseswhenfeaturesarenotusedin
categorizationand/or generalization but are still remembered willrequire a more detailed analysis
andweprovidetheseanalyseswhensuchcasesoccur.Ifadultsattendselectively,theyshouldbetter
encodeandrememberthefeaturesthatcontroltheircategorization.Incontrast,youngchildren(i.e.,
thosewhoareyoungerthan6yearsofage)attenddiffuselyandtheyshouldrememberwellallthe
features.
1.5.Currentstudy
The reviewed above theoretical considerations and evidence suggest a number of important
hypotheses. First, adults, whose attention allocation may differ across tasks and conditions, should
optimizeattentioninsomeconditionsanddistributeattentioninotherconditions.Incontrast,across
conditions,youngchildrenshouldattenddiffusely.Second,dependingonattentionallocation,adults
should extract different features and form different representations. In contrast, across conditions,
youngchildrenshouldextractmultiplefeaturesandformequivalentrepresentations.Andthird,these
differences in attention allocation should transpire in what is remembered: while young children
shouldrememberallormostfeatureswell,adultsshouldhavebettermemoryforfeaturesthatdeter-
minetheircategorizationdecisions.
Thegoalofthecurrentstudywastotestthesehypotheses,thusadvancingourunderstandingof
the development of categorization and the role of selective attention in this process. The reported
studyconsistedofthreeexperimentsconductedwith4-year-olds,7-year-olds,andadults.Thebasic
taskforeachexperimentconsistedofthreephases:instructions,training,andtesting.Asexplained
below, all attentional manipulations were introduced during the instructions and training phases,
whereasthetestingphasewasidenticalacrosstheexperiments.
Duringtraining,participantspredictedthecategoryofagivenitemandreceivedcorrectivefeed-
back.Thereweretwofamily-resemblancecategories,witheachtrainingitemhavingasingledeter-
ministic feature D (which perfectly distinguished the two categories) and multiple probabilistic
featuresP(witheachprovidingimperfectprobabilisticinformationaboutcategorymembership).
Thetestingphaseconsistedofcategorizationandrecognitiontasksandwasadministeredimmedi-
atelyafterthetrainingphase.Duringtestingparticipantswereaskedtodetermine(1)whichcategory
eachitemwasmorelikelytobelongtoand(2)whethereachitemwasoldornew.Nofeedbackwas
providedduringtesting.Categorizationtrialsweredesignedtodeterminewhichfeaturesparticipants
30 W.(Sophia)Deng,V.M.Sloutsky/CognitivePsychology91(2016)24–62
relyonin theirdecisions,whereasrecognitiontrialsweredesignedto determinewhatparticipants
remembered from training, which provides information about how they allocate attention during
trainingandhowtheyrepresentthelearnedcategories.
Threeexperimentsdiffered inhowparticipants’attentionwasdirected todifferenttypesof fea-
tures.InExperiment1,informationaboutPandDfeatureswasexplicitlymentionedtoparticipants.
ThegoalofExperiment1wastoreplicateandfurtherextendDeng andSloutsky’s(2015a)findings
withchildrenandadults,withthegoalofestablishingabaselineforExperiments2and3.
Basedontheconsiderationsreviewedabove,itwaspredictedthatbecauseyoungchildrendonot
optimize attention, their categorization performance and recognition memory should differ from
adults’.Specifically,youngchildrenshouldrelyonmultiplePfeaturesratherthantheDfeatureincat-
egorizationandrememberwellallormostfeatures.Incontrast,adults,whoablyoptimizeattentionin
categorylearning,shouldrelyontheDfeatureandrememberDfeaturebetterthanPfeatures.Perfor-
manceof7-year-olds(incomparisonwiththeothertwoagegroups)willhelpbetterunderstandthe
developmentofcategorization.
InExperiments2and3,wecuedparticipants’attention,withthegoalofexaminingwhetheratten-
tionalcueingresultsinchangesincategorizationperformance(comparedtothepatternestablishedin
Experiment 1) and in changes in underlying representations. Specifically, we directed participants’
attentiontotheDfeatureinExperiment2andtothePfeaturesinExperiment3.Ifweobservechanges
inbothcategorizationperformanceandmemoryforfeatures,thendifferentwaysofcategorizingare
drivenbydifferentunderlyingrepresentations.
