Tracking unique objects
-
Upload
independent -
Category
Documents
-
view
0 -
download
0
Transcript of Tracking unique objects
Copyright 2007 Psychonomic Society Inc 172
WeallliveinadynamicenvironmentcharacterizedbymultipledistinctiveobjectsthatmovethroughtheworldInthestudyofvisualattentionhoweverwhenresearch-ershavestudieddistinctiveobjectsthoseobjectshavenotmoved(egHendersonampHollingworth2003OlivaTorralbaCastelhanoampHenderson2003WolfeOlivaHorowitzButcherampBompas2002)andwhentheyhaveinsteadexaminedtheabilitytoattendtomovingobjectsthoseobjectshavenotbeendistinctiveInthisarticlewewillexploretheroleofattentioninkeepingtrackofiden-tifiableobjectsastheymovethroughtheworld
Inrecentyearsthemultiple-object-tracking(MOT)task(PylyshynampStorm1988)hasemergedasaprimarytoolforstudyingdynamicattentionmdashthatisattentiontoobjectsovertimeInthestandardversionofMOTtheobserverisinitiallypresentedwithadisplayofeightidenticalitems
withfouroftheitemsblinkingonandofftoidentifythemastargetsAlloftheitemsthenmoveforseveralsecondsduringwhichtheobserverhastokeeptrackofthetargetsWhentheobjectsstoptheobserverisaskedtodiscriminatebetweentargetsandnontargetseitherbypointingoutallofthetargetsorbyindicatingwhetheraprobeitemisatargetoranontargetThetypicalresultisthathumanscanaccuratelytrackaboutfourtargetsalthoughperformancestronglyde-pendsonstimulusfactorssuchasthespeedoftheitems(Al-varezampFranconeri2004OksamaampHyoumlnauml2004)
TheMOTparadigmhasprovedtobeaversatileresearchtoolIthasbeenusedtostudythenatureofvisualobjects(Scholl amp Pylyshyn 1999 Scholl Pylyshyn amp Feld-man2001)thecapacityforchangedetection(Bahrami2003Saiki2002)andtheallocationofattentionindepth(ViswanathanampMingolla2002)Furthermoredespiteits
Tracking unique objects
Todd S HorowiTzBrigham and Womenrsquos Hospital Boston Massachusettsand Harvard Medical School Boston Massachusetts
SaraH B KliegerBrigham and Womenrsquos Hospital Boston Massachusetts
david e FencSiKBrigham and Womenrsquos Hospital Boston Massachusettsand Harvard Medical School Boston Massachusetts
Kevin K YangBoston College Chestnut Hill Massachusetts
george a alvarezHarvard University Cambridge Massachusetts
and
JeremY m wolFeBrigham and Womenrsquos Hospital Boston Massachusettsand Harvard Medical School Boston Massachusetts
Iscontentaddressableintherepresentationthatsubservesperformanceinmultiple-object-tracking(MOT)experimentsWedevisedanMOTvariantthatfeatureduniquenameableobjects(cartoonanimals)asstimuliThereweretwopossibleresponsemodesstandardinwhichobserverswereaskedtoreportthelocationsofalltargetitemsandspecificinwhichobservershadtoreportthelocationofaparticularobject(egldquoWhereisthezebrardquo)Ameasureofcapacityderivedfromaccuracyallowedforcomparisonsoftheresultsbetweencondi-tionsWefoundthatcapacityinthespecificcondition(14to26itemsacrossseveralexperiments)wasalwaysreliablylowerthancapacityinthestandardcondition(23to34items)Observerscouldlocatespecificobjectsindicatingacontent-addressablerepresentationHowevercapacitydifferencesbetweenconditionsaswellasdifferingresponsestotheexperimentalmanipulationssuggestthattheremaybetwoseparatesystemsinvolvedintrackingonecarryingonlypositionalinformationandonecarryingidentityinformationaswell
Perception amp Psychophysics2007 69 (2) 172-184
T S Horowitz toddhsearchbwhharvardedu
Tracking UniqUe ObjecTs 173
artificialnaturetheMOTparadigmseemstocapturethedemandsofimportantreal-worldchallengesForinstanceairtrafficcontrollersneedtotrackthechangingpositionsofmultipleplanesinanairspace(seeAllenMcGeorgePearsonampMilne2004)driversneedtopayattentiontoothercarsintheirimmediatevicinityathletesneedtotrackteammatesandopponentsanddaycareprovidersneedtobeawareofthemovementsofseveralchildrenatonce
AlthoughtheMOTtaskcanbeinformativeitdiffersfromsuchreal-worldtasksincriticalandrelatedwaysFirstobjectsintherealworld(carschildrenetc)arerarelyiden-ticaltheycandifferinvisualfeaturesaswellasincharac-teristicmotionSecondthetypeofreal-worldanswerthatwedemandofourvisualsystemistypicallydifferentfromtheanswerdemandedoftheobserverduringMOTInthestandardMOTtaskobserversareaskedtodifferentiatebe-tweentrackedtargetsanduntrackednontargetsHoweveroutintheworldwerarelyaskquestionslikeldquowhichchil-drenareyoutrackingnowrdquoInsteadweoftenwishtoknowthewhereaboutsofaspecificchildcarorplane
InthisarticleweseekevidenceforthisabilityinanMOTtaskIfanobserveristrackingfouruniquetargetsisheorsheabletodistinguishbetweenthosetargetsandreportthelocationofaparticularoneTherearereasonstoarguebothforandagainstsuchldquocontent-addressablerdquorepresentationsunderlyingMOTperformanceWewilladdressthequestionfromthethreedominanttheoreticalperspectives
AprominentapproachtoMOTisPylyshynrsquostheoryofvisualindexesorFINSTs(Pylyshyn1989)VisualindextheoryassumesthatMOTismediatedbyalimited-capacitypreattentiverepresentationconsistingofpointer-likeindexesthatcanbeattachedtoobjectsTheseindexesconveythelocationoftheobjectinquestionandprovidepriorityaccess toattention(SearsampPylyshyn2000)Theycanbelikenedtofingers(hencethetermFINSTfromldquof ingersofinstantiationrdquo)thatcanbeplacedonthetargetsAttheendofthetrialtheobserversimplyaskswhichobjectshisldquofingersrdquopointtoAswithfingersthereareonlyahandfulofindexesavailable
VisualindextheoryseemstobeambivalentaboutthequestionofcontentaddressabilityOneinterpretationofthetheoryholdsthatthetrackingrepresentationisnotcontentaddressablesincetheindexesonlytellyouwheretheyareTheydonotencodeobjectidentityorfeatureinformationAnotherinterpretationofthetheoryhoweveristhattherepresentationmustbecontentaddressablesinceinordertoknowthatitemXisatargetatanytimetiyoumustknowthatitisthesameitemthatwasatargetattimet0Pylyshyn(2004)referstothisasthediscrete reference principle
SofardatacollectedbyPylyshynandhiscolleagueshave supported the first interpretation For instanceSchollPylyshynandFranconeri(1999)showedthatwhenitemsstoppedmovingobserverswereabletoaccuratelyreportthepreviousdirectionandspeedoftargetsbutnotofnontargetsHoweverwhentheshapeorcoloroftheitemswasmaskedobserverswereunabletoaccuratelyreportthepremaskfeaturesofeithertargetsornontargetsFromtheseresultsScholletal(1999)proposedadistinctionbe-tweenspatiotemporalpropertiesofobjectssuchasspeed
directionandtrajectoryandfeaturalpropertiessuchascolororshapeAlthoughfeaturalpropertiescanchangeasobjectsmovethroughdifferentlightingenvironmentsandnonrigidtransformationsspatiotemporalpropertiesarethekeytoobjectcontinuityScholletal(1999)arguedthatvisualindexesencodeonlyspatiotemporalproperties
Inordertoexplicitlytestthediscretereferenceprin-ciplePylyshyn(2004)attachedidentitiestotargetsduringthetargetacquisitionphaseoftheMOTtrialeitherbypresentinganumberinsidethetargetdisksorbyhavingtargetdisksstartoffindifferentcornersofthedisplayTohissurprisehefoundthatobserverswereratherpooratreportingtheidentitiesparticularlythenumericallabelsevenwhentheyhadsuccessfullytrackedthetargetobjectsPartofthisdiscrepancyapparentlyresultedfromobserv-ersinadvertentlyswappingtargetidentitieswhentargetspassedclosetooneanotherPylyshyn(2004)alsospecu-latedthattheldquointernalnamerdquomightnotbeconsciouslyavailableinotherwordsalthoughsomepartofthevisualsystemwasawarethatagivendiskwasthesameobjectasatargetidentifiedduringtheacquisitionphasetherewasnowaytolinkthisinternalrepresentationwithanexternallabelthattheobservercouldreport
InsteadofhypothesizingvisualindexesYantis(1992)stressedtheimportanceofperceptualgroupingAccord-ingtoYantisobserversinanMOTtaskimaginethatthetargetitemsformtheverticesofavirtualpolygonYantishasdemonstratedthatwhenthetargetitemsmoveintoaconfigurationthatcollapsesthepolygontrackingsuf-fersIncontrasttothepreattentivevisualindexaccountYantisrsquosperceptualgroupingaccountassumesthattrack-ingismediatedbyaspatialworkingmemoryrepresen-tationOn thisviewholding items inspatialworkingmemoryservestomakethemitemsofspatialattentionaswell(AwhJonidesampReuter-Lorenz1998)Itisnotclearwhethersucharepresentationcouldbecontentad-dressableOntheonehandsincethisviewisbasedonatop-downstructurethataddsinformationratherthanapreattentivestructurethatdiscardsitonemightexpectthesystemtoknowwhichtargetiswhichOntheotherhandiftherepresentationispurelyspatialinnaturethesystemmightnotknow
FinallyKahnemanandTreismanrsquosobjectfiletheory(KahnemanampTreisman1984KahnemanTreismanampGibbs1992)hasoftenbeeninvokedtoexplainMOTre-sultsTheobjectfileconceptwasdevelopedtoexplaintheperceptualcontinuityofobjectsovertimeAswenotedaboveobjectschangetheirappearanceovertimeastheymovethroughspaceTheinterpretationgiventoanobjectcanalsochangeasthefeaturalinformationchangesThusasanobjectintheskyapproachesitcanbeperceivedasabirdthenaplanethenSupermanallwithoutlosingthesensethatasingleobjecthasbeenapproachingKahne-manandTreismanproposedthatasinglerepresentationtheobjectfileisresponsiblefortheperceptionofconti-nuityWhenanobjectisfirstattendedanobjectfileisopenedforitandallsubsequentinformationaboutthatobjectgoesintotheobjectfileWhentheobjectisat-tendedatsomelatertimeaprocessofimpletionrecallstheappropriateobjectfile
174 HOrOwiTz eT al
IncontrasttovisualindexesobjectfilesareclearlycontentaddressableIntheirseminalarticlepresentingevidencefortheobjectfileconceptKahnemanTreismanandGibbs(1992)developedthereviewingparadigmInthisparadigmobserversseeapreviewdisplaythatin-cludesanumberofobjects(atleasttwo)Theobjectsarecharacterizedbysomefeatureoftenaletter(seeegMitroffSchollampWynn2004)Thefeaturalinformationdisappearsandthensomesortoflinkingdisplayusu-allyconsistingofsmoothorapparentmotionconnectstheobjectsinthepreviewdisplaywithobjectsinatargetdisplayTheobserverthenhastorespondtothetargetdis-playforexamplebyidentifyingaletterKahnemanetalfoundthatiftheletterwasthesameastheonepresentedinthatobjectinthepreviewdisplayobserverswerefastertorespondthanifthatletterhadappearedinanotherobjectinthepreviewdisplayindicatingthatinformationaboutwhatwasinthatobjecthadbeenpreserved
HereweintroduceanewvariationonthebasicMOTtaskthatallowsustolookattrackingofuniqueobjects(arelatedparadigmwasdevelopedbyOksamaampHyoumlnauml2004wewillreturntothisintheDiscussionsection)Inourtasktheobjectsconsistedofuniquecartoonanimals(seeFigure1)Wechosethesestimulibecausetheywerevisuallydistinctandeasilynameable
Theprincipaltechnicalproblemwithsuchstimuliisthatobserverscouldsimplymemorizetheidentitiesofthetargetanimalsatthebeginningofthetrialthenre-spondonthebasisofthoseidentitiesattheendofthetrialwithoutactuallytrackingthetargetswhiletheymovedWemadethisstrategyimpossiblebyhidingtheanimalsbehindcartooncactusesattheendofthetrialThesecac-tuseswerepresentthroughoutthetrialandtheanimalsmovedinfrontofthemAttheendofthetrialthedepth
relationshipswereswitchedandtheanimalsmovedbe-hindthecactusesandremainedthereInotherworkwehavefoundthatthesuddenappearanceofpreviouslyin-visibleoccluderswhetheroneatatimeorallatoncedidnotdisrupttrackingperformance(HorowitzBirnkrantFencsikTranampWolfe2006seealsoSchollampPylyshyn1999)Thusweconsideritunlikelythatamerechangeinthedepthrelationsbetweentrackedanimalsandoccludingcactusesshouldbeproblematic
Asecondtechnicalproblemwiththisarrangementisthatiftheendofeachtrialwerepredictableobserverscouldanticipatethisandlocatethetargetsjustpriortooc-clusionTothwartsuchastrategywemadethetimefromtrackingonsettoocclusionvariablefromtrialtotrialTheanimalsmovedinsuchafashionthattheyallapproachedthestationarycactusesevery1333msecThenumberofsuchcyclesinatrialvariedrandomlysoifobserverswerenotingtargetlocationsonlyattimeswhentheywerelikelytobeoccludedbycactusestheywouldhavetosuccess-fullysearchthedisplayforthetrackedanimalsinaman-nerthatproducedcorrectsearchresultsevery1333msecThisseemsunlikely
AttheendofatrialweaskedobserversoneoftwoquestionsThesequestionswereblockedinExperiments1ndash3andmixedwithinblocksinExperiment4Thestan-dardquestionwasldquoWhereareallthetargetsrdquoObserv-ershadtoclickonthecactusesinfrontofeachtargetThespecificquestionwasldquoWhereisthe_____rdquowiththeblankfilledbyoneofthetargetanimalsTheobserverthenhadtoclickonthecactuswherethatparticularani-malwasldquohidingrdquo
ThesetwoquestionsobviouslyhavedifferentratesofchanceperformanceInordertocompareperformanceonthesetwodifferentquestionswecomputedacommonmetricofcapacitybasedonthenumberofitemstrackedcorrectedforguessing(thedetailsaredescribedintheMethodsection)
InExperiment1wefoundthatcapacitywassubstan-tiallylowerforthespecificquestionthanforthestandardquestioninablockeddesignThisldquocontentdeficitrdquowasreplicatedinfourmoreexperimentsExperiment2ruledoutthepossibilitythatthecontentdeficitisduetointerfer-enceinshort-termmemoryInExperiment3weconfirmedthatthedatadonotchangeifweincreasethevariabilityofthetrackingdurationExperiment4replicatedthecontentdeficitwithamixeddesignusingbothvisualandauditorypostcuesTheremainingexperimentsemployedonlythestandardquestionbuttookadvantageofthepossibilitiesopenedbyourmethodExperiments5and6measuredhowmuchadditionalcapacityisgainedbyusinguniqueratherthanidenticalstimuliFinallyExperiment7measuredca-pacityinaldquofullreportrdquoversionofthestandardtask
MeTHod
observersUnlessotherwisespecifiedeachexperimentinvolved8observers
recruitedfromtheBrighamandWomenrsquosHospitalVisualAtten-tionLaboratoryvolunteerpoolTheobserversrangedinagefrom18to55years(M5283yearsSD597)1allhadvisualacuity
Figure 1 Cartoon animals used as stimuli See Appendix A for a list of stimulus names
Tracking UniqUe ObjecTs 175
ofatleast2025whencorrectedandhadpassedtheIshiharacolorscreenTheobserversgaveIRB-approvedinformedconsentandwerecompensated$10hforparticipationTwelveoftheobserversparticipatedinmorethanoneexperiment
Apparatus and StimuliThevisualstimuliwerepresentedon21-incolorCRTmoni-
tors(SuperScanMc801RasterOpsandMitsubishiDiamondPro91TXM)controlledbyPowerMacintoshG4computers(MacOS922)runningMATLAB521usingthePsychophysicsToolboxroutines(Brainard1997Pelli1997)Monitorspatialresolutionwassetto10243768pixelsThedisplayareasubtended361ordm3271ordmofvisualangleataviewingdistanceof574cmMonitorre-freshratesweresetto75Hz(133msecperframe)
The tracking stimuliwere23 cartoon animals (Figure1 seeAppendixAforalist)in8-bitcolorOccludersweretwotypesofcartooncactusesCartoonimageswerescaledsothatthemaximumdimension(horizontalorvertical)was35ordmforanimalsand53ordmforcactusesThebackgroundwasyellow
Cactuseswereplacedina534gridGridcellswere62ordmonasidesocactuseswereseparatedby09ordmedgetoedgeThetwotypesofcactuseswereassignedtocellsatrandomoneachtrialsothebackgroundcactusarraywasdifferentfromtrialtotrial
Auditoryprobes(Experiments45and6)wererecordedinafemalevoiceandplayedthroughapairofSonyMDRV150headphones
ProcedureOneachtrialeightanimalswereplacedinlocationsrandomly
selectedfromthegrid(Figure2A)Theselectionoftheanimalsvar-iedbyexperimentAtthestartofatrialthetargetanimalsblinkedonandoffsixtimesat1HzTherewerefourtargetsinExperiments1ndash6andeightinExperiment7TheanimalsthenbegantomoveForeachanimaladestinationcactuswasselectedatrandomwithoutre-placementAnimalsmovedtowardtheirdestinationsinstraightlinesatdifferentspeedssuchthateachanimalreacheditsdestinationatthesametime(Figure2B)Eachcycleconsistedof100monitorrefreshesor1333msecThemaximumspeedforananimalwas297ordmsec21(corner-to-cornermovement)andtheminimumspeed
Figure 2 Critical frames from the procedure Note that although we depict 3 3 4 grids in order to save space the grids actually used in the experiments were 5 3 4 At the beginning of a cycle each animal was in front of a cactus (A) each animal was then assigned a different cactus as a destination (b) each trajectory was completed in the same amount of time (one 1333-msec cycle) leading to different speeds for each animal on each cycle each animal ended the cycle in front of the destination cactus (C) except on the last cycle on which the animal was occluded by the cactus (d) for 200 msec Ani-mals were then removed for a 133-msec interval followed by the response phase To see an example of a trial go to searchbwhharvardedunewMoviesnoahmov
176 HOrOwiTz eT al
was46ordmsec21(movementtoanadjacentcactus)Oncethedestina-tionswerereachedanothersetofdestinationswasselectedAtrialconsistedof10ndash15ofthesecycles(exceptforExperiment3whichused5ndash15)ThenumberofcycleswasselectedrandomlyoneachtrialAnimalscouldoccludeoneanotherandmovedinfrontofthebackgroundcactusarray
AttheendofthelastcyclethecactusarraywasmovedtothefrontsuchthattheanimalswereinstantlyoccludedPartsofsomeanimalswouldbevisiblethroughtransparentportionsofthecac-tusimages(Figure2D)After200msectheoccludedanimalswereerasedeliminatinganycuestotheiridentitiesAfteranadditional133msecthecomputercursorappearedandaprobeinstructionwaspresentedatthetopofthescreen
InthestandardconditiontheinstructionwasldquoPleaseclickonthebusheswhereallofthetargetanimalsarehidingrdquoObserversrespondedbymovingthecursortoclickontheappropriatecactusesThecactusthatthecursorwascurrentlyoverwashighlightedwitharedborderOncetheobserverhadmadeanumberofresponsesequaltothenumberoftargetsfeedbackwaspresentedIfalltargetswerecorrectlyidentifiedthemessageldquoYougotallofthemrdquowasprintedatthetopofthescreenOtherwisetheobserverwastoldhowmanytargetsweremissedandthemissedtargetsblinkedonandofffourtimesattheappropriatelocations
InthespecificconditiontheprobeinstructionwasldquoWhereisthetargetrdquowheretargetwasrandomlyselectedfromthenamesofthetargetsforthattrialThetargetimagewasalsopre-sentedattheupperrightofthescreenTheobserverrsquostaskinthisconditionwastomovethecursortothecactuswheretheselectedtargethadbeenattheendofthetrialandclickTheanimalbehindtheselectedcactuswasthenrevealedFeedbackindicatedwhethertheobserverhadselectedthecorrectcactusorwhethertherewasnoanimalbehindthatcactus
Observerstypicallyparticipatedintwoblocksof50trialseachprecededby5practicetrialsWhenconditionswereblockedblockorderwascounterbalancedacrossobserversTheexperimentstypi-callytookbetween1and1-12htocomplete
data AnalysisRawaccuracydataweretransformedaccordingtohigh-threshold
guessingmodelsdevisedforeachconditionInthestandardcondi-tiontheguessingmodel(seeEquation1)wasderivedfromtheonepresentedbyScholletal(2001)WeassumedthatperformancePintermsofthenumberoftargetscorrectlyidentifiedwasdeter-minedbythenumberofobjectsactuallytracked(kforcapacity)plusguessingInEquation1tisthenumberoftargetsandathenumberofpossibleresponseoptionsinthiscasecactusesTheobservermakeskinformedresponsesandthenguessesonthere-maining(t2k)targetsleaving(t2k)targetsand(a2k)possibleresponsesPerformanceisthereforegivenby
P kt k
a k= +
minus( )minus
2
(1)
solvingforkgives
k aP ta P t
= minus+ minus
2
2 (2)
PerformanceinthespecificconditionwasmodeledaccordingtosimilarlogicHereperformancepistheprobabilityofcorrectlyse-lectingthecactuswherethespecifictargetishidingTheprobabilitythattheprobeanimal(selectedatrandomfromthettargets)willbeoneofthekobjectsbeingtrackedisktSoonkttrialsperformanceis10Ontheremaining(12kt)trialstheobserverguessesInthiscasetheobservercaneliminatetheklocationswithtrackedanimalsbecausesheknowstheycontainananimalthatwasnotprobedsosheguessesfromamongtheremaining(a2k)locationsHenceperformanceisgivenbyEquation3
p kt
kt a k
= + minus( ) minus( )1 1 (3)
Whenweagainsolvefork
ka pt a pt apt t
=+ minus minus minus minus( ) minus minus( )1 1 4
2
2
(4)
Equations2and4convertperformanceinbothconditionstoacom-monmetrick
Therawdataunderlyingthecapacitycomputationsforallex-perimentsarereportedinAppendixBThemajorityofouranalysesweresingle-degree-of-freedomplannedcomparisonscarriedoutthroughpairedttests
exPeriMeNT 1
InthisexperimentweintroducetheparadigmandthebasicresultthecontentdeficitInthisversionoftheex-perimentthesetof8uniqueanimalstobepresentedoneachtrialwasselectedatrandomfromthe23availableanimalsAsubsetof4targetanimalswasthenselectedatrandomfromamongthese8NoattemptwasmadetocontroloverlapinsetsfromtrialtotrialThestandardandspecificconditionswereruninseparateblocks
Estimatedcapacitywas16items(SEM503items)inthespecificconditionand29items(SEM503)inthestandardconditionThereweretwonotableresultsinthisexperimentFirstobserverswerereliablyabletoreportontheidentityofmorethanoneiteminthespecificcondition[t(7)550p001one-tailed]indicatingacontent-addressablerepresentationSecondestimatedcapacitywaslowerinthespecificconditionthaninthestandardcondition[t(7)580p0001two-tailed]ThatiswhenaskedtoreportthelocationofaspecifictargetobserversappearedtohavetrackedfeweritemsthanwhentheywereaskedtosimplyclickonallofthetargetlocationsWewillcallthisresultthecontent deficitSubsequentexperimentswilltestpossibleexplanationsforthisdeficit
OnepuzzlingaspectofthesedatawasthatperformanceinthestandardconditionseemedlowinthisexperimentincomparisonwithtypicalMOTexperimentswithidenticalitemsObserversaretypicallyabletotrackmorethanthreeitems(seeOksamaampHyoumlnauml2004Scholletal2001)Our taskmayhavebeenmoredifficultsimplybecausetheanimalsoftenoccludedoneanotherwhichcanreduceperformance(KliegerHorowitzampWolfe2004)ThereisalsoamoreseriouspossibilitySincethestimulussetchangedfromtrialtotrialobserversmayhaveexperiencedinterferenceinworkingmemoryInparticularitemsthatweretargetsononetrialcouldhaveservedasnontargetsonsubsequenttrialsandviceversasuchconditionshaveledtointerferenceinboththevisualsearch(egldquovariablemappingrdquoSchneiderampShiffrin1977)andnegativeprim-ing(MillikenTipperampWeaver1994Tipper1985)para-digmsSuchinterferencemightbeparticularlyproblematicinthespecificconditionThusthechangingstimulussetsmightberesponsibleforboththecontentaddressabilityandthecontentdeficitfromthedataExperiment2wasdesignedtoremedythisproblem
exPeriMeNT 2
Inthisexperimentwesimplyfixedthetargetandnon-targetsetssothatthesamestimuliwereusedthroughout
Tracking UniqUe ObjecTs 177
ablockoftrialsThatisanobservermighttracktheliontherabbittheturtleandthegoatonallthetrialsinablockThismadeiteasierforobserverstorememberwhattheyweresupposedtotrackInadditioniteliminatedthepossibilityof interferenceornegativeprimingarisingfromtargetsandnontargetsswitchingrolesWeexpectedthatperformanceinthestandardconditionwouldimprovetoabove30itemsthequestionwaswhetherthecontentdeficitwouldstillmanifestitselfundertheseconditions
TheprocedurewasidenticaltothatofExperiment1exceptintheselectionofanimalsForeachobserveronesetofeightanimalswasselectedrandomlyatthebegin-ningofeachblockFouroftheseanimalsweredesignatedastargetsandtherestasnontargetsThesetofanimalsandthetargetnontargetassignmentswereheldconstantforthedurationofablock
Asexpectedperformanceinthestandardconditionimprovedtoacreditable34(SEM502)itemsHow-ever the content deficit was still present Observerstrackedonly21(SEM501)itemswhenaskedwhichtargetwaswhereThisdeficitwasreliable[t(7)583p00001]andsimilarinmagnitudetothatobservedinExperiment1ofaround13itemsAmixedANOVAcomparingthetwoexperiments2revealedalargeeffectof condition [F(18)5 1152p 000001]butnoneofexperiment[F(18)510p 10]Theinteractionterminthebetween-experimentsanalysiswasnearzero[F(18)1]
Theroughlyhalf-itemincreaseincapacitywhenwechangedfromavariabletoafixedtargetsetsuggeststhatswitchingtargetsetsfromtrialtotrialcreatedsomesortofinterferenceThiswastruenotonlyinthespecificcon-ditionwhenweprobedtargetidentitiesbutalsointhestandardconditionThisresultisimportantbecauseitsuggestsaroleforobjectidentity(eitherattheperceptualorsemanticlevel)irrespectiveoftaskdemandsAccord-ingtoLeonardandPylyshyn(2003)blinkingthetargetsonandoffshouldautomaticallyassignvisualindexestothemandtheseindexesshouldstickequallywellwhetherornottheseparticularstimuliweretargetsontheprevioustrialSincetheincreaseincapacitywasnotstatisticallyreliablethisconclusionshouldnotbetakentooseriouslyHoweverinsubsequentexperimentsweusedafixedtar-getset
JustasinExperiment1theobserversherecouldreli-ablylocatemultipletargetsaccordingtotheiridentitieswhichwetakeasevidenceforcontentaddressabilityEx-periment2alsoreplicatedthecontentdeficitwithreli-ablybettercapacityinthestandardreportconditionthaninthespecificreportcondition
TheseresultsraiseanumberofmethodologicalissuesFirstourexperimentsdifferfrommostMOTexperimentsintheirtemporalstructureourtrialswerebothlongerthanusualandincludedtemporaluncertaintyHowdidthesefactorsaffectperformanceExperiment3wasdesignedtoaddressthisissue
Asecondimportantissueiswhetherblockingresponsemodesinducedobserverstousedifferentencodingstrate-giesinthetwoconditionsleadingtodifferentialperfor-manceThisissuewillbeaddressedinExperiment4
AthirdissueiswhethertheuseofuniqueidentifiableobjectshelpsorhindersMOTperformanceincomparisonwiththeuseofidenticalobjectsinmostpreviousstudiesofMOTWewilladdressthisissueinExperiments5and6
FinallyExperiment7willaddresstheroleofnontar-getsinhinderingtrackingperformance
exPeriMeNT 3
Aswenotedintheintroductionitisimportanttoen-surethatobserversareattendingtoataskthroughoutatrialOneofourstrategiesweusedtoensurethiswastorandomlyvary trialdurationObserversdidnotknowwheneachtrialwouldendbetweenthe10thand15thcyclesWasthisenoughtemporaluncertaintyhoweverInExperiment3weincreasedtheuncertaintysothatatrialcouldendanywherebetweenthe5thand15thcycles(adurationfrom67to200sec)OtherwisethemethodwasidenticaltothatofExperiment2IfobserverswereactivelytrackingthetargetsonlyafterCycle9inthepre-viousexperimentsweshouldseeadecreaseinperfor-manceinthisexperimentwheretheypotentiallyhavetosustainattentionforalongerperiodoftime
Awiderarrayoftrialendpointsalsoallowedustoex-amineapossibleexplanationforthecontentdeficitWeusedtrialdurationsthatwerelongrelativetopreviousMOTstudiesPerhapslocationndashidentitybindingsdecayovertimeandwewouldobserveasmallerdifferencebe-tweenconditionsifweprobedearlierinthetrial
Figure3showskinExperiment3asafunctionofcyclewithdatafromExperiment2shownforcomparisonMeancapacityforthestandardconditionwas34items(SEM501)versus26items(SEM502)forthespecificcondi-tionrepresentingasignificantadvantageinthestandardcondition[t(7)539p 01]Wealsoconductedananalysisofcovarianceontheresultswithconditionasacategoricalvariableandcycleasacontinuousvariable(covariate)Neitherthemaineffectofcycle[F(17)1]northecycle3conditioninteraction[F(17)5249p516]wassignificant
5 7 9 11 13 150
1
2
3
4
Cycles
k (it
ems)
Figure 3 Capacity as a function of cycle Squares denote the standard and triangles the specific target condition data from experiment 3 are plotted in black and those from experiment 2 in gray error bars indicate standard errors of the means
178 HOrOwiTz eT al
WhencomparingthetwoexperimentsitseemsclearthatincreasingthetemporalvariabilityinExperiment3didnotchangetheresultsnotablyWeconductedacross-experimentANOVAusingdataonlyfromCycles10ndash15inExperiment3Standardversusspecifictargetconditionandcyclewerewithin-subjectsfactorsandexperimentwasabetween-subjectsfactorWefoundnomaineffectofexperiment[F(114)511p 5 31]Thecondition3ex-perimentinteractionapproachedsignificance[F(114)541p506]probablybecausethespecifictargetcapac-itywassomewhathigherinExperiment3Addingtempo-raluncertaintyclearlydidnotdisruptperformanceinthisexperimentinrelationtoExperiment2
ThereisahintinFigure3thatperformanceespeciallyin thespecific targetconditiondroppedas trialdura-tionincreasedThiswouldbecompatiblewithdatafromOksamaandHyoumlnauml(2004)possiblyreflectingavigilancedecrement(Parasuraman1986)IfthedropovertimeweregreaterinthespecificconditionthiswouldsupporttheideathatlocationndashidentitybindingsweredecayingHowevertherewasnoevidenceforthisinanANOVAconductedondatafromalloftheconditionsofExperiment3inwhichneitherthemaineffectofcycle[F(1070)511p538]northecycle3conditioninteraction[F(1070)1]wassignificantIftherewasdecayithadapproachedasymp-totewithinthefirst6ndash7secoftracking
exPeriMeNT 4
OneobvioushypothesistoexplainthecontentdeficitisthatobserversencodeinformationdifferentlyinthetwoconditionsInthestandardconditiononlyspatiotemporalpropertiesareencodedInthespecifictargetconditionobserversknowthattheywillbeaskedaboutfeaturaloridentityinformationsotheyencodethataswellFeaturalinformation(orperhapsthebindingofidentityandlo-cation)takesupmoreresources(Bahrami2003Saiki2002) reducing tracking capacityThus informationaboutfeaturesoridentitiesrequiresmorecapacity
InExperiment4wetestedthisstrategicencodinghy-pothesisbymixingthequestionswithinablockoftri-alsObserversdidnotknowonagiventrialwhethertheywouldbeaskedtolocateallofthetargetsorjustaspecifictargetThereforetheyhadtoadoptthesamestrategiesonbothtypesoftrialsIfthestrategicencodinghypothesisiscorrectthenthecontentdeficitshouldbereducedinthisexperimentPerformanceshouldalsobereducedoverallsinceobserverswouldnotbeoptimallyencodinginfor-mationoneachtrialInotherrespectsthemethodwasidenticaltothatofExperiment2
Pilotworkindicatedthatwhenquestionsweremixedwithinablockof trialshavingto lookawayfromthecactusarraytoreadthequestionatthetopofthescreenimpairedperformanceThereforeinthisexperimentweusedauditoryratherthanvisualprobesInsteadofprintingthequestionatthetopofthescreenobserversheardtheprobequestionsoverheadphones
Meancapacityforthestandardconditionwas34items(SEM501)asopposedto22items(SEM502)forthe
specificconditionrepresentingasignificantadvantageforthestandardcondition[t(7)599p00005]Per-formanceinthisexperimentwithquestionsmixedwithinblockswasnearlyidenticaltothatinExperiment2withblockedquestionsEvenwhenobserverscouldnothaveadopteddifferentencodingstrategies for thedifferentconditionswestillobservedasizable(12-item)contentdeficitThissuggeststhatthecontentdeficitisnotaresultofthestrategicdemandsoftheconditionswedevised
exPeriMeNT 5
ThepreviousexperimentsestablishedaparadigmforstudyingMOTwithunique stimuliOnequestion thatwehadnotyetaddressediswhetheruniqueobjectshelporhinderMOTperformanceincomparisonwithidenticalobjectsForexampleiftheobjectsareidenticalanob-servermightaccidentallyswapatargetandadistractoriftheyapproachtoocloselyWithuniqueobjectsthisislesslikelytohappenSimilarlyifatsomepointtheobserverrealizedthathewasonlytrackingthreetargetsandhadlostonehecouldtheoreticallyrecoverthelosttargetifitwereuniqueOfcoursesuchadvantagesassumeasortofldquoidealobserverrdquowhoalwaysknowsexactlywhatheistrackingIfthetrackingsystemwascompletelyinsensitivetotargetfeaturesoridentityuniqueobjectswouldprovidenoadvan-tageSimilarlytheobservercouldonlyrecoveralosttargetifinformationwereavailableabouthowmanytargetswerecurrentlybeingtrackedandwhichonewasmissing
Inthisexperimentwecomparedtrackinguniqueob-jectswithtrackingidenticalobjectsSinceldquoWhereisthezebrardquoisnotadiagnosticquestionwhenallobjectsarezebrasweonlyusedthestandardresponsemodeTherewerethreeconditionsTheuniqueconditionwasidenti-caltothestandardresponseconditioninExperiment1Intheidenticalconditioneightidenticalanimalswerepresented animal identitywas selectedat randomoneachtrialFinallyinthepairedconditiontherewerefouruniquetargetsagainselectedatrandomForeachtargettherewasanidenticaldistractorProbeswerepresentedauditorilyasinExperiment4Atotalof10observerspar-ticipatedinthisexperiment
Wecancompareperformanceintheuniqueandidenti-calobjectconditionstodeterminewhetherthereisanyadvantageforuniqueobjectsIftheadvantageforuniqueobjectsderivesfromtheabilitytorecoverlosttargetsthisadvantageshouldbeabolishedinthepairedcondition
CapacityvaluesforthethreeconditionsareplottedinFigure4Observerstracked31items(SEM502)intheuniquecondition23(SEM502)inthepairedconditionand25(SEM501)intheidenticalconditionPlannedttestsshowedthatthepairedandidenticalconditionsdidnotdiffer[t(9)513p 10]butthatperformancewassignificantlyworseinboththanintheuniquecondition[forpairedt(9)527p05foridenticalt(9)527p01]
TheseresultsindicatethatobserverscantakeadvantageoftheadditionalinformationprovidedbyuniqueobjectsTheuniqueobjectadvantage(approximately06items
Tracking UniqUe ObjecTs 179
undertheseconditions)isentirelyeliminatediftargetsareidenticaltodistractorssuggestingthatsomesortoferrorrecoveryprocessisresponsible
exPeriMeNT 6
In the previous experiment pairing targets anddistractorsreducedperformanceinthestandardcondi-tionrelativetothatinthefullyuniqueconditionWouldthismanipulationaffectthespecificconditionaswellInExperiment6wemeasuredthecontentdeficitwithbothpairedanduniquestimuliBycomparingthesetwostimulusconditionswecandeterminewhethereliminat-ing featuraldistinctionsbetween target andnontargetsetsimpairstheabilitytotrackspecificobjectsaswellastheabilitytoholdontotargetlocationsProbeswerepresentedauditorilyasinExperiment5Therewere6ob-serversinthisexperiment
AsshowninFigure5weonceagainreplicated thecontentdeficitinthisexperimentsinceobserverscouldtrack11moreitemsinthestandardconditionthaninthe
specificcondition[F(19)51190p001]Pairingtar-getsandnontargetsreducedcapacitybyroughlythesameamountas inExperiment5 [09 itemsF(19)5373p001]Criticallythepairingmanipulationappearedtohavelessofadeleteriouseffectinthespecificconditionthaninthestandardcondition[interactionF(19)5185p005]Inthestandardconditionthepairingmanipu-lationcostobservers12itemsversusalossofonly07itemsinthespecificconditionThisinteractionsuggeststhatdifferentrepresentationsmayunderlieperformanceinthetwoconditions
