Experience-Dependent Perceptual Grouping and Object-Based Attention

Experience-DependentPerceptualGroupingand

Object-BasedAttention

RichardS.Zemel

UniversityofArizonaTucson,AZ85721zemel@u.arizona.edu

MarleneBehrmann

CarnegieMellonUniversity

Pittsburgh,PA15213behrmann+@cmu.edu

MichaelC.Mozer

UniversityofColoradoBoulder,CO80309-0430mozer@cs.colorado.edu

DaphneBavelier

UniversityofRochesterRochester,NY14627

daphne@bcs.rochester.edu

June10,1999

Abstract

Numerousstudieshaveshownthatattentioncanbeallocatedtoobjectsaswellaslocationsinthevisualfieldeveniftheobjectsarepartiallyoccluded.Afundamentalquestionconcernsthenatureofthe‘objects’forwhichthisattentionalbenefitapplies.CurrentstudieshaveshownthatobjectscanbedefinedonthebasisofGestaltgroupingprinciplesaswellasonthebasisoffamiliarity.Boththeeffectsofgroupingaswellasfamiliaritycanbeunderstoodintermsofamoregeneralhypothesis:thatperceptualexperiencewithparticularfeaturecombinationsdetermineswhetherornottwofeatureswillbeintegratedasanobjectofattention.Wepresentdatafromfourstudiesshowingthatrecentlyexperiencednovelfeaturecombinationsgaintheobjectattentionalbenefitandthatthiseffectisrealizedbydifferentfeaturecombinationsunderarangeofexperimentalconditions.Thesestudiesindicatethatobjectattentionisadaptiveandresponsivetothestatisticalstructureoftheenvironment.

Acknowledgements:WethankSusanWatt,JimNelson,andJamesChristopherforhelpwithrunningsub-jectsandanalysis.SupportedbytheMcDonnellFoundationgrantJSMF95-1(RSZ),NIMHProgramProjectgrantMH47566-06(MBandRSZ),NSFawardIBN-9873492(MCM),andaMcDonnellFoundationgrant(DB).

1

AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier2

Introduction

Attentioncanbedirectedtowardobjectsaswellastowardlocationsinthevisualfield,therebyaffordingpreferentialprocessingforthefeaturesofaspecificobject.Evidenceforthisfindingcomesfromstudieswhichshowthatitisdifficulttoattendtotwoobjectssimultaneously.Forexample,whenjudgementsdependontwofeaturesinadisplay,responsesaremorerapidwhenbothfeaturesbelongtothesameobject,evenwhentheobjectsarespatiallysuperimposed(Dun-can,1984;Baylis&Driver,1993;Kramer&Watson,1995).Asecondsourceofevidencearestudiesshowingthatpeoplefinditdifficulttoignorefeaturesthatbelongtoanattendedobject(Kramer&Jacobson,1991;Baylis&Driver,1992;Yantis,1992).

Afundamentalquestionconcernsthenatureofthe‘objects’forwhichthisattentionalbenefitap-plies.Inmostexperimentsdemonstratingobject-basedeffects,groupingprinciplessuchascontin-uation(Moore,Yantis,&Vaughan,1998),collinearity(Lavie&Driver,1996),similarity(Kramer&Jacobson,1991),color(Baylis&Driver,1993)ormotion(Driver&Baylis,1989;Behrmann,Zemel,&Mozer,1999)aresufficienttodefinetheobjectsinthedisplay.VeceraandFarah(1997)havealsoshownthatobject-basedattentionisstrongerforhighlyfamiliarshapes(uprightletters)thanforunfamiliarshapesthatbenefitfromthesamegroupingprinciples(upside-downletters).

Currentstudieshavenotestablishedwhetherthesetwomeansof‘defining’objects—basedongenericgroupingprinciplesorlong-termfamiliarity—aresufficienttopredictwhentwofeatureswillbetreatedasbelongingtothesameobject.Ahypothesisthatprovidesamoregeneralview-point,whichaccountsforbothoftheseotherdefinitions,isthatperceptualexperiencewithpartic-ularfeaturecombinationsdetermineswhetherornottwofeatureswillbeintegratedasanobjectofattention.Wehavepreviouslydevelopedacomputationalmodel,namedMAGIC,basedonthishypothesis(Mozer,Zemel,Behrmann,&Williams,1992).MAGICwastrainedtogroupfeaturesfromasetofimagescontainingmultipleobjectsinwhicheachelementaryfeaturewaslabeledastowhichobjectitbelonged.Aftertraining,MAGICsuccessfullysegregatedfeaturesofnovelimagesintoseparateobjects.ExaminationoftherepresentationsderivedbyMAGICrevealedthatthecriticalaspectsweretheconfigurationsofimagefeatureswithaconsistentlabelingrelativetooneanother.Forexample,themodeldiscoveredthattheelementswithinoneofthetwoper-pendicularsegmentsofaT-junctionwereconsistentlylabeledasbelongingtothesameobject,whileelementsacrosssegmentsweresegregatedintodifferentobjects.Inthisway,MAGICem-bodiesthehypothesisthatperceptualexperiencedefineswhichfeatureswillbegroupedtogetherandwhichfeatureswillnot.Underthishypothesis,genericgroupingprinciplesemergebasedoncompiledexperiencewithavarietyoffeatureandobjectcombinationsinimages.

Inthispaper,weinvestigatetheroleofperceptualexperienceinobject-basedattention,examiningquestionssuchaswhethershort-termexperiencewithashapeissufficienttofacilitateitsbeingprocessedasaunitarywhole;andtheextenttowhichthisexperiencewithashapemayoverrideothercuesastowhethertwofeaturesbelongtoacommonobject.

Inordertoexploretheseissues,weutilizeaparadigmforstudyingobject-basedattentiondevel-opedinpreviouswork(Behrmann,Zemel,&Mozer,1998).Intheseexperiments,subjectsdecidedwhetherthenumberofbumpsappearingattwooffourpossibleendsoftwooverlappingobjects(orbars;seeFigure1)werethesame(Figure1a-c)ordifferent(Figure1d-f).Thetwofeatures(setsofbumps)couldappearontheendsofasingleobject(Figure1aand1d)orontheendsoftwo

AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier3

differentobjects(Figure1band1e).Consistentwiththeobject-costhypothesisthatitisdifficulttoattendtotwoobjectssimultaneously,subjects’responsesweresignificantlyslowertotwofeaturesoftwodifferentobjectsthantotwofeaturesofasingleobject(cf.Duncan,1984).Theseobjectcosts—significantRTdifferencesforrespondingtofeaturesofdifferentobjectsversusfeaturesofasingleobject—provideanassaytodeterminewhenfeaturesaregroupedintoasingleobject.In-structionstothesubjectscarefullyomittedanymentionofobjects,makingthisprobeparticularlyusefulbecauseitdoesnotinvolveanysubjectivedefinitionofobject-hood.

Inourearlierexperiments,wealsoincludedathirdtypeofdisplayinwhichthebumpswereontheoccludedobject(Figure1cand1f)and,again,evaluatedwhethertherewasanycostrelativetothesingleobjecttrials.Occlusionisaparticularlychallengingconditionforanobject-basedaccountofselection—notonlyaretheelementsofasingleoccludedobjectspatiallydistantbuttheyarealsodiscontinuous(seeYantis,1995;Moore,Yantis,&Vaughan,1998forotherstudiesofocclusioneffectsonobject-basedattention).Interestingly,however,therewasnoobjectcostfortheoccludedtrials,reflectedinequivalentRTsforthesinglenonoccludedandthesingleoc-cludeddisplays,bothofwhichdifferedfromthetwoobjecttrials.Theseresultsheldupbothinthe“X-displays”inwhichthebarscrossedtoforman“X”,andalsoin“V-displays”inwhichthesetsofbumpswereallat90degreesfromeachother(see,e.g.,Figure1g-i).Theevidencefromthesestudiessuggeststhatelementsofasingleobject,evenifoccluded,aregroupedtogetherandpreferentiallyprocessedrelativetoelementsofotherobjectsinthescene.

Ourexperimentsestablishedthatfeaturesofanoccludedobjectenjoythesameprocessingad-vantageasfeaturesofanunoccludedobjectrelativetofeaturesoftwodifferentobjects:removingexplicitcontinuityasanobject-definingcuedidnotaffectobject-basedattention.Anadditionalexperimentalsoestablishedaboundaryconditionofthisresult.Whenwechangedtherelationbetweenthetwodiscontinuousfragmentsoftheoccludedobjectsothattheynolongerformedaplausiblesinglebar(Figure2b),theobjectcostreappearedandperformancewasnolongerequiv-alenttothatofasinglenonoccludedobject.