Incontrast,ifweobservechangesonlyincategorizationperformance,butnotinmemoryforfea-
tures,thendifferentwaysofcategorizingaredrivenbydifferentdecisionweightsofdifferentfeatures
indifferentsituations,whereasunderlyingrepresentationsarelikelytoremainthesameacrossthe
situations.AswediscussinSection6,thisinformationisconsequentialforunderstandingthedevel-
opmentofcategorization,andfortheoriesandmodelsofcategorization.
2.Experiment1:establishingabaseline
2.1.Method
2.1.1.Participants
Participants were adults (15 women), 7-year-old3 children (M =83.2months, range 73.2–
age
89.4months; 10 girls), and 4-year-old children (M =54.1months, range 48.3–59.6months; 7 girls),
age
with20participantsperagegroup.AdultparticipantswereTheOhioStateUniversityundergraduatestu-
dentsparticipatingforcoursecreditandtheyweretestedinaquietroominthelaboratoryoncampus.
Childparticipantswererecruitedfromchildcarecentersandpreschools,locatedinmiddle-classsuburbs
ofColumbusandweretestedbyanexperimenterinaquietroomintheirpreschool.Datafromtwoaddi-
tional adults, one additional 7-year-old, and one additional 4-year-old were excluded from analyses
becauseofextremelypoorperformanceintraining(theircategorizationperformancewastwostandard
deviationsbelowthemeanofaccuracyinthelasttentrainingtrials).Datafromoneadditional4-year-old
werealsoexcludedfromanalysesbecauseoftheexperimentbeingdisrupted.
2.1.2.Materials
MaterialsweresimilartothoseusedpreviouslybyDengandSloutsky(2015a)andconsistedofcol-
orfuldrawingsofartificialcreatures.These creatureswereaccompaniedbythenovellabels‘‘flurp‘‘
(CategoryF)and‘‘jalet”(CategoryJ).Thesecategorieshadtwoprototypes(F0andJ0,respectively)that
weredistinctinthecolorandshapeofsevenoftheirfeatures:head,body,hands,feet,antennae,tail,
andabodymark(seeFig.1).
AsshowninTable1,mostofthefeatureswereprobabilisticandtheyjointlyreflectedtheoverall
similarity among the exemplars (we refer to them as the ‘‘P features” or as ‘‘overall appearance”),
3 Inthisandallotherexperimentsreportedhere,theolderchildparticipantsconsistedof6-year-oldsand7-year-olds,withthe
meanagesbeingjustunderseven.Forpurposesofbrevity,werefertothisgroupas‘‘7-year-olds”.
W.(Sophia)Deng,V.M.Sloutsky/CognitivePsychology91(2016)24–62 31
Category Prototype High-Match Switch new-D one-new-P all-new-P
F0 PflurpDflurp PjaletDflurp PflurpDnew PnewDflurp Pall-newDflurp
Flurp
J0 PjaletDjalet PflurpDjalet PjaletDnew PnewDjalet Pall-newDjalet
Jalet
Fig.1. Examplesofstimuliusedinthisstudy.Eachrowdepictsitemswithinacategory,whereaseachcolumnidentifiedan
itemrole(e.g.,switch)anditemtype(e.g.,PjaletDflurp).TheHigh-Matchitemswereusedintrainingandtesting.Theswitch,new-
D,one-new-P,andall-new-Pitemswereusedonlyintesting.Neitherprototypewasshownintrainingortesting.
Table1
CategorystructureusedinExperiments1–3.