Ontheotherhandcapacitywasalreadylowerinthespe-cificconditionIftheeffectofpairingstimuliismultipli-cativeratherthanadditivewewouldthenexpectthattheimpactofpairedstimuliwouldbeproportionallysmallerinthespecificconditionWhenwelog-transformedthecapacityvaluesconvertingproportionaldifferencesintoadditive ones the question3 pairing interaction waseliminated[F(19)1]
Sincewecannotrejectthehypothesisthattheeffectofpairingtargetswithnontargetsisproportionaltocapacitythisexperimentprovidesatbestweakevidencefordiffer-entrepresentations
exPeriMeNT 7
OneadvantageofourmethodisthatwecantestMOTperformanceinacasewithonlytargetsWhywouldwewanttodothisTherearereasonstobelievethatsuppress-ingnontargetsconsumessomeresourcesthatwouldother-wisebeavailablefortrackingtargetsForexamplethereisevidencethatobserverstrackbetterontrialsinwhichnontarget trajectories are repeated from earlier trials(OgawaampYagi2003)IfnontargetsareprocessedtheymayhavetobesuppressedinordertoreduceconfusionwithtargetsandPylyshynandLeonard(2003)showedevidenceforsuchinhibitionMoreoverunpublisheddatafromourlaboratoryindicatethatMOTperformancede-creasesasnontargetsareaddedtothedisplayevenwhenthenumberoftargetsremainsconstant
Accordingly we designed Experiment7 to test thehypothesisthattrackingcapacitywouldincreaseiftherewerenonontargetsThisexperimentwasidenticaltoEx-periment2exceptthatallitemsweretargetsCapacityinthestandardconditionwas58items(SEM502)sig-nificantlygreaterthanthe26items(SEM501)obtainedinthespecificcondition[t(7)5143p000005]
InagreementwithourhypothesiscapacityimprovedmarkedlyinthestandardconditionwhentherewerenodistractorsStandardcapacityinExperiment7wassig-nificantlygreaterthaninExperiments23and4whichweremethodologicallycomparableexceptforthenumberoftargetsandnontargets(allps00005)
Oneexplanationisthatpreviousexperimentsunderes-timatedcapacitybecauseofaceilingeffectTheseexperi-mentswerebasedonthehypothesisthatcapacityvariesacrosstrialswithnounderlyingdifferencesacrossexperi-mentsHoweverthemeasuredcapacityhasbeenlimitedbythenumberoftargetsForexampleifthereareonlyfour
Unique Paired Identical0
1
2
3
4
Condition
k (it
ems)
Figure 4 Capacity as a function of condition in experiment 5
Specific Standard0
1
2
3
4Unique
Paired
Condition
k (it
ems)
Figure 5 effect of pairing target and nontarget objects on ca-pacity in experiment 6 data from the paired conditions are plot-ted as gray bars and those from the unique conditions as open bars
180 HOrOwiTz eT al
targets(egExperiments2ndash4)observerscouldtrackallofthemontrialsinwhichtheircapacitywas4orgreateranytrialswithanactualcapacitygreaterthan4wouldthusappeartohaveacapacityof4InthiscasemeasuredmeancapacitywouldunderestimatethetruemeanInExperi-ment7whichincludedeighttargetsmeasuredcapacitycouldreflectactualcapacityuptoeightitemsleadingtoagreater(andmoreaccurate)estimateofmeancapacity
Wecantestthisceilinghypothesisagainstourhypoth-esisofincreasedmeancapacitybycomparingthecapacitydistributionsinExperiment7withthoseinExperiments2ndash4Accordingtotheceilinghypothesistheproportionoftrialswithacapacitylessthan4shouldbeidenticalacrossexperimentsFurthermoretheproportionoftrialswithacapacityof4orgreaterinExperiment7shouldbethesameastheproportionoftrialswithacapacityequalto4inExperiments2ndash4Incontrastthehypothesisthattrackingcapacityincreaseswhennonontargetsarepres-entpredictsthattheentiredistributionshouldshifttotherightleadingtofewertrialswithcapacitylessthan4inExperiment7thanintheearlierexperiments
We tested these hypotheses by computing capacityforeachtrialusingEquation2andgeneratingadistri-butionacrosstrialsforeachobserverFigure6plotsthedistributionofkforExperiment7andthecorrespondingdistributionforExperiments23and4combined(dis-tributionsfromeachoftheseexperimentswereaveragedacrossobservers)Clearlytheproportionoftrialswithanobservedcapacitylessthan4waslowerwitheighttargetsthanwithfourtargetsThisisconsistentwiththecapacitydistributionshiftingtotherightinExperiment7ratherthanmerespreadingofthek54trialsoverawiderrangeWesuggestthatthesuperiorperformanceinthisexperi-mentmayreflectadditionalcapacityfreedupwhenthereisnoneedtosuppressnontargetsHowevertheshapeofthedistributionforfourtargetsdoessuggestthatcapacitywasunderestimatedinthepreviousexperiments
TheconclusionsarequitedifferentifwelookatthespecifictargetconditionSpecifictargetcapacityinthis
experimentwasbetterthaninExperiment2(uncorrectedttestp05)butnotsignificantlydifferentfromtheca-pacitiesobservedinExperiments3and4( p05)Thusthetrackingoftargetidentitiesinthisconditiondoesnotseem tobe limitedby theneed to suppressnontargetidentities
diSCuSSioN
WehavedevelopedanewMOTparadigmthatallowsustotestthetrackingofuniqueobjectsFourmainfind-ingshaveemergedfromtheseexperimentsFirstthereisacontent-addressablerepresentationoftargetsduringMOT(specificcapacitywas1inallcases)SecondweobservedacontentdeficitThecontent-addressablerep-resentationhasalowercapacity(measuredinitems)thanthelocation-addressablerepresentation(iespecificca-pacitywaslowerthanstandardinallcases)Thirdthevi-sualsystemiscapableoftakingadvantageofdifferencesbetweenuniqueobjectstoimprovetrackingperformanceoverthelevelseenwithidenticalobjects(Experiment5)Thisadvantageappearstobedueprimarilytotheabilitytorecoverlosttargets(Experiment5)Fourthtrackingcapacityappearstobeconstrainedbytheneedtosup-pressnontargetswhentherearenonontargetscapacityincreasesmarkedly(Experiment7)
A Content-Addressable representation in MoTOurdata represent anexistenceproofof a content-
addressable representation inMOTSimilar data havebeenreportedbyOksamaandHyoumlnauml(2004)Intheirmul-tiple-identity-tracking(MIT)taskobserverstrackedmov-inglinedrawings(eitherobjectsorldquopseudo-objectsrdquo)forseveralsecondsAttheendofthetrialthestimuliweremaskedandasingleitemwasmarkedwithablackframeInasubsequentresponsescreenobservershadtoselectthepicturethatcorrespondedtothemarkedobjectOksamaandHyoumlnaumlobtainedamediancapacityoffouritems
Althoughitisdifficulttocomparecapacitymeasuresdirectlyacrossexperimentaltasks3itisinterestingthatbothweandOksamaandHyoumlnauml(2004)demonstratedthatobserversdoknowtosomeextentwhichtargetiswhichThisfindingisincontrasttothoseofPylyshyn(2004)whoseobservershaddifficultyidentifyingtargetsTwokeydifferencesbetweenourparadigmandMITontheonehandandPylyshynrsquos(2004)paradigmontheotherseemrelevanttothisdiscrepancyFirstinourstudyandthatofOksamaandHyoumlnaumlobjectidentitiesweredefinedbyintrinsicfeaturessuchasshapeand(inourexperi-ment)colorInPylyshynrsquos(2004)experimentldquoidentityrdquowasbasedonatangentialfeatureofanobjectsuchasinwhichcornerofthedisplayitbeganthetrialItseemslikelythattheshapeandcolorofanobjectmaybemoretightlyboundmarkersofldquoidentityrdquothaninitialpositionorabrieflypresentedletter
SecondinourexperimentsandthoseofOksamaandHyoumlnauml (2004) thevisualdifferencesbetweenobjectswerecontinuallyavailablethroughoutthetrackingphaseofthetrialonlymaskedduringtheresponsephaseIn
4 targets8 targets
0 2 4 6 80
1
2
3
4
5
6
Capacity
Pro
po
rtio
n o
f Tri
als
Figure 6 Capacity distributions for the standard condition in experiment 7 (gray lines striped area) and in experiments 2 3 and 4 (black lines stippled area) dashed lines indicate standard errors of the means
Tracking UniqUe ObjecTs 181
Pylyshynrsquos(2004)experimentstheobjectidentitieswerepresentedonlybrieflyatthestartofthetrialduringthemajorityofthetrackingphasetheobjectswerephysicallyidenticalItmaybethatthelinkbetweenobjectidentityandspatiotemporalpositiondecayedduringPylyshynrsquos(2004)taskbutwasconstantlyrefreshedinbothoursandOksamaandHyoumlnaumlrsquos
Ourexperimentsalsoprovidedsomeindirectevidenceforacontent-addressablerepresentationForexampleperformancewasconsistentlysuperior inexperimentsinwhichthesameobjectswereusedinthesamerolesfromtrialtotrial(Experiments23and4)thaninex-perimentsinwhichthestimulussetchangedfromtrialtotrial(Experiments15and6)Thisoccurrednotonlyinthespecifictargetconditionsinwhichobjectidentitywasrelevantbutalsointhestandardconditionsinwhichobjectidentitywasnot
one System or TwoIstrackingservedbyasinglerepresentationthatencodes
bothpositionandidentityoraretheretworepresentationsoneforpositionandoneforidentityThisquestionarisesbecauseofthestrikingcapacitydifferencesbetweenourresponseconditionsInallofourexperimentscapacitywassignificantlylowerforspecifictargetsthanfortargetsinthestandardMOTresponsemodeInExperiment4wedisconfirmedthehypothesisthatcapacitydifferencesresultedfromvariedencodingstrategies(seeegDavisWelchHolmesampShepherd2001)sinceweobtainedthesameresultwhenresponsemodesweremixedwithinablockaswedidwhentheywereruninseparateblocks
ThiscapacitydifferenceisnotconclusiveevidencethatseparaterepresentationsorsystemsareinvolvedThereareseveralwaysinwhichaunifiedaccountcouldex-plainthelowercapacityinthespecifictargetconditionForexampleonemightproposeasystemwithaslot-likestructure(VogelWoodmanampLuck2001)inwhichallslotsarenotcreatedequalIfthereweretwoslotswithsuf-ficientresolutiontoencodebothpositionandidentityandtwoslotsthatcouldonlyencodepositionwewouldalsoobtaindifferentcapacityestimatesinourtworesponseconditionsThisrepresentationcouldalsobedescribedintermsofKahnemanandTreismanrsquos(1984)objectfilesifweassumethatsomeoftheobjectfilesareemptyTheimpletionprocessmightinthatcasereturntheinforma-tionthatagivenobjectwasspatiotemporallycontinuouswithatargetobjectbutfailtoretrievethesemanticorperceptualcontentassociatedwiththatobjectfile
IntheMOTliteraturecapacityisgenerallyassumedtorefertoafixednumberofavailablerepresentationsinthevisualsystemeithervisualindexesorobjectfilesHow-everthereissomeevidencethatcapacityinMOTmightbebetterthoughtofasanundifferentiatedresource(AlvarezampCavanagh2005)InsteadofasetofslotsorpointerswemightassumethatthereisafixedamountofinformationthatcanbeencodedObserversmightencodepositionandidentityforalltargetsbutthebindingbetweenpositionsandidentitymaybenoisierthanthepositioninformationitself(ormaydecaymorequicklyExperiment3indicatedthatanysuchdecaywouldhavetoapproachitsasymptote
valuewithin6ndash7sec)leadingtoapparentlylowercapacitywhenresponsesdependonbinding
InsteadofonesystemtherealsomightbetwoAlthoughwetendtoassumethatasingleexperimentalparadigmprobestheperformanceofasinglecorrespondingfunc-tionalsystemitmaybethatMOTnaturallyrecruitsmulti-pleoverlappingsystemsThesimplesttwo-systemaccountmightbeaproposalthattrackingisservedbyaposition-trackingsystemsuchasvisualindexes(FINSTs)butthataccessingidentityndashpositionbindingsrequiresfocalatten-tion(WheelerampTreisman2002)Howeversincespecificcapacityisreliablygreaterthanoneitemwewouldhavetoassumethatmultipleitemscansimultaneouslybethefocusofattention(AwhampPashler2000Davisetal2001)
IfwewishtopreserveasinglefocusofattentionwemightinsteadproposethatbothvisualindexesandobjectfilescanbeusedVisualindexeswouldprovideonlypo-sitioninformationandobjectfilescouldsupportmorecomplexqueries
Itwillobviouslybedifficulttodefinitivelydecidebe-tweenoneandtwosystemsHowevertheevidenceofourfinalthreeexperimentspointstotwoindependentsystemsInExperiments5and6pairingtargetsandnontargetsre-ducedcapacityinthestandardbutnotinthespecifictar-getcondition(althoughwecouldnotrejectthehypothesisofaproportionaldecrease)InExperiment7standardca-pacityincreasedwhennonontargetswerepresentedbutspecificcapacitywasunaffected
MOThasbeenlikenedtoavisualworkingmemorytaskattimezero(CavanaghampAlvarez2005)soitisnaturaltolooktostudiesofworkingmemoryforanalogsofthecontent-addressableandlocation-onlysystemsTheclassictwo-pathwayschemeinvisionisthedistinctionbetweenventralanddorsalstreamsassociatedwithobjectrecogni-tionandspatialprocessingrespectively(UngerleiderampMishkin1982)ThisdistinctionhasbeenextendedintothedomainofworkingmemorybyWilsonOacuteScalaidheandGoldman-Rakic(1993)whoprovidedevidencethatobjectinformationandspatialinformationareneurallysegregatedinmonkeyprefrontalcortexwidelybelievedtobetheneuralsubstrateofworkingmemoryAsimilardissociationinhumanswasreportedusingPETimagingbySmithetal(1995)
Thedivisionofworkingmemoryintomultiplesubsys-temsgoesbackatleasttothemultiple-componentmodelofBaddeleyandHitch(1974)whichproposedseparatecomponents for different modalities a ldquophonologicallooprdquoforauditoryinformationandaldquovisuospatialsketchpadrdquoforvisualinformationInmorerecentversionsofthemodel(BaddeleyampLogie1999Logie1995)thesketchpadhasbeendividedintovisualandspatialsubcompo-nents roughlycorrespondingto theobjectandspatialsystemsdescribedbyWilsonetal(1993)Interestinglythespatialsubcomponent(theldquoinnerscriberdquo)isspecifi-callytaskedwithholdingsequenceorpathinformation(seeParmentierElfordampMayberry2005)makingitalikelysubstrateforMOTperformance
Thewidespreadinsistenceonaseparationbetweenvi-sualorobjectinformationandspatialinformationwouldseem problematic for our findings Standard MOT is
182 HOrOwiTz eT al
clearlyajobforthespatialsubsystem(PostleDrsquoEspositoampCorkin2005forexampleusedaversionofstandardMOTtotieupdorsalprocessinginamemoryinterferenceparadigm)Howevertheredoesnotseemtoberoominthataccountforacontent-addressablerepresentationOurspecifictargettask(aswellastheMITtaskofOksamaampHyoumlnauml2004)isnotapureobjectrecognitiontaskitrequireslocationndashidentitybindingswhichwouldseemtoinvolveintegratinginformationacrossstreamsTheproblemdisappears ifwe take intoaccount recentbe-havioral(OlsonampMarshuetz2005)andfMRI(PostleampDrsquoEsposito1999)investigationsdemonstratingthatobjectworkingmemoryrepresentationsalsocarrysomelocationinformation
NeuroimagingstudiesmightproveusefulhereStan-dardMOThasbeenshowntoactivateparietallobeareasinthedorsalstream(Culhametal1998Jovicichetal2001Seiffert2003)Itwouldbeinformativetoknowwhetherthepatternofactivationchangeswhenuniqueob-jectsareintroducedorwhenthetaskpotentiallyinvolvesobjectidentities
unique object Features in MoTInadditiontoourfindingsaboutcontentaddressability
intrackingwewereabletodemonstratethathavingvisu-allyuniqueobjectsmadetrackingeasierobserverswereabletotrackroughly06moreitemswhenobjectswereuniquethanwhenobjectswereidenticalInExperiments5and6weattributedthisresulttoanerrorrecoverystrat-egyIfIlosetrackofthezebraIcansearchforthezebraonthebasisofitsphysicalfeaturesandcontinuetotrackitThisstrategywouldobviouslybeuselessifallitemsarezebrasThisstrategyisoflimitedutilityifthereareazebratargetandazebradistractoronceIlosethezebraIthenhaveonlya50chanceofpickingupthetargetzebraasopposedtothedistractor
TheerrorrecoveryhypothesisisnottrivialIthassev-eralimportantimplicationsFirstitimpliesthatobserversknowwhattheyarenottrackingThephenomenologyofMOTissuchthatanobserverhasthefeelingofknow-inghowmanytargetsheorsheissuccessfullytrackingTheerrorrecoveryhypothesisimpliesthatwhenatargetislosttheobserverknowswhichone(givenuniquetargets)SecondtheerrorrecoveryhypothesisalsoassumesthatanobservercansearchforamissingtargetwhilecontinuingtotracktheremainingtargetsThisisinagreementwithrecentevidencethatobserverscansuccessfullyperformanMOTtaskandavisualsearchtaskonthesametrial(AlvarezHorowitzArsenioDiMaseampWolfe2005)Finallythehypothesisimpliestheabilitytoaddtothetrackingsetontheflywhiletrackingwithoutanyphysi-calcueTheseimplicationsneedtobeconfirmedthroughfurtherresearch
Howeverifobserversuseuniquephysicalfeaturesofobjectstorecoverlosttargetsinthisfashionitiscuriousthatthereisnosuchbenefitinthespecifictargetcondi-tionIfobserversdorecoverlosttargetsduringthecourse
ofatrialtheyappeartorecoveronlytheirlocationsnottheiridentities
AnalternativetotheerrorrecoveryhypothesisthoughnotamutuallyexclusiveoneisthattheresultsofEx-periment5canbeexplainedbytheperceptualsimilarityoftargetandnontargetitemsYantisrsquos(1992)approachtoMOThighlightstheimportanceofperceptualgroup-ingfactorsthatallowthetargetstobetreatedasaunitbythevisualsystemMostoftheresearchinthislinehasfocusedonspatiotemporalgroupingfactorssuchasthespatialrelationshipsbetweentheobjects(SearsampPylyshyn2000ViswanathanampMingolla2002Yantis1992)
Cangroupingby(nonspatial)featuralpropertiesim-provetrackingAnumberofstudieshavedemonstratedthatobservershave limitedexplicit access to featuralproperties (Bahrami 2003Saiki 2002Scholl etal1999)butthisfindingdoesnotaddressthedirectrolethatfeaturesmayplayWecouldexplaintheresultsofEx-periment5byproposingthatobserverswereabletouseaccidentaldifferencesintheshapeandcolorstatisticsoftargetsandnontargetstohelpsegregatethemvisuallyInthepairedconditiontargetsandnontargetshadidenticalfeaturestatisticssothatadvantagewaslostThishypoth-esiscouldalsoexplainwhyperformancewasbetterwithafixedstimulussetthanwithavariablestimulussetoveranumberoftrialswiththesamestimuliobserverslearnsubtledifferencesbetweenthetwosets
ConclusionsIntheintroductionwenotedthatreal-worlddynamic
attentiontasksdifferfromtraditionalMOTintwowaysFirstobserversintheeverydayworldusuallydealwithmultipleuniqueobjectsSecondtheyoftenwishtoknowwhereagiventargetmightberatherthanwhichobjectsarethetargetsTheexperimentspresentedhererepresentanattempttobridgelaboratoryMOTresearchandtheseeverydaycognitivetasksWehaveshownthatitiseasiertotrackuniqueobjectsthantotrackidenticalobjectsassum-ingthattheobjectsonewishestotrackarealsouniquelydifferentfromnontargetobjectsFurthermoreobserversdoknowwhattheyaretrackingTotakeoneofourex-amplesofecologicallyvalidtrackingtasksifyouwatchyourchildrenandtheirfriendsontheplaygroundyoucanbeawareofwhereyourkidsareatanygivenmomentyoumayalsobeawareofthelocationsoftheirfriendswithoutbeingabletotellonefriendfromtheotherFurthermoreifyoulosetrackofachildyouwillprobablybeawareofwhichoneyouhavelost
AuTHor NoTe
ThisresearchwassupportedbyNIMHGrantR0165576toTSHWethankKristinMichodandSkylerPlaceforassistancewithdatacollec-tionaswellasSteveLuckandananonymousreviewerforconstructivecriticismCorrespondencerelatingtothisarticlemaybesenttoTSHorowitzBrighamandWomenrsquosHospitalVisualAttentionLaboratory64SidneyStreetSuite170BostonMA02139(e-mailtoddhsearchbwhharvardedu)
Tracking UniqUe ObjecTs 183
reFereNCeS
Allen R McGeorge P Pearson D amp Milne A B (2004)AttentionandexpertiseinmultipletargettrackingApplied Cognitive Psychology18337-347
Alvarez G A amp Cavanagh P (2005)IndependentresourcesforattentionaltrackingintheleftandrightvisualhemifieldsPsychologi-cal Science16637-643
Alvarez G A amp Franconeri S (2004November)How many ob-jects can you trackPaperpresentedatthe12thAnnualWorkshoponObjectPerceptionampMemoryMinneapolisMN
Alvarez G A Horowitz T S Arsenio H C DiMase J S amp Wolfe J M (2005)DomultielementvisualtrackingandvisualsearchdrawcontinuouslyonthesamevisualattentionresourcesJournal of Experimental Psychology Human Perception amp Perfor-mance31643-667
Awh E Jonides J amp Reuter-Lorenz P A (1998)RehearsalinspatialworkingmemoryJournal of Experimental Psychology Human Perception amp Performance24780-790
Awh E amp Pashler H (2000)EvidenceforsplitattentionalfociJournal of Experimental Psychology Human Perception amp Perfor-mance26834-846
Baddeley A D amp Hitch G J (1974)WorkingmemoryInGHBower(Ed)The psychology of learning and motivation Advances in research and theory(Vol8pp47-90)NewYorkAcademicPress
Baddeley A D amp Logie R H (1999) Working memoryThemultiple-componentmodelInAMiyakeampPShah(Eds)Models of working memory Mechanisms of active maintenance and executive control(pp28-61)CambridgeCambridgeUniversityPress
Bahrami B (2003)ObjectpropertyencodingandchangeblindnessinmultipleobjecttrackingVisual Cognition10949-963
Brainard D H (1997)ThePsychophysicsToolboxSpatial Vision10433-436
Cavanagh P amp Alvarez G A (2005)TrackingmultipletargetswithmultifocalattentionTrends in Cognitive Sciences9349-354
Culham J C Brandt S A Cavanagh P Kanwisher N G Dale A M amp Tootell R B H (1998)CorticalfMRIactiva-tionproducedbyattentive trackingofmoving targetsJournal of Neurophysiology802657-2670
Davis G Welch V L Holmes A amp Shepherd A (2001)CanattentionselectonlyafixednumberofobjectsatatimePerception301227-1248
Henderson J M amp Hollingworth A (2003)EyemovementsandvisualmemoryDetectingchangestosaccadetargetsinscenesPer-ception amp Psychophysics6558-71
Horowitz T S Birnkrant R S Fencsik D E Tran L amp Wolfe J M (2006)HowdowetrackinvisibleobjectsPsychonomic Bulletin amp Review13516-523
Jovicich J Peters R J Koch C Braun J Chang L amp Ernst T (2001)Brainareasspecificforattentionalloadinamotion-trackingtaskJournal of Cognitive Neuroscience131048-1058
Kahneman D amp Treisman A (1984)ChangingviewsofattentionandautomaticityInRParasuramanampDRDavies(Eds)Varieties of attention(pp29-61)OrlandoAcademicPress
Kahneman D Treisman A amp Gibbs B J (1992)ThereviewingofobjectfilesObject-specificintegrationofinformationCognitive Psychology24175-219
Klieger S B Horowitz T S amp Wolfe J M (2004)Ismultipleobjecttrackingcolorblind[Abstract]Journal of Vision4(8)363a
Leonard C amp Pylyshyn Z W (2003)Measuringtheattentionaldemandofmultipleobjecttracking(MOT)[Abstract]Journal of Vi-sion3(9)582a
Logie R H (1995) Visuo-spatial working memory Hove UKErlbaum
Milliken B Tipper S P amp Weaver B (1994)NegativepriminginaspatiallocalizationtaskFeaturemismatchinganddistractorin-hibitionJournal of Experimental Psychology Human Perception amp Performance20624-646
Mitroff S R Scholl B J amp Wynn K (2004)DivideandconquerHowobjectfilesadaptwhenapersistingobjectsplitsintotwoPsy-chological Science15420-425
Ogawa H amp Yagi A (2003)Primingeffectsinmultipleobjecttrack-ingAnimplicitencodingbasedonglobalspatiotemporalinformation[Abstract]Journal of Vision3(9)339a
Oksama L amp Hyoumlnauml J (2004)Ismultipleobjecttrackingcarriedoutautomaticallybyanearlyvisionmechanismindependentofhigher-ordercognitionAnindividualdifferenceapproachVisual Cognition11631-671
Oliva A Torralba A Castelhano M S amp Henderson J M (2003)Top-downcontrolofvisualattentioninrealworldscenes[Ab-stract]Journal of Vision3(9)3a
Olson I R amp Marshuetz C (2005)RememberingldquowhatrdquobringsalongldquowhererdquoinvisualworkingmemoryPerception amp Psychophysics67185-194
Parasuraman R (1986)VigilancemonitoringandsearchInKRBoffLKaufmanampJPThomas(Eds)Handbook of perception and human performance Vol 2 Cognitive processes and performance(pp1-39)NewYorkWiley
Parmentier F B R Elford G amp Mayberry M (2005)Transi-tionalinformationinspatialserialmemoryPathcharacteristicsaffectrecallperformanceJournal of Experimental Psychology Learning Memory amp Cognition31412-427
Pelli D G (1997)TheVideoToolboxsoftwareforvisualpsychophysicsTransformingnumbersintomoviesSpatial Vision10437-442
Postle B R amp DrsquoEsposito M (1999)ldquoWhatrdquondashthenndashldquowhererdquoinvi-sualworkingmemoryAnevent-relatedfMRIstudyJournal of Cog-nitive Neuroscience11585-597
Postle B R DrsquoEsposito M amp Corkin S (2005)Effectsofver-balandnonverbalinterferenceonspatialandobjectvisualworkingmemoryMemory amp Cognition33203-212
Pylyshyn Z [W] (1989)Theroleoflocationindexesinspatialper-ceptionAsketchoftheFINSTspatial-indexmodelCognition3265-97
Pylyshyn Z W (2004)SomepuzzlingfindingsinmultipleobjecttrackingITrackingwithoutkeepingtrackofobjectidentitiesVisual Cognition11801-822
Pylyshyn Z W amp Leonard C (2003)Inhibitionofnontargetsdur-ingmultipleobjecttracking(MOT)[Abstract]Journal of Vision3(9)585a
Pylyshyn Z W amp Storm R W (1988)Trackingmultipleindepen-denttargetsEvidenceforaparalleltrackingmechanismSpatial Vi-sion3179-197
Saiki J (2002)Multiple-objectpermanencetrackingLimitationinmaintenanceandtransformationofperceptualobjectsProgress in Brain Research140133-148
Schneider W amp Shiffrin R M (1977)ControlledandautomatichumaninformationprocessingIDetectionsearchandattentionPsychological Review841-66
Scholl B J amp Pylyshyn Z W (1999)Trackingmultiple itemsthroughocclusionCluestovisualobjecthoodCognitive Psychology38259-290
Scholl B J Pylyshyn Z W amp Feldman J (2001)Whatisavi-sualobjectEvidencefromtargetmerginginmultipleobjecttrackingCognition80159-177
Scholl B J Pylyshyn Z W amp Franconeri S L (1999May)When are spatiotemporal and featural properties encoded as a result of attentional allocationPaperpresentedattheconferenceoftheAssociationforResearchinVisionandOphthalmologyFortLau-derdaleFL
Sears C R amp Pylyshyn Z W (2000)MultipleobjecttrackingandattentionalprocessingCanadian Journal of Experimental Psychol-ogy541-14
Seiffert A E (2003)Dissociatingneuralcorrelatesofattentionaltrackingandattentiontovisualmotion[Abstract]Journal of Vision3(9)868a
Smith E E Jonides J Koeppe R A Awh E Schumacher E H amp Minoshima S (1995)SpatialversusobjectworkingmemoryPETinvestigationsJournal of Cognitive Neuroscience7337-356
Tipper S P (1985)ThenegativeprimingeffectInhibitoryprimingbyignoredobjectsQuarterly Journal of Experimental Psychology37A571-590
Ungerleider L G amp Mishkin M (1982)Twocorticalvisualsys-
184 HOrOwiTz eT al
temsInDJIngleMAGoodaleampRJWMansfield(Eds)Anal-ysis of visual behavior(pp549-586)CambridgeMAMITPress
Viswanathan L amp Mingolla E (2002)DynamicsofattentionindepthEvidence frommulti-element trackingPerception311415-1437
Vogel E K Woodman G F amp Luck S J (2001)Storageoffea-turesconjunctionsandobjectsinvisualworkingmemoryJournal of Experimental Psychology Human Perception amp Performance2792-114
Wheeler M E amp Treisman A M (2002)Bindinginshort-termvisualmemoryJournal of Experimental Psychology General13148-64
Wilson F A Oacute Scalaidhe S P amp Goldman-Rakic P S (1993)Dissociationofobjectandspatialprocessingdomains inprimateprefrontalcortexScience2601955-1958
Wolfe J M Oliva A Horowitz T S Butcher S J amp Bompas A (2002)SegmentationofobjectsfrombackgroundsinvisualsearchtasksVision Research422985-3004
Yantis S (1992)MultielementvisualtrackingAttentionandpercep-tualorganizationCognitive Psychology24295-340
NoTeS
1Wearemissingageinformationfor1observer2Threeobserverswhoparticipatedinbothexperimentswereelimi-
natedfromthisanalysis3Measuredcapacitycanvarybasedontrackingdurationtherange
ofmovementspeedsandwhetherornotobjectscanoccludeonean-other(asinourexperiment)orbounceoffoneanother(asinOksamaampHyoumlnauml2004)
APPeNdix A Stimulus List
beaver fox horse ostrich rhino turtlecamel giraffe kangaroo pelican rooster wolfcrocodile goat lion rabbit squirrel zebraelephant gorilla moose tiger
APPeNdix b raw data
Table b1 raw Numbers of items reported in the Standard Condition
Total Targets Nontargets Filled
Experiment1 308Experiment2 344 014 359Experiment3 347 015 362Experiment4 339 015 354Experiment5 Unique 312 037 349
Paired 249 080 329Identical 261 086 347
Experiment6 248 078 326Experiment7 599 599
NotemdashTargetsindicatestheaveragenumberofcorrectlyreportedtargetspertrialNontargetstheaveragenumberoftimesobserversclickedonanontargetlocationpertrialandTotal Filledthesumofthesetwocat-egoriesNontargetclickcountswerenotcollectedforExperiment1andtherewerenonontargetsinExperiment7
Table b2 raw Proportions for the Specific Condition by experiment
Errors
Correct Empty Trialsper Responses Transposition Nontarget Space N Observer
Experiment1 44 14 16 26 8 5000Experiment2 56 18 10 16 8 5000Experiment3 68 12 10 10 8 5000Experiment4 58 25 09 08 8 5063Experiment6 41 14 21 23 7 5000Experiment7 33 39 00 28 8 5000
NotemdashTranspositionerrorsrefertotrialsinwhichtheobserverclickedonthewrongtargetlocationTrialsperobservervariedinExperiment4becauseconditionwasselectedrandomlyfromtrialtotrial
(ManuscriptreceivedMarch312005revisionacceptedforpublicationFebruary132006)
Tracking UniqUe ObjecTs 173
artificialnaturetheMOTparadigmseemstocapturethedemandsofimportantreal-worldchallengesForinstanceairtrafficcontrollersneedtotrackthechangingpositionsofmultipleplanesinanairspace(seeAllenMcGeorgePearsonampMilne2004)driversneedtopayattentiontoothercarsintheirimmediatevicinityathletesneedtotrackteammatesandopponentsanddaycareprovidersneedtobeawareofthemovementsofseveralchildrenatonce
AlthoughtheMOTtaskcanbeinformativeitdiffersfromsuchreal-worldtasksincriticalandrelatedwaysFirstobjectsintherealworld(carschildrenetc)arerarelyiden-ticaltheycandifferinvisualfeaturesaswellasincharac-teristicmotionSecondthetypeofreal-worldanswerthatwedemandofourvisualsystemistypicallydifferentfromtheanswerdemandedoftheobserverduringMOTInthestandardMOTtaskobserversareaskedtodifferentiatebe-tweentrackedtargetsanduntrackednontargetsHoweveroutintheworldwerarelyaskquestionslikeldquowhichchil-drenareyoutrackingnowrdquoInsteadweoftenwishtoknowthewhereaboutsofaspecificchildcarorplane
InthisarticleweseekevidenceforthisabilityinanMOTtaskIfanobserveristrackingfouruniquetargetsisheorsheabletodistinguishbetweenthosetargetsandreportthelocationofaparticularoneTherearereasonstoarguebothforandagainstsuchldquocontent-addressablerdquorepresentationsunderlyingMOTperformanceWewilladdressthequestionfromthethreedominanttheoreticalperspectives
AprominentapproachtoMOTisPylyshynrsquostheoryofvisualindexesorFINSTs(Pylyshyn1989)VisualindextheoryassumesthatMOTismediatedbyalimited-capacitypreattentiverepresentationconsistingofpointer-likeindexesthatcanbeattachedtoobjectsTheseindexesconveythelocationoftheobjectinquestionandprovidepriorityaccess toattention(SearsampPylyshyn2000)Theycanbelikenedtofingers(hencethetermFINSTfromldquof ingersofinstantiationrdquo)thatcanbeplacedonthetargetsAttheendofthetrialtheobserversimplyaskswhichobjectshisldquofingersrdquopointtoAswithfingersthereareonlyahandfulofindexesavailable
VisualindextheoryseemstobeambivalentaboutthequestionofcontentaddressabilityOneinterpretationofthetheoryholdsthatthetrackingrepresentationisnotcontentaddressablesincetheindexesonlytellyouwheretheyareTheydonotencodeobjectidentityorfeatureinformationAnotherinterpretationofthetheoryhoweveristhattherepresentationmustbecontentaddressablesinceinordertoknowthatitemXisatargetatanytimetiyoumustknowthatitisthesameitemthatwasatargetattimet0Pylyshyn(2004)referstothisasthediscrete reference principle
SofardatacollectedbyPylyshynandhiscolleagueshave supported the first interpretation For instanceSchollPylyshynandFranconeri(1999)showedthatwhenitemsstoppedmovingobserverswereabletoaccuratelyreportthepreviousdirectionandspeedoftargetsbutnotofnontargetsHoweverwhentheshapeorcoloroftheitemswasmaskedobserverswereunabletoaccuratelyreportthepremaskfeaturesofeithertargetsornontargetsFromtheseresultsScholletal(1999)proposedadistinctionbe-tweenspatiotemporalpropertiesofobjectssuchasspeed
directionandtrajectoryandfeaturalpropertiessuchascolororshapeAlthoughfeaturalpropertiescanchangeasobjectsmovethroughdifferentlightingenvironmentsandnonrigidtransformationsspatiotemporalpropertiesarethekeytoobjectcontinuityScholletal(1999)arguedthatvisualindexesencodeonlyspatiotemporalproperties
Inordertoexplicitlytestthediscretereferenceprin-ciplePylyshyn(2004)attachedidentitiestotargetsduringthetargetacquisitionphaseoftheMOTtrialeitherbypresentinganumberinsidethetargetdisksorbyhavingtargetdisksstartoffindifferentcornersofthedisplayTohissurprisehefoundthatobserverswereratherpooratreportingtheidentitiesparticularlythenumericallabelsevenwhentheyhadsuccessfullytrackedthetargetobjectsPartofthisdiscrepancyapparentlyresultedfromobserv-ersinadvertentlyswappingtargetidentitieswhentargetspassedclosetooneanotherPylyshyn(2004)alsospecu-latedthattheldquointernalnamerdquomightnotbeconsciouslyavailableinotherwordsalthoughsomepartofthevisualsystemwasawarethatagivendiskwasthesameobjectasatargetidentifiedduringtheacquisitionphasetherewasnowaytolinkthisinternalrepresentationwithanexternallabelthattheobservercouldreport
InsteadofhypothesizingvisualindexesYantis(1992)stressedtheimportanceofperceptualgroupingAccord-ingtoYantisobserversinanMOTtaskimaginethatthetargetitemsformtheverticesofavirtualpolygonYantishasdemonstratedthatwhenthetargetitemsmoveintoaconfigurationthatcollapsesthepolygontrackingsuf-fersIncontrasttothepreattentivevisualindexaccountYantisrsquosperceptualgroupingaccountassumesthattrack-ingismediatedbyaspatialworkingmemoryrepresen-tationOn thisviewholding items inspatialworkingmemoryservestomakethemitemsofspatialattentionaswell(AwhJonidesampReuter-Lorenz1998)Itisnotclearwhethersucharepresentationcouldbecontentad-dressableOntheonehandsincethisviewisbasedonatop-downstructurethataddsinformationratherthanapreattentivestructurethatdiscardsitonemightexpectthesystemtoknowwhichtargetiswhichOntheotherhandiftherepresentationispurelyspatialinnaturethesystemmightnotknow
FinallyKahnemanandTreismanrsquosobjectfiletheory(KahnemanampTreisman1984KahnemanTreismanampGibbs1992)hasoftenbeeninvokedtoexplainMOTre-sultsTheobjectfileconceptwasdevelopedtoexplaintheperceptualcontinuityofobjectsovertimeAswenotedaboveobjectschangetheirappearanceovertimeastheymovethroughspaceTheinterpretationgiventoanobjectcanalsochangeasthefeaturalinformationchangesThusasanobjectintheskyapproachesitcanbeperceivedasabirdthenaplanethenSupermanallwithoutlosingthesensethatasingleobjecthasbeenapproachingKahne-manandTreismanproposedthatasinglerepresentationtheobjectfileisresponsiblefortheperceptionofconti-nuityWhenanobjectisfirstattendedanobjectfileisopenedforitandallsubsequentinformationaboutthatobjectgoesintotheobjectfileWhentheobjectisat-tendedatsomelatertimeaprocessofimpletionrecallstheappropriateobjectfile