Aprimaryaimoftheexperimentsinthispaperistodeterminewhatdefinestheconditionsun-derwhichattentionalprocessestreatfragmentsasbelongingtothesameobjectordifferentob-jects.Oneissueconcernswhythenon-alignedfragmentsinFigure2barenottreatedasasin-gleoccludedobject.Oneclassoftheoriesproposesgeneral-purposeprocessesbywhichimagefragmentsareintegratedintoobjects.Forexample,thisresultisconsistentwithatheorythattheparticulargeometricrelationsbetweenfragmentsdetermineswhethertheywillformobjects(Kellman&Shipley,1991).Underthistheoryofrelatability,spatiallyseparatedfragmentsarein-terpolatedwhentheiredgescanbeconnectedbyasmoothmonotoniccurve;whentheedgesarenolongercollinear,thefragmentsarenotrelatableanddonotbelongtoasingleoccludedobject.Adifferenthypothesis,consistentwithMAGIC,isthatthesegeneral-purposemechanismscouldemergefromperceptualexperience.Onthisview,experienceplaysadeterminingroleinpercep-tualorganization.Acorollaryofthisviewisthatshort-termperceptualexperiencemayoverridethegeneral-purpose,compiledgroupingmechanisms.Whenshort-termexperienceisconsistentwithlongertermregularitiesthanheuristicssuchasrelatabilitywillapply,butinothercircum-stancestheywillnot.Considerthesituationinwhichsubjectsareexposedtoashapethatcouldpotentiallylinktogetherthetwonon-alignedfragmentsoftheoccludedobjectintoaplausibleobject(seeFigure2c).Experience-dependentgroupingwouldthenpredictthatevenifthisnovelshapeisratherconvolutedandirregularlyshaped,theobjectadvantagewillapplyevenwhen

AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier4

(a)(b)(c)

(d)(e)(f)

(g)(h)(i)

Figure1:ExamplesofstimuliusedinBehrmannetal.(1998)andhere.Subjectshadtomakesame/differentjudgementsbasedonthenumberofbumpsattwodifferentlocationsineachfigure.Thetopandthirdrowrepresent‘same’judgements,whilethemiddlerowdisplaysare‘different’.Theleft-mostcolumndepictsasingleoccludercondition,inwhichthebumpsareonone,occludingobject,themiddlecolumnshowsthetwoobjectcondition,andtherightcolumnshowsthesingleoccludedcondition.ThefirsttworowsareexamplesofXdisplays,containingtwooverlappingbars,whilethethirdrowshowsexamplesofVdisplays,containingoverlappingVshapes.

AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier5

(a)(b)(c)

Figure2:(a)Reactiontimesondisplayswherethebumpsappearonthetwofragmentsthatcorre-spondtotheendsofanoccludedbarareequivalenttothoseonunoccludedbars,andsignificantlyfasterthanwhenthebumpsappearontwodifferentbars.(b)Whenthefragmentsareshiftedsotheynolongerformaplausibleoccludedbar,theRTsareequivalenttothetwobardisplays.(c)Ifperceptualexperienceplaysadeterminingroleinparsing,thensubjectsexposedtothisshapemaygroupthefragmentsin(b)intoasingleobject.

AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier6

eitherhadtorespondtothefullyvisibleobject,ortothefragmentsthatwouldtypicallynotbegroupedasbelongingtoasingleobject.Inthesecondblockoftrials,Experimentalsubjectssawdisplayscontaininganobjectthatlinkedthesefragmentsupaspartsofasingleobject(Figure3c),whileControlsubjectssawthefragmentsalonewithoutthefullyvisibleobject.

Thekeypredictionconcernsthethirdblockoftrials,inwhichbothgroupsofsubjectsagainsawdisplayslikeFigure2a,b.ThepredictionisthatbecausetheControlgroupsubjects’experiencewillsupportaninterpretationofthisdisplayasthreeseparateobjects(thetwofragmentsandthecentralbar),sothefragmentswillhaveanobjectcost,justastheydidintheinitialblock.Ontheotherhand,thesubjectswhosawthelinkingobjectwillinterpretthetwofragmentsasbelongingtoasingleoccludedshape,asevidencedbytheirrelativelyspeededresponsestothosefragments.

Method

Participants.Atotalofthirty-twosubjectsparticipatedinthisexperiment.Thedataoftwosubjectswereexcludedfromtheanalysisbecauseofhigherrorrates(10%).SubjectsweredrawnfromtheCarnegieMellonUniversitycommunityandwerepaid$5fortheirparticipation.Subjectsrangedinagefrom18to24years.Allhadnormalorcorrectedtonormalvisualacuity.Noneofthesubjectswasawareofthepurposeofthisstudy.

Apparatusandmaterials.ThisexperimentwasconductedonaMacintoshIIcicomputer.Stim-uliwerepresentedona14-inchcolormonitor(Basiccolormonitor:modelM1595LL/A)usingPsychlabexperimentalsoftwareversion1.0(Bub&Gum,1991).Thedisplayswerepresentedasblack-and-whitelinedrawingsonawhitebackground.Viewingdistancewasapproximately50cm.

Therewerefourtypesofdisplays(seeFigure3):

1.(Figure3a,b)Ambiguous.Thisdisplaycouldeitherbeinterpretedasarectangularbaroc-cludingaZ-shapedobject,orasarectangularbarwithtwosmallerrectangularendsbuttedupagainstit.Therectangularbarwas8.7cminlength(10.2)and2.5cminwidth(2.9).Thetwoendswerecreatedbytakingthetwovisiblefragmentsofanorthogonaloccludedbarofthesamedimensionsanddisplacingthembyslightlymorethanthewidthoftherect-angle(3.3).Thisdisplayappearedequallyofteninfourdifferentorientations,asshowninFigure4.Furthermore,thedisplaysfellintotwoconditions,basedonthelocationsofthefeatures(bumps):

(a)Bar-Bumps:Thebumpsappearedontheoppositeendofthesinglecoherentbar.(b)Fragment-Bumps:Thebumpsappearedonthetwofragments.

2.(Figure3c)Bar.ThisdisplaywascreatedbyremovingallbutthebarfromtheAmbiguousdisplays.TheBarappearedintwodifferentorientations,andonlytheBar-Bumpsconditionwasrelevanttothisdisplay.3.(Figure3d)Z.ThisdisplaywascreatedbyremovingthebarfromtheAmbiguousdisplaysandreplacingitwithcontoursconnectingthetworemainingfragments.ThisdisplayalsoappearedequallyofteninthefourdifferentorientationsequivalenttothoseshowninFig-ure4.OnlytheFragment-Bumpsconditionwaspossibleinthisdisplay.

AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier7

Block1Block3

AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier8

Figure4:ExamplesofthefourorientationsofAmbiguousdisplaysusedinExperiment1.

AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier9

Theexperimentcontainedonebetween-subjectsvariable(ExperimentalorControlgroup)andtwowithin-subjectsvariables:block(1and3);andfeaturelocation(Fragment-BumpsandBar-Bumps).Weexpectedthatbothgroupswouldimproveintheiroverallreactiontimesinblock3comparedwithblock1duetoageneralpracticeeffect.Moreimportantly,wepredictedthatiflearningismediatedbyexposuretoalinkingobject,wewouldfinda3-wayinteractioninthedata:theExperimentalgroupwouldprocesstheFragment-BumpsoftheAmbiguousdisplaysmuchfasterthantheControlgroup,butonlyinblock3andnotinblock1.Thisimpliesintermsofobjectcosts—RTdifferencesforfeaturesofdifferentobjectsversusfeaturesofasingleobject—thatthecostsinblock1fortheAmbiguousFragment-Bumpstrialsdisappearinblock3fortheExperimentalgroup,butnotfortheControlgroup.Thusthecriticalcomparisonhereisbetweenthetwogroupsofsubjectswithineachofthetwoconditions.

Procedure.Eachtrialproceededasfollows:Afixationpointappearedfor1second(sec)followedbya500millisecond(msec)delay.Thereafter,thedisplayappearedandremainedonthescreenuntilaresponsewasmade.Aninter-trialintervalof1secoccurredfollowingtheresponseandpriortothenexttrial.Thesameprocedurewasfollowedinallexperimentspresentedinthispaper.Treatmentofresults.Thedatafromthepracticetrialswerediscardedfromtheanalysis.Errortrialswereexcludedfromthereactiontime(RT)analysis.ThemeanRTanderrorsforeachcrossingofgroup,featurelocation,andblockwerecalculatedforeachsubjectandwerethensubjecttoanalysesofvariance.Reactiontimesthatexceededtwostandarddeviationsaboveorbelowasubject’smean(beforetheremovalofthedata)werealsoexcludedfromtheanalysis.

ResultsandDiscussion

Theoverallerrorrateforthisexperimentwaslow,comprising2.5%ofthetotalnumberoftrials.Afurther3.3%ofthedatawasexcludedasexceedingthetwostandarddeviationscutoff.Separateanalysesofvariancewereconductedcrossingjudgementandorientationwiththemainfactorsintheexperiment:block,featurelocation,andgroup.Asistypicallythecaseinthesame-differentparadigm(Nickerson,1965),samejudgementswerefoundtobesignificantlyfasterthandifferentjudgements().However,nosignificantinteractionwasfoundbetweenjudgementandtheothervariables.TherewasalsonodifferenceinRTpatternsasafunctionoforientation().Consequently,thedatawerepooledacrossorientationandjudgementforsubsequentanalyses.

ThecriticalANOVAincludedonebetween-subjectsvariable(ExperimentalorControlgroup)andtwowithin-subjectsvariables(block1and3;Fragment-BumpsandBar-Bumps).ThisANOVAwasconductedfirstwitherrorandthenwithmeanRTasthedependentmeasure.

Figure5showsthemeanRTsandstandarderrorsforthesethreevariables(group,block,andcondition).MoreerrorswereproducedontheFragment-Bumpsinblock1thanonanyothercondition,buttheerroranalysisdidnotrevealanysignificanteffectsofthesethreevariables.)andfeaturelocationTheRTdatacontainedasignificanteffectforblock(

().Themaineffectofblockcorrespondstoanoverallpracticeeffectastheexperimentunfolds,whilethemaineffectoffeaturelocationcorrespondstotheoverallobject

AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier10

Block 1820800780760740720700Block 3Fragment−BumpsBar−BumpsRT (msec)ControlExperimentalControlGROUPExperimentalFigure5:Meanreactiontimesandstandarderrorbarsinblocks1and3forthetwosubjectgroupsasafunctionoffeaturelocation(ontheBaroronthetwoFragments)fortheAmbiguousdisplaysinExperiment1.AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier11

location)inblock3thatwaspresentinblock1(38msvs2ms).