Itemtype Stimulus Probabilisticfeature Deterministicfeature
Head Body Hands Feet Antenna Tail Button
High-Match PflurpDflurp 1 1 1 1 0 0 1
PjaletDjalet 0 0 0 0 1 1 0
Switch PjaletDflurp 0 1 0 1 0 0 1
PflurpDjalet 1 0 1 0 1 1 0
New-D PflurpDnew 1 0 1 0 1 1 N
PjaletDnew 0 1 0 1 0 0 N
One-new-P PnewDflurp 1 1 0 N 1 1 1
PnewDjalet 0 0 1 0 N 0 0
All-new-P Pall-newDflurp N N N N N N 1
Pall-newDjalet N N N N N N 0
Note.Thevalue1=anyofsevendimensionsidenticaltotheprototypeofCategoryF(flurp,seeFig.1).Thevalue0=anyofseven
dimensionsidenticaltotheprototypeofCategoryJ(jalet,seeFig.1).ThevalueN=newfeaturewhichisnotpresentedduring
training.P=probabilisticfeature;D=deterministicfeature.Thistablepresentsanexampleofstimulusstructureforeachitem
type.SeeAppendixAforfulltablesofcategorystructureofallthevariants.VariantsofHigh-Matchitemswereusedinboth
trainingandtesting.Variantsofallotheritemtypeswereusedonlyintesting.
whereasonefeaturewasdeterministicanditperfectlydistinguishedthetwocategories(werefertoit
as‘‘Dfeature”orasa‘‘category-inclusionrule”).Thebodymark(introducedasa‘‘bodybutton”)was
thedeterministicfeature:allmembersofCategoryFhadaraindrop-shapedbuttonwiththevalueof1,
whereasallmembersofCategoryJhadacross-shapedbuttonwiththevalueof0.Alltheotherfea-
tures–thehead,body,hands,feet,antennae,andtail–variedwithineachcategory,thusconstituting
theprobabilisticfeatures.
Asshown in Table 1, some of the itemswere used in training and some in testing. The training
stimuliconsistedofHigh-Matchitems(i.e.,P D andP D ).Theseitemshadthedetermin-
flurp flurp jalet jalet
isticfeature(D)andfourprobabilisticfeatures(P)consistentwithagivenprototype;twootherprob-
abilisticfeatureswereconsistentwiththeoppositeprototype.
The testing stimuli consisted of High-Matchitems presented during training (i.e., P D and
flurp flurp
P D ) and four additional types of items. These included: (1) Switch items (i.e., P D and
jalet jalet jalet flurp
P D ),whichhadthedeterministicfeatureofastudiedcategorybutmostprobabilisticfeatures
flurp jalet
32 W.(Sophia)Deng,V.M.Sloutsky/CognitivePsychology91(2016)24–62
consistent with the opposite prototype; (2) new-D items (i.e., P D and P D ), which had
flurp new jalet new
probabilistic features of a studied category and a novel feature replacingthe deterministic feature;
(3)one-new-Pitems(i.e.,P D andP D ),whichhadallfeaturesofastudiedcategorybut
new flurp new jalet
a novel feature replacing one probabilistic feature; and (4) all-new-P items (i.e., P D and
all-new flurp
P D ), which had the deterministic features from a studied category and all new features
all-new jalet
replacingthestudiedprobabilisticfeatures.
TheHigh-Matchitemswereusedtoexaminehowwelltheparticipantslearnedthecategoriesand
toassesstheirrecognitionaccuracyontheolditems.TheSwitchitemshadmostofthePfeaturesfrom
onecategoryandtheDfeaturefromanother,thusallowingdeterminingwhetherparticipantsintheir
categorizationdecisionsreliedontheoverallsimilarity(i.e.Pfeatures)or onthedeterministicrule
(i.e.,Dfeature).Thenew-Ditemswereusedtoassesswhetherparticipantscouldrelyintheircatego-
rizationonoldPfeatures whentheoldDfeaturewas notavailable.These itemswerealsousedto
examinewhetherparticipantsencodedthecategory-inclusionrule,inwhichcasetheyshouldreject
theseitemsasnewduringthememorytest.Theone-new-Pitemswereusedtoassesswhetherpar-
ticipantscouldrelyintheircategorizationontheoldDfeatureandremainingPfeatureswhenasingle
oldPfeaturewasnotavailable.Theseitemswerealsousedtoexaminewhetherparticipantsencoded
allindividualPfeatures,inwhichcasetheyshouldrejecttheseitemsasnewduringmemorytest.And
finally,theall-new-Pitemswereusedtoassesswhetherparticipantscouldrelyintheircategorization
ontheoldDfeaturewhennoneoftheoldPfeatureswasavailable.Inaddition,theseitemswereused
toassessparticipants’overallmemoryaccuracyforprobabilisticfeatures:iftheyencodedatleastone
suchfeature,theyshouldrejecttheseitemsasnew.Table1presentsexampleofcategorystructure
withPandDbeingcombinedtocreatefivetypesofstimuli,andFig.1showsexamplesofeachkind
ofstimulus.