174 HOrOwiTz eT al
IncontrasttovisualindexesobjectfilesareclearlycontentaddressableIntheirseminalarticlepresentingevidencefortheobjectfileconceptKahnemanTreismanandGibbs(1992)developedthereviewingparadigmInthisparadigmobserversseeapreviewdisplaythatin-cludesanumberofobjects(atleasttwo)Theobjectsarecharacterizedbysomefeatureoftenaletter(seeegMitroffSchollampWynn2004)Thefeaturalinformationdisappearsandthensomesortoflinkingdisplayusu-allyconsistingofsmoothorapparentmotionconnectstheobjectsinthepreviewdisplaywithobjectsinatargetdisplayTheobserverthenhastorespondtothetargetdis-playforexamplebyidentifyingaletterKahnemanetalfoundthatiftheletterwasthesameastheonepresentedinthatobjectinthepreviewdisplayobserverswerefastertorespondthanifthatletterhadappearedinanotherobjectinthepreviewdisplayindicatingthatinformationaboutwhatwasinthatobjecthadbeenpreserved
HereweintroduceanewvariationonthebasicMOTtaskthatallowsustolookattrackingofuniqueobjects(arelatedparadigmwasdevelopedbyOksamaampHyoumlnauml2004wewillreturntothisintheDiscussionsection)Inourtasktheobjectsconsistedofuniquecartoonanimals(seeFigure1)Wechosethesestimulibecausetheywerevisuallydistinctandeasilynameable
Theprincipaltechnicalproblemwithsuchstimuliisthatobserverscouldsimplymemorizetheidentitiesofthetargetanimalsatthebeginningofthetrialthenre-spondonthebasisofthoseidentitiesattheendofthetrialwithoutactuallytrackingthetargetswhiletheymovedWemadethisstrategyimpossiblebyhidingtheanimalsbehindcartooncactusesattheendofthetrialThesecac-tuseswerepresentthroughoutthetrialandtheanimalsmovedinfrontofthemAttheendofthetrialthedepth
relationshipswereswitchedandtheanimalsmovedbe-hindthecactusesandremainedthereInotherworkwehavefoundthatthesuddenappearanceofpreviouslyin-visibleoccluderswhetheroneatatimeorallatoncedidnotdisrupttrackingperformance(HorowitzBirnkrantFencsikTranampWolfe2006seealsoSchollampPylyshyn1999)Thusweconsideritunlikelythatamerechangeinthedepthrelationsbetweentrackedanimalsandoccludingcactusesshouldbeproblematic
Asecondtechnicalproblemwiththisarrangementisthatiftheendofeachtrialwerepredictableobserverscouldanticipatethisandlocatethetargetsjustpriortooc-clusionTothwartsuchastrategywemadethetimefromtrackingonsettoocclusionvariablefromtrialtotrialTheanimalsmovedinsuchafashionthattheyallapproachedthestationarycactusesevery1333msecThenumberofsuchcyclesinatrialvariedrandomlysoifobserverswerenotingtargetlocationsonlyattimeswhentheywerelikelytobeoccludedbycactusestheywouldhavetosuccess-fullysearchthedisplayforthetrackedanimalsinaman-nerthatproducedcorrectsearchresultsevery1333msecThisseemsunlikely
AttheendofatrialweaskedobserversoneoftwoquestionsThesequestionswereblockedinExperiments1ndash3andmixedwithinblocksinExperiment4Thestan-dardquestionwasldquoWhereareallthetargetsrdquoObserv-ershadtoclickonthecactusesinfrontofeachtargetThespecificquestionwasldquoWhereisthe_____rdquowiththeblankfilledbyoneofthetargetanimalsTheobserverthenhadtoclickonthecactuswherethatparticularani-malwasldquohidingrdquo
ThesetwoquestionsobviouslyhavedifferentratesofchanceperformanceInordertocompareperformanceonthesetwodifferentquestionswecomputedacommonmetricofcapacitybasedonthenumberofitemstrackedcorrectedforguessing(thedetailsaredescribedintheMethodsection)
InExperiment1wefoundthatcapacitywassubstan-tiallylowerforthespecificquestionthanforthestandardquestioninablockeddesignThisldquocontentdeficitrdquowasreplicatedinfourmoreexperimentsExperiment2ruledoutthepossibilitythatthecontentdeficitisduetointerfer-enceinshort-termmemoryInExperiment3weconfirmedthatthedatadonotchangeifweincreasethevariabilityofthetrackingdurationExperiment4replicatedthecontentdeficitwithamixeddesignusingbothvisualandauditorypostcuesTheremainingexperimentsemployedonlythestandardquestionbuttookadvantageofthepossibilitiesopenedbyourmethodExperiments5and6measuredhowmuchadditionalcapacityisgainedbyusinguniqueratherthanidenticalstimuliFinallyExperiment7measuredca-pacityinaldquofullreportrdquoversionofthestandardtask
MeTHod
observersUnlessotherwisespecifiedeachexperimentinvolved8observers
recruitedfromtheBrighamandWomenrsquosHospitalVisualAtten-tionLaboratoryvolunteerpoolTheobserversrangedinagefrom18to55years(M5283yearsSD597)1allhadvisualacuity
Figure 1 Cartoon animals used as stimuli See Appendix A for a list of stimulus names
Tracking UniqUe ObjecTs 175
ofatleast2025whencorrectedandhadpassedtheIshiharacolorscreenTheobserversgaveIRB-approvedinformedconsentandwerecompensated$10hforparticipationTwelveoftheobserversparticipatedinmorethanoneexperiment
Apparatus and StimuliThevisualstimuliwerepresentedon21-incolorCRTmoni-
tors(SuperScanMc801RasterOpsandMitsubishiDiamondPro91TXM)controlledbyPowerMacintoshG4computers(MacOS922)runningMATLAB521usingthePsychophysicsToolboxroutines(Brainard1997Pelli1997)Monitorspatialresolutionwassetto10243768pixelsThedisplayareasubtended361ordm3271ordmofvisualangleataviewingdistanceof574cmMonitorre-freshratesweresetto75Hz(133msecperframe)
The tracking stimuliwere23 cartoon animals (Figure1 seeAppendixAforalist)in8-bitcolorOccludersweretwotypesofcartooncactusesCartoonimageswerescaledsothatthemaximumdimension(horizontalorvertical)was35ordmforanimalsand53ordmforcactusesThebackgroundwasyellow
Cactuseswereplacedina534gridGridcellswere62ordmonasidesocactuseswereseparatedby09ordmedgetoedgeThetwotypesofcactuseswereassignedtocellsatrandomoneachtrialsothebackgroundcactusarraywasdifferentfromtrialtotrial
Auditoryprobes(Experiments45and6)wererecordedinafemalevoiceandplayedthroughapairofSonyMDRV150headphones
ProcedureOneachtrialeightanimalswereplacedinlocationsrandomly
selectedfromthegrid(Figure2A)Theselectionoftheanimalsvar-iedbyexperimentAtthestartofatrialthetargetanimalsblinkedonandoffsixtimesat1HzTherewerefourtargetsinExperiments1ndash6andeightinExperiment7TheanimalsthenbegantomoveForeachanimaladestinationcactuswasselectedatrandomwithoutre-placementAnimalsmovedtowardtheirdestinationsinstraightlinesatdifferentspeedssuchthateachanimalreacheditsdestinationatthesametime(Figure2B)Eachcycleconsistedof100monitorrefreshesor1333msecThemaximumspeedforananimalwas297ordmsec21(corner-to-cornermovement)andtheminimumspeed
Figure 2 Critical frames from the procedure Note that although we depict 3 3 4 grids in order to save space the grids actually used in the experiments were 5 3 4 At the beginning of a cycle each animal was in front of a cactus (A) each animal was then assigned a different cactus as a destination (b) each trajectory was completed in the same amount of time (one 1333-msec cycle) leading to different speeds for each animal on each cycle each animal ended the cycle in front of the destination cactus (C) except on the last cycle on which the animal was occluded by the cactus (d) for 200 msec Ani-mals were then removed for a 133-msec interval followed by the response phase To see an example of a trial go to searchbwhharvardedunewMoviesnoahmov
176 HOrOwiTz eT al
was46ordmsec21(movementtoanadjacentcactus)Oncethedestina-tionswerereachedanothersetofdestinationswasselectedAtrialconsistedof10ndash15ofthesecycles(exceptforExperiment3whichused5ndash15)ThenumberofcycleswasselectedrandomlyoneachtrialAnimalscouldoccludeoneanotherandmovedinfrontofthebackgroundcactusarray
AttheendofthelastcyclethecactusarraywasmovedtothefrontsuchthattheanimalswereinstantlyoccludedPartsofsomeanimalswouldbevisiblethroughtransparentportionsofthecac-tusimages(Figure2D)After200msectheoccludedanimalswereerasedeliminatinganycuestotheiridentitiesAfteranadditional133msecthecomputercursorappearedandaprobeinstructionwaspresentedatthetopofthescreen
InthestandardconditiontheinstructionwasldquoPleaseclickonthebusheswhereallofthetargetanimalsarehidingrdquoObserversrespondedbymovingthecursortoclickontheappropriatecactusesThecactusthatthecursorwascurrentlyoverwashighlightedwitharedborderOncetheobserverhadmadeanumberofresponsesequaltothenumberoftargetsfeedbackwaspresentedIfalltargetswerecorrectlyidentifiedthemessageldquoYougotallofthemrdquowasprintedatthetopofthescreenOtherwisetheobserverwastoldhowmanytargetsweremissedandthemissedtargetsblinkedonandofffourtimesattheappropriatelocations
InthespecificconditiontheprobeinstructionwasldquoWhereisthetargetrdquowheretargetwasrandomlyselectedfromthenamesofthetargetsforthattrialThetargetimagewasalsopre-sentedattheupperrightofthescreenTheobserverrsquostaskinthisconditionwastomovethecursortothecactuswheretheselectedtargethadbeenattheendofthetrialandclickTheanimalbehindtheselectedcactuswasthenrevealedFeedbackindicatedwhethertheobserverhadselectedthecorrectcactusorwhethertherewasnoanimalbehindthatcactus
Observerstypicallyparticipatedintwoblocksof50trialseachprecededby5practicetrialsWhenconditionswereblockedblockorderwascounterbalancedacrossobserversTheexperimentstypi-callytookbetween1and1-12htocomplete
data AnalysisRawaccuracydataweretransformedaccordingtohigh-threshold
guessingmodelsdevisedforeachconditionInthestandardcondi-tiontheguessingmodel(seeEquation1)wasderivedfromtheonepresentedbyScholletal(2001)WeassumedthatperformancePintermsofthenumberoftargetscorrectlyidentifiedwasdeter-minedbythenumberofobjectsactuallytracked(kforcapacity)plusguessingInEquation1tisthenumberoftargetsandathenumberofpossibleresponseoptionsinthiscasecactusesTheobservermakeskinformedresponsesandthenguessesonthere-maining(t2k)targetsleaving(t2k)targetsand(a2k)possibleresponsesPerformanceisthereforegivenby
P kt k
a k= +
minus( )minus
2
(1)
solvingforkgives
k aP ta P t
= minus+ minus
2
2 (2)
PerformanceinthespecificconditionwasmodeledaccordingtosimilarlogicHereperformancepistheprobabilityofcorrectlyse-lectingthecactuswherethespecifictargetishidingTheprobabilitythattheprobeanimal(selectedatrandomfromthettargets)willbeoneofthekobjectsbeingtrackedisktSoonkttrialsperformanceis10Ontheremaining(12kt)trialstheobserverguessesInthiscasetheobservercaneliminatetheklocationswithtrackedanimalsbecausesheknowstheycontainananimalthatwasnotprobedsosheguessesfromamongtheremaining(a2k)locationsHenceperformanceisgivenbyEquation3
p kt
kt a k
= + minus( ) minus( )1 1 (3)
Whenweagainsolvefork
ka pt a pt apt t
=+ minus minus minus minus( ) minus minus( )1 1 4
2
2
(4)
Equations2and4convertperformanceinbothconditionstoacom-monmetrick
Therawdataunderlyingthecapacitycomputationsforallex-perimentsarereportedinAppendixBThemajorityofouranalysesweresingle-degree-of-freedomplannedcomparisonscarriedoutthroughpairedttests
exPeriMeNT 1
InthisexperimentweintroducetheparadigmandthebasicresultthecontentdeficitInthisversionoftheex-perimentthesetof8uniqueanimalstobepresentedoneachtrialwasselectedatrandomfromthe23availableanimalsAsubsetof4targetanimalswasthenselectedatrandomfromamongthese8NoattemptwasmadetocontroloverlapinsetsfromtrialtotrialThestandardandspecificconditionswereruninseparateblocks
Estimatedcapacitywas16items(SEM503items)inthespecificconditionand29items(SEM503)inthestandardconditionThereweretwonotableresultsinthisexperimentFirstobserverswerereliablyabletoreportontheidentityofmorethanoneiteminthespecificcondition[t(7)550p001one-tailed]indicatingacontent-addressablerepresentationSecondestimatedcapacitywaslowerinthespecificconditionthaninthestandardcondition[t(7)580p0001two-tailed]ThatiswhenaskedtoreportthelocationofaspecifictargetobserversappearedtohavetrackedfeweritemsthanwhentheywereaskedtosimplyclickonallofthetargetlocationsWewillcallthisresultthecontent deficitSubsequentexperimentswilltestpossibleexplanationsforthisdeficit
OnepuzzlingaspectofthesedatawasthatperformanceinthestandardconditionseemedlowinthisexperimentincomparisonwithtypicalMOTexperimentswithidenticalitemsObserversaretypicallyabletotrackmorethanthreeitems(seeOksamaampHyoumlnauml2004Scholletal2001)Our taskmayhavebeenmoredifficultsimplybecausetheanimalsoftenoccludedoneanotherwhichcanreduceperformance(KliegerHorowitzampWolfe2004)ThereisalsoamoreseriouspossibilitySincethestimulussetchangedfromtrialtotrialobserversmayhaveexperiencedinterferenceinworkingmemoryInparticularitemsthatweretargetsononetrialcouldhaveservedasnontargetsonsubsequenttrialsandviceversasuchconditionshaveledtointerferenceinboththevisualsearch(egldquovariablemappingrdquoSchneiderampShiffrin1977)andnegativeprim-ing(MillikenTipperampWeaver1994Tipper1985)para-digmsSuchinterferencemightbeparticularlyproblematicinthespecificconditionThusthechangingstimulussetsmightberesponsibleforboththecontentaddressabilityandthecontentdeficitfromthedataExperiment2wasdesignedtoremedythisproblem
exPeriMeNT 2
Inthisexperimentwesimplyfixedthetargetandnon-targetsetssothatthesamestimuliwereusedthroughout
Tracking UniqUe ObjecTs 177
ablockoftrialsThatisanobservermighttracktheliontherabbittheturtleandthegoatonallthetrialsinablockThismadeiteasierforobserverstorememberwhattheyweresupposedtotrackInadditioniteliminatedthepossibilityof interferenceornegativeprimingarisingfromtargetsandnontargetsswitchingrolesWeexpectedthatperformanceinthestandardconditionwouldimprovetoabove30itemsthequestionwaswhetherthecontentdeficitwouldstillmanifestitselfundertheseconditions
TheprocedurewasidenticaltothatofExperiment1exceptintheselectionofanimalsForeachobserveronesetofeightanimalswasselectedrandomlyatthebegin-ningofeachblockFouroftheseanimalsweredesignatedastargetsandtherestasnontargetsThesetofanimalsandthetargetnontargetassignmentswereheldconstantforthedurationofablock
Asexpectedperformanceinthestandardconditionimprovedtoacreditable34(SEM502)itemsHow-ever the content deficit was still present Observerstrackedonly21(SEM501)itemswhenaskedwhichtargetwaswhereThisdeficitwasreliable[t(7)583p00001]andsimilarinmagnitudetothatobservedinExperiment1ofaround13itemsAmixedANOVAcomparingthetwoexperiments2revealedalargeeffectof condition [F(18)5 1152p 000001]butnoneofexperiment[F(18)510p 10]Theinteractionterminthebetween-experimentsanalysiswasnearzero[F(18)1]
Theroughlyhalf-itemincreaseincapacitywhenwechangedfromavariabletoafixedtargetsetsuggeststhatswitchingtargetsetsfromtrialtotrialcreatedsomesortofinterferenceThiswastruenotonlyinthespecificcon-ditionwhenweprobedtargetidentitiesbutalsointhestandardconditionThisresultisimportantbecauseitsuggestsaroleforobjectidentity(eitherattheperceptualorsemanticlevel)irrespectiveoftaskdemandsAccord-ingtoLeonardandPylyshyn(2003)blinkingthetargetsonandoffshouldautomaticallyassignvisualindexestothemandtheseindexesshouldstickequallywellwhetherornottheseparticularstimuliweretargetsontheprevioustrialSincetheincreaseincapacitywasnotstatisticallyreliablethisconclusionshouldnotbetakentooseriouslyHoweverinsubsequentexperimentsweusedafixedtar-getset
JustasinExperiment1theobserversherecouldreli-ablylocatemultipletargetsaccordingtotheiridentitieswhichwetakeasevidenceforcontentaddressabilityEx-periment2alsoreplicatedthecontentdeficitwithreli-ablybettercapacityinthestandardreportconditionthaninthespecificreportcondition
TheseresultsraiseanumberofmethodologicalissuesFirstourexperimentsdifferfrommostMOTexperimentsintheirtemporalstructureourtrialswerebothlongerthanusualandincludedtemporaluncertaintyHowdidthesefactorsaffectperformanceExperiment3wasdesignedtoaddressthisissue
Asecondimportantissueiswhetherblockingresponsemodesinducedobserverstousedifferentencodingstrate-giesinthetwoconditionsleadingtodifferentialperfor-manceThisissuewillbeaddressedinExperiment4
AthirdissueiswhethertheuseofuniqueidentifiableobjectshelpsorhindersMOTperformanceincomparisonwiththeuseofidenticalobjectsinmostpreviousstudiesofMOTWewilladdressthisissueinExperiments5and6
FinallyExperiment7willaddresstheroleofnontar-getsinhinderingtrackingperformance
exPeriMeNT 3
Aswenotedintheintroductionitisimportanttoen-surethatobserversareattendingtoataskthroughoutatrialOneofourstrategiesweusedtoensurethiswastorandomlyvary trialdurationObserversdidnotknowwheneachtrialwouldendbetweenthe10thand15thcyclesWasthisenoughtemporaluncertaintyhoweverInExperiment3weincreasedtheuncertaintysothatatrialcouldendanywherebetweenthe5thand15thcycles(adurationfrom67to200sec)OtherwisethemethodwasidenticaltothatofExperiment2IfobserverswereactivelytrackingthetargetsonlyafterCycle9inthepre-viousexperimentsweshouldseeadecreaseinperfor-manceinthisexperimentwheretheypotentiallyhavetosustainattentionforalongerperiodoftime
Awiderarrayoftrialendpointsalsoallowedustoex-amineapossibleexplanationforthecontentdeficitWeusedtrialdurationsthatwerelongrelativetopreviousMOTstudiesPerhapslocationndashidentitybindingsdecayovertimeandwewouldobserveasmallerdifferencebe-tweenconditionsifweprobedearlierinthetrial
Figure3showskinExperiment3asafunctionofcyclewithdatafromExperiment2shownforcomparisonMeancapacityforthestandardconditionwas34items(SEM501)versus26items(SEM502)forthespecificcondi-tionrepresentingasignificantadvantageinthestandardcondition[t(7)539p 01]Wealsoconductedananalysisofcovarianceontheresultswithconditionasacategoricalvariableandcycleasacontinuousvariable(covariate)Neitherthemaineffectofcycle[F(17)1]northecycle3conditioninteraction[F(17)5249p516]wassignificant
5 7 9 11 13 150
1
2
3
4
Cycles
k (it
ems)
Figure 3 Capacity as a function of cycle Squares denote the standard and triangles the specific target condition data from experiment 3 are plotted in black and those from experiment 2 in gray error bars indicate standard errors of the means
178 HOrOwiTz eT al
WhencomparingthetwoexperimentsitseemsclearthatincreasingthetemporalvariabilityinExperiment3didnotchangetheresultsnotablyWeconductedacross-experimentANOVAusingdataonlyfromCycles10ndash15inExperiment3Standardversusspecifictargetconditionandcyclewerewithin-subjectsfactorsandexperimentwasabetween-subjectsfactorWefoundnomaineffectofexperiment[F(114)511p 5 31]Thecondition3ex-perimentinteractionapproachedsignificance[F(114)541p506]probablybecausethespecifictargetcapac-itywassomewhathigherinExperiment3Addingtempo-raluncertaintyclearlydidnotdisruptperformanceinthisexperimentinrelationtoExperiment2
ThereisahintinFigure3thatperformanceespeciallyin thespecific targetconditiondroppedas trialdura-tionincreasedThiswouldbecompatiblewithdatafromOksamaandHyoumlnauml(2004)possiblyreflectingavigilancedecrement(Parasuraman1986)IfthedropovertimeweregreaterinthespecificconditionthiswouldsupporttheideathatlocationndashidentitybindingsweredecayingHowevertherewasnoevidenceforthisinanANOVAconductedondatafromalloftheconditionsofExperiment3inwhichneitherthemaineffectofcycle[F(1070)511p538]northecycle3conditioninteraction[F(1070)1]wassignificantIftherewasdecayithadapproachedasymp-totewithinthefirst6ndash7secoftracking
exPeriMeNT 4
OneobvioushypothesistoexplainthecontentdeficitisthatobserversencodeinformationdifferentlyinthetwoconditionsInthestandardconditiononlyspatiotemporalpropertiesareencodedInthespecifictargetconditionobserversknowthattheywillbeaskedaboutfeaturaloridentityinformationsotheyencodethataswellFeaturalinformation(orperhapsthebindingofidentityandlo-cation)takesupmoreresources(Bahrami2003Saiki2002) reducing tracking capacityThus informationaboutfeaturesoridentitiesrequiresmorecapacity
InExperiment4wetestedthisstrategicencodinghy-pothesisbymixingthequestionswithinablockoftri-alsObserversdidnotknowonagiventrialwhethertheywouldbeaskedtolocateallofthetargetsorjustaspecifictargetThereforetheyhadtoadoptthesamestrategiesonbothtypesoftrialsIfthestrategicencodinghypothesisiscorrectthenthecontentdeficitshouldbereducedinthisexperimentPerformanceshouldalsobereducedoverallsinceobserverswouldnotbeoptimallyencodinginfor-mationoneachtrialInotherrespectsthemethodwasidenticaltothatofExperiment2
Pilotworkindicatedthatwhenquestionsweremixedwithinablockof trialshavingto lookawayfromthecactusarraytoreadthequestionatthetopofthescreenimpairedperformanceThereforeinthisexperimentweusedauditoryratherthanvisualprobesInsteadofprintingthequestionatthetopofthescreenobserversheardtheprobequestionsoverheadphones
Meancapacityforthestandardconditionwas34items(SEM501)asopposedto22items(SEM502)forthe
specificconditionrepresentingasignificantadvantageforthestandardcondition[t(7)599p00005]Per-formanceinthisexperimentwithquestionsmixedwithinblockswasnearlyidenticaltothatinExperiment2withblockedquestionsEvenwhenobserverscouldnothaveadopteddifferentencodingstrategies for thedifferentconditionswestillobservedasizable(12-item)contentdeficitThissuggeststhatthecontentdeficitisnotaresultofthestrategicdemandsoftheconditionswedevised
exPeriMeNT 5
ThepreviousexperimentsestablishedaparadigmforstudyingMOTwithunique stimuliOnequestion thatwehadnotyetaddressediswhetheruniqueobjectshelporhinderMOTperformanceincomparisonwithidenticalobjectsForexampleiftheobjectsareidenticalanob-servermightaccidentallyswapatargetandadistractoriftheyapproachtoocloselyWithuniqueobjectsthisislesslikelytohappenSimilarlyifatsomepointtheobserverrealizedthathewasonlytrackingthreetargetsandhadlostonehecouldtheoreticallyrecoverthelosttargetifitwereuniqueOfcoursesuchadvantagesassumeasortofldquoidealobserverrdquowhoalwaysknowsexactlywhatheistrackingIfthetrackingsystemwascompletelyinsensitivetotargetfeaturesoridentityuniqueobjectswouldprovidenoadvan-tageSimilarlytheobservercouldonlyrecoveralosttargetifinformationwereavailableabouthowmanytargetswerecurrentlybeingtrackedandwhichonewasmissing
Inthisexperimentwecomparedtrackinguniqueob-jectswithtrackingidenticalobjectsSinceldquoWhereisthezebrardquoisnotadiagnosticquestionwhenallobjectsarezebrasweonlyusedthestandardresponsemodeTherewerethreeconditionsTheuniqueconditionwasidenti-caltothestandardresponseconditioninExperiment1Intheidenticalconditioneightidenticalanimalswerepresented animal identitywas selectedat randomoneachtrialFinallyinthepairedconditiontherewerefouruniquetargetsagainselectedatrandomForeachtargettherewasanidenticaldistractorProbeswerepresentedauditorilyasinExperiment4Atotalof10observerspar-ticipatedinthisexperiment
Wecancompareperformanceintheuniqueandidenti-calobjectconditionstodeterminewhetherthereisanyadvantageforuniqueobjectsIftheadvantageforuniqueobjectsderivesfromtheabilitytorecoverlosttargetsthisadvantageshouldbeabolishedinthepairedcondition
CapacityvaluesforthethreeconditionsareplottedinFigure4Observerstracked31items(SEM502)intheuniquecondition23(SEM502)inthepairedconditionand25(SEM501)intheidenticalconditionPlannedttestsshowedthatthepairedandidenticalconditionsdidnotdiffer[t(9)513p 10]butthatperformancewassignificantlyworseinboththanintheuniquecondition[forpairedt(9)527p05foridenticalt(9)527p01]
TheseresultsindicatethatobserverscantakeadvantageoftheadditionalinformationprovidedbyuniqueobjectsTheuniqueobjectadvantage(approximately06items
Tracking UniqUe ObjecTs 179
undertheseconditions)isentirelyeliminatediftargetsareidenticaltodistractorssuggestingthatsomesortoferrorrecoveryprocessisresponsible
exPeriMeNT 6
In the previous experiment pairing targets anddistractorsreducedperformanceinthestandardcondi-tionrelativetothatinthefullyuniqueconditionWouldthismanipulationaffectthespecificconditionaswellInExperiment6wemeasuredthecontentdeficitwithbothpairedanduniquestimuliBycomparingthesetwostimulusconditionswecandeterminewhethereliminat-ing featuraldistinctionsbetween target andnontargetsetsimpairstheabilitytotrackspecificobjectsaswellastheabilitytoholdontotargetlocationsProbeswerepresentedauditorilyasinExperiment5Therewere6ob-serversinthisexperiment
AsshowninFigure5weonceagainreplicated thecontentdeficitinthisexperimentsinceobserverscouldtrack11moreitemsinthestandardconditionthaninthe
specificcondition[F(19)51190p001]Pairingtar-getsandnontargetsreducedcapacitybyroughlythesameamountas inExperiment5 [09 itemsF(19)5373p001]Criticallythepairingmanipulationappearedtohavelessofadeleteriouseffectinthespecificconditionthaninthestandardcondition[interactionF(19)5185p005]Inthestandardconditionthepairingmanipu-lationcostobservers12itemsversusalossofonly07itemsinthespecificconditionThisinteractionsuggeststhatdifferentrepresentationsmayunderlieperformanceinthetwoconditions
Ontheotherhandcapacitywasalreadylowerinthespe-cificconditionIftheeffectofpairingstimuliismultipli-cativeratherthanadditivewewouldthenexpectthattheimpactofpairedstimuliwouldbeproportionallysmallerinthespecificconditionWhenwelog-transformedthecapacityvaluesconvertingproportionaldifferencesintoadditive ones the question3 pairing interaction waseliminated[F(19)1]
Sincewecannotrejectthehypothesisthattheeffectofpairingtargetswithnontargetsisproportionaltocapacitythisexperimentprovidesatbestweakevidencefordiffer-entrepresentations
exPeriMeNT 7
OneadvantageofourmethodisthatwecantestMOTperformanceinacasewithonlytargetsWhywouldwewanttodothisTherearereasonstobelievethatsuppress-ingnontargetsconsumessomeresourcesthatwouldother-wisebeavailablefortrackingtargetsForexamplethereisevidencethatobserverstrackbetterontrialsinwhichnontarget trajectories are repeated from earlier trials(OgawaampYagi2003)IfnontargetsareprocessedtheymayhavetobesuppressedinordertoreduceconfusionwithtargetsandPylyshynandLeonard(2003)showedevidenceforsuchinhibitionMoreoverunpublisheddatafromourlaboratoryindicatethatMOTperformancede-creasesasnontargetsareaddedtothedisplayevenwhenthenumberoftargetsremainsconstant
Accordingly we designed Experiment7 to test thehypothesisthattrackingcapacitywouldincreaseiftherewerenonontargetsThisexperimentwasidenticaltoEx-periment2exceptthatallitemsweretargetsCapacityinthestandardconditionwas58items(SEM502)sig-nificantlygreaterthanthe26items(SEM501)obtainedinthespecificcondition[t(7)5143p000005]
InagreementwithourhypothesiscapacityimprovedmarkedlyinthestandardconditionwhentherewerenodistractorsStandardcapacityinExperiment7wassig-nificantlygreaterthaninExperiments23and4whichweremethodologicallycomparableexceptforthenumberoftargetsandnontargets(allps00005)
Oneexplanationisthatpreviousexperimentsunderes-timatedcapacitybecauseofaceilingeffectTheseexperi-mentswerebasedonthehypothesisthatcapacityvariesacrosstrialswithnounderlyingdifferencesacrossexperi-mentsHoweverthemeasuredcapacityhasbeenlimitedbythenumberoftargetsForexampleifthereareonlyfour
Unique Paired Identical0
1
2
3
4
Condition
k (it
ems)
Figure 4 Capacity as a function of condition in experiment 5
Specific Standard0
1
2
3
4Unique
Paired
Condition
k (it
ems)
Figure 5 effect of pairing target and nontarget objects on ca-pacity in experiment 6 data from the paired conditions are plot-ted as gray bars and those from the unique conditions as open bars
180 HOrOwiTz eT al
targets(egExperiments2ndash4)observerscouldtrackallofthemontrialsinwhichtheircapacitywas4orgreateranytrialswithanactualcapacitygreaterthan4wouldthusappeartohaveacapacityof4InthiscasemeasuredmeancapacitywouldunderestimatethetruemeanInExperi-ment7whichincludedeighttargetsmeasuredcapacitycouldreflectactualcapacityuptoeightitemsleadingtoagreater(andmoreaccurate)estimateofmeancapacity
Wecantestthisceilinghypothesisagainstourhypoth-esisofincreasedmeancapacitybycomparingthecapacitydistributionsinExperiment7withthoseinExperiments2ndash4Accordingtotheceilinghypothesistheproportionoftrialswithacapacitylessthan4shouldbeidenticalacrossexperimentsFurthermoretheproportionoftrialswithacapacityof4orgreaterinExperiment7shouldbethesameastheproportionoftrialswithacapacityequalto4inExperiments2ndash4Incontrastthehypothesisthattrackingcapacityincreaseswhennonontargetsarepres-entpredictsthattheentiredistributionshouldshifttotherightleadingtofewertrialswithcapacitylessthan4inExperiment7thanintheearlierexperiments
We tested these hypotheses by computing capacityforeachtrialusingEquation2andgeneratingadistri-butionacrosstrialsforeachobserverFigure6plotsthedistributionofkforExperiment7andthecorrespondingdistributionforExperiments23and4combined(dis-tributionsfromeachoftheseexperimentswereaveragedacrossobservers)Clearlytheproportionoftrialswithanobservedcapacitylessthan4waslowerwitheighttargetsthanwithfourtargetsThisisconsistentwiththecapacitydistributionshiftingtotherightinExperiment7ratherthanmerespreadingofthek54trialsoverawiderrangeWesuggestthatthesuperiorperformanceinthisexperi-mentmayreflectadditionalcapacityfreedupwhenthereisnoneedtosuppressnontargetsHowevertheshapeofthedistributionforfourtargetsdoessuggestthatcapacitywasunderestimatedinthepreviousexperiments
TheconclusionsarequitedifferentifwelookatthespecifictargetconditionSpecifictargetcapacityinthis
experimentwasbetterthaninExperiment2(uncorrectedttestp05)butnotsignificantlydifferentfromtheca-pacitiesobservedinExperiments3and4( p05)Thusthetrackingoftargetidentitiesinthisconditiondoesnotseem tobe limitedby theneed to suppressnontargetidentities
diSCuSSioN
WehavedevelopedanewMOTparadigmthatallowsustotestthetrackingofuniqueobjectsFourmainfind-ingshaveemergedfromtheseexperimentsFirstthereisacontent-addressablerepresentationoftargetsduringMOT(specificcapacitywas1inallcases)SecondweobservedacontentdeficitThecontent-addressablerep-resentationhasalowercapacity(measuredinitems)thanthelocation-addressablerepresentation(iespecificca-pacitywaslowerthanstandardinallcases)Thirdthevi-sualsystemiscapableoftakingadvantageofdifferencesbetweenuniqueobjectstoimprovetrackingperformanceoverthelevelseenwithidenticalobjects(Experiment5)Thisadvantageappearstobedueprimarilytotheabilitytorecoverlosttargets(Experiment5)Fourthtrackingcapacityappearstobeconstrainedbytheneedtosup-pressnontargetswhentherearenonontargetscapacityincreasesmarkedly(Experiment7)
A Content-Addressable representation in MoTOurdata represent anexistenceproofof a content-
addressable representation inMOTSimilar data havebeenreportedbyOksamaandHyoumlnauml(2004)Intheirmul-tiple-identity-tracking(MIT)taskobserverstrackedmov-inglinedrawings(eitherobjectsorldquopseudo-objectsrdquo)forseveralsecondsAttheendofthetrialthestimuliweremaskedandasingleitemwasmarkedwithablackframeInasubsequentresponsescreenobservershadtoselectthepicturethatcorrespondedtothemarkedobjectOksamaandHyoumlnaumlobtainedamediancapacityoffouritems
Althoughitisdifficulttocomparecapacitymeasuresdirectlyacrossexperimentaltasks3itisinterestingthatbothweandOksamaandHyoumlnauml(2004)demonstratedthatobserversdoknowtosomeextentwhichtargetiswhichThisfindingisincontrasttothoseofPylyshyn(2004)whoseobservershaddifficultyidentifyingtargetsTwokeydifferencesbetweenourparadigmandMITontheonehandandPylyshynrsquos(2004)paradigmontheotherseemrelevanttothisdiscrepancyFirstinourstudyandthatofOksamaandHyoumlnaumlobjectidentitiesweredefinedbyintrinsicfeaturessuchasshapeand(inourexperi-ment)colorInPylyshynrsquos(2004)experimentldquoidentityrdquowasbasedonatangentialfeatureofanobjectsuchasinwhichcornerofthedisplayitbeganthetrialItseemslikelythattheshapeandcolorofanobjectmaybemoretightlyboundmarkersofldquoidentityrdquothaninitialpositionorabrieflypresentedletter
SecondinourexperimentsandthoseofOksamaandHyoumlnauml (2004) thevisualdifferencesbetweenobjectswerecontinuallyavailablethroughoutthetrackingphaseofthetrialonlymaskedduringtheresponsephaseIn
4 targets8 targets
0 2 4 6 80
1
2
3
4
5
6
Capacity
Pro
po
rtio
n o
f Tri
als
Figure 6 Capacity distributions for the standard condition in experiment 7 (gray lines striped area) and in experiments 2 3 and 4 (black lines stippled area) dashed lines indicate standard errors of the means
Tracking UniqUe ObjecTs 181
Pylyshynrsquos(2004)experimentstheobjectidentitieswerepresentedonlybrieflyatthestartofthetrialduringthemajorityofthetrackingphasetheobjectswerephysicallyidenticalItmaybethatthelinkbetweenobjectidentityandspatiotemporalpositiondecayedduringPylyshynrsquos(2004)taskbutwasconstantlyrefreshedinbothoursandOksamaandHyoumlnaumlrsquos
Ourexperimentsalsoprovidedsomeindirectevidenceforacontent-addressablerepresentationForexampleperformancewasconsistentlysuperior inexperimentsinwhichthesameobjectswereusedinthesamerolesfromtrialtotrial(Experiments23and4)thaninex-perimentsinwhichthestimulussetchangedfromtrialtotrial(Experiments15and6)Thisoccurrednotonlyinthespecifictargetconditionsinwhichobjectidentitywasrelevantbutalsointhestandardconditionsinwhichobjectidentitywasnot
one System or TwoIstrackingservedbyasinglerepresentationthatencodes
bothpositionandidentityoraretheretworepresentationsoneforpositionandoneforidentityThisquestionarisesbecauseofthestrikingcapacitydifferencesbetweenourresponseconditionsInallofourexperimentscapacitywassignificantlylowerforspecifictargetsthanfortargetsinthestandardMOTresponsemodeInExperiment4wedisconfirmedthehypothesisthatcapacitydifferencesresultedfromvariedencodingstrategies(seeegDavisWelchHolmesampShepherd2001)sinceweobtainedthesameresultwhenresponsemodesweremixedwithinablockaswedidwhentheywereruninseparateblocks