Theeffectofexperiencecanbeclearlyseenbyanalyzingthedataacrossblocks.Foreachgroup,thedifferenceintheBar-BumpsRTsbetweenblocks1and3describesabaselinepracticeeffect.FortheControlgroup,thismeandifferencewas21msec,whilefortheExperimentalgroupitwas27msec.

Ifthesecondblockoftrialsdoesnotdifferentiallyaffectthetwofeature-locationconditions,thenonewouldpredictthatthespeed-upsfortheFragment-Bumpdisplayswouldbeapproximatelythesameasthesevalues.ThisisthecasefortheControlgroup,wherethedifferenceinFragment-BumpRTsbetweenblocks1and3was15msec.However,theFragment-Bumpsspeed-upfortheExperimentalgroupwassignificantlygreater:63msec.

Twomainconclusionsmaybedrawnfromtheseresults.First,theresultsreplicatethefindinginBehrmannetal.(1998)thatdisplacedfragmentsarenottreatedasanoccludedobjectintheAmbiguousdisplays.Thisismanifestedinthesignificanteffectoffeaturelocationinblock1,wheresubjectsarefasteronBar-BumpsthanFragment-Bumps,suggestingthatthefragmentsarenotbeingperceivedasasingleobject.Thisfindingisconsistentwiththeprincipleofrelatability(Kellman&Shipley,1992),whichpredictsthatthecontoursofthetwofragmentswillnotbeinterpolatedbecauseofthemisalignedgeometricrelationshipbetweenthem.

Thesecondconclusionisthemoreinterestingone:exposuretoanovelobjectchangedtheprocess-ingoftheambiguousvisualinput.Thespeed-upfromblocks1to3intheControlgroupsubjectsissimilarinthetwofeature-locationconditions,indicatingthatviewingthedisplaysinblock2(BarandFragments)hadasimilareffectontheprocessingofthesetwoconditions.FortheExperimen-talgroup,however,viewingtheblock2displays(BarandZobject)hadadifferentialeffectontheFragment-BumpsandBar-Bumpsconditions.Forthisgroupofsubjects,theRTmeansarealmostidenticalforthesetwofeaturelocationsinblock3(723vs.721msec),indicatinganobject-basedeffectintheAmbiguousdisplaysthatisasstrongfortheFragmentsasforthefullyvisibleBar.

Experiment2:V

EndsTransfer

TheresultsofExperiment1demonstratethatexposuretoanovelshapethatlinkstogetherfeaturefragmentsaffectsthelaterprocessingofdisplaysinwhichfragmentsmaybeinterpretedaspartofasingleobject.Thissuggeststhatsubjectsperformaformofamodalcompletiongivenknowledgeofanobjectthatcanlinkthefragmentsinadisplay.Anaturalnextquestionconcernsthenecessityofocclusion:Doesthiscompletionrequirethepresenceofanoccludingobject?

ConsiderforexampleBregman’swell-knownBdisplays,whereonerecognizesasmatteringofedgefragmentsasasetofblock-letterBsoncetheoccludingblobsareaddedtotheimage(seeFigure6).ThequestioniswhetherconditionsexistunderwhichthefragmentsoftheBsmaybesufficienttoallowforcompletionwithoutthepresenceoftheoccludingblobs.AcompletionmechanismlargelydrivenbyperceptualexperiencewouldpredictthatrepeatedexperiencewithBshapescouldinfluenceperceptualorganizationsuchthattheeffectoftheobjectcanbedetectedinanattentiontask,evenwithouttheoccludingblobs.Experiment2wasdesignedtoaddressthisprediction.

AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier12

Figure6:Evidenceofocclusionfacilitatescompletionoftheoccludedshapesontheright,whiletheyaremoredifficulttoperceiveontheleft(afterBregman,1981).

Toexaminetheseissues,wedevisedaseriesofexperimentsusingthesamemethodologyasEx-periment1,andsimilarstimulitoourearlierexperiments.AsinExperiment1,subjectsfirstsawablockoftrialswithanambiguousdisplay,thenablockusingafulldisplaythatfavoredaparticu-larinterpretationoftheambiguousdisplay,followedbyanotherblockoftheambiguousdisplay.InExperiment2,thestimuliwerederivedfromtheVdisplays(Figure1g-i),forwhichwefoundanobject-basedattentioneffect(Behrmannetal.,1998).Thecentralpredictioninthenewexperi-mentwasthatexposuretoVdisplayswouldaffecttheprocessingofdisplaysinwhichtherewasnoocclusioninformationintheimage,andthefragmentscouldbeconsistentwithmanydifferentshapeconfigurations.TheseEndsdisplays(seeFigure7)weresimilartotheFragmentsdisplaysinExperiment1,exceptthatthefragmentsherecorrespondedtotheendsoftwooverlappingVshapes(theVdisplay).TheVdisplaythusactedliketheZdisplayinExperiment1,andweusedtheEndsdisplaystoexaminetheinfluenceofexperienceonsubjects’performance.

Method

Participants.Eighteensubjects,ninemaleandninefemale,between18and23yearsofagewerere-cruitedfromtheundergraduatesubjectpoolattheUniversityofArizona.Allsubjectshadnormalorcorrectedvisualacuitybyselfreport,andwereunawareofthepurposeoftheexperiment.Apparatusandmaterials.ThesameapparatususedinExperiment1wasusedhere,exceptthatweuseda13-inchcolormonitor.

TheEndsdisplayscontainedfourrectanglesof2.5cmx1.5cm,orientedat45degrees(seeFig-ure7).TheEndsweremadebyremovingthecenterofadisplaywhichcontainedtwoV’slyingatoponeanother(seeFigure1g-i).ThediagonalextentofthisdisplaymatchedthedimensionsofthebarinthedisplaysofExperiment1(8.7cmlongby2.5cmwide).Thehorizontallinedrawn

AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier13

fromthemidpointofonerectangularEndtothemidpointofthehorizontally-alignedotherEndwas6.5cm.Oneachtrial,thefeatures(bumps)appearedontwoofthefourEnds,ineitheratwoorthree-bumpconfiguration.Thebumpconfigurationswerethesameasinthepreviousexperiment.

Thedisplaysusedinthisexperimentfellintotwoconditions,basedonfeaturelocation:

1.Diagonal-Bumps(Figure7a-b):thebumpsappearedonthediagonallyoppositeends.FortheVdisplays,thisconfigurationcorrespondstothebumpslyingontwodifferentobjects.FortheXdisplays,itcorrespondstothemlyingonasingleobject(eithertheoccludingoroccludedbar).2.Vertical-Bumps(Figure7c-d):thebumpsappearedeitherontheendpairsontherightorlefthandsideofthedisplay.Heretheobjectrelationshipisreversed.FortheVdisplays,thisconfigurationcorrespondstoasingleobject,whilefortheXdisplays,itcorrespondstothetwo-objectcondition.Therewasanequalnumberof‘same’and‘different’judgmentsineachofthetwoconditions,asinthepreviousexperiment,andthelocationsofthebumpswereevenlycounterbalanced(diagonal-leftorrightforDiagonal-Bumps,vertical-leftorrightforVertical-Bumps).TheVdisplayshadoneotherdegreeoffreedom,orientation:whethertheleftorright-facingVwasontop.Thisvariablewasalsocounter-balanced.ThetotalnumberofdisplaysfortheEndswas16;thisnumberwasdoubledtoequalthenumberofVdisplays.

AsinExperiment1,thesubject’staskwassimplytodecidewhetherthenumberofbumpsonthetwoendswasthesameordifferent.Responseswereindicatedwiththe[Z]or[/]keyswiththeleftandrightindexfingersonthestandardkeyboard.Theassignmentofkeysto‘same’or‘different’responseswascounterbalancedacrosssubjects.Reactiontimes(RT)tomakethedecisionwasrecordedinmillisecondsandaccuracywasnoted.

Design.Theexperimentwasrunin3blocks.Inthefirstblock,thesubjectssawonlytheEnds.Inthenextblock,thesubjectssawonlyVdisplays.ThesetwoblocksconstitutetheInitialepochforthesetwodisplaysrespectively.Inthefinalblock(theTestepoch),EndsandVdisplayswererandomlyintermixed.NotethatsubjectsdidnotseeanyexamplesoftheVdisplaysbeforeblock2.Thedesignwasentirelywithin-subject,withtherelevantindependentvariablesbeingfea-turelocation(Diagonal-Bumps,Vertical-Bumps),display(VsorEnds),andepoch(Initial,Test).Orientationandjudgementweretheotherindependentvariables.ThedesignissummarizedinFigure7.

Thethreeexperimentalblockseachconsistedof192trials,withafewminutesbreakbetweeneachblock.Trialswererandomizedwithinablock.Priortostartingtheexperiment,subjectsweregiven32practicetrials,twoofeachoftheEndstrials.Timingandresponsemeasurementswerethesameasinthepreviousexperiment.

AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier14

Block1:Endsdisplays

(a)

Block2:Vdisplays

(b)(c)(d)

(e)

Block3:EndsandVs

(f)(g)(h)

Figure7:DesignandstimuliusedinExperiment2.Inblock1,subjectsperformedsame/differentjudgementstoEndsdisplays,examplesofwhichareshownhere(a-d).ThesesameEndsdisplayswerealsousedinExperiment3.Inblock2,subjectsperformedthesametaskonVdisplays(e-h).Displaysa,b,eandfareexamplesofDiagonal-Bumpsdisplays,whilec,d,g,andhareVertical-Bumps;a,c,e,andgaresamejudgements,b,d,f,andharedifferent.Inblock3,bothtypesofdisplays,EndsandVs,werethenmixed.

AdaptiveObjectAttentionInitial Epoch820Zemel,Behrmann,Mozer,&BavelierTest Epoch15

800Diagonal−BumpsVertical−Bumps780RT (msec)760740720700EndsVsDisplayEndsVsFigure8:MeanRTsasafunctionofdisplay(Ends,Vs),epoch(Initial,Test),andfeaturelocation(Diagonal-Bumps,Vertical-Bumps)inExperiment2.ThefirstpanelofthedisplayshowsthemeanRTstoEndsandVsduringblocks1and2respectively.ThesecondpanelshowsthemeanRTstothesesamedisplaysinthethirdblock,whichconsistedofbothdisplaytypes.NotethatDiagonalcorrespondstothebumpsbeingonseparateobjects,whileVerticalcorrespondstothembeingonasingleobject.AdaptiveObjectAttentionlentmeanRTs(differenceof4msec)isnotsurprising.

Zemel,Behrmann,Mozer,&Bavelier16

Secondly,forVdisplaysintheInitialepoch,RTsfortheVertical(single-object)conditionweresignificantlyfasterthantheDiagonal(two-object)condition:.TheRTdifferencewas35msec.ThisresultreplicatestheobjecteffectfortheVsfoundinBehrmannetal.(1998).

MostnotableisthecomparisonbetweentheInitialandTestepochs.Overall,fortheVdisplays,featurelocationwashighlysignificant().Therewasanon-significant

),andtherewasnospeed-upfromInitialtoTestepochforthesedisplays(

interactionbetweenepochandfeaturelocation().ThustheobjecteffectfortheVsisobservedintheTestepoch(differenceof37msecinfeaturelocation)aswellastheInitialone.

ThecriticalresultwithrespecttoourhypothesisisthesignificantinteractionbetweenfeaturelocationandepochfortheEndsdisplays().WhereasfeaturelocationwasnotsignificantfortheEndsintheInitialepoch—a4msecdifferenceinthewrongdirection—itwassignificantintheTestepoch—,)—a38msecdifference.FortheEnds

)butepochwas(overall,featurelocationwasnotsignificant(

).

ThisresultclearlyshowsthatexposuretotheVdisplaysinblock2inducesanobjecteffectintheEndsdisplaysinthefollowingblock,withinasinglegroupofsubjectswithinonetestingsession.Theequivalenceofthediagonalandverticalfeaturelocationsforthesesubjectsinthefirstblockisundonebytheirperceptualexperienceduringblock2.Thus,eventhoughtheEndsareambiguousinandofthemselves,theyaretreatedasequivalenttothesingle-objectandtwo-objectconditionsoftheVdisplaysbyvirtueofsubjects’experiencewiththeVdisplays.

ThisfindingisanalogoustothedisambiguatingeffectthattheZdisplayshadontheExperimentalgroupsubjectsinExperiment1.Inthiscase,however,theeffectofthenovelobjectonthegroupingoffragmentsisobervedintheattentiontaskevenwithoutthepresenceofanoccludingshape.Thisresultissurprising,becauseexperiencewiththeVdisplaysinducedendstobegroupedtogethereventhoughallperceptualinformationinthedisplayindicatesthattheyarenotrelated,i.e.,thefourEndsareeachclosedrectangles.

Onepossibleaccountforthisfindingisthattheshort-termeffectofexperiencewiththeobjectissostrongthatitcanover-rideevidencethattheEndsareclosedself-containedobjects,andproduceaperceptofthelinkingVshape.ThisaccountwouldsuggestthattheBsinFigure6awouldbeperceivedfollowingsomeexposuretotheBshapes.Analternativeexplanationfortheresultsofthisexperimentisthattheobjectattentionprocessisimprecise,andcanbetrickedintolinkingtheEndsevenwhentheimageevidenceisnotconsistentwiththeinterpretationofthetwofragmentsasbelongingtoasingleobject.Whereasthefirstaccountisdrivenbyawhole-objectmatchingprocess,thissecondaccountreliesonalimitedanalysisoflocalfeatures.WereturntothisissueintheGeneralDiscussion.

Inanycase,thefindingprovidesstrongevidencethatsubjects’perceptualexperiencealterssub-sequentvisualprocessing:theeffectofexperienceissufficientlyrobustsothateventhoughinfor-mationintheimagemightbeinterpretedtothecontrary,thatthebumpsarenotpartofalargerobjectbyvirtueoftheclosingbar,theyarestillgroupedtogether.

AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier17

Experiment3:X

EndsTransfer

Thepreviousexperimentconfirmedthehypothesisthatimagefragmentsmaybecometreatedorinterpretedinaparticularwaydependingonthespecificexperienceofthesubject.Exposingasubjecttoimagesthatlinkparticularfragmentsintoobjectsaffectstheirsubsequentorganizationofthesefragments,bothinthepresenceofandabsenceofocclusion.Intheocclusion-presentAmbiguousdisplaysofExperiment1,thegroupingoffeaturesonthefragmentswasneutral,po-tentiallyinterpretableasbelongingtoasingleoccludedobjectortwoseparateobjects.FollowingexperiencewiththeZdisplays,subjectsgroupedtheFragment-Bumpstogether.Intheocclusion-absentEndsdisplaysofExperiment2,priortoanyexposuretotheVdisplays,subjectsdidnotshowanyRTdifferencestoVertical-BumpsorDiagonal-Bumps.HereHowever,afterexperiencewiththeVdisplays,cuesthattheEndswereunrelatedwereovercome,andtheverticallyalignedfeaturesoftheEndsdisplaywererespondedtofasterthanthefeaturesthatwerealigneddiago-nally.

Oneimportantissueconcernswhethersubjectsaresomehowpredisposedtogroupingverticallyalignedfeatures,andtheslightestexperiencewithshapesinwhichfeaturesgroupverticallyissufficienttoinducetheverticalbiasintheneutralEndsdisplays.Astrongerdemonstrationoftheeffectofexperiencewouldutilizeadifferentobjectdisplay,inwhichthefeaturesaregroupeddiagonallyinsteadofvertically,andshowthatthisreversesthegroupingoffeaturesinthesameneutralEndsdisplays.WeaddressedthisissueinExperiment3byusingtheXdisplaysasthedisambiguatingdisplaysinsteadoftheVdisplays.Thisleadstothedirectlyoppositeprediction:verticallyalignedbumpsnowbelongtotwodifferentobjectswhilediagonallyalignedbumpsbelongtothesameobject,sosubjectsshouldnowshowanobjectadvantageforDiagonal-BumpsontheendsasopposedtotheadvantageobtainedforVertical-BumpendsinExperiment2.GiventhatweknowfromExperiment2thatinitiallythereisnodifferencebetweenVertical-BumpsandDiagonal-BumpsintheEndsdisplays,westartedthisexperimentbyexposingsubjectstotheXsdirectly,withoutprobingtheEndsalonefirst.

Method

Participants.Twenty-foursubjects,halfmaleandhalffemale,between18and25yearsofagewererecruitedfromtheundergraduatesubjectpoolattheUniversityofToronto.Allsubjectswereright-handedandhadnormalorcorrectedvisualacuitybyselfreport.

Apparatusandmaterials.ThesameapparatusasusedinExperiment1wasusedhere.ThedisplaysincludedtheEndsusedinExperiment2aswellasthefullXdisplays,showninFigure1a-f.ThedimensionsoftheXdisplayswereidenticaltothoseoftheVsandEndsdisplayasthesedisplayswereconstructedfromthefullXdisplays.Asinthepreviousexperiment,thedisplaysfallintotwofeatureconditions,dependingonwhetherthetwosetsofbumpsfallonthediagonal(Diagonal-Bumps)orontheverticalleftorright(Vertical-Bumps).Notethathereasopposedtothepreviousexperiment,Diagonal-Bumpscorrespondstoasingleobject(inthefullXdisplay),whileVertical-Bumpscorrespondstotwodifferentobjects.Thetaskandresponsemeasureswerethesameasinthepreviousexperiments.

AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier18

Procedure.Theexperimentwasrunin4blockswithafewminutesbreakbetweeneachblock.ThefirsttwoblockscontainedXdisplays,andthenexttwocontainedEndsdisplays.Eachblockconsistedof128trials:fourreplicationsofthefullsetofXdisplays,oreightreplicationsoftheEndsdisplays.Trialswererandomizedwithinablock.Atthebeginningoftheexperiment,sub-jectswereshownprintoutscontainingexamplesoftheXdisplaysandwereinstructedtomakesame/differentjudgementsonthenumberofbumps.Eachtrialproceededasintheprevioustwoexperiments.Priortostartingtheexperiment,subjectsweregiven32practicetrials,oneofeachoftheXdisplays.

Design.Thedesignwasentirelywithin-subject,withthestatisticallyindependentvariablesbeingdisplaytype(Xs,Ends),featurelocation(Diagonal-Bumps,Vertical-Bumps),judgement(same,different).OrientationwasanotherindependentvariablefortheXdisplays.