2.1.3.Designandprocedure
The experiment consisted of instructions,training, and testing (see Fig. 2). The procedureswere
similarforbothadultsandchildren,exceptthewaytheinstructionswerepresented,thequestions
wereasked,andtheresponseswererecorded.Adultsreadtheinstructionsandquestionsonthecom-
puter screen and pressed the keyboard to make responses, whereas for children, a trained experi-
menter presented instructions and the questions verbally and recorded children’s responses by
pressing the keyboard. The experiment took approximately 10min for adults and approximately
15minfor children.Mostchildrenandadultsfinishedtheexperimentand, asevidencedfromchil-
dren’s high recognition accuracy (see below), their response patterns did not stem from confusion
orfatigue.
Instructionsandtraining.Beforetraining,participantswerepresentedwithacoverstoryabouttwo
categoriesofcreaturesfromotherplanets.InformationaboutPandDfeatureswasexplicitlygivento
the participants (see Fig.2A). They were told that all flurps (or jalets) had a raindrop-shaped (or a
cross-shaped) button and most of the flurps’ (or jalets’) features (at this point, the deterministic
andprobabilisticfeatureswerepresented,oneatatime).Thisinformationwasrepeatedinthecorrec-
tivefeedbackoneachtrialduringtrainingusingthefollowingscript:Thisonelookslikeaflurp(ora
jalet)andithasflurp’s(orjalet’s)button.Afterinstructions,participantsweregiven30trainingtrials
(15 trials per category). On each trial, participants predicted the category label of a stimulus given
information about all other features and each trial was accompanied by corrective feedback (see
Fig.2B). The order of the training trials was randomized across participants. Testing was not men-
tioneduntiltheendofthetrainingphase.
Testing.Thetestingphasewasadministeredimmediatelyaftertrainingandincludedcategorization
andrecognitiontasks.Duringthetestingphase,adultsandchildrenwerepresentedwith40testtrials
(8trialsperitem-type;withequalnumbercomingfromeachofthetwocategories)andwereaskedto
determine(1)whichcategoryeachcreaturewasmorelikelytobelongtoand(2)whethereachcrea-
ture was old (i.e., exactly the one presented during the training phase) or new (see Fig.2C). As we
explain in the results section, the ways participants categorize and remember different item types
providecriticalinformationabouthowtheyrepresentedthecategories.
Eachtrialincludedacategorizationandrecognitionquestionandtheorderofthequestionswas
counterbalancedbetweenparticipantsandtheorderofthe40testitemswasrandomizedacrosspar-
W.(Sophia)Deng,V.M.Sloutsky/CognitivePsychology91(2016)24–62 33
Fig.2. SchematicrepresentationoftheprocedureinExperiments1–3.
ticipants.Allrecognitionquestionsreferredtothe‘‘thefirstpartofthegame”(i.e.,thetrainingphase),
withparticipantsbeingaskedwhetheraniteminquestionwaspresentedduringthefirstpartofthe
gameorwasanewitem.Nofeedbackwasprovidedduringtesting.
Forcategorizationtesting,theprimaryanalysesfocusedontheproportionofresponsesinaccor-
dancewiththeDfeature(i.e.,rule-basedresponses).Forrecognitionmemory,theprimaryanalyses
focused on the difference between the proportion of hits (i.e., correctly identifying the High-Match
itemsthatwerepresentedduringtrainingas‘‘old”)andfalsealarms(i.e.,incorrectlyidentifyingother
itemtypesthatwerenotpresentedduringtrainingas‘‘old”).
Iftherearedevelopmentaldifferencesinattentionallocation,thesedifferencesshouldtranspirein
bothcategorizationandrecognitionperformance.Inparticular,ifparticipantsselectivelyattendtothe
mostdiagnosticfeaturedistinguishingtwocategories,theyshouldrelyontheDfeatureincategoriza-
tionandremembertheDfeaturebetterthanPfeatures.However,ifparticipantsattenddiffuselyto