ThiscapacitydifferenceisnotconclusiveevidencethatseparaterepresentationsorsystemsareinvolvedThereareseveralwaysinwhichaunifiedaccountcouldex-plainthelowercapacityinthespecifictargetconditionForexampleonemightproposeasystemwithaslot-likestructure(VogelWoodmanampLuck2001)inwhichallslotsarenotcreatedequalIfthereweretwoslotswithsuf-ficientresolutiontoencodebothpositionandidentityandtwoslotsthatcouldonlyencodepositionwewouldalsoobtaindifferentcapacityestimatesinourtworesponseconditionsThisrepresentationcouldalsobedescribedintermsofKahnemanandTreismanrsquos(1984)objectfilesifweassumethatsomeoftheobjectfilesareemptyTheimpletionprocessmightinthatcasereturntheinforma-tionthatagivenobjectwasspatiotemporallycontinuouswithatargetobjectbutfailtoretrievethesemanticorperceptualcontentassociatedwiththatobjectfile
IntheMOTliteraturecapacityisgenerallyassumedtorefertoafixednumberofavailablerepresentationsinthevisualsystemeithervisualindexesorobjectfilesHow-everthereissomeevidencethatcapacityinMOTmightbebetterthoughtofasanundifferentiatedresource(AlvarezampCavanagh2005)InsteadofasetofslotsorpointerswemightassumethatthereisafixedamountofinformationthatcanbeencodedObserversmightencodepositionandidentityforalltargetsbutthebindingbetweenpositionsandidentitymaybenoisierthanthepositioninformationitself(ormaydecaymorequicklyExperiment3indicatedthatanysuchdecaywouldhavetoapproachitsasymptote
valuewithin6ndash7sec)leadingtoapparentlylowercapacitywhenresponsesdependonbinding
InsteadofonesystemtherealsomightbetwoAlthoughwetendtoassumethatasingleexperimentalparadigmprobestheperformanceofasinglecorrespondingfunc-tionalsystemitmaybethatMOTnaturallyrecruitsmulti-pleoverlappingsystemsThesimplesttwo-systemaccountmightbeaproposalthattrackingisservedbyaposition-trackingsystemsuchasvisualindexes(FINSTs)butthataccessingidentityndashpositionbindingsrequiresfocalatten-tion(WheelerampTreisman2002)Howeversincespecificcapacityisreliablygreaterthanoneitemwewouldhavetoassumethatmultipleitemscansimultaneouslybethefocusofattention(AwhampPashler2000Davisetal2001)
IfwewishtopreserveasinglefocusofattentionwemightinsteadproposethatbothvisualindexesandobjectfilescanbeusedVisualindexeswouldprovideonlypo-sitioninformationandobjectfilescouldsupportmorecomplexqueries
Itwillobviouslybedifficulttodefinitivelydecidebe-tweenoneandtwosystemsHowevertheevidenceofourfinalthreeexperimentspointstotwoindependentsystemsInExperiments5and6pairingtargetsandnontargetsre-ducedcapacityinthestandardbutnotinthespecifictar-getcondition(althoughwecouldnotrejectthehypothesisofaproportionaldecrease)InExperiment7standardca-pacityincreasedwhennonontargetswerepresentedbutspecificcapacitywasunaffected
MOThasbeenlikenedtoavisualworkingmemorytaskattimezero(CavanaghampAlvarez2005)soitisnaturaltolooktostudiesofworkingmemoryforanalogsofthecontent-addressableandlocation-onlysystemsTheclassictwo-pathwayschemeinvisionisthedistinctionbetweenventralanddorsalstreamsassociatedwithobjectrecogni-tionandspatialprocessingrespectively(UngerleiderampMishkin1982)ThisdistinctionhasbeenextendedintothedomainofworkingmemorybyWilsonOacuteScalaidheandGoldman-Rakic(1993)whoprovidedevidencethatobjectinformationandspatialinformationareneurallysegregatedinmonkeyprefrontalcortexwidelybelievedtobetheneuralsubstrateofworkingmemoryAsimilardissociationinhumanswasreportedusingPETimagingbySmithetal(1995)
Thedivisionofworkingmemoryintomultiplesubsys-temsgoesbackatleasttothemultiple-componentmodelofBaddeleyandHitch(1974)whichproposedseparatecomponents for different modalities a ldquophonologicallooprdquoforauditoryinformationandaldquovisuospatialsketchpadrdquoforvisualinformationInmorerecentversionsofthemodel(BaddeleyampLogie1999Logie1995)thesketchpadhasbeendividedintovisualandspatialsubcompo-nents roughlycorrespondingto theobjectandspatialsystemsdescribedbyWilsonetal(1993)Interestinglythespatialsubcomponent(theldquoinnerscriberdquo)isspecifi-callytaskedwithholdingsequenceorpathinformation(seeParmentierElfordampMayberry2005)makingitalikelysubstrateforMOTperformance
Thewidespreadinsistenceonaseparationbetweenvi-sualorobjectinformationandspatialinformationwouldseem problematic for our findings Standard MOT is
182 HOrOwiTz eT al
clearlyajobforthespatialsubsystem(PostleDrsquoEspositoampCorkin2005forexampleusedaversionofstandardMOTtotieupdorsalprocessinginamemoryinterferenceparadigm)Howevertheredoesnotseemtoberoominthataccountforacontent-addressablerepresentationOurspecifictargettask(aswellastheMITtaskofOksamaampHyoumlnauml2004)isnotapureobjectrecognitiontaskitrequireslocationndashidentitybindingswhichwouldseemtoinvolveintegratinginformationacrossstreamsTheproblemdisappears ifwe take intoaccount recentbe-havioral(OlsonampMarshuetz2005)andfMRI(PostleampDrsquoEsposito1999)investigationsdemonstratingthatobjectworkingmemoryrepresentationsalsocarrysomelocationinformation
NeuroimagingstudiesmightproveusefulhereStan-dardMOThasbeenshowntoactivateparietallobeareasinthedorsalstream(Culhametal1998Jovicichetal2001Seiffert2003)Itwouldbeinformativetoknowwhetherthepatternofactivationchangeswhenuniqueob-jectsareintroducedorwhenthetaskpotentiallyinvolvesobjectidentities
unique object Features in MoTInadditiontoourfindingsaboutcontentaddressability
intrackingwewereabletodemonstratethathavingvisu-allyuniqueobjectsmadetrackingeasierobserverswereabletotrackroughly06moreitemswhenobjectswereuniquethanwhenobjectswereidenticalInExperiments5and6weattributedthisresulttoanerrorrecoverystrat-egyIfIlosetrackofthezebraIcansearchforthezebraonthebasisofitsphysicalfeaturesandcontinuetotrackitThisstrategywouldobviouslybeuselessifallitemsarezebrasThisstrategyisoflimitedutilityifthereareazebratargetandazebradistractoronceIlosethezebraIthenhaveonlya50chanceofpickingupthetargetzebraasopposedtothedistractor
TheerrorrecoveryhypothesisisnottrivialIthassev-eralimportantimplicationsFirstitimpliesthatobserversknowwhattheyarenottrackingThephenomenologyofMOTissuchthatanobserverhasthefeelingofknow-inghowmanytargetsheorsheissuccessfullytrackingTheerrorrecoveryhypothesisimpliesthatwhenatargetislosttheobserverknowswhichone(givenuniquetargets)SecondtheerrorrecoveryhypothesisalsoassumesthatanobservercansearchforamissingtargetwhilecontinuingtotracktheremainingtargetsThisisinagreementwithrecentevidencethatobserverscansuccessfullyperformanMOTtaskandavisualsearchtaskonthesametrial(AlvarezHorowitzArsenioDiMaseampWolfe2005)Finallythehypothesisimpliestheabilitytoaddtothetrackingsetontheflywhiletrackingwithoutanyphysi-calcueTheseimplicationsneedtobeconfirmedthroughfurtherresearch
Howeverifobserversuseuniquephysicalfeaturesofobjectstorecoverlosttargetsinthisfashionitiscuriousthatthereisnosuchbenefitinthespecifictargetcondi-tionIfobserversdorecoverlosttargetsduringthecourse
ofatrialtheyappeartorecoveronlytheirlocationsnottheiridentities
AnalternativetotheerrorrecoveryhypothesisthoughnotamutuallyexclusiveoneisthattheresultsofEx-periment5canbeexplainedbytheperceptualsimilarityoftargetandnontargetitemsYantisrsquos(1992)approachtoMOThighlightstheimportanceofperceptualgroup-ingfactorsthatallowthetargetstobetreatedasaunitbythevisualsystemMostoftheresearchinthislinehasfocusedonspatiotemporalgroupingfactorssuchasthespatialrelationshipsbetweentheobjects(SearsampPylyshyn2000ViswanathanampMingolla2002Yantis1992)
Cangroupingby(nonspatial)featuralpropertiesim-provetrackingAnumberofstudieshavedemonstratedthatobservershave limitedexplicit access to featuralproperties (Bahrami 2003Saiki 2002Scholl etal1999)butthisfindingdoesnotaddressthedirectrolethatfeaturesmayplayWecouldexplaintheresultsofEx-periment5byproposingthatobserverswereabletouseaccidentaldifferencesintheshapeandcolorstatisticsoftargetsandnontargetstohelpsegregatethemvisuallyInthepairedconditiontargetsandnontargetshadidenticalfeaturestatisticssothatadvantagewaslostThishypoth-esiscouldalsoexplainwhyperformancewasbetterwithafixedstimulussetthanwithavariablestimulussetoveranumberoftrialswiththesamestimuliobserverslearnsubtledifferencesbetweenthetwosets
ConclusionsIntheintroductionwenotedthatreal-worlddynamic
attentiontasksdifferfromtraditionalMOTintwowaysFirstobserversintheeverydayworldusuallydealwithmultipleuniqueobjectsSecondtheyoftenwishtoknowwhereagiventargetmightberatherthanwhichobjectsarethetargetsTheexperimentspresentedhererepresentanattempttobridgelaboratoryMOTresearchandtheseeverydaycognitivetasksWehaveshownthatitiseasiertotrackuniqueobjectsthantotrackidenticalobjectsassum-ingthattheobjectsonewishestotrackarealsouniquelydifferentfromnontargetobjectsFurthermoreobserversdoknowwhattheyaretrackingTotakeoneofourex-amplesofecologicallyvalidtrackingtasksifyouwatchyourchildrenandtheirfriendsontheplaygroundyoucanbeawareofwhereyourkidsareatanygivenmomentyoumayalsobeawareofthelocationsoftheirfriendswithoutbeingabletotellonefriendfromtheotherFurthermoreifyoulosetrackofachildyouwillprobablybeawareofwhichoneyouhavelost
AuTHor NoTe
ThisresearchwassupportedbyNIMHGrantR0165576toTSHWethankKristinMichodandSkylerPlaceforassistancewithdatacollec-tionaswellasSteveLuckandananonymousreviewerforconstructivecriticismCorrespondencerelatingtothisarticlemaybesenttoTSHorowitzBrighamandWomenrsquosHospitalVisualAttentionLaboratory64SidneyStreetSuite170BostonMA02139(e-mailtoddhsearchbwhharvardedu)
Tracking UniqUe ObjecTs 183
reFereNCeS
Allen R McGeorge P Pearson D amp Milne A B (2004)AttentionandexpertiseinmultipletargettrackingApplied Cognitive Psychology18337-347
Alvarez G A amp Cavanagh P (2005)IndependentresourcesforattentionaltrackingintheleftandrightvisualhemifieldsPsychologi-cal Science16637-643
Alvarez G A amp Franconeri S (2004November)How many ob-jects can you trackPaperpresentedatthe12thAnnualWorkshoponObjectPerceptionampMemoryMinneapolisMN
Alvarez G A Horowitz T S Arsenio H C DiMase J S amp Wolfe J M (2005)DomultielementvisualtrackingandvisualsearchdrawcontinuouslyonthesamevisualattentionresourcesJournal of Experimental Psychology Human Perception amp Perfor-mance31643-667
Awh E Jonides J amp Reuter-Lorenz P A (1998)RehearsalinspatialworkingmemoryJournal of Experimental Psychology Human Perception amp Performance24780-790
Awh E amp Pashler H (2000)EvidenceforsplitattentionalfociJournal of Experimental Psychology Human Perception amp Perfor-mance26834-846
Baddeley A D amp Hitch G J (1974)WorkingmemoryInGHBower(Ed)The psychology of learning and motivation Advances in research and theory(Vol8pp47-90)NewYorkAcademicPress
Baddeley A D amp Logie R H (1999) Working memoryThemultiple-componentmodelInAMiyakeampPShah(Eds)Models of working memory Mechanisms of active maintenance and executive control(pp28-61)CambridgeCambridgeUniversityPress
Bahrami B (2003)ObjectpropertyencodingandchangeblindnessinmultipleobjecttrackingVisual Cognition10949-963
Brainard D H (1997)ThePsychophysicsToolboxSpatial Vision10433-436
Cavanagh P amp Alvarez G A (2005)TrackingmultipletargetswithmultifocalattentionTrends in Cognitive Sciences9349-354
Culham J C Brandt S A Cavanagh P Kanwisher N G Dale A M amp Tootell R B H (1998)CorticalfMRIactiva-tionproducedbyattentive trackingofmoving targetsJournal of Neurophysiology802657-2670
Davis G Welch V L Holmes A amp Shepherd A (2001)CanattentionselectonlyafixednumberofobjectsatatimePerception301227-1248
Henderson J M amp Hollingworth A (2003)EyemovementsandvisualmemoryDetectingchangestosaccadetargetsinscenesPer-ception amp Psychophysics6558-71
Horowitz T S Birnkrant R S Fencsik D E Tran L amp Wolfe J M (2006)HowdowetrackinvisibleobjectsPsychonomic Bulletin amp Review13516-523
Jovicich J Peters R J Koch C Braun J Chang L amp Ernst T (2001)Brainareasspecificforattentionalloadinamotion-trackingtaskJournal of Cognitive Neuroscience131048-1058
Kahneman D amp Treisman A (1984)ChangingviewsofattentionandautomaticityInRParasuramanampDRDavies(Eds)Varieties of attention(pp29-61)OrlandoAcademicPress
Kahneman D Treisman A amp Gibbs B J (1992)ThereviewingofobjectfilesObject-specificintegrationofinformationCognitive Psychology24175-219
Klieger S B Horowitz T S amp Wolfe J M (2004)Ismultipleobjecttrackingcolorblind[Abstract]Journal of Vision4(8)363a
Leonard C amp Pylyshyn Z W (2003)Measuringtheattentionaldemandofmultipleobjecttracking(MOT)[Abstract]Journal of Vi-sion3(9)582a
Logie R H (1995) Visuo-spatial working memory Hove UKErlbaum
Milliken B Tipper S P amp Weaver B (1994)NegativepriminginaspatiallocalizationtaskFeaturemismatchinganddistractorin-hibitionJournal of Experimental Psychology Human Perception amp Performance20624-646
Mitroff S R Scholl B J amp Wynn K (2004)DivideandconquerHowobjectfilesadaptwhenapersistingobjectsplitsintotwoPsy-chological Science15420-425
Ogawa H amp Yagi A (2003)Primingeffectsinmultipleobjecttrack-ingAnimplicitencodingbasedonglobalspatiotemporalinformation[Abstract]Journal of Vision3(9)339a
Oksama L amp Hyoumlnauml J (2004)Ismultipleobjecttrackingcarriedoutautomaticallybyanearlyvisionmechanismindependentofhigher-ordercognitionAnindividualdifferenceapproachVisual Cognition11631-671
Oliva A Torralba A Castelhano M S amp Henderson J M (2003)Top-downcontrolofvisualattentioninrealworldscenes[Ab-stract]Journal of Vision3(9)3a
Olson I R amp Marshuetz C (2005)RememberingldquowhatrdquobringsalongldquowhererdquoinvisualworkingmemoryPerception amp Psychophysics67185-194
Parasuraman R (1986)VigilancemonitoringandsearchInKRBoffLKaufmanampJPThomas(Eds)Handbook of perception and human performance Vol 2 Cognitive processes and performance(pp1-39)NewYorkWiley
Parmentier F B R Elford G amp Mayberry M (2005)Transi-tionalinformationinspatialserialmemoryPathcharacteristicsaffectrecallperformanceJournal of Experimental Psychology Learning Memory amp Cognition31412-427
Pelli D G (1997)TheVideoToolboxsoftwareforvisualpsychophysicsTransformingnumbersintomoviesSpatial Vision10437-442
Postle B R amp DrsquoEsposito M (1999)ldquoWhatrdquondashthenndashldquowhererdquoinvi-sualworkingmemoryAnevent-relatedfMRIstudyJournal of Cog-nitive Neuroscience11585-597
Postle B R DrsquoEsposito M amp Corkin S (2005)Effectsofver-balandnonverbalinterferenceonspatialandobjectvisualworkingmemoryMemory amp Cognition33203-212
Pylyshyn Z [W] (1989)Theroleoflocationindexesinspatialper-ceptionAsketchoftheFINSTspatial-indexmodelCognition3265-97
Pylyshyn Z W (2004)SomepuzzlingfindingsinmultipleobjecttrackingITrackingwithoutkeepingtrackofobjectidentitiesVisual Cognition11801-822
Pylyshyn Z W amp Leonard C (2003)Inhibitionofnontargetsdur-ingmultipleobjecttracking(MOT)[Abstract]Journal of Vision3(9)585a
Pylyshyn Z W amp Storm R W (1988)Trackingmultipleindepen-denttargetsEvidenceforaparalleltrackingmechanismSpatial Vi-sion3179-197
Saiki J (2002)Multiple-objectpermanencetrackingLimitationinmaintenanceandtransformationofperceptualobjectsProgress in Brain Research140133-148
Schneider W amp Shiffrin R M (1977)ControlledandautomatichumaninformationprocessingIDetectionsearchandattentionPsychological Review841-66
Scholl B J amp Pylyshyn Z W (1999)Trackingmultiple itemsthroughocclusionCluestovisualobjecthoodCognitive Psychology38259-290
Scholl B J Pylyshyn Z W amp Feldman J (2001)Whatisavi-sualobjectEvidencefromtargetmerginginmultipleobjecttrackingCognition80159-177
Scholl B J Pylyshyn Z W amp Franconeri S L (1999May)When are spatiotemporal and featural properties encoded as a result of attentional allocationPaperpresentedattheconferenceoftheAssociationforResearchinVisionandOphthalmologyFortLau-derdaleFL
Sears C R amp Pylyshyn Z W (2000)MultipleobjecttrackingandattentionalprocessingCanadian Journal of Experimental Psychol-ogy541-14
Seiffert A E (2003)Dissociatingneuralcorrelatesofattentionaltrackingandattentiontovisualmotion[Abstract]Journal of Vision3(9)868a
Smith E E Jonides J Koeppe R A Awh E Schumacher E H amp Minoshima S (1995)SpatialversusobjectworkingmemoryPETinvestigationsJournal of Cognitive Neuroscience7337-356
Tipper S P (1985)ThenegativeprimingeffectInhibitoryprimingbyignoredobjectsQuarterly Journal of Experimental Psychology37A571-590
Ungerleider L G amp Mishkin M (1982)Twocorticalvisualsys-
184 HOrOwiTz eT al
temsInDJIngleMAGoodaleampRJWMansfield(Eds)Anal-ysis of visual behavior(pp549-586)CambridgeMAMITPress
Viswanathan L amp Mingolla E (2002)DynamicsofattentionindepthEvidence frommulti-element trackingPerception311415-1437
Vogel E K Woodman G F amp Luck S J (2001)Storageoffea-turesconjunctionsandobjectsinvisualworkingmemoryJournal of Experimental Psychology Human Perception amp Performance2792-114
Wheeler M E amp Treisman A M (2002)Bindinginshort-termvisualmemoryJournal of Experimental Psychology General13148-64
Wilson F A Oacute Scalaidhe S P amp Goldman-Rakic P S (1993)Dissociationofobjectandspatialprocessingdomains inprimateprefrontalcortexScience2601955-1958
Wolfe J M Oliva A Horowitz T S Butcher S J amp Bompas A (2002)SegmentationofobjectsfrombackgroundsinvisualsearchtasksVision Research422985-3004
Yantis S (1992)MultielementvisualtrackingAttentionandpercep-tualorganizationCognitive Psychology24295-340
NoTeS
1Wearemissingageinformationfor1observer2Threeobserverswhoparticipatedinbothexperimentswereelimi-
natedfromthisanalysis3Measuredcapacitycanvarybasedontrackingdurationtherange
ofmovementspeedsandwhetherornotobjectscanoccludeonean-other(asinourexperiment)orbounceoffoneanother(asinOksamaampHyoumlnauml2004)
APPeNdix A Stimulus List
beaver fox horse ostrich rhino turtlecamel giraffe kangaroo pelican rooster wolfcrocodile goat lion rabbit squirrel zebraelephant gorilla moose tiger
APPeNdix b raw data
Table b1 raw Numbers of items reported in the Standard Condition
Total Targets Nontargets Filled
Experiment1 308Experiment2 344 014 359Experiment3 347 015 362Experiment4 339 015 354Experiment5 Unique 312 037 349
Paired 249 080 329Identical 261 086 347
Experiment6 248 078 326Experiment7 599 599
NotemdashTargetsindicatestheaveragenumberofcorrectlyreportedtargetspertrialNontargetstheaveragenumberoftimesobserversclickedonanontargetlocationpertrialandTotal Filledthesumofthesetwocat-egoriesNontargetclickcountswerenotcollectedforExperiment1andtherewerenonontargetsinExperiment7
Table b2 raw Proportions for the Specific Condition by experiment
Errors
Correct Empty Trialsper Responses Transposition Nontarget Space N Observer
Experiment1 44 14 16 26 8 5000Experiment2 56 18 10 16 8 5000Experiment3 68 12 10 10 8 5000Experiment4 58 25 09 08 8 5063Experiment6 41 14 21 23 7 5000Experiment7 33 39 00 28 8 5000
NotemdashTranspositionerrorsrefertotrialsinwhichtheobserverclickedonthewrongtargetlocationTrialsperobservervariedinExperiment4becauseconditionwasselectedrandomlyfromtrialtotrial
(ManuscriptreceivedMarch312005revisionacceptedforpublicationFebruary132006)
174 HOrOwiTz eT al
IncontrasttovisualindexesobjectfilesareclearlycontentaddressableIntheirseminalarticlepresentingevidencefortheobjectfileconceptKahnemanTreismanandGibbs(1992)developedthereviewingparadigmInthisparadigmobserversseeapreviewdisplaythatin-cludesanumberofobjects(atleasttwo)Theobjectsarecharacterizedbysomefeatureoftenaletter(seeegMitroffSchollampWynn2004)Thefeaturalinformationdisappearsandthensomesortoflinkingdisplayusu-allyconsistingofsmoothorapparentmotionconnectstheobjectsinthepreviewdisplaywithobjectsinatargetdisplayTheobserverthenhastorespondtothetargetdis-playforexamplebyidentifyingaletterKahnemanetalfoundthatiftheletterwasthesameastheonepresentedinthatobjectinthepreviewdisplayobserverswerefastertorespondthanifthatletterhadappearedinanotherobjectinthepreviewdisplayindicatingthatinformationaboutwhatwasinthatobjecthadbeenpreserved
HereweintroduceanewvariationonthebasicMOTtaskthatallowsustolookattrackingofuniqueobjects(arelatedparadigmwasdevelopedbyOksamaampHyoumlnauml2004wewillreturntothisintheDiscussionsection)Inourtasktheobjectsconsistedofuniquecartoonanimals(seeFigure1)Wechosethesestimulibecausetheywerevisuallydistinctandeasilynameable
Theprincipaltechnicalproblemwithsuchstimuliisthatobserverscouldsimplymemorizetheidentitiesofthetargetanimalsatthebeginningofthetrialthenre-spondonthebasisofthoseidentitiesattheendofthetrialwithoutactuallytrackingthetargetswhiletheymovedWemadethisstrategyimpossiblebyhidingtheanimalsbehindcartooncactusesattheendofthetrialThesecac-tuseswerepresentthroughoutthetrialandtheanimalsmovedinfrontofthemAttheendofthetrialthedepth
relationshipswereswitchedandtheanimalsmovedbe-hindthecactusesandremainedthereInotherworkwehavefoundthatthesuddenappearanceofpreviouslyin-visibleoccluderswhetheroneatatimeorallatoncedidnotdisrupttrackingperformance(HorowitzBirnkrantFencsikTranampWolfe2006seealsoSchollampPylyshyn1999)Thusweconsideritunlikelythatamerechangeinthedepthrelationsbetweentrackedanimalsandoccludingcactusesshouldbeproblematic
Asecondtechnicalproblemwiththisarrangementisthatiftheendofeachtrialwerepredictableobserverscouldanticipatethisandlocatethetargetsjustpriortooc-clusionTothwartsuchastrategywemadethetimefromtrackingonsettoocclusionvariablefromtrialtotrialTheanimalsmovedinsuchafashionthattheyallapproachedthestationarycactusesevery1333msecThenumberofsuchcyclesinatrialvariedrandomlysoifobserverswerenotingtargetlocationsonlyattimeswhentheywerelikelytobeoccludedbycactusestheywouldhavetosuccess-fullysearchthedisplayforthetrackedanimalsinaman-nerthatproducedcorrectsearchresultsevery1333msecThisseemsunlikely
AttheendofatrialweaskedobserversoneoftwoquestionsThesequestionswereblockedinExperiments1ndash3andmixedwithinblocksinExperiment4Thestan-dardquestionwasldquoWhereareallthetargetsrdquoObserv-ershadtoclickonthecactusesinfrontofeachtargetThespecificquestionwasldquoWhereisthe_____rdquowiththeblankfilledbyoneofthetargetanimalsTheobserverthenhadtoclickonthecactuswherethatparticularani-malwasldquohidingrdquo
ThesetwoquestionsobviouslyhavedifferentratesofchanceperformanceInordertocompareperformanceonthesetwodifferentquestionswecomputedacommonmetricofcapacitybasedonthenumberofitemstrackedcorrectedforguessing(thedetailsaredescribedintheMethodsection)
InExperiment1wefoundthatcapacitywassubstan-tiallylowerforthespecificquestionthanforthestandardquestioninablockeddesignThisldquocontentdeficitrdquowasreplicatedinfourmoreexperimentsExperiment2ruledoutthepossibilitythatthecontentdeficitisduetointerfer-enceinshort-termmemoryInExperiment3weconfirmedthatthedatadonotchangeifweincreasethevariabilityofthetrackingdurationExperiment4replicatedthecontentdeficitwithamixeddesignusingbothvisualandauditorypostcuesTheremainingexperimentsemployedonlythestandardquestionbuttookadvantageofthepossibilitiesopenedbyourmethodExperiments5and6measuredhowmuchadditionalcapacityisgainedbyusinguniqueratherthanidenticalstimuliFinallyExperiment7measuredca-pacityinaldquofullreportrdquoversionofthestandardtask
MeTHod
observersUnlessotherwisespecifiedeachexperimentinvolved8observers
recruitedfromtheBrighamandWomenrsquosHospitalVisualAtten-tionLaboratoryvolunteerpoolTheobserversrangedinagefrom18to55years(M5283yearsSD597)1allhadvisualacuity
Figure 1 Cartoon animals used as stimuli See Appendix A for a list of stimulus names
Tracking UniqUe ObjecTs 175
ofatleast2025whencorrectedandhadpassedtheIshiharacolorscreenTheobserversgaveIRB-approvedinformedconsentandwerecompensated$10hforparticipationTwelveoftheobserversparticipatedinmorethanoneexperiment
Apparatus and StimuliThevisualstimuliwerepresentedon21-incolorCRTmoni-
tors(SuperScanMc801RasterOpsandMitsubishiDiamondPro91TXM)controlledbyPowerMacintoshG4computers(MacOS922)runningMATLAB521usingthePsychophysicsToolboxroutines(Brainard1997Pelli1997)Monitorspatialresolutionwassetto10243768pixelsThedisplayareasubtended361ordm3271ordmofvisualangleataviewingdistanceof574cmMonitorre-freshratesweresetto75Hz(133msecperframe)
The tracking stimuliwere23 cartoon animals (Figure1 seeAppendixAforalist)in8-bitcolorOccludersweretwotypesofcartooncactusesCartoonimageswerescaledsothatthemaximumdimension(horizontalorvertical)was35ordmforanimalsand53ordmforcactusesThebackgroundwasyellow
Cactuseswereplacedina534gridGridcellswere62ordmonasidesocactuseswereseparatedby09ordmedgetoedgeThetwotypesofcactuseswereassignedtocellsatrandomoneachtrialsothebackgroundcactusarraywasdifferentfromtrialtotrial
Auditoryprobes(Experiments45and6)wererecordedinafemalevoiceandplayedthroughapairofSonyMDRV150headphones
ProcedureOneachtrialeightanimalswereplacedinlocationsrandomly
selectedfromthegrid(Figure2A)Theselectionoftheanimalsvar-iedbyexperimentAtthestartofatrialthetargetanimalsblinkedonandoffsixtimesat1HzTherewerefourtargetsinExperiments1ndash6andeightinExperiment7TheanimalsthenbegantomoveForeachanimaladestinationcactuswasselectedatrandomwithoutre-placementAnimalsmovedtowardtheirdestinationsinstraightlinesatdifferentspeedssuchthateachanimalreacheditsdestinationatthesametime(Figure2B)Eachcycleconsistedof100monitorrefreshesor1333msecThemaximumspeedforananimalwas297ordmsec21(corner-to-cornermovement)andtheminimumspeed
Figure 2 Critical frames from the procedure Note that although we depict 3 3 4 grids in order to save space the grids actually used in the experiments were 5 3 4 At the beginning of a cycle each animal was in front of a cactus (A) each animal was then assigned a different cactus as a destination (b) each trajectory was completed in the same amount of time (one 1333-msec cycle) leading to different speeds for each animal on each cycle each animal ended the cycle in front of the destination cactus (C) except on the last cycle on which the animal was occluded by the cactus (d) for 200 msec Ani-mals were then removed for a 133-msec interval followed by the response phase To see an example of a trial go to searchbwhharvardedunewMoviesnoahmov
176 HOrOwiTz eT al
was46ordmsec21(movementtoanadjacentcactus)Oncethedestina-tionswerereachedanothersetofdestinationswasselectedAtrialconsistedof10ndash15ofthesecycles(exceptforExperiment3whichused5ndash15)ThenumberofcycleswasselectedrandomlyoneachtrialAnimalscouldoccludeoneanotherandmovedinfrontofthebackgroundcactusarray
AttheendofthelastcyclethecactusarraywasmovedtothefrontsuchthattheanimalswereinstantlyoccludedPartsofsomeanimalswouldbevisiblethroughtransparentportionsofthecac-tusimages(Figure2D)After200msectheoccludedanimalswereerasedeliminatinganycuestotheiridentitiesAfteranadditional133msecthecomputercursorappearedandaprobeinstructionwaspresentedatthetopofthescreen
InthestandardconditiontheinstructionwasldquoPleaseclickonthebusheswhereallofthetargetanimalsarehidingrdquoObserversrespondedbymovingthecursortoclickontheappropriatecactusesThecactusthatthecursorwascurrentlyoverwashighlightedwitharedborderOncetheobserverhadmadeanumberofresponsesequaltothenumberoftargetsfeedbackwaspresentedIfalltargetswerecorrectlyidentifiedthemessageldquoYougotallofthemrdquowasprintedatthetopofthescreenOtherwisetheobserverwastoldhowmanytargetsweremissedandthemissedtargetsblinkedonandofffourtimesattheappropriatelocations
InthespecificconditiontheprobeinstructionwasldquoWhereisthetargetrdquowheretargetwasrandomlyselectedfromthenamesofthetargetsforthattrialThetargetimagewasalsopre-sentedattheupperrightofthescreenTheobserverrsquostaskinthisconditionwastomovethecursortothecactuswheretheselectedtargethadbeenattheendofthetrialandclickTheanimalbehindtheselectedcactuswasthenrevealedFeedbackindicatedwhethertheobserverhadselectedthecorrectcactusorwhethertherewasnoanimalbehindthatcactus
Observerstypicallyparticipatedintwoblocksof50trialseachprecededby5practicetrialsWhenconditionswereblockedblockorderwascounterbalancedacrossobserversTheexperimentstypi-callytookbetween1and1-12htocomplete
data AnalysisRawaccuracydataweretransformedaccordingtohigh-threshold
guessingmodelsdevisedforeachconditionInthestandardcondi-tiontheguessingmodel(seeEquation1)wasderivedfromtheonepresentedbyScholletal(2001)WeassumedthatperformancePintermsofthenumberoftargetscorrectlyidentifiedwasdeter-minedbythenumberofobjectsactuallytracked(kforcapacity)plusguessingInEquation1tisthenumberoftargetsandathenumberofpossibleresponseoptionsinthiscasecactusesTheobservermakeskinformedresponsesandthenguessesonthere-maining(t2k)targetsleaving(t2k)targetsand(a2k)possibleresponsesPerformanceisthereforegivenby
P kt k
a k= +
minus( )minus
2
(1)
solvingforkgives
k aP ta P t
= minus+ minus
2
2 (2)
PerformanceinthespecificconditionwasmodeledaccordingtosimilarlogicHereperformancepistheprobabilityofcorrectlyse-lectingthecactuswherethespecifictargetishidingTheprobabilitythattheprobeanimal(selectedatrandomfromthettargets)willbeoneofthekobjectsbeingtrackedisktSoonkttrialsperformanceis10Ontheremaining(12kt)trialstheobserverguessesInthiscasetheobservercaneliminatetheklocationswithtrackedanimalsbecausesheknowstheycontainananimalthatwasnotprobedsosheguessesfromamongtheremaining(a2k)locationsHenceperformanceisgivenbyEquation3
p kt
kt a k
= + minus( ) minus( )1 1 (3)
Whenweagainsolvefork
ka pt a pt apt t
=+ minus minus minus minus( ) minus minus( )1 1 4
2
2
(4)
Equations2and4convertperformanceinbothconditionstoacom-monmetrick
Therawdataunderlyingthecapacitycomputationsforallex-perimentsarereportedinAppendixBThemajorityofouranalysesweresingle-degree-of-freedomplannedcomparisonscarriedoutthroughpairedttests
exPeriMeNT 1
InthisexperimentweintroducetheparadigmandthebasicresultthecontentdeficitInthisversionoftheex-perimentthesetof8uniqueanimalstobepresentedoneachtrialwasselectedatrandomfromthe23availableanimalsAsubsetof4targetanimalswasthenselectedatrandomfromamongthese8NoattemptwasmadetocontroloverlapinsetsfromtrialtotrialThestandardandspecificconditionswereruninseparateblocks
Estimatedcapacitywas16items(SEM503items)inthespecificconditionand29items(SEM503)inthestandardconditionThereweretwonotableresultsinthisexperimentFirstobserverswerereliablyabletoreportontheidentityofmorethanoneiteminthespecificcondition[t(7)550p001one-tailed]indicatingacontent-addressablerepresentationSecondestimatedcapacitywaslowerinthespecificconditionthaninthestandardcondition[t(7)580p0001two-tailed]ThatiswhenaskedtoreportthelocationofaspecifictargetobserversappearedtohavetrackedfeweritemsthanwhentheywereaskedtosimplyclickonallofthetargetlocationsWewillcallthisresultthecontent deficitSubsequentexperimentswilltestpossibleexplanationsforthisdeficit
OnepuzzlingaspectofthesedatawasthatperformanceinthestandardconditionseemedlowinthisexperimentincomparisonwithtypicalMOTexperimentswithidenticalitemsObserversaretypicallyabletotrackmorethanthreeitems(seeOksamaampHyoumlnauml2004Scholletal2001)Our taskmayhavebeenmoredifficultsimplybecausetheanimalsoftenoccludedoneanotherwhichcanreduceperformance(KliegerHorowitzampWolfe2004)ThereisalsoamoreseriouspossibilitySincethestimulussetchangedfromtrialtotrialobserversmayhaveexperiencedinterferenceinworkingmemoryInparticularitemsthatweretargetsononetrialcouldhaveservedasnontargetsonsubsequenttrialsandviceversasuchconditionshaveledtointerferenceinboththevisualsearch(egldquovariablemappingrdquoSchneiderampShiffrin1977)andnegativeprim-ing(MillikenTipperampWeaver1994Tipper1985)para-digmsSuchinterferencemightbeparticularlyproblematicinthespecificconditionThusthechangingstimulussetsmightberesponsibleforboththecontentaddressabilityandthecontentdeficitfromthedataExperiment2wasdesignedtoremedythisproblem
exPeriMeNT 2
Inthisexperimentwesimplyfixedthetargetandnon-targetsetssothatthesamestimuliwereusedthroughout
Tracking UniqUe ObjecTs 177
ablockoftrialsThatisanobservermighttracktheliontherabbittheturtleandthegoatonallthetrialsinablockThismadeiteasierforobserverstorememberwhattheyweresupposedtotrackInadditioniteliminatedthepossibilityof interferenceornegativeprimingarisingfromtargetsandnontargetsswitchingrolesWeexpectedthatperformanceinthestandardconditionwouldimprovetoabove30itemsthequestionwaswhetherthecontentdeficitwouldstillmanifestitselfundertheseconditions
TheprocedurewasidenticaltothatofExperiment1exceptintheselectionofanimalsForeachobserveronesetofeightanimalswasselectedrandomlyatthebegin-ningofeachblockFouroftheseanimalsweredesignatedastargetsandtherestasnontargetsThesetofanimalsandthetargetnontargetassignmentswereheldconstantforthedurationofablock
Asexpectedperformanceinthestandardconditionimprovedtoacreditable34(SEM502)itemsHow-ever the content deficit was still present Observerstrackedonly21(SEM501)itemswhenaskedwhichtargetwaswhereThisdeficitwasreliable[t(7)583p00001]andsimilarinmagnitudetothatobservedinExperiment1ofaround13itemsAmixedANOVAcomparingthetwoexperiments2revealedalargeeffectof condition [F(18)5 1152p 000001]butnoneofexperiment[F(18)510p 10]Theinteractionterminthebetween-experimentsanalysiswasnearzero[F(18)1]
Theroughlyhalf-itemincreaseincapacitywhenwechangedfromavariabletoafixedtargetsetsuggeststhatswitchingtargetsetsfromtrialtotrialcreatedsomesortofinterferenceThiswastruenotonlyinthespecificcon-ditionwhenweprobedtargetidentitiesbutalsointhestandardconditionThisresultisimportantbecauseitsuggestsaroleforobjectidentity(eitherattheperceptualorsemanticlevel)irrespectiveoftaskdemandsAccord-ingtoLeonardandPylyshyn(2003)blinkingthetargetsonandoffshouldautomaticallyassignvisualindexestothemandtheseindexesshouldstickequallywellwhetherornottheseparticularstimuliweretargetsontheprevioustrialSincetheincreaseincapacitywasnotstatisticallyreliablethisconclusionshouldnotbetakentooseriouslyHoweverinsubsequentexperimentsweusedafixedtar-getset