Resultsanddiscussion

Asinthetwopreviousexperiments,separateanalysesofvarianceconductedcrossingjudgement(andorientationinthecaseoftheXs)withthemainfactors—displaytypeandfeaturelocation—revealednosignificantinteractions(all),sothedatawerepooledacrossorientationsandjudgementsforsubsequentanalyses.

AnanalysisofvariancewithmeancorrectRTsasthedependentmeasureanddisplay(Ends,X)andfeaturelocation(Diagonal,Vertical)wasconducted.TheRTdataareillustratedinFigure9.Theprimaryresultfromthisstudyisthehighlysignificanteffectoffeaturelocation,

.Thisfinding,togetherwiththelackofanysignificantinteractionbetweenfeature

locationanddisplaytype,,indicatesthattheobjecteffectholdsidenticallyforboththeXdisplaysandtheEndsdisplays:theDiagonal-Bumps(lyingonasinglebar)areprocessedmorequicklythantheVertical-Bumps(lyingonseparatebars).

Tosummarize,theresultsofthisexperimentarestraightforward.ThedifferencebetweenthesingleobjectandtwoobjectconditionintheXdisplayisreplicated.Inaddition,thisdifferencealsoappliestotheEndsdisplays:afterbeingexposedtoXdisplays,subjectsrespondrelativelyquicklytotheEndsdisplayswhichcorrespondtothesingleobjectinthefullXdisplaysandlessquicklytotheEndsdisplayswhichcorrespondtothetwoobjectconditioninthefullXdisplays.WeknowfromExperiment2thatnaivesubjectswhodonothaveexperiencewithfullXdisplaysdonottreattheDiagonalorVerticalbumpedEndsdifferentially.Thus,eventhoughtheEndsconditionsareambiguousinandofthemselves,theycometobetreatedasequivalenttothesingleandtwoconditionsoftheXdisplaysbyvirtueofsubjects’experiencewiththeseXdisplays.Takentogether,Experiments2and3providecomplementaryresultsillustratingtheabilityofperceptualexperiencewithanobjecttoaffectanattentiontaskwherethedisplayscontainonlyfragments.Inbothcases,thefragmentswereclosedshapes,sotheimagesnotonlylackedinfor-mationaboutocclusionbutcontainedcontradictoryevidenceagainstanocclusioninterpretation,yettheresultsdemonstratethattheyweretreatedaspartsofobjectsduetoexperience.

AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier19

820800Diagonal−BumpsVertical−Bumps780RT (msec)760740720700Xs DisplayEnds Figure9:Meanreactiontimesasafunctionofdisplay(XsorEnds)andfeaturelocation(Di-agonalorVerticalfeaturelocations)forExperiment3.SubjectsinExperiment3performedthesame/differentnumber-of-bumpstaskontwoblocksoftheXdisplaysfollowedbytwoblocksofEndsdisplays.NotethatheretheDiagonal-Bumpslieonasingleobject,whiletheVertical-Bumpslieondifferentobjects.InExperiment2,forsubjectswhohadnotyetseeneitherXorVdisplays,meanRTsontheEndsdisplayswere778and782msecforDiagonal-BumpsandVertical-Bumps,respectively.ThustheobjecteffectapparentherefortheEndsdisplayemergesafterexposuretotheXs.

AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier20

Experiment4:LocationSpecificityofPerceptualExperience

Theexperimentspresentedaboveshowthatexposuretodisplayscontaininganobjectcaninduceanobjecteffectinambiguousdisplaysthatinandofthemselvesdonotcontainanyevidenceofanocclusionrelationship.SubjectsexposedtoXdisplaysrespondfastertodiagonalendsthantoverticalonesintheEndsdisplays,correspondingtotheobjecteffectintheXdisplays.Incontrast,ifasubjectisexposedtotheVdisplays,theverticallyalignedendsarerespondedtofasterthanthediagonalends.TheEndsdisplay,then,isinitiallyambiguousbutislaterparsedaccordingtothesubject’srecentperceptualexperience.

Animportantquestionthenconcernsthenatureoftherepresentationsthatareactivatedasafunctionofexperiencewithaparticulardisplay.Anumberofpossibilitiesexist:

1.Onepossibilityisthatparticularpairsoflocationsonthescreenobtainaprocessingadvantageduetoexperiencewithaspecificshape.Inourexperiments,thedisplayswereaconsistentsizeandlocationonthescreen.Specificobjectsthencouldinducegroupingsofscreenloca-tionswherefeaturesofthatobjectappear.Clearly,atleastpairsofpositionsmustbeprimedbytheobject,becausetheXandVdisplaysinvolvedthesamesetofindividualpositions,andonlydifferedwithrespecttowhichpairsofpositionsbelongedtothesameobjects.Onthisview,theEndsdisplaysdonotneedtoactivateanyobjectrepresentationsinordertoproducetheobjecteffectsobservedinExperiments2and3.2.Anotherpossibilityisthattheperceptualorganizationprocessismediatedbyshaperep-resentations,buttheserepresentationsarehighlyviewpointspecific,i.e.,tiedtoparticularlocationsonthescreen.Thisviewisconsistentwithrecenttheoriesabouttherelationshipbetweenobject-basedandspatialattention(e.g,Goldsmith,1998;Mozeretal.,1992;Vecera&Farah,1994;seeGeneralDiscussion),inthattheobjectsactivateparticularspatialloca-tions,andattentionisthenallocatedtothosespatialpositions.Onthisview,exposuretotheV(orX)displaysprimesparticulargroupingsofspatiallocationsthatcorrespondtotheobjects,andthisprimingisthenapparentintheEndsdisplays.3.Athirdpossibilityisthatthegroupingprocessutilizesshaperepresentationsthatarenotviewpointspecific.Onthisview,experiencewithaparticularshapedoesnotonlyeffec-tivelyprimethegroupingofitsfeaturesinthespecificspatiallocationsinwhichtheyap-pear.Inaddition,theexperiencemayalsoprimegroupingoffeaturesinotherlocationsthatcorrespondtothatsameobjectinadifferentposition,scale,ororientationthantheoriginalone.Oritmayprimesomemoreabstractrepresentationoftheobject,onenottiedtoanyparticularexemplar.Theaimofthisexperimentwastotestthehypothesiscontainedinthethirdviewdescribedabove:thattheexperience-inducedobjecteffectisnotviewpointspecific,butrathergeneralizestoinstan-tiationsofanobjectthatareneveractuallypresentedinanyfull-objectdisplays.

WetestedthisbymanipulatingtherelativepositionswithintheEndsdisplays,whilenotchang-ingtheXdisplays.WemodifiedthedesignusedinExperiment2—inwhichthesubjectssawablockofEndsdisplays,thenablockofVdisplays,followedbyanotherblockofEnds—intwo

AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier21

ways:(1)ThesecondblockcontainedXratherthanVdisplays.ThereforeanytransfertotheEndsbasedonperceptualexperiencewiththeXleadstothesamepredictionasinExperiment3:theEnds-Diagonalwillbefasterinblock3thanEnds-Vertical,sincethediagonalbumpscorrespondtoasingleobjectinthedisambiguatingfullobjectdisplay.(2)Thefeatures(bumps)intheEndsdisplayswereindifferentlocationsthantheXdisplays.ThereweretwotypesofEndsdisplays,oneinwhichthedistancebetweenthefeatureswaslargerandtheothersmallerthantheXdis-plays,sothefeaturelocationsdidnotmatchbetweentheEndsandX.Thecriticalquestionhereiswhethersubjectsstillobtainthebenefitofblock2suchthatEnds-Diagonalwillbefasterinblock3thanEnds-VerticalevenwhenthefeaturelocationsintheXandEndsdisplaysdonotcorrespond.NotethatthemanipulationofthefeaturelocationsintheEndsdisplaysdoesnotexploretherangeofpossibleviewpointvariations,includinglocation,orientation,andsize.Withoutexplicitlycon-sideringmanipulationsalongeachofthesedimensions,thisapproachstillallowsanexplorationofthebasicissueofwhethertheobjectbenefitfromtheXdisplayswillapplytotheEndsdisplaysevenwhenthefeaturelocationsdonotmatch.

Method

Participants.Twenty-foursubjects,tenmaleandfourteenfemale,between18and23yearsofagewererecruitedfromtheundergraduatesubjectpoolattheUniversityofArizona.Allsubjectshadnormalorcorrectedvisualacuitybyselfreport,andwereunawareofthepurposeoftheexperiment.

Apparatusandmaterials.ThesameapparatususedinExperiment2wasusedhere.ThedisplaysincludedtheXdisplaysusedinExperiment3aswellasmodifiedversionsoftheEndsdisplaysusedinExperiments2and3.TwovariationsoftheoriginaldisplayswerecreatedbyshiftingthelocationsoftheEnds:intheEnds-Smallset,theEndswereallmovedtowardsthecenterofthedisplayby1cm(1.2);intheEnds-Largeset,theEndsweremovedawayfromthedisplaycenterbythesameamount(seeFigure10).Asaresultofthismanipulation,thebumpsintheEndsdisplaysarenotinthesamelocationsasinthefull,disambiguatingdisplays,unlikethepreviousexperiments.