JustasinExperiment1theobserversherecouldreli-ablylocatemultipletargetsaccordingtotheiridentitieswhichwetakeasevidenceforcontentaddressabilityEx-periment2alsoreplicatedthecontentdeficitwithreli-ablybettercapacityinthestandardreportconditionthaninthespecificreportcondition
TheseresultsraiseanumberofmethodologicalissuesFirstourexperimentsdifferfrommostMOTexperimentsintheirtemporalstructureourtrialswerebothlongerthanusualandincludedtemporaluncertaintyHowdidthesefactorsaffectperformanceExperiment3wasdesignedtoaddressthisissue
Asecondimportantissueiswhetherblockingresponsemodesinducedobserverstousedifferentencodingstrate-giesinthetwoconditionsleadingtodifferentialperfor-manceThisissuewillbeaddressedinExperiment4
AthirdissueiswhethertheuseofuniqueidentifiableobjectshelpsorhindersMOTperformanceincomparisonwiththeuseofidenticalobjectsinmostpreviousstudiesofMOTWewilladdressthisissueinExperiments5and6
FinallyExperiment7willaddresstheroleofnontar-getsinhinderingtrackingperformance
exPeriMeNT 3
Aswenotedintheintroductionitisimportanttoen-surethatobserversareattendingtoataskthroughoutatrialOneofourstrategiesweusedtoensurethiswastorandomlyvary trialdurationObserversdidnotknowwheneachtrialwouldendbetweenthe10thand15thcyclesWasthisenoughtemporaluncertaintyhoweverInExperiment3weincreasedtheuncertaintysothatatrialcouldendanywherebetweenthe5thand15thcycles(adurationfrom67to200sec)OtherwisethemethodwasidenticaltothatofExperiment2IfobserverswereactivelytrackingthetargetsonlyafterCycle9inthepre-viousexperimentsweshouldseeadecreaseinperfor-manceinthisexperimentwheretheypotentiallyhavetosustainattentionforalongerperiodoftime
Awiderarrayoftrialendpointsalsoallowedustoex-amineapossibleexplanationforthecontentdeficitWeusedtrialdurationsthatwerelongrelativetopreviousMOTstudiesPerhapslocationndashidentitybindingsdecayovertimeandwewouldobserveasmallerdifferencebe-tweenconditionsifweprobedearlierinthetrial
Figure3showskinExperiment3asafunctionofcyclewithdatafromExperiment2shownforcomparisonMeancapacityforthestandardconditionwas34items(SEM501)versus26items(SEM502)forthespecificcondi-tionrepresentingasignificantadvantageinthestandardcondition[t(7)539p 01]Wealsoconductedananalysisofcovarianceontheresultswithconditionasacategoricalvariableandcycleasacontinuousvariable(covariate)Neitherthemaineffectofcycle[F(17)1]northecycle3conditioninteraction[F(17)5249p516]wassignificant
5 7 9 11 13 150
1
2
3
4
Cycles
k (it
ems)
Figure 3 Capacity as a function of cycle Squares denote the standard and triangles the specific target condition data from experiment 3 are plotted in black and those from experiment 2 in gray error bars indicate standard errors of the means
178 HOrOwiTz eT al
WhencomparingthetwoexperimentsitseemsclearthatincreasingthetemporalvariabilityinExperiment3didnotchangetheresultsnotablyWeconductedacross-experimentANOVAusingdataonlyfromCycles10ndash15inExperiment3Standardversusspecifictargetconditionandcyclewerewithin-subjectsfactorsandexperimentwasabetween-subjectsfactorWefoundnomaineffectofexperiment[F(114)511p 5 31]Thecondition3ex-perimentinteractionapproachedsignificance[F(114)541p506]probablybecausethespecifictargetcapac-itywassomewhathigherinExperiment3Addingtempo-raluncertaintyclearlydidnotdisruptperformanceinthisexperimentinrelationtoExperiment2
ThereisahintinFigure3thatperformanceespeciallyin thespecific targetconditiondroppedas trialdura-tionincreasedThiswouldbecompatiblewithdatafromOksamaandHyoumlnauml(2004)possiblyreflectingavigilancedecrement(Parasuraman1986)IfthedropovertimeweregreaterinthespecificconditionthiswouldsupporttheideathatlocationndashidentitybindingsweredecayingHowevertherewasnoevidenceforthisinanANOVAconductedondatafromalloftheconditionsofExperiment3inwhichneitherthemaineffectofcycle[F(1070)511p538]northecycle3conditioninteraction[F(1070)1]wassignificantIftherewasdecayithadapproachedasymp-totewithinthefirst6ndash7secoftracking
exPeriMeNT 4
OneobvioushypothesistoexplainthecontentdeficitisthatobserversencodeinformationdifferentlyinthetwoconditionsInthestandardconditiononlyspatiotemporalpropertiesareencodedInthespecifictargetconditionobserversknowthattheywillbeaskedaboutfeaturaloridentityinformationsotheyencodethataswellFeaturalinformation(orperhapsthebindingofidentityandlo-cation)takesupmoreresources(Bahrami2003Saiki2002) reducing tracking capacityThus informationaboutfeaturesoridentitiesrequiresmorecapacity
InExperiment4wetestedthisstrategicencodinghy-pothesisbymixingthequestionswithinablockoftri-alsObserversdidnotknowonagiventrialwhethertheywouldbeaskedtolocateallofthetargetsorjustaspecifictargetThereforetheyhadtoadoptthesamestrategiesonbothtypesoftrialsIfthestrategicencodinghypothesisiscorrectthenthecontentdeficitshouldbereducedinthisexperimentPerformanceshouldalsobereducedoverallsinceobserverswouldnotbeoptimallyencodinginfor-mationoneachtrialInotherrespectsthemethodwasidenticaltothatofExperiment2
Pilotworkindicatedthatwhenquestionsweremixedwithinablockof trialshavingto lookawayfromthecactusarraytoreadthequestionatthetopofthescreenimpairedperformanceThereforeinthisexperimentweusedauditoryratherthanvisualprobesInsteadofprintingthequestionatthetopofthescreenobserversheardtheprobequestionsoverheadphones
Meancapacityforthestandardconditionwas34items(SEM501)asopposedto22items(SEM502)forthe
specificconditionrepresentingasignificantadvantageforthestandardcondition[t(7)599p00005]Per-formanceinthisexperimentwithquestionsmixedwithinblockswasnearlyidenticaltothatinExperiment2withblockedquestionsEvenwhenobserverscouldnothaveadopteddifferentencodingstrategies for thedifferentconditionswestillobservedasizable(12-item)contentdeficitThissuggeststhatthecontentdeficitisnotaresultofthestrategicdemandsoftheconditionswedevised
exPeriMeNT 5
ThepreviousexperimentsestablishedaparadigmforstudyingMOTwithunique stimuliOnequestion thatwehadnotyetaddressediswhetheruniqueobjectshelporhinderMOTperformanceincomparisonwithidenticalobjectsForexampleiftheobjectsareidenticalanob-servermightaccidentallyswapatargetandadistractoriftheyapproachtoocloselyWithuniqueobjectsthisislesslikelytohappenSimilarlyifatsomepointtheobserverrealizedthathewasonlytrackingthreetargetsandhadlostonehecouldtheoreticallyrecoverthelosttargetifitwereuniqueOfcoursesuchadvantagesassumeasortofldquoidealobserverrdquowhoalwaysknowsexactlywhatheistrackingIfthetrackingsystemwascompletelyinsensitivetotargetfeaturesoridentityuniqueobjectswouldprovidenoadvan-tageSimilarlytheobservercouldonlyrecoveralosttargetifinformationwereavailableabouthowmanytargetswerecurrentlybeingtrackedandwhichonewasmissing
Inthisexperimentwecomparedtrackinguniqueob-jectswithtrackingidenticalobjectsSinceldquoWhereisthezebrardquoisnotadiagnosticquestionwhenallobjectsarezebrasweonlyusedthestandardresponsemodeTherewerethreeconditionsTheuniqueconditionwasidenti-caltothestandardresponseconditioninExperiment1Intheidenticalconditioneightidenticalanimalswerepresented animal identitywas selectedat randomoneachtrialFinallyinthepairedconditiontherewerefouruniquetargetsagainselectedatrandomForeachtargettherewasanidenticaldistractorProbeswerepresentedauditorilyasinExperiment4Atotalof10observerspar-ticipatedinthisexperiment
Wecancompareperformanceintheuniqueandidenti-calobjectconditionstodeterminewhetherthereisanyadvantageforuniqueobjectsIftheadvantageforuniqueobjectsderivesfromtheabilitytorecoverlosttargetsthisadvantageshouldbeabolishedinthepairedcondition
CapacityvaluesforthethreeconditionsareplottedinFigure4Observerstracked31items(SEM502)intheuniquecondition23(SEM502)inthepairedconditionand25(SEM501)intheidenticalconditionPlannedttestsshowedthatthepairedandidenticalconditionsdidnotdiffer[t(9)513p 10]butthatperformancewassignificantlyworseinboththanintheuniquecondition[forpairedt(9)527p05foridenticalt(9)527p01]
TheseresultsindicatethatobserverscantakeadvantageoftheadditionalinformationprovidedbyuniqueobjectsTheuniqueobjectadvantage(approximately06items
Tracking UniqUe ObjecTs 179
undertheseconditions)isentirelyeliminatediftargetsareidenticaltodistractorssuggestingthatsomesortoferrorrecoveryprocessisresponsible
exPeriMeNT 6
In the previous experiment pairing targets anddistractorsreducedperformanceinthestandardcondi-tionrelativetothatinthefullyuniqueconditionWouldthismanipulationaffectthespecificconditionaswellInExperiment6wemeasuredthecontentdeficitwithbothpairedanduniquestimuliBycomparingthesetwostimulusconditionswecandeterminewhethereliminat-ing featuraldistinctionsbetween target andnontargetsetsimpairstheabilitytotrackspecificobjectsaswellastheabilitytoholdontotargetlocationsProbeswerepresentedauditorilyasinExperiment5Therewere6ob-serversinthisexperiment
AsshowninFigure5weonceagainreplicated thecontentdeficitinthisexperimentsinceobserverscouldtrack11moreitemsinthestandardconditionthaninthe
specificcondition[F(19)51190p001]Pairingtar-getsandnontargetsreducedcapacitybyroughlythesameamountas inExperiment5 [09 itemsF(19)5373p001]Criticallythepairingmanipulationappearedtohavelessofadeleteriouseffectinthespecificconditionthaninthestandardcondition[interactionF(19)5185p005]Inthestandardconditionthepairingmanipu-lationcostobservers12itemsversusalossofonly07itemsinthespecificconditionThisinteractionsuggeststhatdifferentrepresentationsmayunderlieperformanceinthetwoconditions
Ontheotherhandcapacitywasalreadylowerinthespe-cificconditionIftheeffectofpairingstimuliismultipli-cativeratherthanadditivewewouldthenexpectthattheimpactofpairedstimuliwouldbeproportionallysmallerinthespecificconditionWhenwelog-transformedthecapacityvaluesconvertingproportionaldifferencesintoadditive ones the question3 pairing interaction waseliminated[F(19)1]
Sincewecannotrejectthehypothesisthattheeffectofpairingtargetswithnontargetsisproportionaltocapacitythisexperimentprovidesatbestweakevidencefordiffer-entrepresentations
exPeriMeNT 7
OneadvantageofourmethodisthatwecantestMOTperformanceinacasewithonlytargetsWhywouldwewanttodothisTherearereasonstobelievethatsuppress-ingnontargetsconsumessomeresourcesthatwouldother-wisebeavailablefortrackingtargetsForexamplethereisevidencethatobserverstrackbetterontrialsinwhichnontarget trajectories are repeated from earlier trials(OgawaampYagi2003)IfnontargetsareprocessedtheymayhavetobesuppressedinordertoreduceconfusionwithtargetsandPylyshynandLeonard(2003)showedevidenceforsuchinhibitionMoreoverunpublisheddatafromourlaboratoryindicatethatMOTperformancede-creasesasnontargetsareaddedtothedisplayevenwhenthenumberoftargetsremainsconstant
Accordingly we designed Experiment7 to test thehypothesisthattrackingcapacitywouldincreaseiftherewerenonontargetsThisexperimentwasidenticaltoEx-periment2exceptthatallitemsweretargetsCapacityinthestandardconditionwas58items(SEM502)sig-nificantlygreaterthanthe26items(SEM501)obtainedinthespecificcondition[t(7)5143p000005]
InagreementwithourhypothesiscapacityimprovedmarkedlyinthestandardconditionwhentherewerenodistractorsStandardcapacityinExperiment7wassig-nificantlygreaterthaninExperiments23and4whichweremethodologicallycomparableexceptforthenumberoftargetsandnontargets(allps00005)
Oneexplanationisthatpreviousexperimentsunderes-timatedcapacitybecauseofaceilingeffectTheseexperi-mentswerebasedonthehypothesisthatcapacityvariesacrosstrialswithnounderlyingdifferencesacrossexperi-mentsHoweverthemeasuredcapacityhasbeenlimitedbythenumberoftargetsForexampleifthereareonlyfour
Unique Paired Identical0
1
2
3
4
Condition
k (it
ems)
Figure 4 Capacity as a function of condition in experiment 5
Specific Standard0
1
2
3
4Unique
Paired
Condition
k (it
ems)
Figure 5 effect of pairing target and nontarget objects on ca-pacity in experiment 6 data from the paired conditions are plot-ted as gray bars and those from the unique conditions as open bars
180 HOrOwiTz eT al
targets(egExperiments2ndash4)observerscouldtrackallofthemontrialsinwhichtheircapacitywas4orgreateranytrialswithanactualcapacitygreaterthan4wouldthusappeartohaveacapacityof4InthiscasemeasuredmeancapacitywouldunderestimatethetruemeanInExperi-ment7whichincludedeighttargetsmeasuredcapacitycouldreflectactualcapacityuptoeightitemsleadingtoagreater(andmoreaccurate)estimateofmeancapacity
Wecantestthisceilinghypothesisagainstourhypoth-esisofincreasedmeancapacitybycomparingthecapacitydistributionsinExperiment7withthoseinExperiments2ndash4Accordingtotheceilinghypothesistheproportionoftrialswithacapacitylessthan4shouldbeidenticalacrossexperimentsFurthermoretheproportionoftrialswithacapacityof4orgreaterinExperiment7shouldbethesameastheproportionoftrialswithacapacityequalto4inExperiments2ndash4Incontrastthehypothesisthattrackingcapacityincreaseswhennonontargetsarepres-entpredictsthattheentiredistributionshouldshifttotherightleadingtofewertrialswithcapacitylessthan4inExperiment7thanintheearlierexperiments
We tested these hypotheses by computing capacityforeachtrialusingEquation2andgeneratingadistri-butionacrosstrialsforeachobserverFigure6plotsthedistributionofkforExperiment7andthecorrespondingdistributionforExperiments23and4combined(dis-tributionsfromeachoftheseexperimentswereaveragedacrossobservers)Clearlytheproportionoftrialswithanobservedcapacitylessthan4waslowerwitheighttargetsthanwithfourtargetsThisisconsistentwiththecapacitydistributionshiftingtotherightinExperiment7ratherthanmerespreadingofthek54trialsoverawiderrangeWesuggestthatthesuperiorperformanceinthisexperi-mentmayreflectadditionalcapacityfreedupwhenthereisnoneedtosuppressnontargetsHowevertheshapeofthedistributionforfourtargetsdoessuggestthatcapacitywasunderestimatedinthepreviousexperiments
TheconclusionsarequitedifferentifwelookatthespecifictargetconditionSpecifictargetcapacityinthis
experimentwasbetterthaninExperiment2(uncorrectedttestp05)butnotsignificantlydifferentfromtheca-pacitiesobservedinExperiments3and4( p05)Thusthetrackingoftargetidentitiesinthisconditiondoesnotseem tobe limitedby theneed to suppressnontargetidentities
diSCuSSioN
WehavedevelopedanewMOTparadigmthatallowsustotestthetrackingofuniqueobjectsFourmainfind-ingshaveemergedfromtheseexperimentsFirstthereisacontent-addressablerepresentationoftargetsduringMOT(specificcapacitywas1inallcases)SecondweobservedacontentdeficitThecontent-addressablerep-resentationhasalowercapacity(measuredinitems)thanthelocation-addressablerepresentation(iespecificca-pacitywaslowerthanstandardinallcases)Thirdthevi-sualsystemiscapableoftakingadvantageofdifferencesbetweenuniqueobjectstoimprovetrackingperformanceoverthelevelseenwithidenticalobjects(Experiment5)Thisadvantageappearstobedueprimarilytotheabilitytorecoverlosttargets(Experiment5)Fourthtrackingcapacityappearstobeconstrainedbytheneedtosup-pressnontargetswhentherearenonontargetscapacityincreasesmarkedly(Experiment7)
A Content-Addressable representation in MoTOurdata represent anexistenceproofof a content-
addressable representation inMOTSimilar data havebeenreportedbyOksamaandHyoumlnauml(2004)Intheirmul-tiple-identity-tracking(MIT)taskobserverstrackedmov-inglinedrawings(eitherobjectsorldquopseudo-objectsrdquo)forseveralsecondsAttheendofthetrialthestimuliweremaskedandasingleitemwasmarkedwithablackframeInasubsequentresponsescreenobservershadtoselectthepicturethatcorrespondedtothemarkedobjectOksamaandHyoumlnaumlobtainedamediancapacityoffouritems
Althoughitisdifficulttocomparecapacitymeasuresdirectlyacrossexperimentaltasks3itisinterestingthatbothweandOksamaandHyoumlnauml(2004)demonstratedthatobserversdoknowtosomeextentwhichtargetiswhichThisfindingisincontrasttothoseofPylyshyn(2004)whoseobservershaddifficultyidentifyingtargetsTwokeydifferencesbetweenourparadigmandMITontheonehandandPylyshynrsquos(2004)paradigmontheotherseemrelevanttothisdiscrepancyFirstinourstudyandthatofOksamaandHyoumlnaumlobjectidentitiesweredefinedbyintrinsicfeaturessuchasshapeand(inourexperi-ment)colorInPylyshynrsquos(2004)experimentldquoidentityrdquowasbasedonatangentialfeatureofanobjectsuchasinwhichcornerofthedisplayitbeganthetrialItseemslikelythattheshapeandcolorofanobjectmaybemoretightlyboundmarkersofldquoidentityrdquothaninitialpositionorabrieflypresentedletter
SecondinourexperimentsandthoseofOksamaandHyoumlnauml (2004) thevisualdifferencesbetweenobjectswerecontinuallyavailablethroughoutthetrackingphaseofthetrialonlymaskedduringtheresponsephaseIn
4 targets8 targets
0 2 4 6 80
1
2
3
4
5
6
Capacity
Pro
po
rtio
n o
f Tri
als
Figure 6 Capacity distributions for the standard condition in experiment 7 (gray lines striped area) and in experiments 2 3 and 4 (black lines stippled area) dashed lines indicate standard errors of the means
Tracking UniqUe ObjecTs 181
Pylyshynrsquos(2004)experimentstheobjectidentitieswerepresentedonlybrieflyatthestartofthetrialduringthemajorityofthetrackingphasetheobjectswerephysicallyidenticalItmaybethatthelinkbetweenobjectidentityandspatiotemporalpositiondecayedduringPylyshynrsquos(2004)taskbutwasconstantlyrefreshedinbothoursandOksamaandHyoumlnaumlrsquos
Ourexperimentsalsoprovidedsomeindirectevidenceforacontent-addressablerepresentationForexampleperformancewasconsistentlysuperior inexperimentsinwhichthesameobjectswereusedinthesamerolesfromtrialtotrial(Experiments23and4)thaninex-perimentsinwhichthestimulussetchangedfromtrialtotrial(Experiments15and6)Thisoccurrednotonlyinthespecifictargetconditionsinwhichobjectidentitywasrelevantbutalsointhestandardconditionsinwhichobjectidentitywasnot
one System or TwoIstrackingservedbyasinglerepresentationthatencodes
bothpositionandidentityoraretheretworepresentationsoneforpositionandoneforidentityThisquestionarisesbecauseofthestrikingcapacitydifferencesbetweenourresponseconditionsInallofourexperimentscapacitywassignificantlylowerforspecifictargetsthanfortargetsinthestandardMOTresponsemodeInExperiment4wedisconfirmedthehypothesisthatcapacitydifferencesresultedfromvariedencodingstrategies(seeegDavisWelchHolmesampShepherd2001)sinceweobtainedthesameresultwhenresponsemodesweremixedwithinablockaswedidwhentheywereruninseparateblocks
ThiscapacitydifferenceisnotconclusiveevidencethatseparaterepresentationsorsystemsareinvolvedThereareseveralwaysinwhichaunifiedaccountcouldex-plainthelowercapacityinthespecifictargetconditionForexampleonemightproposeasystemwithaslot-likestructure(VogelWoodmanampLuck2001)inwhichallslotsarenotcreatedequalIfthereweretwoslotswithsuf-ficientresolutiontoencodebothpositionandidentityandtwoslotsthatcouldonlyencodepositionwewouldalsoobtaindifferentcapacityestimatesinourtworesponseconditionsThisrepresentationcouldalsobedescribedintermsofKahnemanandTreismanrsquos(1984)objectfilesifweassumethatsomeoftheobjectfilesareemptyTheimpletionprocessmightinthatcasereturntheinforma-tionthatagivenobjectwasspatiotemporallycontinuouswithatargetobjectbutfailtoretrievethesemanticorperceptualcontentassociatedwiththatobjectfile
IntheMOTliteraturecapacityisgenerallyassumedtorefertoafixednumberofavailablerepresentationsinthevisualsystemeithervisualindexesorobjectfilesHow-everthereissomeevidencethatcapacityinMOTmightbebetterthoughtofasanundifferentiatedresource(AlvarezampCavanagh2005)InsteadofasetofslotsorpointerswemightassumethatthereisafixedamountofinformationthatcanbeencodedObserversmightencodepositionandidentityforalltargetsbutthebindingbetweenpositionsandidentitymaybenoisierthanthepositioninformationitself(ormaydecaymorequicklyExperiment3indicatedthatanysuchdecaywouldhavetoapproachitsasymptote
valuewithin6ndash7sec)leadingtoapparentlylowercapacitywhenresponsesdependonbinding
InsteadofonesystemtherealsomightbetwoAlthoughwetendtoassumethatasingleexperimentalparadigmprobestheperformanceofasinglecorrespondingfunc-tionalsystemitmaybethatMOTnaturallyrecruitsmulti-pleoverlappingsystemsThesimplesttwo-systemaccountmightbeaproposalthattrackingisservedbyaposition-trackingsystemsuchasvisualindexes(FINSTs)butthataccessingidentityndashpositionbindingsrequiresfocalatten-tion(WheelerampTreisman2002)Howeversincespecificcapacityisreliablygreaterthanoneitemwewouldhavetoassumethatmultipleitemscansimultaneouslybethefocusofattention(AwhampPashler2000Davisetal2001)
IfwewishtopreserveasinglefocusofattentionwemightinsteadproposethatbothvisualindexesandobjectfilescanbeusedVisualindexeswouldprovideonlypo-sitioninformationandobjectfilescouldsupportmorecomplexqueries
Itwillobviouslybedifficulttodefinitivelydecidebe-tweenoneandtwosystemsHowevertheevidenceofourfinalthreeexperimentspointstotwoindependentsystemsInExperiments5and6pairingtargetsandnontargetsre-ducedcapacityinthestandardbutnotinthespecifictar-getcondition(althoughwecouldnotrejectthehypothesisofaproportionaldecrease)InExperiment7standardca-pacityincreasedwhennonontargetswerepresentedbutspecificcapacitywasunaffected
MOThasbeenlikenedtoavisualworkingmemorytaskattimezero(CavanaghampAlvarez2005)soitisnaturaltolooktostudiesofworkingmemoryforanalogsofthecontent-addressableandlocation-onlysystemsTheclassictwo-pathwayschemeinvisionisthedistinctionbetweenventralanddorsalstreamsassociatedwithobjectrecogni-tionandspatialprocessingrespectively(UngerleiderampMishkin1982)ThisdistinctionhasbeenextendedintothedomainofworkingmemorybyWilsonOacuteScalaidheandGoldman-Rakic(1993)whoprovidedevidencethatobjectinformationandspatialinformationareneurallysegregatedinmonkeyprefrontalcortexwidelybelievedtobetheneuralsubstrateofworkingmemoryAsimilardissociationinhumanswasreportedusingPETimagingbySmithetal(1995)
Thedivisionofworkingmemoryintomultiplesubsys-temsgoesbackatleasttothemultiple-componentmodelofBaddeleyandHitch(1974)whichproposedseparatecomponents for different modalities a ldquophonologicallooprdquoforauditoryinformationandaldquovisuospatialsketchpadrdquoforvisualinformationInmorerecentversionsofthemodel(BaddeleyampLogie1999Logie1995)thesketchpadhasbeendividedintovisualandspatialsubcompo-nents roughlycorrespondingto theobjectandspatialsystemsdescribedbyWilsonetal(1993)Interestinglythespatialsubcomponent(theldquoinnerscriberdquo)isspecifi-callytaskedwithholdingsequenceorpathinformation(seeParmentierElfordampMayberry2005)makingitalikelysubstrateforMOTperformance
Thewidespreadinsistenceonaseparationbetweenvi-sualorobjectinformationandspatialinformationwouldseem problematic for our findings Standard MOT is
182 HOrOwiTz eT al
clearlyajobforthespatialsubsystem(PostleDrsquoEspositoampCorkin2005forexampleusedaversionofstandardMOTtotieupdorsalprocessinginamemoryinterferenceparadigm)Howevertheredoesnotseemtoberoominthataccountforacontent-addressablerepresentationOurspecifictargettask(aswellastheMITtaskofOksamaampHyoumlnauml2004)isnotapureobjectrecognitiontaskitrequireslocationndashidentitybindingswhichwouldseemtoinvolveintegratinginformationacrossstreamsTheproblemdisappears ifwe take intoaccount recentbe-havioral(OlsonampMarshuetz2005)andfMRI(PostleampDrsquoEsposito1999)investigationsdemonstratingthatobjectworkingmemoryrepresentationsalsocarrysomelocationinformation
NeuroimagingstudiesmightproveusefulhereStan-dardMOThasbeenshowntoactivateparietallobeareasinthedorsalstream(Culhametal1998Jovicichetal2001Seiffert2003)Itwouldbeinformativetoknowwhetherthepatternofactivationchangeswhenuniqueob-jectsareintroducedorwhenthetaskpotentiallyinvolvesobjectidentities
unique object Features in MoTInadditiontoourfindingsaboutcontentaddressability
intrackingwewereabletodemonstratethathavingvisu-allyuniqueobjectsmadetrackingeasierobserverswereabletotrackroughly06moreitemswhenobjectswereuniquethanwhenobjectswereidenticalInExperiments5and6weattributedthisresulttoanerrorrecoverystrat-egyIfIlosetrackofthezebraIcansearchforthezebraonthebasisofitsphysicalfeaturesandcontinuetotrackitThisstrategywouldobviouslybeuselessifallitemsarezebrasThisstrategyisoflimitedutilityifthereareazebratargetandazebradistractoronceIlosethezebraIthenhaveonlya50chanceofpickingupthetargetzebraasopposedtothedistractor
TheerrorrecoveryhypothesisisnottrivialIthassev-eralimportantimplicationsFirstitimpliesthatobserversknowwhattheyarenottrackingThephenomenologyofMOTissuchthatanobserverhasthefeelingofknow-inghowmanytargetsheorsheissuccessfullytrackingTheerrorrecoveryhypothesisimpliesthatwhenatargetislosttheobserverknowswhichone(givenuniquetargets)SecondtheerrorrecoveryhypothesisalsoassumesthatanobservercansearchforamissingtargetwhilecontinuingtotracktheremainingtargetsThisisinagreementwithrecentevidencethatobserverscansuccessfullyperformanMOTtaskandavisualsearchtaskonthesametrial(AlvarezHorowitzArsenioDiMaseampWolfe2005)Finallythehypothesisimpliestheabilitytoaddtothetrackingsetontheflywhiletrackingwithoutanyphysi-calcueTheseimplicationsneedtobeconfirmedthroughfurtherresearch
Howeverifobserversuseuniquephysicalfeaturesofobjectstorecoverlosttargetsinthisfashionitiscuriousthatthereisnosuchbenefitinthespecifictargetcondi-tionIfobserversdorecoverlosttargetsduringthecourse
ofatrialtheyappeartorecoveronlytheirlocationsnottheiridentities
AnalternativetotheerrorrecoveryhypothesisthoughnotamutuallyexclusiveoneisthattheresultsofEx-periment5canbeexplainedbytheperceptualsimilarityoftargetandnontargetitemsYantisrsquos(1992)approachtoMOThighlightstheimportanceofperceptualgroup-ingfactorsthatallowthetargetstobetreatedasaunitbythevisualsystemMostoftheresearchinthislinehasfocusedonspatiotemporalgroupingfactorssuchasthespatialrelationshipsbetweentheobjects(SearsampPylyshyn2000ViswanathanampMingolla2002Yantis1992)
Cangroupingby(nonspatial)featuralpropertiesim-provetrackingAnumberofstudieshavedemonstratedthatobservershave limitedexplicit access to featuralproperties (Bahrami 2003Saiki 2002Scholl etal1999)butthisfindingdoesnotaddressthedirectrolethatfeaturesmayplayWecouldexplaintheresultsofEx-periment5byproposingthatobserverswereabletouseaccidentaldifferencesintheshapeandcolorstatisticsoftargetsandnontargetstohelpsegregatethemvisuallyInthepairedconditiontargetsandnontargetshadidenticalfeaturestatisticssothatadvantagewaslostThishypoth-esiscouldalsoexplainwhyperformancewasbetterwithafixedstimulussetthanwithavariablestimulussetoveranumberoftrialswiththesamestimuliobserverslearnsubtledifferencesbetweenthetwosets
ConclusionsIntheintroductionwenotedthatreal-worlddynamic
attentiontasksdifferfromtraditionalMOTintwowaysFirstobserversintheeverydayworldusuallydealwithmultipleuniqueobjectsSecondtheyoftenwishtoknowwhereagiventargetmightberatherthanwhichobjectsarethetargetsTheexperimentspresentedhererepresentanattempttobridgelaboratoryMOTresearchandtheseeverydaycognitivetasksWehaveshownthatitiseasiertotrackuniqueobjectsthantotrackidenticalobjectsassum-ingthattheobjectsonewishestotrackarealsouniquelydifferentfromnontargetobjectsFurthermoreobserversdoknowwhattheyaretrackingTotakeoneofourex-amplesofecologicallyvalidtrackingtasksifyouwatchyourchildrenandtheirfriendsontheplaygroundyoucanbeawareofwhereyourkidsareatanygivenmomentyoumayalsobeawareofthelocationsoftheirfriendswithoutbeingabletotellonefriendfromtheotherFurthermoreifyoulosetrackofachildyouwillprobablybeawareofwhichoneyouhavelost
AuTHor NoTe
ThisresearchwassupportedbyNIMHGrantR0165576toTSHWethankKristinMichodandSkylerPlaceforassistancewithdatacollec-tionaswellasSteveLuckandananonymousreviewerforconstructivecriticismCorrespondencerelatingtothisarticlemaybesenttoTSHorowitzBrighamandWomenrsquosHospitalVisualAttentionLaboratory64SidneyStreetSuite170BostonMA02139(e-mailtoddhsearchbwhharvardedu)
Tracking UniqUe ObjecTs 183
reFereNCeS
Allen R McGeorge P Pearson D amp Milne A B (2004)AttentionandexpertiseinmultipletargettrackingApplied Cognitive Psychology18337-347
Alvarez G A amp Cavanagh P (2005)IndependentresourcesforattentionaltrackingintheleftandrightvisualhemifieldsPsychologi-cal Science16637-643
Alvarez G A amp Franconeri S (2004November)How many ob-jects can you trackPaperpresentedatthe12thAnnualWorkshoponObjectPerceptionampMemoryMinneapolisMN
Alvarez G A Horowitz T S Arsenio H C DiMase J S amp Wolfe J M (2005)DomultielementvisualtrackingandvisualsearchdrawcontinuouslyonthesamevisualattentionresourcesJournal of Experimental Psychology Human Perception amp Perfor-mance31643-667
Awh E Jonides J amp Reuter-Lorenz P A (1998)RehearsalinspatialworkingmemoryJournal of Experimental Psychology Human Perception amp Performance24780-790
Awh E amp Pashler H (2000)EvidenceforsplitattentionalfociJournal of Experimental Psychology Human Perception amp Perfor-mance26834-846
Baddeley A D amp Hitch G J (1974)WorkingmemoryInGHBower(Ed)The psychology of learning and motivation Advances in research and theory(Vol8pp47-90)NewYorkAcademicPress
Baddeley A D amp Logie R H (1999) Working memoryThemultiple-componentmodelInAMiyakeampPShah(Eds)Models of working memory Mechanisms of active maintenance and executive control(pp28-61)CambridgeCambridgeUniversityPress
Bahrami B (2003)ObjectpropertyencodingandchangeblindnessinmultipleobjecttrackingVisual Cognition10949-963
Brainard D H (1997)ThePsychophysicsToolboxSpatial Vision10433-436
Cavanagh P amp Alvarez G A (2005)TrackingmultipletargetswithmultifocalattentionTrends in Cognitive Sciences9349-354
Culham J C Brandt S A Cavanagh P Kanwisher N G Dale A M amp Tootell R B H (1998)CorticalfMRIactiva-tionproducedbyattentive trackingofmoving targetsJournal of Neurophysiology802657-2670
Davis G Welch V L Holmes A amp Shepherd A (2001)CanattentionselectonlyafixednumberofobjectsatatimePerception301227-1248
Henderson J M amp Hollingworth A (2003)EyemovementsandvisualmemoryDetectingchangestosaccadetargetsinscenesPer-ception amp Psychophysics6558-71
Horowitz T S Birnkrant R S Fencsik D E Tran L amp Wolfe J M (2006)HowdowetrackinvisibleobjectsPsychonomic Bulletin amp Review13516-523
Jovicich J Peters R J Koch C Braun J Chang L amp Ernst T (2001)Brainareasspecificforattentionalloadinamotion-trackingtaskJournal of Cognitive Neuroscience131048-1058
Kahneman D amp Treisman A (1984)ChangingviewsofattentionandautomaticityInRParasuramanampDRDavies(Eds)Varieties of attention(pp29-61)OrlandoAcademicPress
Kahneman D Treisman A amp Gibbs B J (1992)ThereviewingofobjectfilesObject-specificintegrationofinformationCognitive Psychology24175-219
Klieger S B Horowitz T S amp Wolfe J M (2004)Ismultipleobjecttrackingcolorblind[Abstract]Journal of Vision4(8)363a
Leonard C amp Pylyshyn Z W (2003)Measuringtheattentionaldemandofmultipleobjecttracking(MOT)[Abstract]Journal of Vi-sion3(9)582a
Logie R H (1995) Visuo-spatial working memory Hove UKErlbaum
Milliken B Tipper S P amp Weaver B (1994)NegativepriminginaspatiallocalizationtaskFeaturemismatchinganddistractorin-hibitionJournal of Experimental Psychology Human Perception amp Performance20624-646
Mitroff S R Scholl B J amp Wynn K (2004)DivideandconquerHowobjectfilesadaptwhenapersistingobjectsplitsintotwoPsy-chological Science15420-425
Ogawa H amp Yagi A (2003)Primingeffectsinmultipleobjecttrack-ingAnimplicitencodingbasedonglobalspatiotemporalinformation[Abstract]Journal of Vision3(9)339a
Oksama L amp Hyoumlnauml J (2004)Ismultipleobjecttrackingcarriedoutautomaticallybyanearlyvisionmechanismindependentofhigher-ordercognitionAnindividualdifferenceapproachVisual Cognition11631-671
Oliva A Torralba A Castelhano M S amp Henderson J M (2003)Top-downcontrolofvisualattentioninrealworldscenes[Ab-stract]Journal of Vision3(9)3a
Olson I R amp Marshuetz C (2005)RememberingldquowhatrdquobringsalongldquowhererdquoinvisualworkingmemoryPerception amp Psychophysics67185-194
Parasuraman R (1986)VigilancemonitoringandsearchInKRBoffLKaufmanampJPThomas(Eds)Handbook of perception and human performance Vol 2 Cognitive processes and performance(pp1-39)NewYorkWiley
Parmentier F B R Elford G amp Mayberry M (2005)Transi-tionalinformationinspatialserialmemoryPathcharacteristicsaffectrecallperformanceJournal of Experimental Psychology Learning Memory amp Cognition31412-427
Pelli D G (1997)TheVideoToolboxsoftwareforvisualpsychophysicsTransformingnumbersintomoviesSpatial Vision10437-442
Postle B R amp DrsquoEsposito M (1999)ldquoWhatrdquondashthenndashldquowhererdquoinvi-sualworkingmemoryAnevent-relatedfMRIstudyJournal of Cog-nitive Neuroscience11585-597
Postle B R DrsquoEsposito M amp Corkin S (2005)Effectsofver-balandnonverbalinterferenceonspatialandobjectvisualworkingmemoryMemory amp Cognition33203-212
Pylyshyn Z [W] (1989)Theroleoflocationindexesinspatialper-ceptionAsketchoftheFINSTspatial-indexmodelCognition3265-97
Pylyshyn Z W (2004)SomepuzzlingfindingsinmultipleobjecttrackingITrackingwithoutkeepingtrackofobjectidentitiesVisual Cognition11801-822
Pylyshyn Z W amp Leonard C (2003)Inhibitionofnontargetsdur-ingmultipleobjecttracking(MOT)[Abstract]Journal of Vision3(9)585a
Pylyshyn Z W amp Storm R W (1988)Trackingmultipleindepen-denttargetsEvidenceforaparalleltrackingmechanismSpatial Vi-sion3179-197
Saiki J (2002)Multiple-objectpermanencetrackingLimitationinmaintenanceandtransformationofperceptualobjectsProgress in Brain Research140133-148
Schneider W amp Shiffrin R M (1977)ControlledandautomatichumaninformationprocessingIDetectionsearchandattentionPsychological Review841-66
Scholl B J amp Pylyshyn Z W (1999)Trackingmultiple itemsthroughocclusionCluestovisualobjecthoodCognitive Psychology38259-290
Scholl B J Pylyshyn Z W amp Feldman J (2001)Whatisavi-sualobjectEvidencefromtargetmerginginmultipleobjecttrackingCognition80159-177
Scholl B J Pylyshyn Z W amp Franconeri S L (1999May)When are spatiotemporal and featural properties encoded as a result of attentional allocationPaperpresentedattheconferenceoftheAssociationforResearchinVisionandOphthalmologyFortLau-derdaleFL
Sears C R amp Pylyshyn Z W (2000)MultipleobjecttrackingandattentionalprocessingCanadian Journal of Experimental Psychol-ogy541-14
Seiffert A E (2003)Dissociatingneuralcorrelatesofattentionaltrackingandattentiontovisualmotion[Abstract]Journal of Vision3(9)868a
Smith E E Jonides J Koeppe R A Awh E Schumacher E H amp Minoshima S (1995)SpatialversusobjectworkingmemoryPETinvestigationsJournal of Cognitive Neuroscience7337-356
Tipper S P (1985)ThenegativeprimingeffectInhibitoryprimingbyignoredobjectsQuarterly Journal of Experimental Psychology37A571-590
Ungerleider L G amp Mishkin M (1982)Twocorticalvisualsys-
184 HOrOwiTz eT al
temsInDJIngleMAGoodaleampRJWMansfield(Eds)Anal-ysis of visual behavior(pp549-586)CambridgeMAMITPress
Viswanathan L amp Mingolla E (2002)DynamicsofattentionindepthEvidence frommulti-element trackingPerception311415-1437
Vogel E K Woodman G F amp Luck S J (2001)Storageoffea-turesconjunctionsandobjectsinvisualworkingmemoryJournal of Experimental Psychology Human Perception amp Performance2792-114
Wheeler M E amp Treisman A M (2002)Bindinginshort-termvisualmemoryJournal of Experimental Psychology General13148-64
Wilson F A Oacute Scalaidhe S P amp Goldman-Rakic P S (1993)Dissociationofobjectandspatialprocessingdomains inprimateprefrontalcortexScience2601955-1958
Wolfe J M Oliva A Horowitz T S Butcher S J amp Bompas A (2002)SegmentationofobjectsfrombackgroundsinvisualsearchtasksVision Research422985-3004
Yantis S (1992)MultielementvisualtrackingAttentionandpercep-tualorganizationCognitive Psychology24295-340
NoTeS
1Wearemissingageinformationfor1observer2Threeobserverswhoparticipatedinbothexperimentswereelimi-
natedfromthisanalysis3Measuredcapacitycanvarybasedontrackingdurationtherange
ofmovementspeedsandwhetherornotobjectscanoccludeonean-other(asinourexperiment)orbounceoffoneanother(asinOksamaampHyoumlnauml2004)
APPeNdix A Stimulus List
beaver fox horse ostrich rhino turtlecamel giraffe kangaroo pelican rooster wolfcrocodile goat lion rabbit squirrel zebraelephant gorilla moose tiger
APPeNdix b raw data
Table b1 raw Numbers of items reported in the Standard Condition
Total Targets Nontargets Filled
Experiment1 308Experiment2 344 014 359Experiment3 347 015 362Experiment4 339 015 354Experiment5 Unique 312 037 349
Paired 249 080 329Identical 261 086 347
Experiment6 248 078 326Experiment7 599 599
NotemdashTargetsindicatestheaveragenumberofcorrectlyreportedtargetspertrialNontargetstheaveragenumberoftimesobserversclickedonanontargetlocationpertrialandTotal Filledthesumofthesetwocat-egoriesNontargetclickcountswerenotcollectedforExperiment1andtherewerenonontargetsinExperiment7
Table b2 raw Proportions for the Specific Condition by experiment
Errors
Correct Empty Trialsper Responses Transposition Nontarget Space N Observer
Experiment1 44 14 16 26 8 5000Experiment2 56 18 10 16 8 5000Experiment3 68 12 10 10 8 5000Experiment4 58 25 09 08 8 5063Experiment6 41 14 21 23 7 5000Experiment7 33 39 00 28 8 5000
NotemdashTranspositionerrorsrefertotrialsinwhichtheobserverclickedonthewrongtargetlocationTrialsperobservervariedinExperiment4becauseconditionwasselectedrandomlyfromtrialtotrial
(ManuscriptreceivedMarch312005revisionacceptedforpublicationFebruary132006)
Tracking UniqUe ObjecTs 175
ofatleast2025whencorrectedandhadpassedtheIshiharacolorscreenTheobserversgaveIRB-approvedinformedconsentandwerecompensated$10hforparticipationTwelveoftheobserversparticipatedinmorethanoneexperiment
Apparatus and StimuliThevisualstimuliwerepresentedon21-incolorCRTmoni-
tors(SuperScanMc801RasterOpsandMitsubishiDiamondPro91TXM)controlledbyPowerMacintoshG4computers(MacOS922)runningMATLAB521usingthePsychophysicsToolboxroutines(Brainard1997Pelli1997)Monitorspatialresolutionwassetto10243768pixelsThedisplayareasubtended361ordm3271ordmofvisualangleataviewingdistanceof574cmMonitorre-freshratesweresetto75Hz(133msecperframe)
The tracking stimuliwere23 cartoon animals (Figure1 seeAppendixAforalist)in8-bitcolorOccludersweretwotypesofcartooncactusesCartoonimageswerescaledsothatthemaximumdimension(horizontalorvertical)was35ordmforanimalsand53ordmforcactusesThebackgroundwasyellow
Cactuseswereplacedina534gridGridcellswere62ordmonasidesocactuseswereseparatedby09ordmedgetoedgeThetwotypesofcactuseswereassignedtocellsatrandomoneachtrialsothebackgroundcactusarraywasdifferentfromtrialtotrial
Auditoryprobes(Experiments45and6)wererecordedinafemalevoiceandplayedthroughapairofSonyMDRV150headphones
ProcedureOneachtrialeightanimalswereplacedinlocationsrandomly
selectedfromthegrid(Figure2A)Theselectionoftheanimalsvar-iedbyexperimentAtthestartofatrialthetargetanimalsblinkedonandoffsixtimesat1HzTherewerefourtargetsinExperiments1ndash6andeightinExperiment7TheanimalsthenbegantomoveForeachanimaladestinationcactuswasselectedatrandomwithoutre-placementAnimalsmovedtowardtheirdestinationsinstraightlinesatdifferentspeedssuchthateachanimalreacheditsdestinationatthesametime(Figure2B)Eachcycleconsistedof100monitorrefreshesor1333msecThemaximumspeedforananimalwas297ordmsec21(corner-to-cornermovement)andtheminimumspeed
Figure 2 Critical frames from the procedure Note that although we depict 3 3 4 grids in order to save space the grids actually used in the experiments were 5 3 4 At the beginning of a cycle each animal was in front of a cactus (A) each animal was then assigned a different cactus as a destination (b) each trajectory was completed in the same amount of time (one 1333-msec cycle) leading to different speeds for each animal on each cycle each animal ended the cycle in front of the destination cactus (C) except on the last cycle on which the animal was occluded by the cactus (d) for 200 msec Ani-mals were then removed for a 133-msec interval followed by the response phase To see an example of a trial go to searchbwhharvardedunewMoviesnoahmov
176 HOrOwiTz eT al
was46ordmsec21(movementtoanadjacentcactus)Oncethedestina-tionswerereachedanothersetofdestinationswasselectedAtrialconsistedof10ndash15ofthesecycles(exceptforExperiment3whichused5ndash15)ThenumberofcycleswasselectedrandomlyoneachtrialAnimalscouldoccludeoneanotherandmovedinfrontofthebackgroundcactusarray
AttheendofthelastcyclethecactusarraywasmovedtothefrontsuchthattheanimalswereinstantlyoccludedPartsofsomeanimalswouldbevisiblethroughtransparentportionsofthecac-tusimages(Figure2D)After200msectheoccludedanimalswereerasedeliminatinganycuestotheiridentitiesAfteranadditional133msecthecomputercursorappearedandaprobeinstructionwaspresentedatthetopofthescreen
InthestandardconditiontheinstructionwasldquoPleaseclickonthebusheswhereallofthetargetanimalsarehidingrdquoObserversrespondedbymovingthecursortoclickontheappropriatecactusesThecactusthatthecursorwascurrentlyoverwashighlightedwitharedborderOncetheobserverhadmadeanumberofresponsesequaltothenumberoftargetsfeedbackwaspresentedIfalltargetswerecorrectlyidentifiedthemessageldquoYougotallofthemrdquowasprintedatthetopofthescreenOtherwisetheobserverwastoldhowmanytargetsweremissedandthemissedtargetsblinkedonandofffourtimesattheappropriatelocations
InthespecificconditiontheprobeinstructionwasldquoWhereisthetargetrdquowheretargetwasrandomlyselectedfromthenamesofthetargetsforthattrialThetargetimagewasalsopre-sentedattheupperrightofthescreenTheobserverrsquostaskinthisconditionwastomovethecursortothecactuswheretheselectedtargethadbeenattheendofthetrialandclickTheanimalbehindtheselectedcactuswasthenrevealedFeedbackindicatedwhethertheobserverhadselectedthecorrectcactusorwhethertherewasnoanimalbehindthatcactus
Observerstypicallyparticipatedintwoblocksof50trialseachprecededby5practicetrialsWhenconditionswereblockedblockorderwascounterbalancedacrossobserversTheexperimentstypi-callytookbetween1and1-12htocomplete
data AnalysisRawaccuracydataweretransformedaccordingtohigh-threshold
guessingmodelsdevisedforeachconditionInthestandardcondi-tiontheguessingmodel(seeEquation1)wasderivedfromtheonepresentedbyScholletal(2001)WeassumedthatperformancePintermsofthenumberoftargetscorrectlyidentifiedwasdeter-minedbythenumberofobjectsactuallytracked(kforcapacity)plusguessingInEquation1tisthenumberoftargetsandathenumberofpossibleresponseoptionsinthiscasecactusesTheobservermakeskinformedresponsesandthenguessesonthere-maining(t2k)targetsleaving(t2k)targetsand(a2k)possibleresponsesPerformanceisthereforegivenby
P kt k
a k= +
minus( )minus
2
(1)
solvingforkgives
k aP ta P t
= minus+ minus
2
2 (2)
PerformanceinthespecificconditionwasmodeledaccordingtosimilarlogicHereperformancepistheprobabilityofcorrectlyse-lectingthecactuswherethespecifictargetishidingTheprobabilitythattheprobeanimal(selectedatrandomfromthettargets)willbeoneofthekobjectsbeingtrackedisktSoonkttrialsperformanceis10Ontheremaining(12kt)trialstheobserverguessesInthiscasetheobservercaneliminatetheklocationswithtrackedanimalsbecausesheknowstheycontainananimalthatwasnotprobedsosheguessesfromamongtheremaining(a2k)locationsHenceperformanceisgivenbyEquation3
p kt
kt a k
= + minus( ) minus( )1 1 (3)
Whenweagainsolvefork
ka pt a pt apt t
=+ minus minus minus minus( ) minus minus( )1 1 4
2
2
(4)
Equations2and4convertperformanceinbothconditionstoacom-monmetrick
Therawdataunderlyingthecapacitycomputationsforallex-perimentsarereportedinAppendixBThemajorityofouranalysesweresingle-degree-of-freedomplannedcomparisonscarriedoutthroughpairedttests
exPeriMeNT 1
InthisexperimentweintroducetheparadigmandthebasicresultthecontentdeficitInthisversionoftheex-perimentthesetof8uniqueanimalstobepresentedoneachtrialwasselectedatrandomfromthe23availableanimalsAsubsetof4targetanimalswasthenselectedatrandomfromamongthese8NoattemptwasmadetocontroloverlapinsetsfromtrialtotrialThestandardandspecificconditionswereruninseparateblocks
Estimatedcapacitywas16items(SEM503items)inthespecificconditionand29items(SEM503)inthestandardconditionThereweretwonotableresultsinthisexperimentFirstobserverswerereliablyabletoreportontheidentityofmorethanoneiteminthespecificcondition[t(7)550p001one-tailed]indicatingacontent-addressablerepresentationSecondestimatedcapacitywaslowerinthespecificconditionthaninthestandardcondition[t(7)580p0001two-tailed]ThatiswhenaskedtoreportthelocationofaspecifictargetobserversappearedtohavetrackedfeweritemsthanwhentheywereaskedtosimplyclickonallofthetargetlocationsWewillcallthisresultthecontent deficitSubsequentexperimentswilltestpossibleexplanationsforthisdeficit
OnepuzzlingaspectofthesedatawasthatperformanceinthestandardconditionseemedlowinthisexperimentincomparisonwithtypicalMOTexperimentswithidenticalitemsObserversaretypicallyabletotrackmorethanthreeitems(seeOksamaampHyoumlnauml2004Scholletal2001)Our taskmayhavebeenmoredifficultsimplybecausetheanimalsoftenoccludedoneanotherwhichcanreduceperformance(KliegerHorowitzampWolfe2004)ThereisalsoamoreseriouspossibilitySincethestimulussetchangedfromtrialtotrialobserversmayhaveexperiencedinterferenceinworkingmemoryInparticularitemsthatweretargetsononetrialcouldhaveservedasnontargetsonsubsequenttrialsandviceversasuchconditionshaveledtointerferenceinboththevisualsearch(egldquovariablemappingrdquoSchneiderampShiffrin1977)andnegativeprim-ing(MillikenTipperampWeaver1994Tipper1985)para-digmsSuchinterferencemightbeparticularlyproblematicinthespecificconditionThusthechangingstimulussetsmightberesponsibleforboththecontentaddressabilityandthecontentdeficitfromthedataExperiment2wasdesignedtoremedythisproblem
exPeriMeNT 2
Inthisexperimentwesimplyfixedthetargetandnon-targetsetssothatthesamestimuliwereusedthroughout
Tracking UniqUe ObjecTs 177
ablockoftrialsThatisanobservermighttracktheliontherabbittheturtleandthegoatonallthetrialsinablockThismadeiteasierforobserverstorememberwhattheyweresupposedtotrackInadditioniteliminatedthepossibilityof interferenceornegativeprimingarisingfromtargetsandnontargetsswitchingrolesWeexpectedthatperformanceinthestandardconditionwouldimprovetoabove30itemsthequestionwaswhetherthecontentdeficitwouldstillmanifestitselfundertheseconditions
TheprocedurewasidenticaltothatofExperiment1exceptintheselectionofanimalsForeachobserveronesetofeightanimalswasselectedrandomlyatthebegin-ningofeachblockFouroftheseanimalsweredesignatedastargetsandtherestasnontargetsThesetofanimalsandthetargetnontargetassignmentswereheldconstantforthedurationofablock
Asexpectedperformanceinthestandardconditionimprovedtoacreditable34(SEM502)itemsHow-ever the content deficit was still present Observerstrackedonly21(SEM501)itemswhenaskedwhichtargetwaswhereThisdeficitwasreliable[t(7)583p00001]andsimilarinmagnitudetothatobservedinExperiment1ofaround13itemsAmixedANOVAcomparingthetwoexperiments2revealedalargeeffectof condition [F(18)5 1152p 000001]butnoneofexperiment[F(18)510p 10]Theinteractionterminthebetween-experimentsanalysiswasnearzero[F(18)1]
Theroughlyhalf-itemincreaseincapacitywhenwechangedfromavariabletoafixedtargetsetsuggeststhatswitchingtargetsetsfromtrialtotrialcreatedsomesortofinterferenceThiswastruenotonlyinthespecificcon-ditionwhenweprobedtargetidentitiesbutalsointhestandardconditionThisresultisimportantbecauseitsuggestsaroleforobjectidentity(eitherattheperceptualorsemanticlevel)irrespectiveoftaskdemandsAccord-ingtoLeonardandPylyshyn(2003)blinkingthetargetsonandoffshouldautomaticallyassignvisualindexestothemandtheseindexesshouldstickequallywellwhetherornottheseparticularstimuliweretargetsontheprevioustrialSincetheincreaseincapacitywasnotstatisticallyreliablethisconclusionshouldnotbetakentooseriouslyHoweverinsubsequentexperimentsweusedafixedtar-getset
JustasinExperiment1theobserversherecouldreli-ablylocatemultipletargetsaccordingtotheiridentitieswhichwetakeasevidenceforcontentaddressabilityEx-periment2alsoreplicatedthecontentdeficitwithreli-ablybettercapacityinthestandardreportconditionthaninthespecificreportcondition
TheseresultsraiseanumberofmethodologicalissuesFirstourexperimentsdifferfrommostMOTexperimentsintheirtemporalstructureourtrialswerebothlongerthanusualandincludedtemporaluncertaintyHowdidthesefactorsaffectperformanceExperiment3wasdesignedtoaddressthisissue
Asecondimportantissueiswhetherblockingresponsemodesinducedobserverstousedifferentencodingstrate-giesinthetwoconditionsleadingtodifferentialperfor-manceThisissuewillbeaddressedinExperiment4
AthirdissueiswhethertheuseofuniqueidentifiableobjectshelpsorhindersMOTperformanceincomparisonwiththeuseofidenticalobjectsinmostpreviousstudiesofMOTWewilladdressthisissueinExperiments5and6
FinallyExperiment7willaddresstheroleofnontar-getsinhinderingtrackingperformance
exPeriMeNT 3
Aswenotedintheintroductionitisimportanttoen-surethatobserversareattendingtoataskthroughoutatrialOneofourstrategiesweusedtoensurethiswastorandomlyvary trialdurationObserversdidnotknowwheneachtrialwouldendbetweenthe10thand15thcyclesWasthisenoughtemporaluncertaintyhoweverInExperiment3weincreasedtheuncertaintysothatatrialcouldendanywherebetweenthe5thand15thcycles(adurationfrom67to200sec)OtherwisethemethodwasidenticaltothatofExperiment2IfobserverswereactivelytrackingthetargetsonlyafterCycle9inthepre-viousexperimentsweshouldseeadecreaseinperfor-manceinthisexperimentwheretheypotentiallyhavetosustainattentionforalongerperiodoftime
Awiderarrayoftrialendpointsalsoallowedustoex-amineapossibleexplanationforthecontentdeficitWeusedtrialdurationsthatwerelongrelativetopreviousMOTstudiesPerhapslocationndashidentitybindingsdecayovertimeandwewouldobserveasmallerdifferencebe-tweenconditionsifweprobedearlierinthetrial
Figure3showskinExperiment3asafunctionofcyclewithdatafromExperiment2shownforcomparisonMeancapacityforthestandardconditionwas34items(SEM501)versus26items(SEM502)forthespecificcondi-tionrepresentingasignificantadvantageinthestandardcondition[t(7)539p 01]Wealsoconductedananalysisofcovarianceontheresultswithconditionasacategoricalvariableandcycleasacontinuousvariable(covariate)Neitherthemaineffectofcycle[F(17)1]northecycle3conditioninteraction[F(17)5249p516]wassignificant
5 7 9 11 13 150
1
2
3
4
Cycles
k (it
ems)
Figure 3 Capacity as a function of cycle Squares denote the standard and triangles the specific target condition data from experiment 3 are plotted in black and those from experiment 2 in gray error bars indicate standard errors of the means
178 HOrOwiTz eT al
WhencomparingthetwoexperimentsitseemsclearthatincreasingthetemporalvariabilityinExperiment3didnotchangetheresultsnotablyWeconductedacross-experimentANOVAusingdataonlyfromCycles10ndash15inExperiment3Standardversusspecifictargetconditionandcyclewerewithin-subjectsfactorsandexperimentwasabetween-subjectsfactorWefoundnomaineffectofexperiment[F(114)511p 5 31]Thecondition3ex-perimentinteractionapproachedsignificance[F(114)541p506]probablybecausethespecifictargetcapac-itywassomewhathigherinExperiment3Addingtempo-raluncertaintyclearlydidnotdisruptperformanceinthisexperimentinrelationtoExperiment2
ThereisahintinFigure3thatperformanceespeciallyin thespecific targetconditiondroppedas trialdura-tionincreasedThiswouldbecompatiblewithdatafromOksamaandHyoumlnauml(2004)possiblyreflectingavigilancedecrement(Parasuraman1986)IfthedropovertimeweregreaterinthespecificconditionthiswouldsupporttheideathatlocationndashidentitybindingsweredecayingHowevertherewasnoevidenceforthisinanANOVAconductedondatafromalloftheconditionsofExperiment3inwhichneitherthemaineffectofcycle[F(1070)511p538]northecycle3conditioninteraction[F(1070)1]wassignificantIftherewasdecayithadapproachedasymp-totewithinthefirst6ndash7secoftracking
exPeriMeNT 4
OneobvioushypothesistoexplainthecontentdeficitisthatobserversencodeinformationdifferentlyinthetwoconditionsInthestandardconditiononlyspatiotemporalpropertiesareencodedInthespecifictargetconditionobserversknowthattheywillbeaskedaboutfeaturaloridentityinformationsotheyencodethataswellFeaturalinformation(orperhapsthebindingofidentityandlo-cation)takesupmoreresources(Bahrami2003Saiki2002) reducing tracking capacityThus informationaboutfeaturesoridentitiesrequiresmorecapacity
InExperiment4wetestedthisstrategicencodinghy-pothesisbymixingthequestionswithinablockoftri-alsObserversdidnotknowonagiventrialwhethertheywouldbeaskedtolocateallofthetargetsorjustaspecifictargetThereforetheyhadtoadoptthesamestrategiesonbothtypesoftrialsIfthestrategicencodinghypothesisiscorrectthenthecontentdeficitshouldbereducedinthisexperimentPerformanceshouldalsobereducedoverallsinceobserverswouldnotbeoptimallyencodinginfor-mationoneachtrialInotherrespectsthemethodwasidenticaltothatofExperiment2
Pilotworkindicatedthatwhenquestionsweremixedwithinablockof trialshavingto lookawayfromthecactusarraytoreadthequestionatthetopofthescreenimpairedperformanceThereforeinthisexperimentweusedauditoryratherthanvisualprobesInsteadofprintingthequestionatthetopofthescreenobserversheardtheprobequestionsoverheadphones
Meancapacityforthestandardconditionwas34items(SEM501)asopposedto22items(SEM502)forthe
specificconditionrepresentingasignificantadvantageforthestandardcondition[t(7)599p00005]Per-formanceinthisexperimentwithquestionsmixedwithinblockswasnearlyidenticaltothatinExperiment2withblockedquestionsEvenwhenobserverscouldnothaveadopteddifferentencodingstrategies for thedifferentconditionswestillobservedasizable(12-item)contentdeficitThissuggeststhatthecontentdeficitisnotaresultofthestrategicdemandsoftheconditionswedevised
exPeriMeNT 5
ThepreviousexperimentsestablishedaparadigmforstudyingMOTwithunique stimuliOnequestion thatwehadnotyetaddressediswhetheruniqueobjectshelporhinderMOTperformanceincomparisonwithidenticalobjectsForexampleiftheobjectsareidenticalanob-servermightaccidentallyswapatargetandadistractoriftheyapproachtoocloselyWithuniqueobjectsthisislesslikelytohappenSimilarlyifatsomepointtheobserverrealizedthathewasonlytrackingthreetargetsandhadlostonehecouldtheoreticallyrecoverthelosttargetifitwereuniqueOfcoursesuchadvantagesassumeasortofldquoidealobserverrdquowhoalwaysknowsexactlywhatheistrackingIfthetrackingsystemwascompletelyinsensitivetotargetfeaturesoridentityuniqueobjectswouldprovidenoadvan-tageSimilarlytheobservercouldonlyrecoveralosttargetifinformationwereavailableabouthowmanytargetswerecurrentlybeingtrackedandwhichonewasmissing
Inthisexperimentwecomparedtrackinguniqueob-jectswithtrackingidenticalobjectsSinceldquoWhereisthezebrardquoisnotadiagnosticquestionwhenallobjectsarezebrasweonlyusedthestandardresponsemodeTherewerethreeconditionsTheuniqueconditionwasidenti-caltothestandardresponseconditioninExperiment1Intheidenticalconditioneightidenticalanimalswerepresented animal identitywas selectedat randomoneachtrialFinallyinthepairedconditiontherewerefouruniquetargetsagainselectedatrandomForeachtargettherewasanidenticaldistractorProbeswerepresentedauditorilyasinExperiment4Atotalof10observerspar-ticipatedinthisexperiment
Wecancompareperformanceintheuniqueandidenti-calobjectconditionstodeterminewhetherthereisanyadvantageforuniqueobjectsIftheadvantageforuniqueobjectsderivesfromtheabilitytorecoverlosttargetsthisadvantageshouldbeabolishedinthepairedcondition
CapacityvaluesforthethreeconditionsareplottedinFigure4Observerstracked31items(SEM502)intheuniquecondition23(SEM502)inthepairedconditionand25(SEM501)intheidenticalconditionPlannedttestsshowedthatthepairedandidenticalconditionsdidnotdiffer[t(9)513p 10]butthatperformancewassignificantlyworseinboththanintheuniquecondition[forpairedt(9)527p05foridenticalt(9)527p01]
TheseresultsindicatethatobserverscantakeadvantageoftheadditionalinformationprovidedbyuniqueobjectsTheuniqueobjectadvantage(approximately06items
Tracking UniqUe ObjecTs 179
undertheseconditions)isentirelyeliminatediftargetsareidenticaltodistractorssuggestingthatsomesortoferrorrecoveryprocessisresponsible
exPeriMeNT 6
In the previous experiment pairing targets anddistractorsreducedperformanceinthestandardcondi-tionrelativetothatinthefullyuniqueconditionWouldthismanipulationaffectthespecificconditionaswellInExperiment6wemeasuredthecontentdeficitwithbothpairedanduniquestimuliBycomparingthesetwostimulusconditionswecandeterminewhethereliminat-ing featuraldistinctionsbetween target andnontargetsetsimpairstheabilitytotrackspecificobjectsaswellastheabilitytoholdontotargetlocationsProbeswerepresentedauditorilyasinExperiment5Therewere6ob-serversinthisexperiment
AsshowninFigure5weonceagainreplicated thecontentdeficitinthisexperimentsinceobserverscouldtrack11moreitemsinthestandardconditionthaninthe
specificcondition[F(19)51190p001]Pairingtar-getsandnontargetsreducedcapacitybyroughlythesameamountas inExperiment5 [09 itemsF(19)5373p001]Criticallythepairingmanipulationappearedtohavelessofadeleteriouseffectinthespecificconditionthaninthestandardcondition[interactionF(19)5185p005]Inthestandardconditionthepairingmanipu-lationcostobservers12itemsversusalossofonly07itemsinthespecificconditionThisinteractionsuggeststhatdifferentrepresentationsmayunderlieperformanceinthetwoconditions
Ontheotherhandcapacitywasalreadylowerinthespe-cificconditionIftheeffectofpairingstimuliismultipli-cativeratherthanadditivewewouldthenexpectthattheimpactofpairedstimuliwouldbeproportionallysmallerinthespecificconditionWhenwelog-transformedthecapacityvaluesconvertingproportionaldifferencesintoadditive ones the question3 pairing interaction waseliminated[F(19)1]
Sincewecannotrejectthehypothesisthattheeffectofpairingtargetswithnontargetsisproportionaltocapacitythisexperimentprovidesatbestweakevidencefordiffer-entrepresentations
exPeriMeNT 7
OneadvantageofourmethodisthatwecantestMOTperformanceinacasewithonlytargetsWhywouldwewanttodothisTherearereasonstobelievethatsuppress-ingnontargetsconsumessomeresourcesthatwouldother-wisebeavailablefortrackingtargetsForexamplethereisevidencethatobserverstrackbetterontrialsinwhichnontarget trajectories are repeated from earlier trials(OgawaampYagi2003)IfnontargetsareprocessedtheymayhavetobesuppressedinordertoreduceconfusionwithtargetsandPylyshynandLeonard(2003)showedevidenceforsuchinhibitionMoreoverunpublisheddatafromourlaboratoryindicatethatMOTperformancede-creasesasnontargetsareaddedtothedisplayevenwhenthenumberoftargetsremainsconstant
Accordingly we designed Experiment7 to test thehypothesisthattrackingcapacitywouldincreaseiftherewerenonontargetsThisexperimentwasidenticaltoEx-periment2exceptthatallitemsweretargetsCapacityinthestandardconditionwas58items(SEM502)sig-nificantlygreaterthanthe26items(SEM501)obtainedinthespecificcondition[t(7)5143p000005]
InagreementwithourhypothesiscapacityimprovedmarkedlyinthestandardconditionwhentherewerenodistractorsStandardcapacityinExperiment7wassig-nificantlygreaterthaninExperiments23and4whichweremethodologicallycomparableexceptforthenumberoftargetsandnontargets(allps00005)
Oneexplanationisthatpreviousexperimentsunderes-timatedcapacitybecauseofaceilingeffectTheseexperi-mentswerebasedonthehypothesisthatcapacityvariesacrosstrialswithnounderlyingdifferencesacrossexperi-mentsHoweverthemeasuredcapacityhasbeenlimitedbythenumberoftargetsForexampleifthereareonlyfour
Unique Paired Identical0
1
2
3
4
Condition
k (it
ems)
Figure 4 Capacity as a function of condition in experiment 5
Specific Standard0
1
2
3
4Unique
Paired
Condition
k (it
ems)
Figure 5 effect of pairing target and nontarget objects on ca-pacity in experiment 6 data from the paired conditions are plot-ted as gray bars and those from the unique conditions as open bars
180 HOrOwiTz eT al
targets(egExperiments2ndash4)observerscouldtrackallofthemontrialsinwhichtheircapacitywas4orgreateranytrialswithanactualcapacitygreaterthan4wouldthusappeartohaveacapacityof4InthiscasemeasuredmeancapacitywouldunderestimatethetruemeanInExperi-ment7whichincludedeighttargetsmeasuredcapacitycouldreflectactualcapacityuptoeightitemsleadingtoagreater(andmoreaccurate)estimateofmeancapacity
Wecantestthisceilinghypothesisagainstourhypoth-esisofincreasedmeancapacitybycomparingthecapacitydistributionsinExperiment7withthoseinExperiments2ndash4Accordingtotheceilinghypothesistheproportionoftrialswithacapacitylessthan4shouldbeidenticalacrossexperimentsFurthermoretheproportionoftrialswithacapacityof4orgreaterinExperiment7shouldbethesameastheproportionoftrialswithacapacityequalto4inExperiments2ndash4Incontrastthehypothesisthattrackingcapacityincreaseswhennonontargetsarepres-entpredictsthattheentiredistributionshouldshifttotherightleadingtofewertrialswithcapacitylessthan4inExperiment7thanintheearlierexperiments
We tested these hypotheses by computing capacityforeachtrialusingEquation2andgeneratingadistri-butionacrosstrialsforeachobserverFigure6plotsthedistributionofkforExperiment7andthecorrespondingdistributionforExperiments23and4combined(dis-tributionsfromeachoftheseexperimentswereaveragedacrossobservers)Clearlytheproportionoftrialswithanobservedcapacitylessthan4waslowerwitheighttargetsthanwithfourtargetsThisisconsistentwiththecapacitydistributionshiftingtotherightinExperiment7ratherthanmerespreadingofthek54trialsoverawiderrangeWesuggestthatthesuperiorperformanceinthisexperi-mentmayreflectadditionalcapacityfreedupwhenthereisnoneedtosuppressnontargetsHowevertheshapeofthedistributionforfourtargetsdoessuggestthatcapacitywasunderestimatedinthepreviousexperiments
TheconclusionsarequitedifferentifwelookatthespecifictargetconditionSpecifictargetcapacityinthis
experimentwasbetterthaninExperiment2(uncorrectedttestp05)butnotsignificantlydifferentfromtheca-pacitiesobservedinExperiments3and4( p05)Thusthetrackingoftargetidentitiesinthisconditiondoesnotseem tobe limitedby theneed to suppressnontargetidentities
diSCuSSioN
WehavedevelopedanewMOTparadigmthatallowsustotestthetrackingofuniqueobjectsFourmainfind-ingshaveemergedfromtheseexperimentsFirstthereisacontent-addressablerepresentationoftargetsduringMOT(specificcapacitywas1inallcases)SecondweobservedacontentdeficitThecontent-addressablerep-resentationhasalowercapacity(measuredinitems)thanthelocation-addressablerepresentation(iespecificca-pacitywaslowerthanstandardinallcases)Thirdthevi-sualsystemiscapableoftakingadvantageofdifferencesbetweenuniqueobjectstoimprovetrackingperformanceoverthelevelseenwithidenticalobjects(Experiment5)Thisadvantageappearstobedueprimarilytotheabilitytorecoverlosttargets(Experiment5)Fourthtrackingcapacityappearstobeconstrainedbytheneedtosup-pressnontargetswhentherearenonontargetscapacityincreasesmarkedly(Experiment7)
A Content-Addressable representation in MoTOurdata represent anexistenceproofof a content-
addressable representation inMOTSimilar data havebeenreportedbyOksamaandHyoumlnauml(2004)Intheirmul-tiple-identity-tracking(MIT)taskobserverstrackedmov-inglinedrawings(eitherobjectsorldquopseudo-objectsrdquo)forseveralsecondsAttheendofthetrialthestimuliweremaskedandasingleitemwasmarkedwithablackframeInasubsequentresponsescreenobservershadtoselectthepicturethatcorrespondedtothemarkedobjectOksamaandHyoumlnaumlobtainedamediancapacityoffouritems
Althoughitisdifficulttocomparecapacitymeasuresdirectlyacrossexperimentaltasks3itisinterestingthatbothweandOksamaandHyoumlnauml(2004)demonstratedthatobserversdoknowtosomeextentwhichtargetiswhichThisfindingisincontrasttothoseofPylyshyn(2004)whoseobservershaddifficultyidentifyingtargetsTwokeydifferencesbetweenourparadigmandMITontheonehandandPylyshynrsquos(2004)paradigmontheotherseemrelevanttothisdiscrepancyFirstinourstudyandthatofOksamaandHyoumlnaumlobjectidentitiesweredefinedbyintrinsicfeaturessuchasshapeand(inourexperi-ment)colorInPylyshynrsquos(2004)experimentldquoidentityrdquowasbasedonatangentialfeatureofanobjectsuchasinwhichcornerofthedisplayitbeganthetrialItseemslikelythattheshapeandcolorofanobjectmaybemoretightlyboundmarkersofldquoidentityrdquothaninitialpositionorabrieflypresentedletter
SecondinourexperimentsandthoseofOksamaandHyoumlnauml (2004) thevisualdifferencesbetweenobjectswerecontinuallyavailablethroughoutthetrackingphaseofthetrialonlymaskedduringtheresponsephaseIn
4 targets8 targets
0 2 4 6 80
1
2
3
4
5
6
Capacity
Pro
po
rtio
n o
f Tri
als
Figure 6 Capacity distributions for the standard condition in experiment 7 (gray lines striped area) and in experiments 2 3 and 4 (black lines stippled area) dashed lines indicate standard errors of the means
Tracking UniqUe ObjecTs 181
Pylyshynrsquos(2004)experimentstheobjectidentitieswerepresentedonlybrieflyatthestartofthetrialduringthemajorityofthetrackingphasetheobjectswerephysicallyidenticalItmaybethatthelinkbetweenobjectidentityandspatiotemporalpositiondecayedduringPylyshynrsquos(2004)taskbutwasconstantlyrefreshedinbothoursandOksamaandHyoumlnaumlrsquos
Ourexperimentsalsoprovidedsomeindirectevidenceforacontent-addressablerepresentationForexampleperformancewasconsistentlysuperior inexperimentsinwhichthesameobjectswereusedinthesamerolesfromtrialtotrial(Experiments23and4)thaninex-perimentsinwhichthestimulussetchangedfromtrialtotrial(Experiments15and6)Thisoccurrednotonlyinthespecifictargetconditionsinwhichobjectidentitywasrelevantbutalsointhestandardconditionsinwhichobjectidentitywasnot
one System or TwoIstrackingservedbyasinglerepresentationthatencodes
bothpositionandidentityoraretheretworepresentationsoneforpositionandoneforidentityThisquestionarisesbecauseofthestrikingcapacitydifferencesbetweenourresponseconditionsInallofourexperimentscapacitywassignificantlylowerforspecifictargetsthanfortargetsinthestandardMOTresponsemodeInExperiment4wedisconfirmedthehypothesisthatcapacitydifferencesresultedfromvariedencodingstrategies(seeegDavisWelchHolmesampShepherd2001)sinceweobtainedthesameresultwhenresponsemodesweremixedwithinablockaswedidwhentheywereruninseparateblocks
ThiscapacitydifferenceisnotconclusiveevidencethatseparaterepresentationsorsystemsareinvolvedThereareseveralwaysinwhichaunifiedaccountcouldex-plainthelowercapacityinthespecifictargetconditionForexampleonemightproposeasystemwithaslot-likestructure(VogelWoodmanampLuck2001)inwhichallslotsarenotcreatedequalIfthereweretwoslotswithsuf-ficientresolutiontoencodebothpositionandidentityandtwoslotsthatcouldonlyencodepositionwewouldalsoobtaindifferentcapacityestimatesinourtworesponseconditionsThisrepresentationcouldalsobedescribedintermsofKahnemanandTreismanrsquos(1984)objectfilesifweassumethatsomeoftheobjectfilesareemptyTheimpletionprocessmightinthatcasereturntheinforma-tionthatagivenobjectwasspatiotemporallycontinuouswithatargetobjectbutfailtoretrievethesemanticorperceptualcontentassociatedwiththatobjectfile
IntheMOTliteraturecapacityisgenerallyassumedtorefertoafixednumberofavailablerepresentationsinthevisualsystemeithervisualindexesorobjectfilesHow-everthereissomeevidencethatcapacityinMOTmightbebetterthoughtofasanundifferentiatedresource(AlvarezampCavanagh2005)InsteadofasetofslotsorpointerswemightassumethatthereisafixedamountofinformationthatcanbeencodedObserversmightencodepositionandidentityforalltargetsbutthebindingbetweenpositionsandidentitymaybenoisierthanthepositioninformationitself(ormaydecaymorequicklyExperiment3indicatedthatanysuchdecaywouldhavetoapproachitsasymptote