Procedure.Atthebeginningoftheexperiment,subjectswereshownprintoutscontainingexamplesoftheLargeandSmallEndsdisplaysandwereinstructedtomakesame/differentjudgementsonthenumberofbumps.Eachtrialproceededasinthepreviousexperiments.Asbefore,thesubject’staskwassimplytodecidewhetherthenumberofbumpsonthetwoendswasthesameordifferent.Theexperimentwasrunin3blockswithafewminutesbreakbetweeneachblock.Trialswererandomizedwithinablock.Priortostartingtheexperiment,subjectsweregiven32practicetrials,oneofeachofthefullsetofEndsdisplays.

Design.Thedesignwasentirelywithin-subject.TheimportantindependentvariablesareexactlyasinExperiment2,exceptthatthereisanadditionalvariable(Large,Small)fortheEndsdisplays.AsinExperiment2,theexperimentwasconductedinaseriesofthreeblocks.Inthefirstblock,thesubjectssawonlytheEnds,withLargeandSmallrandomlyinter-mixed.Inthenextblock,thesubjectssawonlyXdisplays.TheXdisplayswereintermediateinposition,andmatchedneithertheLargenorSmalldisplays.ThefinalblockconsistedsolelyofEndstrials,againwithLargeandSmallintermixed.

AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier22

(a)(b)

Figure10:ExamplesofmodifiedEndsdisplaysusedinExperiments4.(a)Ends-Small.(b)Ends-Large.ThedashedlinesinbothdisplaysdepicttheoriginalEndsstimuli(thesearethesamesizein(a)and(b)).

AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier23

840820800Block 1 Block 2 Block 3DiagonalVerticalRT (msec)780760740720700LargeSmallXsDisplayLargeSmallFigure11:MeanRTsasafunctionofdisplay(Ends-Large,Ends-Small,X),Block(1,2,3),andfeaturelocation(Diagonal,Vertical)inExperiment4.NotethatDiagonalcorrespondstoasingleobjectwhileVerticalcorrespondstothetwoobjectcondition.AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier24

processedsignificantlyfasterthanEnds-Large(24msec):.Moreim-portantly,forbothEndsdisplays,diagonal-bumpswereprocessedfasterthanvertical-bumps.Inthethirdblock,themeanRTdifferencebetweenDiagonalandVerticalwas14and22msecfortheLargeandSmalldisplays,respectively.Overall,theeffectoffeaturelocationwasmarginallysignificant().Whenthedataforthetwodisplayswereconsideredsepa-),whiletherately,thedifferenceforEnds-Largeapproachedsignificance(

).differenceforEnds-Smallwassignificant(

Thisstudydemonstratesthattheeffectsofperceptuallearninginthistaskgeneralizetosome

degreetootherscreenlocations.ExposuretoablockofXdisplaysledtofasterprocessingofthediagonalEnds—theEndspairsconsistentwiththeobjectsintheXdisplays—inthesubsequentblockeventhoughtheexactfeaturelocationsdidnotmatchinthetwoblocks.

Thisfindingarguesagainstthefirsttwoofthethreealternativespresentedabove,thatthelearningisscreenlocationspecificortiedtolocation-specificshaperepresentations.Instead,itismoreconsistentwiththeideathatthelearningisaffectingamoreabstractformofrepresentation.Wereturntothisissuebelow.

GeneralDiscussion

Numerousstudieshaveshownthattheparsingofavisualsceneisanimportantfactoraffectingthedistributionofattention.Spatiallocationsclearlyplayanimportantroleinthisparsingprocess(Posner,Snyder,&Davidson,1980;Tsal&Lavie,1988).Manyotherstudieshaveshownthatattentioncanbedirectedtoobjectsratherthantolocationsperse(Duncan,1984;Egly,Driver,&Rafal,1994;Kramer&Jacobson,1991).

Acentralissueiswhatdeterminestheobjectorgroupingofelementsforattentionalselection.Severalstudieshavereinforcedthepivotalroleofperceptualorganization.Thisrolehasbeendemonstratedindistractorstudies,whichshowthatitisdifficulttoignoreinformationthatbe-longstothesameobjectorgroupastask-relevantinformation.Forexample,byvirtueofcommonfate,identificationofacentraltargetwasmoreaffectedbydistantdistractorsthatmovedinthesamedirectionasthetargetthanbynearbystaticdistractors(Driver&Baylis,1989).Similarly,theresponse-compatibilityeffect(enhancementfromsimilar,andinhibitionfromdissimilardistrac-tors)wasreducedwhenthetargetanddistractorswereembeddedinorgroupedwithdifferentobjectscomparedtowhentheyweregroupedonthesameobject(Kramer&Jacobson,1991);forotherexamples,see(Baylis&Driver,1992;Bundesen,1990).

However,genericgroupingprinciplesdonotsufficetodefinetheobjectsofattention,aspastexperienceorfamiliarityappearstoplayarole.Forexample,whensubjectsdecidedwhethertwo”x”sappearonthesameorontwodifferentsuperimposedlettersornon-letters,performancewassuperioronlettersthanonnonletters(Vecera,1993),andsubjectswerefasteronuprightlettersthanupside-downletters(Vecera&Farah,1997).Thesestudiesdemonstratethatattentioncanalsobeallocatedpreferentiallytohighlyfamiliarshapes.

Theresultspresentedinthispaperextendtheseeffectstoapplytorecentlyviewednovelshapes.Theprimaryfindingintheseexperimentsisthatobjectattentionbenefitsareobtainedfornewly-

AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier25

learnedobjects,andforfragmentsofsuchobjectstowhichstandardcompletionheuristics,e.g.,relatability,donotapply.Relativelybriefexposuretoanovel,odd-shapedlinkingobjectsufficetoinduceobject-basedattentiontofragmentsthatcanbeinterpretedasthevisiblepartsofthatobjectunderocclusion.Asecondprimaryfindingisthattheseexperience-dependentobjectbenefitscanapplytofragmentsevenwithoutanyevidenceofocclusion.Theeffectsofexperiencewerestrongenoughtoovercomeevidencethatthefragmentswereseparateobjects(i.e.,thepresenceofterminatorsineachEnd).Also,theresultsofthisstudyshowthatuniformconnectedness(Palmer&Rock,1994)isnotnecessaryforobjectattention.WhereasKramerandWatson(1995)foundnoobjecteffectswhenanobject’suniformconnectednesswasdisruptedbyanewregionofdifferentcolorortexture,thefindingsheredemonstratearobustobjecteffect.

Predictionsbasedonsubjectiveorganization

Consideredinconjunctionwiththeearlierresultsonobject-basedattention,thesefindingshigh-lighttheroleofperceptualorganizationintheallocationofattention.Inalloftheexperimentspresentedhere,wecontrastedsubjects’responsestothesamestimuluspatternasafunctionoftheirperceptualexperience.Ineverystudy,therecentexperiencehadasignificantinfluenceontheirorganizationofthedisplays,asassayedbytheirresponses.Short-termshapefamiliarity,aswellaslong-termfamiliarityandgenericgroupingprinciples,affectthesceneorganizationandattentionalallocation.

Alogicalextensionofthisfindingisthatothermethodsofalteringsubjectiveorganizationshouldbeabletoinduceanobjectbenefit.Forexample,taskinstructionscouldbeusedtosuggestapar-ticularparsingofthescene.Yantis(1992)askedsubjectstotrackfiveoutoftenrandomlymovingdots,andtoindicate,afterallthedotshadstoppedmoving,whetherornotaparticulardotwasamemberofthetargetset.Hefoundthatsubjectswhowereencouragedtogroupthetargetdotsasahigherorderformor”object”performedbetterintheearlyphasesoftheexperimentthanthosewhosawthesamestimulibutdidnotreceivesuchencouragement.Inamoredirectlyrel-evantstudy,Chen(1998)foundanobjecteffectwhensubjectswereinstructedtoviewadisplayastwoseparateobjects,butnoeffectwhentheinstructionssuggestedasingle-objectinterpreta-tion,fortheidenticalstimulusconfiguration.BaylisandDriver(1992)alsousedtaskinstructionstogetsubjectstointerpretthesamedisplaysindifferentways.Theseresultsindicatethattheobjecteffectsinducedbyperceptualexperiencemaybeachievedbysimplymodifyingthetaskinstructions.

Thehypothesisthatsubjectiveorganizationisadeterminingfactorinobjectattentionpredictsthatothermanipulationswillalsoinduceanobjecteffect.Evenshorter-termfamiliaritymaysuffice;theobjecteffectobservedinthestudiespresentedheremaybeobtainedinaprimingstudy,inwhichthedisambiguatingstimulusisusedasaprime.Ontheotherhand,nofamiliarityatallmaybenecessary;entirelynovelobjectsthatadheretostandardgroupingprinciplesshouldalsobenefitfromobjectattention.

AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier26

Underlyingmechanisms

Whatmechanismsunderlietheresultsdescribedhere,andthegrowingbodyofobjectandspatialattentionfindings?Forsometime,spaceandobjectattentionhadbeenconsideredtobemutuallyexclusivealternatives(Kanwisher&Driver,1992).Morerecently,attemptshavebeenmadetoreconcilethetwoformsofattention.Eglyandcolleagues(Eglyetal.,1994)havearguedthatbothprocessesco-exist.Forexample,theyreportedthatacostinRTandaccuracyisincurredwhenattentionisshiftedbetweenacueandatargetbothwhenthetargetappearedatasecondlocationinthecuedobject(within-object,objectattention)oratanequidistantlocationbutinadifferentobject(between-object,spatialattention).Datafavoringthesimultaneousoperationofspace-andobject-basedprocessesalsocomefromastudybyUmiltaetal.(1995)whocuedavertexofacubewhicheitherremainedstationaryorrotated.Subjectsnotonlyshowedfacilitationwhenthetargetappearedinthesamespatialorretinallocationasthecue(stationary)butalsowhenthetargetappearedinadifferentretinallocationbutintheequivalentobject-definedlocationasthecue(rotatedcondition).Similarfindingsarerevealedinstudiesoninhibitionofreturninbothlocation-andobject-basedcoordinates(e.g.,Gibson&Egeth,1994;Tipper&Weaver,1996).Onestandardaccountofobject-andspace-basedattentionisthatlow-levelvisualroutinesidentifyregionsofsalienceorcoherenceinthevisualfieldpre-attentivelyandinparallel.Theseregionsarethensubjectedtofurtheranalysisbyfocalattentionprocessingforlaterobjectidentification(Julesz,1981;Koch&Ullman,1985;Neisser,1967;Treisman,1982;1988).Thisviewofatwo-stagefeedforwardmodelinwhichspatialattentionfollowsobject-basedattentionhasbeenproposedtoaccountfornumerousfindingsinthevisualsearchliteratureaswellasfindingsinwhichgroup-ing,basedonfeaturesimilarityorproximity,occursearly,inparallelandindependentofspatialattention(Driver,Baylis,&Rafal,1992;Marshall&Halligan,1994;Moore&Egeth,1997).Thissequential,hierarchicalmodel,however,hasbeenincreasinglychallenged.Underthismodel,theobjectsofobject-basedattentionaredefinedbylow-levelvisualroutines.Resultsdemonstratingtheeffectsoffamiliarspecificobjectsonobject-basedattention(Vecera&Farah,1997),andresultsshowingthatspecificobjectsinfluenceimagesegregation(Peterson,1994)callthisaccountintoquestion.Inaddition,object-basedattentionfindingsinwhichthedifferentobjectsshareacom-monregioninspace(e.g.,Duncan,1984;Lavie&Driver,1996;Behrmannetal.,1998;andtheresultsreportedhere)arealsodifficulttoreconcilewiththisfeedforwardapproach.

Analternativetothesimplefeedforwardschemeisoneinwhichobject-andspace-basedpro-cessesoperateinparallelandmutuallyinfluenceeachother(Farah,1990;Humphreys&Riddoch,1991;Humphreysetal.,1996).Theinteractionoccursthroughatopographicallyorganizedgroupedarray,whichrepresentsthecurrentlyactivebottom-upinputfromtheenvironmentaswellastop-downactivationfrommatchinghigher-leveldescriptions.Throughthisexplicitarray,spatiotopicinformationandgroupinginformationarebothpresentandsimultaneouslyinfluencevisualpro-cessing.VeceraandFarah(1994)claimedthatsuchanarray-likerepresentationmustexist;usingtheEglyetal.(1994)paradigm,theyshowednotonlythatthereisacostassociatedwithshiftingattentionwithinandbetweenobjectsbutalsothatthecostofshiftingattentionbetweenobjectsincreasedasthespatialdistancebetweentheobjectsincreased.Similarly,asisusuallythecaseinthedistractorparadigm,KramerandJacobson(1991)showedthattheresponsecompatibilityeffectswerediminishedwhenthespatialdistancebetweenthegroupedelementswasincreased.Takentogether,thesefindingssuggestthatbothspace-andobject-selectionareoperativeand,assuch,aremorecompatiblewithaparallelaccountratherthanwithaserialtwo-stagemechanism.

AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier27

Withinthisparallelaccountofattention,theissueofhowthegroupedarrayoperatesisstillopen.Theoriginalproposalwasthatacombinationofgenericgroupingprinciplesandshape-specificinformationacttolabelthearraylocations(Vecera&Farah,1994;Krameretal.,1997).Geometricpropertiesofthedisplay,suchastherelatabilityoffragments(Kellman&Shipley,1992),wouldbefundamentalelementsofthegroupingcomponent.

Withrespecttotheinfluenceoffamiliarshapes,thestandardconceptionisthatthisinvolvesrep-resentationsofwholeobjects.Withinthisview,severalpossibilitiesexist.Anobjectcouldbe:(a)exactexemplar,specifictoparticularspatiallocationsandorientations;(b)fuzzyexemplar,specifyingaparticularshape,butlessspecificinitsspatialinstantiation;or(c)spatiallyinvariantobjectrepresentation.Theresultsoftheexperimentspresentedheredonotbearonthespatiallyinvariantrepresentationhypothesis.Limitedevidenceexistsforthisview:theresultsofVeceraandFarah’s(1994)studyimplicatedspatially-invariantobjectrepresentations,butotherstudieshavenotfoundevidenceforthem(forfurtherdiscussion,seeKrameretal,1997).Experiment4inthispaperprovidesevidencethattheobjecteffectcantransfertodifferentfeaturelocations,whichmakestheexactexemplarrepresentationunlikely.Instead,theseresultsareconsistentwiththefuzzyexemplarrepresentation,astherewassomespatialoverlapbetweenthelearnedfeaturelocationsandthegeneralizedlocations.Inaddition,thefactthatthedegreeoftransferwasgreatertotheSmall-Endsisalsoconsistentwiththefuzzyexemplar,undertheassumptionthattheobjectattentionbenefitextendstoalllocationsencompassedbytheviewedexemplar.

Analternativeconceptionoftheshape-specificcomponentofthegroupedarrayinvolveslearnedconfigurationsoflocalfeaturesratherthanwholeobjects.Thismechanismisconsistentwiththecomputationalmodel,MAGIC(Mozeretal.,1992).Underthisinterpretation,thedisambiguatingdisplaysprimarilyservetofacilitatethegroupingofparticularpairsofEndsinthedisplays;andthisgroupingthenappliestotheambiguousdisplays.ThisaccountissupportedbyfactthatthegroupingoftheEndsappearstooperateeveninthepresenceofterminators,whichprovideevidencethattheEndsarecompleteobjectsthemselves.Thisfeature-basedrepresentationcanalsoaccountfortheresultsofExperiment4,assumingafeature-basedanalogofthefuzzyexemplarmodelproposedforthewhole-objectrepresentation.

Finally,wenotethatallofthesedifferentmechanismscanbelearnedfromstatisticalstructureintheenvironment.Wholeobjects,orparticularlocalfeatureconfigurations,canbothbeextractedbasedonexperiencewithvariousfeaturecombinationsinimages.Theevidenceforexperience-dependenceprovidedinthispaperfurtherindicatesthatsuchhigher-orderstatisticalregularitiesplayacriticalroleinvisualperception.

Issuesforfurtherstudy

Thestudiespresentedhereleadtomanyquestionsrequiringfurtherresearch.Oneimportantissueconcernsthenumberofexposurestothedisambiguatingstimulusthatarerequiredtoobtaintheobjectadvantageintheambiguousdisplays.Inalloftheexperimentsdescribedhere,oneblock(consistingof32-128trialswiththerelevantstimulus)wassufficienttoobtaintheeffect.Itmaybepossiblethatmanyfewerexposuresarerequired.Arelatedquestioniswhetherobjecteffectscanbeinducedsimplybyinstructions.Forexample,suggestinganinterpretationoftheEndsdisplaysastwodiagonalcrossedbarswherethecentralportionofthebarshavebeenwoventhrough

AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier28

thefabricofawhitescreen,mightreducetherequirednumberofexposurestoobtaintheobjectadvantage.

Asecondimportantissueconcernsthedurationoftheeffectsofperceptuallearningshownhere.Aninterestingstudywouldtestsubjectsatdifferenttimeintervalsafterexposuretothedisam-biguatingstimulitodeterminehowlongthisexperienceexertedaneffectontheprocessingofambiguousstimuli(e.g.,Treisman&deSchepper,1996).

Athirdissueforfurtherstudyconsidershowthetaskthatthesubjectsperformmayinfluencethedegreeofeffectofperceptualexperience.Intheexperimentspresentedhere,thetaskdidnotrequireanyinterpretationofthedisplayintermsofobjects.Instead,thesubjectssimplyhadtofindandcomparethenumberofbumps;theobjectsinthedisplaywereirrelevanttothetask.Otherstudieshavealsofoundeffectsofexperienceonobjectattentionevenwhentheexperi-enceisnottask-relevant(e.g.,Goldsmith,1998).Thisincidentalformoflearningisincontrasttomoststudiesoftheeffectofexperiencewithnovelobjectsonfutureprocessing(Edelman&Bulthoff,¨1992;Gauthier&Tarr,1997),wherethetask(e.g.,familiarityjudgements,identification)explicitlyrequiredobjectidentification.Similarly,studiesofperceptuallearninghavedemon-stratedstrongerlearningwhenthestimuliaretask-relevant(e.g.,Ahissar&Hochstein,1993;Shiu&Pashler,1992),whileChunandJiang(1998)haveshownthataconsistentconfigurationofdis-tractorscanspeedvisualsearchwhenitistask-relevant(indicativeoftargetlocation).Basedonthesestudies,onewouldpredictthatmakingtheobjectsrelevanttothetaskobjectidentificationwouldleadtostronger,andperhapslonger-lastingeffectsofexperience.