valuewithin6ndash7sec)leadingtoapparentlylowercapacitywhenresponsesdependonbinding
InsteadofonesystemtherealsomightbetwoAlthoughwetendtoassumethatasingleexperimentalparadigmprobestheperformanceofasinglecorrespondingfunc-tionalsystemitmaybethatMOTnaturallyrecruitsmulti-pleoverlappingsystemsThesimplesttwo-systemaccountmightbeaproposalthattrackingisservedbyaposition-trackingsystemsuchasvisualindexes(FINSTs)butthataccessingidentityndashpositionbindingsrequiresfocalatten-tion(WheelerampTreisman2002)Howeversincespecificcapacityisreliablygreaterthanoneitemwewouldhavetoassumethatmultipleitemscansimultaneouslybethefocusofattention(AwhampPashler2000Davisetal2001)
IfwewishtopreserveasinglefocusofattentionwemightinsteadproposethatbothvisualindexesandobjectfilescanbeusedVisualindexeswouldprovideonlypo-sitioninformationandobjectfilescouldsupportmorecomplexqueries
Itwillobviouslybedifficulttodefinitivelydecidebe-tweenoneandtwosystemsHowevertheevidenceofourfinalthreeexperimentspointstotwoindependentsystemsInExperiments5and6pairingtargetsandnontargetsre-ducedcapacityinthestandardbutnotinthespecifictar-getcondition(althoughwecouldnotrejectthehypothesisofaproportionaldecrease)InExperiment7standardca-pacityincreasedwhennonontargetswerepresentedbutspecificcapacitywasunaffected
MOThasbeenlikenedtoavisualworkingmemorytaskattimezero(CavanaghampAlvarez2005)soitisnaturaltolooktostudiesofworkingmemoryforanalogsofthecontent-addressableandlocation-onlysystemsTheclassictwo-pathwayschemeinvisionisthedistinctionbetweenventralanddorsalstreamsassociatedwithobjectrecogni-tionandspatialprocessingrespectively(UngerleiderampMishkin1982)ThisdistinctionhasbeenextendedintothedomainofworkingmemorybyWilsonOacuteScalaidheandGoldman-Rakic(1993)whoprovidedevidencethatobjectinformationandspatialinformationareneurallysegregatedinmonkeyprefrontalcortexwidelybelievedtobetheneuralsubstrateofworkingmemoryAsimilardissociationinhumanswasreportedusingPETimagingbySmithetal(1995)
Thedivisionofworkingmemoryintomultiplesubsys-temsgoesbackatleasttothemultiple-componentmodelofBaddeleyandHitch(1974)whichproposedseparatecomponents for different modalities a ldquophonologicallooprdquoforauditoryinformationandaldquovisuospatialsketchpadrdquoforvisualinformationInmorerecentversionsofthemodel(BaddeleyampLogie1999Logie1995)thesketchpadhasbeendividedintovisualandspatialsubcompo-nents roughlycorrespondingto theobjectandspatialsystemsdescribedbyWilsonetal(1993)Interestinglythespatialsubcomponent(theldquoinnerscriberdquo)isspecifi-callytaskedwithholdingsequenceorpathinformation(seeParmentierElfordampMayberry2005)makingitalikelysubstrateforMOTperformance
Thewidespreadinsistenceonaseparationbetweenvi-sualorobjectinformationandspatialinformationwouldseem problematic for our findings Standard MOT is
182 HOrOwiTz eT al
clearlyajobforthespatialsubsystem(PostleDrsquoEspositoampCorkin2005forexampleusedaversionofstandardMOTtotieupdorsalprocessinginamemoryinterferenceparadigm)Howevertheredoesnotseemtoberoominthataccountforacontent-addressablerepresentationOurspecifictargettask(aswellastheMITtaskofOksamaampHyoumlnauml2004)isnotapureobjectrecognitiontaskitrequireslocationndashidentitybindingswhichwouldseemtoinvolveintegratinginformationacrossstreamsTheproblemdisappears ifwe take intoaccount recentbe-havioral(OlsonampMarshuetz2005)andfMRI(PostleampDrsquoEsposito1999)investigationsdemonstratingthatobjectworkingmemoryrepresentationsalsocarrysomelocationinformation
NeuroimagingstudiesmightproveusefulhereStan-dardMOThasbeenshowntoactivateparietallobeareasinthedorsalstream(Culhametal1998Jovicichetal2001Seiffert2003)Itwouldbeinformativetoknowwhetherthepatternofactivationchangeswhenuniqueob-jectsareintroducedorwhenthetaskpotentiallyinvolvesobjectidentities
unique object Features in MoTInadditiontoourfindingsaboutcontentaddressability
intrackingwewereabletodemonstratethathavingvisu-allyuniqueobjectsmadetrackingeasierobserverswereabletotrackroughly06moreitemswhenobjectswereuniquethanwhenobjectswereidenticalInExperiments5and6weattributedthisresulttoanerrorrecoverystrat-egyIfIlosetrackofthezebraIcansearchforthezebraonthebasisofitsphysicalfeaturesandcontinuetotrackitThisstrategywouldobviouslybeuselessifallitemsarezebrasThisstrategyisoflimitedutilityifthereareazebratargetandazebradistractoronceIlosethezebraIthenhaveonlya50chanceofpickingupthetargetzebraasopposedtothedistractor
TheerrorrecoveryhypothesisisnottrivialIthassev-eralimportantimplicationsFirstitimpliesthatobserversknowwhattheyarenottrackingThephenomenologyofMOTissuchthatanobserverhasthefeelingofknow-inghowmanytargetsheorsheissuccessfullytrackingTheerrorrecoveryhypothesisimpliesthatwhenatargetislosttheobserverknowswhichone(givenuniquetargets)SecondtheerrorrecoveryhypothesisalsoassumesthatanobservercansearchforamissingtargetwhilecontinuingtotracktheremainingtargetsThisisinagreementwithrecentevidencethatobserverscansuccessfullyperformanMOTtaskandavisualsearchtaskonthesametrial(AlvarezHorowitzArsenioDiMaseampWolfe2005)Finallythehypothesisimpliestheabilitytoaddtothetrackingsetontheflywhiletrackingwithoutanyphysi-calcueTheseimplicationsneedtobeconfirmedthroughfurtherresearch
Howeverifobserversuseuniquephysicalfeaturesofobjectstorecoverlosttargetsinthisfashionitiscuriousthatthereisnosuchbenefitinthespecifictargetcondi-tionIfobserversdorecoverlosttargetsduringthecourse
ofatrialtheyappeartorecoveronlytheirlocationsnottheiridentities
AnalternativetotheerrorrecoveryhypothesisthoughnotamutuallyexclusiveoneisthattheresultsofEx-periment5canbeexplainedbytheperceptualsimilarityoftargetandnontargetitemsYantisrsquos(1992)approachtoMOThighlightstheimportanceofperceptualgroup-ingfactorsthatallowthetargetstobetreatedasaunitbythevisualsystemMostoftheresearchinthislinehasfocusedonspatiotemporalgroupingfactorssuchasthespatialrelationshipsbetweentheobjects(SearsampPylyshyn2000ViswanathanampMingolla2002Yantis1992)
Cangroupingby(nonspatial)featuralpropertiesim-provetrackingAnumberofstudieshavedemonstratedthatobservershave limitedexplicit access to featuralproperties (Bahrami 2003Saiki 2002Scholl etal1999)butthisfindingdoesnotaddressthedirectrolethatfeaturesmayplayWecouldexplaintheresultsofEx-periment5byproposingthatobserverswereabletouseaccidentaldifferencesintheshapeandcolorstatisticsoftargetsandnontargetstohelpsegregatethemvisuallyInthepairedconditiontargetsandnontargetshadidenticalfeaturestatisticssothatadvantagewaslostThishypoth-esiscouldalsoexplainwhyperformancewasbetterwithafixedstimulussetthanwithavariablestimulussetoveranumberoftrialswiththesamestimuliobserverslearnsubtledifferencesbetweenthetwosets
ConclusionsIntheintroductionwenotedthatreal-worlddynamic
attentiontasksdifferfromtraditionalMOTintwowaysFirstobserversintheeverydayworldusuallydealwithmultipleuniqueobjectsSecondtheyoftenwishtoknowwhereagiventargetmightberatherthanwhichobjectsarethetargetsTheexperimentspresentedhererepresentanattempttobridgelaboratoryMOTresearchandtheseeverydaycognitivetasksWehaveshownthatitiseasiertotrackuniqueobjectsthantotrackidenticalobjectsassum-ingthattheobjectsonewishestotrackarealsouniquelydifferentfromnontargetobjectsFurthermoreobserversdoknowwhattheyaretrackingTotakeoneofourex-amplesofecologicallyvalidtrackingtasksifyouwatchyourchildrenandtheirfriendsontheplaygroundyoucanbeawareofwhereyourkidsareatanygivenmomentyoumayalsobeawareofthelocationsoftheirfriendswithoutbeingabletotellonefriendfromtheotherFurthermoreifyoulosetrackofachildyouwillprobablybeawareofwhichoneyouhavelost
AuTHor NoTe
ThisresearchwassupportedbyNIMHGrantR0165576toTSHWethankKristinMichodandSkylerPlaceforassistancewithdatacollec-tionaswellasSteveLuckandananonymousreviewerforconstructivecriticismCorrespondencerelatingtothisarticlemaybesenttoTSHorowitzBrighamandWomenrsquosHospitalVisualAttentionLaboratory64SidneyStreetSuite170BostonMA02139(e-mailtoddhsearchbwhharvardedu)
Tracking UniqUe ObjecTs 183
reFereNCeS
Allen R McGeorge P Pearson D amp Milne A B (2004)AttentionandexpertiseinmultipletargettrackingApplied Cognitive Psychology18337-347
Alvarez G A amp Cavanagh P (2005)IndependentresourcesforattentionaltrackingintheleftandrightvisualhemifieldsPsychologi-cal Science16637-643
Alvarez G A amp Franconeri S (2004November)How many ob-jects can you trackPaperpresentedatthe12thAnnualWorkshoponObjectPerceptionampMemoryMinneapolisMN
Alvarez G A Horowitz T S Arsenio H C DiMase J S amp Wolfe J M (2005)DomultielementvisualtrackingandvisualsearchdrawcontinuouslyonthesamevisualattentionresourcesJournal of Experimental Psychology Human Perception amp Perfor-mance31643-667
Awh E Jonides J amp Reuter-Lorenz P A (1998)RehearsalinspatialworkingmemoryJournal of Experimental Psychology Human Perception amp Performance24780-790
Awh E amp Pashler H (2000)EvidenceforsplitattentionalfociJournal of Experimental Psychology Human Perception amp Perfor-mance26834-846
Baddeley A D amp Hitch G J (1974)WorkingmemoryInGHBower(Ed)The psychology of learning and motivation Advances in research and theory(Vol8pp47-90)NewYorkAcademicPress
Baddeley A D amp Logie R H (1999) Working memoryThemultiple-componentmodelInAMiyakeampPShah(Eds)Models of working memory Mechanisms of active maintenance and executive control(pp28-61)CambridgeCambridgeUniversityPress
Bahrami B (2003)ObjectpropertyencodingandchangeblindnessinmultipleobjecttrackingVisual Cognition10949-963
Brainard D H (1997)ThePsychophysicsToolboxSpatial Vision10433-436
Cavanagh P amp Alvarez G A (2005)TrackingmultipletargetswithmultifocalattentionTrends in Cognitive Sciences9349-354
Culham J C Brandt S A Cavanagh P Kanwisher N G Dale A M amp Tootell R B H (1998)CorticalfMRIactiva-tionproducedbyattentive trackingofmoving targetsJournal of Neurophysiology802657-2670
Davis G Welch V L Holmes A amp Shepherd A (2001)CanattentionselectonlyafixednumberofobjectsatatimePerception301227-1248
Henderson J M amp Hollingworth A (2003)EyemovementsandvisualmemoryDetectingchangestosaccadetargetsinscenesPer-ception amp Psychophysics6558-71
Horowitz T S Birnkrant R S Fencsik D E Tran L amp Wolfe J M (2006)HowdowetrackinvisibleobjectsPsychonomic Bulletin amp Review13516-523
Jovicich J Peters R J Koch C Braun J Chang L amp Ernst T (2001)Brainareasspecificforattentionalloadinamotion-trackingtaskJournal of Cognitive Neuroscience131048-1058
Kahneman D amp Treisman A (1984)ChangingviewsofattentionandautomaticityInRParasuramanampDRDavies(Eds)Varieties of attention(pp29-61)OrlandoAcademicPress
Kahneman D Treisman A amp Gibbs B J (1992)ThereviewingofobjectfilesObject-specificintegrationofinformationCognitive Psychology24175-219
Klieger S B Horowitz T S amp Wolfe J M (2004)Ismultipleobjecttrackingcolorblind[Abstract]Journal of Vision4(8)363a
Leonard C amp Pylyshyn Z W (2003)Measuringtheattentionaldemandofmultipleobjecttracking(MOT)[Abstract]Journal of Vi-sion3(9)582a
Logie R H (1995) Visuo-spatial working memory Hove UKErlbaum
Milliken B Tipper S P amp Weaver B (1994)NegativepriminginaspatiallocalizationtaskFeaturemismatchinganddistractorin-hibitionJournal of Experimental Psychology Human Perception amp Performance20624-646
Mitroff S R Scholl B J amp Wynn K (2004)DivideandconquerHowobjectfilesadaptwhenapersistingobjectsplitsintotwoPsy-chological Science15420-425
Ogawa H amp Yagi A (2003)Primingeffectsinmultipleobjecttrack-ingAnimplicitencodingbasedonglobalspatiotemporalinformation[Abstract]Journal of Vision3(9)339a
Oksama L amp Hyoumlnauml J (2004)Ismultipleobjecttrackingcarriedoutautomaticallybyanearlyvisionmechanismindependentofhigher-ordercognitionAnindividualdifferenceapproachVisual Cognition11631-671
Oliva A Torralba A Castelhano M S amp Henderson J M (2003)Top-downcontrolofvisualattentioninrealworldscenes[Ab-stract]Journal of Vision3(9)3a
Olson I R amp Marshuetz C (2005)RememberingldquowhatrdquobringsalongldquowhererdquoinvisualworkingmemoryPerception amp Psychophysics67185-194
Parasuraman R (1986)VigilancemonitoringandsearchInKRBoffLKaufmanampJPThomas(Eds)Handbook of perception and human performance Vol 2 Cognitive processes and performance(pp1-39)NewYorkWiley
Parmentier F B R Elford G amp Mayberry M (2005)Transi-tionalinformationinspatialserialmemoryPathcharacteristicsaffectrecallperformanceJournal of Experimental Psychology Learning Memory amp Cognition31412-427
Pelli D G (1997)TheVideoToolboxsoftwareforvisualpsychophysicsTransformingnumbersintomoviesSpatial Vision10437-442
Postle B R amp DrsquoEsposito M (1999)ldquoWhatrdquondashthenndashldquowhererdquoinvi-sualworkingmemoryAnevent-relatedfMRIstudyJournal of Cog-nitive Neuroscience11585-597
Postle B R DrsquoEsposito M amp Corkin S (2005)Effectsofver-balandnonverbalinterferenceonspatialandobjectvisualworkingmemoryMemory amp Cognition33203-212
Pylyshyn Z [W] (1989)Theroleoflocationindexesinspatialper-ceptionAsketchoftheFINSTspatial-indexmodelCognition3265-97
Pylyshyn Z W (2004)SomepuzzlingfindingsinmultipleobjecttrackingITrackingwithoutkeepingtrackofobjectidentitiesVisual Cognition11801-822
Pylyshyn Z W amp Leonard C (2003)Inhibitionofnontargetsdur-ingmultipleobjecttracking(MOT)[Abstract]Journal of Vision3(9)585a
Pylyshyn Z W amp Storm R W (1988)Trackingmultipleindepen-denttargetsEvidenceforaparalleltrackingmechanismSpatial Vi-sion3179-197
Saiki J (2002)Multiple-objectpermanencetrackingLimitationinmaintenanceandtransformationofperceptualobjectsProgress in Brain Research140133-148
Schneider W amp Shiffrin R M (1977)ControlledandautomatichumaninformationprocessingIDetectionsearchandattentionPsychological Review841-66
Scholl B J amp Pylyshyn Z W (1999)Trackingmultiple itemsthroughocclusionCluestovisualobjecthoodCognitive Psychology38259-290
Scholl B J Pylyshyn Z W amp Feldman J (2001)Whatisavi-sualobjectEvidencefromtargetmerginginmultipleobjecttrackingCognition80159-177
Scholl B J Pylyshyn Z W amp Franconeri S L (1999May)When are spatiotemporal and featural properties encoded as a result of attentional allocationPaperpresentedattheconferenceoftheAssociationforResearchinVisionandOphthalmologyFortLau-derdaleFL
Sears C R amp Pylyshyn Z W (2000)MultipleobjecttrackingandattentionalprocessingCanadian Journal of Experimental Psychol-ogy541-14
Seiffert A E (2003)Dissociatingneuralcorrelatesofattentionaltrackingandattentiontovisualmotion[Abstract]Journal of Vision3(9)868a
Smith E E Jonides J Koeppe R A Awh E Schumacher E H amp Minoshima S (1995)SpatialversusobjectworkingmemoryPETinvestigationsJournal of Cognitive Neuroscience7337-356
Tipper S P (1985)ThenegativeprimingeffectInhibitoryprimingbyignoredobjectsQuarterly Journal of Experimental Psychology37A571-590
Ungerleider L G amp Mishkin M (1982)Twocorticalvisualsys-
184 HOrOwiTz eT al
temsInDJIngleMAGoodaleampRJWMansfield(Eds)Anal-ysis of visual behavior(pp549-586)CambridgeMAMITPress
Viswanathan L amp Mingolla E (2002)DynamicsofattentionindepthEvidence frommulti-element trackingPerception311415-1437
Vogel E K Woodman G F amp Luck S J (2001)Storageoffea-turesconjunctionsandobjectsinvisualworkingmemoryJournal of Experimental Psychology Human Perception amp Performance2792-114
Wheeler M E amp Treisman A M (2002)Bindinginshort-termvisualmemoryJournal of Experimental Psychology General13148-64
Wilson F A Oacute Scalaidhe S P amp Goldman-Rakic P S (1993)Dissociationofobjectandspatialprocessingdomains inprimateprefrontalcortexScience2601955-1958
Wolfe J M Oliva A Horowitz T S Butcher S J amp Bompas A (2002)SegmentationofobjectsfrombackgroundsinvisualsearchtasksVision Research422985-3004
Yantis S (1992)MultielementvisualtrackingAttentionandpercep-tualorganizationCognitive Psychology24295-340
NoTeS
1Wearemissingageinformationfor1observer2Threeobserverswhoparticipatedinbothexperimentswereelimi-
natedfromthisanalysis3Measuredcapacitycanvarybasedontrackingdurationtherange
ofmovementspeedsandwhetherornotobjectscanoccludeonean-other(asinourexperiment)orbounceoffoneanother(asinOksamaampHyoumlnauml2004)
APPeNdix A Stimulus List
beaver fox horse ostrich rhino turtlecamel giraffe kangaroo pelican rooster wolfcrocodile goat lion rabbit squirrel zebraelephant gorilla moose tiger
APPeNdix b raw data
Table b1 raw Numbers of items reported in the Standard Condition
Total Targets Nontargets Filled
Experiment1 308Experiment2 344 014 359Experiment3 347 015 362Experiment4 339 015 354Experiment5 Unique 312 037 349
Paired 249 080 329Identical 261 086 347
Experiment6 248 078 326Experiment7 599 599
NotemdashTargetsindicatestheaveragenumberofcorrectlyreportedtargetspertrialNontargetstheaveragenumberoftimesobserversclickedonanontargetlocationpertrialandTotal Filledthesumofthesetwocat-egoriesNontargetclickcountswerenotcollectedforExperiment1andtherewerenonontargetsinExperiment7
Table b2 raw Proportions for the Specific Condition by experiment
Errors
Correct Empty Trialsper Responses Transposition Nontarget Space N Observer
Experiment1 44 14 16 26 8 5000Experiment2 56 18 10 16 8 5000Experiment3 68 12 10 10 8 5000Experiment4 58 25 09 08 8 5063Experiment6 41 14 21 23 7 5000Experiment7 33 39 00 28 8 5000
NotemdashTranspositionerrorsrefertotrialsinwhichtheobserverclickedonthewrongtargetlocationTrialsperobservervariedinExperiment4becauseconditionwasselectedrandomlyfromtrialtotrial
(ManuscriptreceivedMarch312005revisionacceptedforpublicationFebruary132006)
176 HOrOwiTz eT al
was46ordmsec21(movementtoanadjacentcactus)Oncethedestina-tionswerereachedanothersetofdestinationswasselectedAtrialconsistedof10ndash15ofthesecycles(exceptforExperiment3whichused5ndash15)ThenumberofcycleswasselectedrandomlyoneachtrialAnimalscouldoccludeoneanotherandmovedinfrontofthebackgroundcactusarray
AttheendofthelastcyclethecactusarraywasmovedtothefrontsuchthattheanimalswereinstantlyoccludedPartsofsomeanimalswouldbevisiblethroughtransparentportionsofthecac-tusimages(Figure2D)After200msectheoccludedanimalswereerasedeliminatinganycuestotheiridentitiesAfteranadditional133msecthecomputercursorappearedandaprobeinstructionwaspresentedatthetopofthescreen
InthestandardconditiontheinstructionwasldquoPleaseclickonthebusheswhereallofthetargetanimalsarehidingrdquoObserversrespondedbymovingthecursortoclickontheappropriatecactusesThecactusthatthecursorwascurrentlyoverwashighlightedwitharedborderOncetheobserverhadmadeanumberofresponsesequaltothenumberoftargetsfeedbackwaspresentedIfalltargetswerecorrectlyidentifiedthemessageldquoYougotallofthemrdquowasprintedatthetopofthescreenOtherwisetheobserverwastoldhowmanytargetsweremissedandthemissedtargetsblinkedonandofffourtimesattheappropriatelocations
InthespecificconditiontheprobeinstructionwasldquoWhereisthetargetrdquowheretargetwasrandomlyselectedfromthenamesofthetargetsforthattrialThetargetimagewasalsopre-sentedattheupperrightofthescreenTheobserverrsquostaskinthisconditionwastomovethecursortothecactuswheretheselectedtargethadbeenattheendofthetrialandclickTheanimalbehindtheselectedcactuswasthenrevealedFeedbackindicatedwhethertheobserverhadselectedthecorrectcactusorwhethertherewasnoanimalbehindthatcactus
Observerstypicallyparticipatedintwoblocksof50trialseachprecededby5practicetrialsWhenconditionswereblockedblockorderwascounterbalancedacrossobserversTheexperimentstypi-callytookbetween1and1-12htocomplete
data AnalysisRawaccuracydataweretransformedaccordingtohigh-threshold
guessingmodelsdevisedforeachconditionInthestandardcondi-tiontheguessingmodel(seeEquation1)wasderivedfromtheonepresentedbyScholletal(2001)WeassumedthatperformancePintermsofthenumberoftargetscorrectlyidentifiedwasdeter-minedbythenumberofobjectsactuallytracked(kforcapacity)plusguessingInEquation1tisthenumberoftargetsandathenumberofpossibleresponseoptionsinthiscasecactusesTheobservermakeskinformedresponsesandthenguessesonthere-maining(t2k)targetsleaving(t2k)targetsand(a2k)possibleresponsesPerformanceisthereforegivenby
P kt k
a k= +
minus( )minus
2
(1)
solvingforkgives
k aP ta P t
= minus+ minus
2
2 (2)
PerformanceinthespecificconditionwasmodeledaccordingtosimilarlogicHereperformancepistheprobabilityofcorrectlyse-lectingthecactuswherethespecifictargetishidingTheprobabilitythattheprobeanimal(selectedatrandomfromthettargets)willbeoneofthekobjectsbeingtrackedisktSoonkttrialsperformanceis10Ontheremaining(12kt)trialstheobserverguessesInthiscasetheobservercaneliminatetheklocationswithtrackedanimalsbecausesheknowstheycontainananimalthatwasnotprobedsosheguessesfromamongtheremaining(a2k)locationsHenceperformanceisgivenbyEquation3
p kt
kt a k
= + minus( ) minus( )1 1 (3)
Whenweagainsolvefork
ka pt a pt apt t
=+ minus minus minus minus( ) minus minus( )1 1 4
2
2
(4)
Equations2and4convertperformanceinbothconditionstoacom-monmetrick
Therawdataunderlyingthecapacitycomputationsforallex-perimentsarereportedinAppendixBThemajorityofouranalysesweresingle-degree-of-freedomplannedcomparisonscarriedoutthroughpairedttests
exPeriMeNT 1
InthisexperimentweintroducetheparadigmandthebasicresultthecontentdeficitInthisversionoftheex-perimentthesetof8uniqueanimalstobepresentedoneachtrialwasselectedatrandomfromthe23availableanimalsAsubsetof4targetanimalswasthenselectedatrandomfromamongthese8NoattemptwasmadetocontroloverlapinsetsfromtrialtotrialThestandardandspecificconditionswereruninseparateblocks
Estimatedcapacitywas16items(SEM503items)inthespecificconditionand29items(SEM503)inthestandardconditionThereweretwonotableresultsinthisexperimentFirstobserverswerereliablyabletoreportontheidentityofmorethanoneiteminthespecificcondition[t(7)550p001one-tailed]indicatingacontent-addressablerepresentationSecondestimatedcapacitywaslowerinthespecificconditionthaninthestandardcondition[t(7)580p0001two-tailed]ThatiswhenaskedtoreportthelocationofaspecifictargetobserversappearedtohavetrackedfeweritemsthanwhentheywereaskedtosimplyclickonallofthetargetlocationsWewillcallthisresultthecontent deficitSubsequentexperimentswilltestpossibleexplanationsforthisdeficit
OnepuzzlingaspectofthesedatawasthatperformanceinthestandardconditionseemedlowinthisexperimentincomparisonwithtypicalMOTexperimentswithidenticalitemsObserversaretypicallyabletotrackmorethanthreeitems(seeOksamaampHyoumlnauml2004Scholletal2001)Our taskmayhavebeenmoredifficultsimplybecausetheanimalsoftenoccludedoneanotherwhichcanreduceperformance(KliegerHorowitzampWolfe2004)ThereisalsoamoreseriouspossibilitySincethestimulussetchangedfromtrialtotrialobserversmayhaveexperiencedinterferenceinworkingmemoryInparticularitemsthatweretargetsononetrialcouldhaveservedasnontargetsonsubsequenttrialsandviceversasuchconditionshaveledtointerferenceinboththevisualsearch(egldquovariablemappingrdquoSchneiderampShiffrin1977)andnegativeprim-ing(MillikenTipperampWeaver1994Tipper1985)para-digmsSuchinterferencemightbeparticularlyproblematicinthespecificconditionThusthechangingstimulussetsmightberesponsibleforboththecontentaddressabilityandthecontentdeficitfromthedataExperiment2wasdesignedtoremedythisproblem
exPeriMeNT 2
Inthisexperimentwesimplyfixedthetargetandnon-targetsetssothatthesamestimuliwereusedthroughout
Tracking UniqUe ObjecTs 177
ablockoftrialsThatisanobservermighttracktheliontherabbittheturtleandthegoatonallthetrialsinablockThismadeiteasierforobserverstorememberwhattheyweresupposedtotrackInadditioniteliminatedthepossibilityof interferenceornegativeprimingarisingfromtargetsandnontargetsswitchingrolesWeexpectedthatperformanceinthestandardconditionwouldimprovetoabove30itemsthequestionwaswhetherthecontentdeficitwouldstillmanifestitselfundertheseconditions
TheprocedurewasidenticaltothatofExperiment1exceptintheselectionofanimalsForeachobserveronesetofeightanimalswasselectedrandomlyatthebegin-ningofeachblockFouroftheseanimalsweredesignatedastargetsandtherestasnontargetsThesetofanimalsandthetargetnontargetassignmentswereheldconstantforthedurationofablock
Asexpectedperformanceinthestandardconditionimprovedtoacreditable34(SEM502)itemsHow-ever the content deficit was still present Observerstrackedonly21(SEM501)itemswhenaskedwhichtargetwaswhereThisdeficitwasreliable[t(7)583p00001]andsimilarinmagnitudetothatobservedinExperiment1ofaround13itemsAmixedANOVAcomparingthetwoexperiments2revealedalargeeffectof condition [F(18)5 1152p 000001]butnoneofexperiment[F(18)510p 10]Theinteractionterminthebetween-experimentsanalysiswasnearzero[F(18)1]
Theroughlyhalf-itemincreaseincapacitywhenwechangedfromavariabletoafixedtargetsetsuggeststhatswitchingtargetsetsfromtrialtotrialcreatedsomesortofinterferenceThiswastruenotonlyinthespecificcon-ditionwhenweprobedtargetidentitiesbutalsointhestandardconditionThisresultisimportantbecauseitsuggestsaroleforobjectidentity(eitherattheperceptualorsemanticlevel)irrespectiveoftaskdemandsAccord-ingtoLeonardandPylyshyn(2003)blinkingthetargetsonandoffshouldautomaticallyassignvisualindexestothemandtheseindexesshouldstickequallywellwhetherornottheseparticularstimuliweretargetsontheprevioustrialSincetheincreaseincapacitywasnotstatisticallyreliablethisconclusionshouldnotbetakentooseriouslyHoweverinsubsequentexperimentsweusedafixedtar-getset
JustasinExperiment1theobserversherecouldreli-ablylocatemultipletargetsaccordingtotheiridentitieswhichwetakeasevidenceforcontentaddressabilityEx-periment2alsoreplicatedthecontentdeficitwithreli-ablybettercapacityinthestandardreportconditionthaninthespecificreportcondition
TheseresultsraiseanumberofmethodologicalissuesFirstourexperimentsdifferfrommostMOTexperimentsintheirtemporalstructureourtrialswerebothlongerthanusualandincludedtemporaluncertaintyHowdidthesefactorsaffectperformanceExperiment3wasdesignedtoaddressthisissue
Asecondimportantissueiswhetherblockingresponsemodesinducedobserverstousedifferentencodingstrate-giesinthetwoconditionsleadingtodifferentialperfor-manceThisissuewillbeaddressedinExperiment4
AthirdissueiswhethertheuseofuniqueidentifiableobjectshelpsorhindersMOTperformanceincomparisonwiththeuseofidenticalobjectsinmostpreviousstudiesofMOTWewilladdressthisissueinExperiments5and6
FinallyExperiment7willaddresstheroleofnontar-getsinhinderingtrackingperformance
exPeriMeNT 3
Aswenotedintheintroductionitisimportanttoen-surethatobserversareattendingtoataskthroughoutatrialOneofourstrategiesweusedtoensurethiswastorandomlyvary trialdurationObserversdidnotknowwheneachtrialwouldendbetweenthe10thand15thcyclesWasthisenoughtemporaluncertaintyhoweverInExperiment3weincreasedtheuncertaintysothatatrialcouldendanywherebetweenthe5thand15thcycles(adurationfrom67to200sec)OtherwisethemethodwasidenticaltothatofExperiment2IfobserverswereactivelytrackingthetargetsonlyafterCycle9inthepre-viousexperimentsweshouldseeadecreaseinperfor-manceinthisexperimentwheretheypotentiallyhavetosustainattentionforalongerperiodoftime
Awiderarrayoftrialendpointsalsoallowedustoex-amineapossibleexplanationforthecontentdeficitWeusedtrialdurationsthatwerelongrelativetopreviousMOTstudiesPerhapslocationndashidentitybindingsdecayovertimeandwewouldobserveasmallerdifferencebe-tweenconditionsifweprobedearlierinthetrial
Figure3showskinExperiment3asafunctionofcyclewithdatafromExperiment2shownforcomparisonMeancapacityforthestandardconditionwas34items(SEM501)versus26items(SEM502)forthespecificcondi-tionrepresentingasignificantadvantageinthestandardcondition[t(7)539p 01]Wealsoconductedananalysisofcovarianceontheresultswithconditionasacategoricalvariableandcycleasacontinuousvariable(covariate)Neitherthemaineffectofcycle[F(17)1]northecycle3conditioninteraction[F(17)5249p516]wassignificant
5 7 9 11 13 150
1
2
3
4
Cycles
k (it
ems)
Figure 3 Capacity as a function of cycle Squares denote the standard and triangles the specific target condition data from experiment 3 are plotted in black and those from experiment 2 in gray error bars indicate standard errors of the means
178 HOrOwiTz eT al
WhencomparingthetwoexperimentsitseemsclearthatincreasingthetemporalvariabilityinExperiment3didnotchangetheresultsnotablyWeconductedacross-experimentANOVAusingdataonlyfromCycles10ndash15inExperiment3Standardversusspecifictargetconditionandcyclewerewithin-subjectsfactorsandexperimentwasabetween-subjectsfactorWefoundnomaineffectofexperiment[F(114)511p 5 31]Thecondition3ex-perimentinteractionapproachedsignificance[F(114)541p506]probablybecausethespecifictargetcapac-itywassomewhathigherinExperiment3Addingtempo-raluncertaintyclearlydidnotdisruptperformanceinthisexperimentinrelationtoExperiment2
ThereisahintinFigure3thatperformanceespeciallyin thespecific targetconditiondroppedas trialdura-tionincreasedThiswouldbecompatiblewithdatafromOksamaandHyoumlnauml(2004)possiblyreflectingavigilancedecrement(Parasuraman1986)IfthedropovertimeweregreaterinthespecificconditionthiswouldsupporttheideathatlocationndashidentitybindingsweredecayingHowevertherewasnoevidenceforthisinanANOVAconductedondatafromalloftheconditionsofExperiment3inwhichneitherthemaineffectofcycle[F(1070)511p538]northecycle3conditioninteraction[F(1070)1]wassignificantIftherewasdecayithadapproachedasymp-totewithinthefirst6ndash7secoftracking
exPeriMeNT 4
OneobvioushypothesistoexplainthecontentdeficitisthatobserversencodeinformationdifferentlyinthetwoconditionsInthestandardconditiononlyspatiotemporalpropertiesareencodedInthespecifictargetconditionobserversknowthattheywillbeaskedaboutfeaturaloridentityinformationsotheyencodethataswellFeaturalinformation(orperhapsthebindingofidentityandlo-cation)takesupmoreresources(Bahrami2003Saiki2002) reducing tracking capacityThus informationaboutfeaturesoridentitiesrequiresmorecapacity
InExperiment4wetestedthisstrategicencodinghy-pothesisbymixingthequestionswithinablockoftri-alsObserversdidnotknowonagiventrialwhethertheywouldbeaskedtolocateallofthetargetsorjustaspecifictargetThereforetheyhadtoadoptthesamestrategiesonbothtypesoftrialsIfthestrategicencodinghypothesisiscorrectthenthecontentdeficitshouldbereducedinthisexperimentPerformanceshouldalsobereducedoverallsinceobserverswouldnotbeoptimallyencodinginfor-mationoneachtrialInotherrespectsthemethodwasidenticaltothatofExperiment2
Pilotworkindicatedthatwhenquestionsweremixedwithinablockof trialshavingto lookawayfromthecactusarraytoreadthequestionatthetopofthescreenimpairedperformanceThereforeinthisexperimentweusedauditoryratherthanvisualprobesInsteadofprintingthequestionatthetopofthescreenobserversheardtheprobequestionsoverheadphones
Meancapacityforthestandardconditionwas34items(SEM501)asopposedto22items(SEM502)forthe
specificconditionrepresentingasignificantadvantageforthestandardcondition[t(7)599p00005]Per-formanceinthisexperimentwithquestionsmixedwithinblockswasnearlyidenticaltothatinExperiment2withblockedquestionsEvenwhenobserverscouldnothaveadopteddifferentencodingstrategies for thedifferentconditionswestillobservedasizable(12-item)contentdeficitThissuggeststhatthecontentdeficitisnotaresultofthestrategicdemandsoftheconditionswedevised
exPeriMeNT 5
ThepreviousexperimentsestablishedaparadigmforstudyingMOTwithunique stimuliOnequestion thatwehadnotyetaddressediswhetheruniqueobjectshelporhinderMOTperformanceincomparisonwithidenticalobjectsForexampleiftheobjectsareidenticalanob-servermightaccidentallyswapatargetandadistractoriftheyapproachtoocloselyWithuniqueobjectsthisislesslikelytohappenSimilarlyifatsomepointtheobserverrealizedthathewasonlytrackingthreetargetsandhadlostonehecouldtheoreticallyrecoverthelosttargetifitwereuniqueOfcoursesuchadvantagesassumeasortofldquoidealobserverrdquowhoalwaysknowsexactlywhatheistrackingIfthetrackingsystemwascompletelyinsensitivetotargetfeaturesoridentityuniqueobjectswouldprovidenoadvan-tageSimilarlytheobservercouldonlyrecoveralosttargetifinformationwereavailableabouthowmanytargetswerecurrentlybeingtrackedandwhichonewasmissing
Inthisexperimentwecomparedtrackinguniqueob-jectswithtrackingidenticalobjectsSinceldquoWhereisthezebrardquoisnotadiagnosticquestionwhenallobjectsarezebrasweonlyusedthestandardresponsemodeTherewerethreeconditionsTheuniqueconditionwasidenti-caltothestandardresponseconditioninExperiment1Intheidenticalconditioneightidenticalanimalswerepresented animal identitywas selectedat randomoneachtrialFinallyinthepairedconditiontherewerefouruniquetargetsagainselectedatrandomForeachtargettherewasanidenticaldistractorProbeswerepresentedauditorilyasinExperiment4Atotalof10observerspar-ticipatedinthisexperiment
Wecancompareperformanceintheuniqueandidenti-calobjectconditionstodeterminewhetherthereisanyadvantageforuniqueobjectsIftheadvantageforuniqueobjectsderivesfromtheabilitytorecoverlosttargetsthisadvantageshouldbeabolishedinthepairedcondition
CapacityvaluesforthethreeconditionsareplottedinFigure4Observerstracked31items(SEM502)intheuniquecondition23(SEM502)inthepairedconditionand25(SEM501)intheidenticalconditionPlannedttestsshowedthatthepairedandidenticalconditionsdidnotdiffer[t(9)513p 10]butthatperformancewassignificantlyworseinboththanintheuniquecondition[forpairedt(9)527p05foridenticalt(9)527p01]