References

Ahissar,M.andHochstein,S.(1993).Attentionalcontrolofearlyperceptuallearning.ProceedingsoftheNationalAcademyofScience,USA,90:5718–5722.Baylis,G.andDriver,J.(1992).Visualparsingandresponsecompetition:Theeffectofgroupingfactors.PerceptionandPsychophysics,51:145–162.Baylis,G.C.andDriver,J.(1993).Visualattentionandobjects:Evidenceforhierachicalcodingoflocation.JournalofExperimentalPsychology:HumanPerceptionandPerformance,19:451–470.Behrmann,M.,Zemel,R.S.,andMozer,M.C.(1998).Object-basedattentionandocclusion:Evi-dencefromnormalparticipantsandacomputationalmodel.JournalofExperimentalPsychology:HumanPerceptionandPerformance,24:1011–1036.Ben-Av,M.B.,Sagi,D.,andBraun,J.(1992).Visualattentionandperceptualgrouping.PerceptionandPsychophysics,52:277–294.Bregman,A.S.(1981).Askingthe‘Whatfor’questioninauditoryperception.InKubovy,M.andPomerantz,J.R.,editors,PerceptualOrganization,pages99–118.LawrenceErlbaumAssociates,Hillsdale,NJ.Bub,D.andGum,T.(1991).Psychlabexperimentalsoftware.Technicalreport,McGillUniversity,Montreal.Bundesen,C.(1990).Atheoryofvisualattention.PsychologicalReview,97:523–547.

Chen,Z.(1998).Switchingattentionwithinandbetweenobjects:Theroleofsubjectiveorganiza-tion.CanadianJournalofExperimentalPsychology,52:7–16.

AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier29

Chun,M.M.andJiang,Y.(1998).Contextualcueing:Implicitlearningandmemoryofvisualcontextguidesspatialattention.CognitivePsychology,36:28–71.Driver,J.andBaylis,G.(1989).Movementandvisualattention:Thespotlightmetaphorbreaksdown.JournalofExperimentalPsychology:HumanPerceptionandPerformance,17:561–570.Driver,J.,Baylis,G.,andRafal,R.D.(1992).Preservedfigure-groundsegregationandsymmetryperceptioninvisualneglect.Nature,360:73–76.Duncan,J.(1984).Selectiveattentionandtheorganizationofvisualinformation.JournalofExperi-mentalPsychology:General,113:501–517.Edelman,S.andBulthoff,¨H.H.(1992).Orientationdependenceintherecognitionoffamiliarandnovelviewsofthree-dimensionalobjects.VisionResearch,32:2395–2400.Egly,R.,Driver,J.,andRafal,R.D.(1994).Shiftingvisualattentionbetweenobjectsandlocations:Evidencefromnormalandparietallesionsubjects.JournalofExperimentalPsychology:General,123:161–177.Farah,M.J.(1990).VisualAgnosia:DisordersofObjectRecognitionandWhatTheyTellUsAboutNormalVision.MITPress,Cambridge,MA.Gauthier,I.andTarr,M.J.(1997).Orientationprimingofnovelshapesinthecontextofviewpoint-dependentrecognition.Perception,26:51–73.Gibson,B.S.andEgeth,H.(1994).Inhibitionofreturntoobject-basedandenvironment-basedlocations.PerceptionandPsychophysics,55:323–339.Goldsmith,M.(1998).What’sinalocation?:Comparingobject-basedandspace-basedmodelsoffeatureintegrationinvisualsearch.JournalofExperimentalPsychology:General,127:189–219.Humphreys,G.W.,Olson,A.,Romani,C.,andRiddoch,M.J.(1993).Competitivemechanismsofselectionbyspaceandobject:Aneuropsychologicalapproach.InKramer,A.F.,Coles,M.G.H.,andLogan,G.,editors,Convergingoperationsinthestudyofvisualselectiveattention.AmericanPsychologicalAssociation,Washington,DC.Humphreys,G.W.andRiddoch,M.J.(1991).Interactionsbetweenobjectandspacesystemsrevealedthroughneuropsychology.InMeyer,D.E.andKornblum,S.,editors,AttentionandPerformanceXIV.LawrenceErlbaumAssociates,Hillsdale,NJ.Julesz,B.(1981).Textons,theelementsoftextureperception,andtheirinteractions.Nature,290:91–97.Kanisza,G.andGerbino,W.(1982).Amodalcompletion:Seeingorthinking?InBeck,J.,editor,Organizationandrepresentationinperception.Erlbaum,Hillsdale,NJ.Kanwisher,N.andDriver,J.(1992).Objects,attributesandvisualattention.CurrentDirectionsinPsychologicalScience,1:26–31.Kellman,P.andShipley,T.F.(1992).Percivingobjectsacrossgapsinspaceandtime.CurrentDirectionsinPsychologicalScience,1:193–199.Kellman,P.J.andShipley,T.F.(1991).Atheoryofvisualinterpolationinobjectperception.Cog-nitivePsychology,23:141–221.Koch,C.andUllman,S.(1985).Shiftsinselectivevisualattention:Towardstheunderlyingneuralcircuitry.HumanNeurobiology,4:219–227.Kramer,A.F.andJacobson,A.(1991).Perceptualorganizationandfocusedattention:Theroleofobjectsandproximityinvisualprocessing.PerceptionandPsychophysics,50:267–284.

AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier30

Kramer,A.F.andWatson,S.E.(1995).Object-basedvisualselectionandtheprincipleofuniformconnectedness.InConvergingoperationsinthestudyofvisualattention,pages395–414.AmericanPsychologicalAssociation.Kramer,A.F.,Weber,T.A.,andWatson,S.E.(1997).Object-basedattentionalselection:Grouped-arraysorspatially-invariantrepresentations.JournalofExperimentalPsychology:General,126:3–13.Lavie,N.andDriver,J.(1996).Onthespatialextentofattentioninobject-basedvisualselection.PerceptionandPsychophysics,58:1238–1251.Mack,A.,Tang,B.,Tuma,R.,Kahn,S.,andRock,I.(1992).Perceptualorganizationandattention.CognitivePsychology,24:475–501.Marshall,J.C.andHalligan,P.W.(1994).Theyinandyangofvisuo-spatialneglect:Acasestudy.Neuropsychology,32:1036–1057.Moore,C.M.andEgeth,H.(1997).Perceptionwithoutattention:Evidenceofgroupingunderconditionsofinattention.JournalofExperimentalPsychology:HumanPerceptionandPerformance,23:339–352.Moore,C.M.,Yantis,S.,andVaughan,B.(1998).Object-basedvisualselection:Evidencefromperceptualcompletion.PsychologicalScience,9:104–110.Mozer,M.C.,Zemel,R.S.,Behrmann,M.,andWilliams,C.K.I.(1992).Learningtosegmentimagesusingdynamicfeaturebinding.NeuralComputation,pages650–665.Neisser,U.(1967).Cognitivepsychology.Appleton-Century-Crofts,NewYork.

Nickerson,R.F.(1965).Repsonsetimesfor”same”-”different”judgements.PerceptualandMotorSkills,20:15–18.Palmer,S.andRock,I.(1994).Rethinkingperceptualorganization:Theroleofuniformconnect-edness.PsychonomicBulletinandReview,1:29–55.Peterson,M.A.(1994).Objectrecognitionprocessescananddooperatebeforefigure-groundorganization.CurrentDirectionsinPsychologicalScience,3:105–111.Shipley,T.F.andKellman,P.J.(1992).Perceptionofpartiallyoccludedobjectsandillusoryfigures:Evidenceforanidentityhypothesis.JournalofExperimentalPsychology:HumanPerceptionandPerformance,18:106–120.Shiu,L.-P.andPashler,H.(1992).Improvementinlineorientationdiscriminationisretinallylocalbutdependentoncognitiveset.PerceptionandPsychophysics,52:582–588.Tipper,S.P.andWeaver,B.(1996).Themediumofattention:Location-based,object-centeredorscene-based.InVisualAttention.OxfordUniversityPress,Oxford,UK.Treisman,A.(1982).Perceptualgroupingandattentioninvisualsearchforfeaturesandforobjects.JournalofExperimentalPsychology:HumanPerceptionandPerformance,8:194–214.Treisman,A.(1988).Featuresandobjects:TheFourteenthBartlettMemorialLecture.QuarterlyJournalofExperimentalPsychology,40A(2):201–237.Treisman,A.andDeSchepper,B.(1996).Objecttokens,attention,andvisualmemory.InAttentionandPerformance,XVI,pages15–45.MITPress,Cambridge,MA.Tsal,Y.andLavie,N.(1988).Attendingtocolorandshape:Thespecialroleoflocationinselectivevisualprocessing.PerceptionandPsychophysics,44:15–21.

AdaptiveObjectAttentionZemel,Behrmann,Mozer,&Bavelier31

Umilta,C.,Castiello,U.,Fontana,M.,andVestri,A.(1995).Object-centredorientingofattention.VisualCognition,2:165–182.Vecera,S.P.(1993).Objectknowledgeinfluencesvisualimagesegmentation.InProceedingsofthe15thAnnualConverenceoftheCognitiveScienceSociety,pages1040–1045,Hillsdale,NJ.Erlbaum.Vecera,S.P.andFarah,M.J.(1994).Doesvisualattentionselectobjectsorlocation?JournalofExperimentalPsychology:General,123:146–160.Vecera,S.P.andFarah,M.J.(1997).Isimagesegmentationabottom-uporaninteractiveprocess?PerceptionandPsychophysics,59:1280–1296.Yantis,S.(1992).Multielementvisualtracking:Attentionandperceptualorganization.CognitivePsychology,24:295–340.Yantis,S.(1995).Perceivedcontinuityofoccludedvisualobjects.PsychologicalScience,6:182–186.