TheseresultsindicatethatobserverscantakeadvantageoftheadditionalinformationprovidedbyuniqueobjectsTheuniqueobjectadvantage(approximately06items
Tracking UniqUe ObjecTs 179
undertheseconditions)isentirelyeliminatediftargetsareidenticaltodistractorssuggestingthatsomesortoferrorrecoveryprocessisresponsible
exPeriMeNT 6
In the previous experiment pairing targets anddistractorsreducedperformanceinthestandardcondi-tionrelativetothatinthefullyuniqueconditionWouldthismanipulationaffectthespecificconditionaswellInExperiment6wemeasuredthecontentdeficitwithbothpairedanduniquestimuliBycomparingthesetwostimulusconditionswecandeterminewhethereliminat-ing featuraldistinctionsbetween target andnontargetsetsimpairstheabilitytotrackspecificobjectsaswellastheabilitytoholdontotargetlocationsProbeswerepresentedauditorilyasinExperiment5Therewere6ob-serversinthisexperiment
AsshowninFigure5weonceagainreplicated thecontentdeficitinthisexperimentsinceobserverscouldtrack11moreitemsinthestandardconditionthaninthe
specificcondition[F(19)51190p001]Pairingtar-getsandnontargetsreducedcapacitybyroughlythesameamountas inExperiment5 [09 itemsF(19)5373p001]Criticallythepairingmanipulationappearedtohavelessofadeleteriouseffectinthespecificconditionthaninthestandardcondition[interactionF(19)5185p005]Inthestandardconditionthepairingmanipu-lationcostobservers12itemsversusalossofonly07itemsinthespecificconditionThisinteractionsuggeststhatdifferentrepresentationsmayunderlieperformanceinthetwoconditions
Ontheotherhandcapacitywasalreadylowerinthespe-cificconditionIftheeffectofpairingstimuliismultipli-cativeratherthanadditivewewouldthenexpectthattheimpactofpairedstimuliwouldbeproportionallysmallerinthespecificconditionWhenwelog-transformedthecapacityvaluesconvertingproportionaldifferencesintoadditive ones the question3 pairing interaction waseliminated[F(19)1]
Sincewecannotrejectthehypothesisthattheeffectofpairingtargetswithnontargetsisproportionaltocapacitythisexperimentprovidesatbestweakevidencefordiffer-entrepresentations
exPeriMeNT 7
OneadvantageofourmethodisthatwecantestMOTperformanceinacasewithonlytargetsWhywouldwewanttodothisTherearereasonstobelievethatsuppress-ingnontargetsconsumessomeresourcesthatwouldother-wisebeavailablefortrackingtargetsForexamplethereisevidencethatobserverstrackbetterontrialsinwhichnontarget trajectories are repeated from earlier trials(OgawaampYagi2003)IfnontargetsareprocessedtheymayhavetobesuppressedinordertoreduceconfusionwithtargetsandPylyshynandLeonard(2003)showedevidenceforsuchinhibitionMoreoverunpublisheddatafromourlaboratoryindicatethatMOTperformancede-creasesasnontargetsareaddedtothedisplayevenwhenthenumberoftargetsremainsconstant
Accordingly we designed Experiment7 to test thehypothesisthattrackingcapacitywouldincreaseiftherewerenonontargetsThisexperimentwasidenticaltoEx-periment2exceptthatallitemsweretargetsCapacityinthestandardconditionwas58items(SEM502)sig-nificantlygreaterthanthe26items(SEM501)obtainedinthespecificcondition[t(7)5143p000005]
InagreementwithourhypothesiscapacityimprovedmarkedlyinthestandardconditionwhentherewerenodistractorsStandardcapacityinExperiment7wassig-nificantlygreaterthaninExperiments23and4whichweremethodologicallycomparableexceptforthenumberoftargetsandnontargets(allps00005)
Oneexplanationisthatpreviousexperimentsunderes-timatedcapacitybecauseofaceilingeffectTheseexperi-mentswerebasedonthehypothesisthatcapacityvariesacrosstrialswithnounderlyingdifferencesacrossexperi-mentsHoweverthemeasuredcapacityhasbeenlimitedbythenumberoftargetsForexampleifthereareonlyfour
Unique Paired Identical0
1
2
3
4
Condition
k (it
ems)
Figure 4 Capacity as a function of condition in experiment 5
Specific Standard0
1
2
3
4Unique
Paired
Condition
k (it
ems)
Figure 5 effect of pairing target and nontarget objects on ca-pacity in experiment 6 data from the paired conditions are plot-ted as gray bars and those from the unique conditions as open bars
180 HOrOwiTz eT al
targets(egExperiments2ndash4)observerscouldtrackallofthemontrialsinwhichtheircapacitywas4orgreateranytrialswithanactualcapacitygreaterthan4wouldthusappeartohaveacapacityof4InthiscasemeasuredmeancapacitywouldunderestimatethetruemeanInExperi-ment7whichincludedeighttargetsmeasuredcapacitycouldreflectactualcapacityuptoeightitemsleadingtoagreater(andmoreaccurate)estimateofmeancapacity
Wecantestthisceilinghypothesisagainstourhypoth-esisofincreasedmeancapacitybycomparingthecapacitydistributionsinExperiment7withthoseinExperiments2ndash4Accordingtotheceilinghypothesistheproportionoftrialswithacapacitylessthan4shouldbeidenticalacrossexperimentsFurthermoretheproportionoftrialswithacapacityof4orgreaterinExperiment7shouldbethesameastheproportionoftrialswithacapacityequalto4inExperiments2ndash4Incontrastthehypothesisthattrackingcapacityincreaseswhennonontargetsarepres-entpredictsthattheentiredistributionshouldshifttotherightleadingtofewertrialswithcapacitylessthan4inExperiment7thanintheearlierexperiments
We tested these hypotheses by computing capacityforeachtrialusingEquation2andgeneratingadistri-butionacrosstrialsforeachobserverFigure6plotsthedistributionofkforExperiment7andthecorrespondingdistributionforExperiments23and4combined(dis-tributionsfromeachoftheseexperimentswereaveragedacrossobservers)Clearlytheproportionoftrialswithanobservedcapacitylessthan4waslowerwitheighttargetsthanwithfourtargetsThisisconsistentwiththecapacitydistributionshiftingtotherightinExperiment7ratherthanmerespreadingofthek54trialsoverawiderrangeWesuggestthatthesuperiorperformanceinthisexperi-mentmayreflectadditionalcapacityfreedupwhenthereisnoneedtosuppressnontargetsHowevertheshapeofthedistributionforfourtargetsdoessuggestthatcapacitywasunderestimatedinthepreviousexperiments
TheconclusionsarequitedifferentifwelookatthespecifictargetconditionSpecifictargetcapacityinthis
experimentwasbetterthaninExperiment2(uncorrectedttestp05)butnotsignificantlydifferentfromtheca-pacitiesobservedinExperiments3and4( p05)Thusthetrackingoftargetidentitiesinthisconditiondoesnotseem tobe limitedby theneed to suppressnontargetidentities
diSCuSSioN
WehavedevelopedanewMOTparadigmthatallowsustotestthetrackingofuniqueobjectsFourmainfind-ingshaveemergedfromtheseexperimentsFirstthereisacontent-addressablerepresentationoftargetsduringMOT(specificcapacitywas1inallcases)SecondweobservedacontentdeficitThecontent-addressablerep-resentationhasalowercapacity(measuredinitems)thanthelocation-addressablerepresentation(iespecificca-pacitywaslowerthanstandardinallcases)Thirdthevi-sualsystemiscapableoftakingadvantageofdifferencesbetweenuniqueobjectstoimprovetrackingperformanceoverthelevelseenwithidenticalobjects(Experiment5)Thisadvantageappearstobedueprimarilytotheabilitytorecoverlosttargets(Experiment5)Fourthtrackingcapacityappearstobeconstrainedbytheneedtosup-pressnontargetswhentherearenonontargetscapacityincreasesmarkedly(Experiment7)
A Content-Addressable representation in MoTOurdata represent anexistenceproofof a content-
addressable representation inMOTSimilar data havebeenreportedbyOksamaandHyoumlnauml(2004)Intheirmul-tiple-identity-tracking(MIT)taskobserverstrackedmov-inglinedrawings(eitherobjectsorldquopseudo-objectsrdquo)forseveralsecondsAttheendofthetrialthestimuliweremaskedandasingleitemwasmarkedwithablackframeInasubsequentresponsescreenobservershadtoselectthepicturethatcorrespondedtothemarkedobjectOksamaandHyoumlnaumlobtainedamediancapacityoffouritems
Althoughitisdifficulttocomparecapacitymeasuresdirectlyacrossexperimentaltasks3itisinterestingthatbothweandOksamaandHyoumlnauml(2004)demonstratedthatobserversdoknowtosomeextentwhichtargetiswhichThisfindingisincontrasttothoseofPylyshyn(2004)whoseobservershaddifficultyidentifyingtargetsTwokeydifferencesbetweenourparadigmandMITontheonehandandPylyshynrsquos(2004)paradigmontheotherseemrelevanttothisdiscrepancyFirstinourstudyandthatofOksamaandHyoumlnaumlobjectidentitiesweredefinedbyintrinsicfeaturessuchasshapeand(inourexperi-ment)colorInPylyshynrsquos(2004)experimentldquoidentityrdquowasbasedonatangentialfeatureofanobjectsuchasinwhichcornerofthedisplayitbeganthetrialItseemslikelythattheshapeandcolorofanobjectmaybemoretightlyboundmarkersofldquoidentityrdquothaninitialpositionorabrieflypresentedletter
SecondinourexperimentsandthoseofOksamaandHyoumlnauml (2004) thevisualdifferencesbetweenobjectswerecontinuallyavailablethroughoutthetrackingphaseofthetrialonlymaskedduringtheresponsephaseIn
4 targets8 targets
0 2 4 6 80
1
2
3
4
5
6
Capacity
Pro
po
rtio
n o
f Tri
als
Figure 6 Capacity distributions for the standard condition in experiment 7 (gray lines striped area) and in experiments 2 3 and 4 (black lines stippled area) dashed lines indicate standard errors of the means
Tracking UniqUe ObjecTs 181
Pylyshynrsquos(2004)experimentstheobjectidentitieswerepresentedonlybrieflyatthestartofthetrialduringthemajorityofthetrackingphasetheobjectswerephysicallyidenticalItmaybethatthelinkbetweenobjectidentityandspatiotemporalpositiondecayedduringPylyshynrsquos(2004)taskbutwasconstantlyrefreshedinbothoursandOksamaandHyoumlnaumlrsquos
Ourexperimentsalsoprovidedsomeindirectevidenceforacontent-addressablerepresentationForexampleperformancewasconsistentlysuperior inexperimentsinwhichthesameobjectswereusedinthesamerolesfromtrialtotrial(Experiments23and4)thaninex-perimentsinwhichthestimulussetchangedfromtrialtotrial(Experiments15and6)Thisoccurrednotonlyinthespecifictargetconditionsinwhichobjectidentitywasrelevantbutalsointhestandardconditionsinwhichobjectidentitywasnot
one System or TwoIstrackingservedbyasinglerepresentationthatencodes
bothpositionandidentityoraretheretworepresentationsoneforpositionandoneforidentityThisquestionarisesbecauseofthestrikingcapacitydifferencesbetweenourresponseconditionsInallofourexperimentscapacitywassignificantlylowerforspecifictargetsthanfortargetsinthestandardMOTresponsemodeInExperiment4wedisconfirmedthehypothesisthatcapacitydifferencesresultedfromvariedencodingstrategies(seeegDavisWelchHolmesampShepherd2001)sinceweobtainedthesameresultwhenresponsemodesweremixedwithinablockaswedidwhentheywereruninseparateblocks
ThiscapacitydifferenceisnotconclusiveevidencethatseparaterepresentationsorsystemsareinvolvedThereareseveralwaysinwhichaunifiedaccountcouldex-plainthelowercapacityinthespecifictargetconditionForexampleonemightproposeasystemwithaslot-likestructure(VogelWoodmanampLuck2001)inwhichallslotsarenotcreatedequalIfthereweretwoslotswithsuf-ficientresolutiontoencodebothpositionandidentityandtwoslotsthatcouldonlyencodepositionwewouldalsoobtaindifferentcapacityestimatesinourtworesponseconditionsThisrepresentationcouldalsobedescribedintermsofKahnemanandTreismanrsquos(1984)objectfilesifweassumethatsomeoftheobjectfilesareemptyTheimpletionprocessmightinthatcasereturntheinforma-tionthatagivenobjectwasspatiotemporallycontinuouswithatargetobjectbutfailtoretrievethesemanticorperceptualcontentassociatedwiththatobjectfile
IntheMOTliteraturecapacityisgenerallyassumedtorefertoafixednumberofavailablerepresentationsinthevisualsystemeithervisualindexesorobjectfilesHow-everthereissomeevidencethatcapacityinMOTmightbebetterthoughtofasanundifferentiatedresource(AlvarezampCavanagh2005)InsteadofasetofslotsorpointerswemightassumethatthereisafixedamountofinformationthatcanbeencodedObserversmightencodepositionandidentityforalltargetsbutthebindingbetweenpositionsandidentitymaybenoisierthanthepositioninformationitself(ormaydecaymorequicklyExperiment3indicatedthatanysuchdecaywouldhavetoapproachitsasymptote
valuewithin6ndash7sec)leadingtoapparentlylowercapacitywhenresponsesdependonbinding
InsteadofonesystemtherealsomightbetwoAlthoughwetendtoassumethatasingleexperimentalparadigmprobestheperformanceofasinglecorrespondingfunc-tionalsystemitmaybethatMOTnaturallyrecruitsmulti-pleoverlappingsystemsThesimplesttwo-systemaccountmightbeaproposalthattrackingisservedbyaposition-trackingsystemsuchasvisualindexes(FINSTs)butthataccessingidentityndashpositionbindingsrequiresfocalatten-tion(WheelerampTreisman2002)Howeversincespecificcapacityisreliablygreaterthanoneitemwewouldhavetoassumethatmultipleitemscansimultaneouslybethefocusofattention(AwhampPashler2000Davisetal2001)
IfwewishtopreserveasinglefocusofattentionwemightinsteadproposethatbothvisualindexesandobjectfilescanbeusedVisualindexeswouldprovideonlypo-sitioninformationandobjectfilescouldsupportmorecomplexqueries
Itwillobviouslybedifficulttodefinitivelydecidebe-tweenoneandtwosystemsHowevertheevidenceofourfinalthreeexperimentspointstotwoindependentsystemsInExperiments5and6pairingtargetsandnontargetsre-ducedcapacityinthestandardbutnotinthespecifictar-getcondition(althoughwecouldnotrejectthehypothesisofaproportionaldecrease)InExperiment7standardca-pacityincreasedwhennonontargetswerepresentedbutspecificcapacitywasunaffected
MOThasbeenlikenedtoavisualworkingmemorytaskattimezero(CavanaghampAlvarez2005)soitisnaturaltolooktostudiesofworkingmemoryforanalogsofthecontent-addressableandlocation-onlysystemsTheclassictwo-pathwayschemeinvisionisthedistinctionbetweenventralanddorsalstreamsassociatedwithobjectrecogni-tionandspatialprocessingrespectively(UngerleiderampMishkin1982)ThisdistinctionhasbeenextendedintothedomainofworkingmemorybyWilsonOacuteScalaidheandGoldman-Rakic(1993)whoprovidedevidencethatobjectinformationandspatialinformationareneurallysegregatedinmonkeyprefrontalcortexwidelybelievedtobetheneuralsubstrateofworkingmemoryAsimilardissociationinhumanswasreportedusingPETimagingbySmithetal(1995)
Thedivisionofworkingmemoryintomultiplesubsys-temsgoesbackatleasttothemultiple-componentmodelofBaddeleyandHitch(1974)whichproposedseparatecomponents for different modalities a ldquophonologicallooprdquoforauditoryinformationandaldquovisuospatialsketchpadrdquoforvisualinformationInmorerecentversionsofthemodel(BaddeleyampLogie1999Logie1995)thesketchpadhasbeendividedintovisualandspatialsubcompo-nents roughlycorrespondingto theobjectandspatialsystemsdescribedbyWilsonetal(1993)Interestinglythespatialsubcomponent(theldquoinnerscriberdquo)isspecifi-callytaskedwithholdingsequenceorpathinformation(seeParmentierElfordampMayberry2005)makingitalikelysubstrateforMOTperformance
Thewidespreadinsistenceonaseparationbetweenvi-sualorobjectinformationandspatialinformationwouldseem problematic for our findings Standard MOT is
182 HOrOwiTz eT al
clearlyajobforthespatialsubsystem(PostleDrsquoEspositoampCorkin2005forexampleusedaversionofstandardMOTtotieupdorsalprocessinginamemoryinterferenceparadigm)Howevertheredoesnotseemtoberoominthataccountforacontent-addressablerepresentationOurspecifictargettask(aswellastheMITtaskofOksamaampHyoumlnauml2004)isnotapureobjectrecognitiontaskitrequireslocationndashidentitybindingswhichwouldseemtoinvolveintegratinginformationacrossstreamsTheproblemdisappears ifwe take intoaccount recentbe-havioral(OlsonampMarshuetz2005)andfMRI(PostleampDrsquoEsposito1999)investigationsdemonstratingthatobjectworkingmemoryrepresentationsalsocarrysomelocationinformation
NeuroimagingstudiesmightproveusefulhereStan-dardMOThasbeenshowntoactivateparietallobeareasinthedorsalstream(Culhametal1998Jovicichetal2001Seiffert2003)Itwouldbeinformativetoknowwhetherthepatternofactivationchangeswhenuniqueob-jectsareintroducedorwhenthetaskpotentiallyinvolvesobjectidentities
unique object Features in MoTInadditiontoourfindingsaboutcontentaddressability
intrackingwewereabletodemonstratethathavingvisu-allyuniqueobjectsmadetrackingeasierobserverswereabletotrackroughly06moreitemswhenobjectswereuniquethanwhenobjectswereidenticalInExperiments5and6weattributedthisresulttoanerrorrecoverystrat-egyIfIlosetrackofthezebraIcansearchforthezebraonthebasisofitsphysicalfeaturesandcontinuetotrackitThisstrategywouldobviouslybeuselessifallitemsarezebrasThisstrategyisoflimitedutilityifthereareazebratargetandazebradistractoronceIlosethezebraIthenhaveonlya50chanceofpickingupthetargetzebraasopposedtothedistractor
TheerrorrecoveryhypothesisisnottrivialIthassev-eralimportantimplicationsFirstitimpliesthatobserversknowwhattheyarenottrackingThephenomenologyofMOTissuchthatanobserverhasthefeelingofknow-inghowmanytargetsheorsheissuccessfullytrackingTheerrorrecoveryhypothesisimpliesthatwhenatargetislosttheobserverknowswhichone(givenuniquetargets)SecondtheerrorrecoveryhypothesisalsoassumesthatanobservercansearchforamissingtargetwhilecontinuingtotracktheremainingtargetsThisisinagreementwithrecentevidencethatobserverscansuccessfullyperformanMOTtaskandavisualsearchtaskonthesametrial(AlvarezHorowitzArsenioDiMaseampWolfe2005)Finallythehypothesisimpliestheabilitytoaddtothetrackingsetontheflywhiletrackingwithoutanyphysi-calcueTheseimplicationsneedtobeconfirmedthroughfurtherresearch
Howeverifobserversuseuniquephysicalfeaturesofobjectstorecoverlosttargetsinthisfashionitiscuriousthatthereisnosuchbenefitinthespecifictargetcondi-tionIfobserversdorecoverlosttargetsduringthecourse
ofatrialtheyappeartorecoveronlytheirlocationsnottheiridentities
AnalternativetotheerrorrecoveryhypothesisthoughnotamutuallyexclusiveoneisthattheresultsofEx-periment5canbeexplainedbytheperceptualsimilarityoftargetandnontargetitemsYantisrsquos(1992)approachtoMOThighlightstheimportanceofperceptualgroup-ingfactorsthatallowthetargetstobetreatedasaunitbythevisualsystemMostoftheresearchinthislinehasfocusedonspatiotemporalgroupingfactorssuchasthespatialrelationshipsbetweentheobjects(SearsampPylyshyn2000ViswanathanampMingolla2002Yantis1992)
Cangroupingby(nonspatial)featuralpropertiesim-provetrackingAnumberofstudieshavedemonstratedthatobservershave limitedexplicit access to featuralproperties (Bahrami 2003Saiki 2002Scholl etal1999)butthisfindingdoesnotaddressthedirectrolethatfeaturesmayplayWecouldexplaintheresultsofEx-periment5byproposingthatobserverswereabletouseaccidentaldifferencesintheshapeandcolorstatisticsoftargetsandnontargetstohelpsegregatethemvisuallyInthepairedconditiontargetsandnontargetshadidenticalfeaturestatisticssothatadvantagewaslostThishypoth-esiscouldalsoexplainwhyperformancewasbetterwithafixedstimulussetthanwithavariablestimulussetoveranumberoftrialswiththesamestimuliobserverslearnsubtledifferencesbetweenthetwosets
ConclusionsIntheintroductionwenotedthatreal-worlddynamic
attentiontasksdifferfromtraditionalMOTintwowaysFirstobserversintheeverydayworldusuallydealwithmultipleuniqueobjectsSecondtheyoftenwishtoknowwhereagiventargetmightberatherthanwhichobjectsarethetargetsTheexperimentspresentedhererepresentanattempttobridgelaboratoryMOTresearchandtheseeverydaycognitivetasksWehaveshownthatitiseasiertotrackuniqueobjectsthantotrackidenticalobjectsassum-ingthattheobjectsonewishestotrackarealsouniquelydifferentfromnontargetobjectsFurthermoreobserversdoknowwhattheyaretrackingTotakeoneofourex-amplesofecologicallyvalidtrackingtasksifyouwatchyourchildrenandtheirfriendsontheplaygroundyoucanbeawareofwhereyourkidsareatanygivenmomentyoumayalsobeawareofthelocationsoftheirfriendswithoutbeingabletotellonefriendfromtheotherFurthermoreifyoulosetrackofachildyouwillprobablybeawareofwhichoneyouhavelost
AuTHor NoTe
ThisresearchwassupportedbyNIMHGrantR0165576toTSHWethankKristinMichodandSkylerPlaceforassistancewithdatacollec-tionaswellasSteveLuckandananonymousreviewerforconstructivecriticismCorrespondencerelatingtothisarticlemaybesenttoTSHorowitzBrighamandWomenrsquosHospitalVisualAttentionLaboratory64SidneyStreetSuite170BostonMA02139(e-mailtoddhsearchbwhharvardedu)
Tracking UniqUe ObjecTs 183
reFereNCeS
Allen R McGeorge P Pearson D amp Milne A B (2004)AttentionandexpertiseinmultipletargettrackingApplied Cognitive Psychology18337-347
Alvarez G A amp Cavanagh P (2005)IndependentresourcesforattentionaltrackingintheleftandrightvisualhemifieldsPsychologi-cal Science16637-643
Alvarez G A amp Franconeri S (2004November)How many ob-jects can you trackPaperpresentedatthe12thAnnualWorkshoponObjectPerceptionampMemoryMinneapolisMN
Alvarez G A Horowitz T S Arsenio H C DiMase J S amp Wolfe J M (2005)DomultielementvisualtrackingandvisualsearchdrawcontinuouslyonthesamevisualattentionresourcesJournal of Experimental Psychology Human Perception amp Perfor-mance31643-667
Awh E Jonides J amp Reuter-Lorenz P A (1998)RehearsalinspatialworkingmemoryJournal of Experimental Psychology Human Perception amp Performance24780-790
Awh E amp Pashler H (2000)EvidenceforsplitattentionalfociJournal of Experimental Psychology Human Perception amp Perfor-mance26834-846
Baddeley A D amp Hitch G J (1974)WorkingmemoryInGHBower(Ed)The psychology of learning and motivation Advances in research and theory(Vol8pp47-90)NewYorkAcademicPress
Baddeley A D amp Logie R H (1999) Working memoryThemultiple-componentmodelInAMiyakeampPShah(Eds)Models of working memory Mechanisms of active maintenance and executive control(pp28-61)CambridgeCambridgeUniversityPress
Bahrami B (2003)ObjectpropertyencodingandchangeblindnessinmultipleobjecttrackingVisual Cognition10949-963
Brainard D H (1997)ThePsychophysicsToolboxSpatial Vision10433-436
Cavanagh P amp Alvarez G A (2005)TrackingmultipletargetswithmultifocalattentionTrends in Cognitive Sciences9349-354
Culham J C Brandt S A Cavanagh P Kanwisher N G Dale A M amp Tootell R B H (1998)CorticalfMRIactiva-tionproducedbyattentive trackingofmoving targetsJournal of Neurophysiology802657-2670
Davis G Welch V L Holmes A amp Shepherd A (2001)CanattentionselectonlyafixednumberofobjectsatatimePerception301227-1248
Henderson J M amp Hollingworth A (2003)EyemovementsandvisualmemoryDetectingchangestosaccadetargetsinscenesPer-ception amp Psychophysics6558-71
Horowitz T S Birnkrant R S Fencsik D E Tran L amp Wolfe J M (2006)HowdowetrackinvisibleobjectsPsychonomic Bulletin amp Review13516-523
Jovicich J Peters R J Koch C Braun J Chang L amp Ernst T (2001)Brainareasspecificforattentionalloadinamotion-trackingtaskJournal of Cognitive Neuroscience131048-1058
Kahneman D amp Treisman A (1984)ChangingviewsofattentionandautomaticityInRParasuramanampDRDavies(Eds)Varieties of attention(pp29-61)OrlandoAcademicPress
Kahneman D Treisman A amp Gibbs B J (1992)ThereviewingofobjectfilesObject-specificintegrationofinformationCognitive Psychology24175-219
Klieger S B Horowitz T S amp Wolfe J M (2004)Ismultipleobjecttrackingcolorblind[Abstract]Journal of Vision4(8)363a
Leonard C amp Pylyshyn Z W (2003)Measuringtheattentionaldemandofmultipleobjecttracking(MOT)[Abstract]Journal of Vi-sion3(9)582a
Logie R H (1995) Visuo-spatial working memory Hove UKErlbaum
Milliken B Tipper S P amp Weaver B (1994)NegativepriminginaspatiallocalizationtaskFeaturemismatchinganddistractorin-hibitionJournal of Experimental Psychology Human Perception amp Performance20624-646
Mitroff S R Scholl B J amp Wynn K (2004)DivideandconquerHowobjectfilesadaptwhenapersistingobjectsplitsintotwoPsy-chological Science15420-425
Ogawa H amp Yagi A (2003)Primingeffectsinmultipleobjecttrack-ingAnimplicitencodingbasedonglobalspatiotemporalinformation[Abstract]Journal of Vision3(9)339a
Oksama L amp Hyoumlnauml J (2004)Ismultipleobjecttrackingcarriedoutautomaticallybyanearlyvisionmechanismindependentofhigher-ordercognitionAnindividualdifferenceapproachVisual Cognition11631-671
Oliva A Torralba A Castelhano M S amp Henderson J M (2003)Top-downcontrolofvisualattentioninrealworldscenes[Ab-stract]Journal of Vision3(9)3a
Olson I R amp Marshuetz C (2005)RememberingldquowhatrdquobringsalongldquowhererdquoinvisualworkingmemoryPerception amp Psychophysics67185-194
Parasuraman R (1986)VigilancemonitoringandsearchInKRBoffLKaufmanampJPThomas(Eds)Handbook of perception and human performance Vol 2 Cognitive processes and performance(pp1-39)NewYorkWiley
Parmentier F B R Elford G amp Mayberry M (2005)Transi-tionalinformationinspatialserialmemoryPathcharacteristicsaffectrecallperformanceJournal of Experimental Psychology Learning Memory amp Cognition31412-427
Pelli D G (1997)TheVideoToolboxsoftwareforvisualpsychophysicsTransformingnumbersintomoviesSpatial Vision10437-442
Postle B R amp DrsquoEsposito M (1999)ldquoWhatrdquondashthenndashldquowhererdquoinvi-sualworkingmemoryAnevent-relatedfMRIstudyJournal of Cog-nitive Neuroscience11585-597
Postle B R DrsquoEsposito M amp Corkin S (2005)Effectsofver-balandnonverbalinterferenceonspatialandobjectvisualworkingmemoryMemory amp Cognition33203-212
Pylyshyn Z [W] (1989)Theroleoflocationindexesinspatialper-ceptionAsketchoftheFINSTspatial-indexmodelCognition3265-97
Pylyshyn Z W (2004)SomepuzzlingfindingsinmultipleobjecttrackingITrackingwithoutkeepingtrackofobjectidentitiesVisual Cognition11801-822
Pylyshyn Z W amp Leonard C (2003)Inhibitionofnontargetsdur-ingmultipleobjecttracking(MOT)[Abstract]Journal of Vision3(9)585a
Pylyshyn Z W amp Storm R W (1988)Trackingmultipleindepen-denttargetsEvidenceforaparalleltrackingmechanismSpatial Vi-sion3179-197
Saiki J (2002)Multiple-objectpermanencetrackingLimitationinmaintenanceandtransformationofperceptualobjectsProgress in Brain Research140133-148
Schneider W amp Shiffrin R M (1977)ControlledandautomatichumaninformationprocessingIDetectionsearchandattentionPsychological Review841-66
Scholl B J amp Pylyshyn Z W (1999)Trackingmultiple itemsthroughocclusionCluestovisualobjecthoodCognitive Psychology38259-290
Scholl B J Pylyshyn Z W amp Feldman J (2001)Whatisavi-sualobjectEvidencefromtargetmerginginmultipleobjecttrackingCognition80159-177
Scholl B J Pylyshyn Z W amp Franconeri S L (1999May)When are spatiotemporal and featural properties encoded as a result of attentional allocationPaperpresentedattheconferenceoftheAssociationforResearchinVisionandOphthalmologyFortLau-derdaleFL
Sears C R amp Pylyshyn Z W (2000)MultipleobjecttrackingandattentionalprocessingCanadian Journal of Experimental Psychol-ogy541-14
Seiffert A E (2003)Dissociatingneuralcorrelatesofattentionaltrackingandattentiontovisualmotion[Abstract]Journal of Vision3(9)868a
Smith E E Jonides J Koeppe R A Awh E Schumacher E H amp Minoshima S (1995)SpatialversusobjectworkingmemoryPETinvestigationsJournal of Cognitive Neuroscience7337-356
Tipper S P (1985)ThenegativeprimingeffectInhibitoryprimingbyignoredobjectsQuarterly Journal of Experimental Psychology37A571-590
Ungerleider L G amp Mishkin M (1982)Twocorticalvisualsys-
184 HOrOwiTz eT al
temsInDJIngleMAGoodaleampRJWMansfield(Eds)Anal-ysis of visual behavior(pp549-586)CambridgeMAMITPress
Viswanathan L amp Mingolla E (2002)DynamicsofattentionindepthEvidence frommulti-element trackingPerception311415-1437
Vogel E K Woodman G F amp Luck S J (2001)Storageoffea-turesconjunctionsandobjectsinvisualworkingmemoryJournal of Experimental Psychology Human Perception amp Performance2792-114
Wheeler M E amp Treisman A M (2002)Bindinginshort-termvisualmemoryJournal of Experimental Psychology General13148-64
Wilson F A Oacute Scalaidhe S P amp Goldman-Rakic P S (1993)Dissociationofobjectandspatialprocessingdomains inprimateprefrontalcortexScience2601955-1958
Wolfe J M Oliva A Horowitz T S Butcher S J amp Bompas A (2002)SegmentationofobjectsfrombackgroundsinvisualsearchtasksVision Research422985-3004
Yantis S (1992)MultielementvisualtrackingAttentionandpercep-tualorganizationCognitive Psychology24295-340
NoTeS
1Wearemissingageinformationfor1observer2Threeobserverswhoparticipatedinbothexperimentswereelimi-
natedfromthisanalysis3Measuredcapacitycanvarybasedontrackingdurationtherange
ofmovementspeedsandwhetherornotobjectscanoccludeonean-other(asinourexperiment)orbounceoffoneanother(asinOksamaampHyoumlnauml2004)
APPeNdix A Stimulus List
beaver fox horse ostrich rhino turtlecamel giraffe kangaroo pelican rooster wolfcrocodile goat lion rabbit squirrel zebraelephant gorilla moose tiger
APPeNdix b raw data
Table b1 raw Numbers of items reported in the Standard Condition
Total Targets Nontargets Filled
Experiment1 308Experiment2 344 014 359Experiment3 347 015 362Experiment4 339 015 354Experiment5 Unique 312 037 349
Paired 249 080 329Identical 261 086 347
Experiment6 248 078 326Experiment7 599 599
NotemdashTargetsindicatestheaveragenumberofcorrectlyreportedtargetspertrialNontargetstheaveragenumberoftimesobserversclickedonanontargetlocationpertrialandTotal Filledthesumofthesetwocat-egoriesNontargetclickcountswerenotcollectedforExperiment1andtherewerenonontargetsinExperiment7
Table b2 raw Proportions for the Specific Condition by experiment
Errors
Correct Empty Trialsper Responses Transposition Nontarget Space N Observer
Experiment1 44 14 16 26 8 5000Experiment2 56 18 10 16 8 5000Experiment3 68 12 10 10 8 5000Experiment4 58 25 09 08 8 5063Experiment6 41 14 21 23 7 5000Experiment7 33 39 00 28 8 5000
NotemdashTranspositionerrorsrefertotrialsinwhichtheobserverclickedonthewrongtargetlocationTrialsperobservervariedinExperiment4becauseconditionwasselectedrandomlyfromtrialtotrial
(ManuscriptreceivedMarch312005revisionacceptedforpublicationFebruary132006)
Tracking UniqUe ObjecTs 177
ablockoftrialsThatisanobservermighttracktheliontherabbittheturtleandthegoatonallthetrialsinablockThismadeiteasierforobserverstorememberwhattheyweresupposedtotrackInadditioniteliminatedthepossibilityof interferenceornegativeprimingarisingfromtargetsandnontargetsswitchingrolesWeexpectedthatperformanceinthestandardconditionwouldimprovetoabove30itemsthequestionwaswhetherthecontentdeficitwouldstillmanifestitselfundertheseconditions
TheprocedurewasidenticaltothatofExperiment1exceptintheselectionofanimalsForeachobserveronesetofeightanimalswasselectedrandomlyatthebegin-ningofeachblockFouroftheseanimalsweredesignatedastargetsandtherestasnontargetsThesetofanimalsandthetargetnontargetassignmentswereheldconstantforthedurationofablock
Asexpectedperformanceinthestandardconditionimprovedtoacreditable34(SEM502)itemsHow-ever the content deficit was still present Observerstrackedonly21(SEM501)itemswhenaskedwhichtargetwaswhereThisdeficitwasreliable[t(7)583p00001]andsimilarinmagnitudetothatobservedinExperiment1ofaround13itemsAmixedANOVAcomparingthetwoexperiments2revealedalargeeffectof condition [F(18)5 1152p 000001]butnoneofexperiment[F(18)510p 10]Theinteractionterminthebetween-experimentsanalysiswasnearzero[F(18)1]
Theroughlyhalf-itemincreaseincapacitywhenwechangedfromavariabletoafixedtargetsetsuggeststhatswitchingtargetsetsfromtrialtotrialcreatedsomesortofinterferenceThiswastruenotonlyinthespecificcon-ditionwhenweprobedtargetidentitiesbutalsointhestandardconditionThisresultisimportantbecauseitsuggestsaroleforobjectidentity(eitherattheperceptualorsemanticlevel)irrespectiveoftaskdemandsAccord-ingtoLeonardandPylyshyn(2003)blinkingthetargetsonandoffshouldautomaticallyassignvisualindexestothemandtheseindexesshouldstickequallywellwhetherornottheseparticularstimuliweretargetsontheprevioustrialSincetheincreaseincapacitywasnotstatisticallyreliablethisconclusionshouldnotbetakentooseriouslyHoweverinsubsequentexperimentsweusedafixedtar-getset
JustasinExperiment1theobserversherecouldreli-ablylocatemultipletargetsaccordingtotheiridentitieswhichwetakeasevidenceforcontentaddressabilityEx-periment2alsoreplicatedthecontentdeficitwithreli-ablybettercapacityinthestandardreportconditionthaninthespecificreportcondition
TheseresultsraiseanumberofmethodologicalissuesFirstourexperimentsdifferfrommostMOTexperimentsintheirtemporalstructureourtrialswerebothlongerthanusualandincludedtemporaluncertaintyHowdidthesefactorsaffectperformanceExperiment3wasdesignedtoaddressthisissue
Asecondimportantissueiswhetherblockingresponsemodesinducedobserverstousedifferentencodingstrate-giesinthetwoconditionsleadingtodifferentialperfor-manceThisissuewillbeaddressedinExperiment4
AthirdissueiswhethertheuseofuniqueidentifiableobjectshelpsorhindersMOTperformanceincomparisonwiththeuseofidenticalobjectsinmostpreviousstudiesofMOTWewilladdressthisissueinExperiments5and6
FinallyExperiment7willaddresstheroleofnontar-getsinhinderingtrackingperformance
exPeriMeNT 3
Aswenotedintheintroductionitisimportanttoen-surethatobserversareattendingtoataskthroughoutatrialOneofourstrategiesweusedtoensurethiswastorandomlyvary trialdurationObserversdidnotknowwheneachtrialwouldendbetweenthe10thand15thcyclesWasthisenoughtemporaluncertaintyhoweverInExperiment3weincreasedtheuncertaintysothatatrialcouldendanywherebetweenthe5thand15thcycles(adurationfrom67to200sec)OtherwisethemethodwasidenticaltothatofExperiment2IfobserverswereactivelytrackingthetargetsonlyafterCycle9inthepre-viousexperimentsweshouldseeadecreaseinperfor-manceinthisexperimentwheretheypotentiallyhavetosustainattentionforalongerperiodoftime
Awiderarrayoftrialendpointsalsoallowedustoex-amineapossibleexplanationforthecontentdeficitWeusedtrialdurationsthatwerelongrelativetopreviousMOTstudiesPerhapslocationndashidentitybindingsdecayovertimeandwewouldobserveasmallerdifferencebe-tweenconditionsifweprobedearlierinthetrial
Figure3showskinExperiment3asafunctionofcyclewithdatafromExperiment2shownforcomparisonMeancapacityforthestandardconditionwas34items(SEM501)versus26items(SEM502)forthespecificcondi-tionrepresentingasignificantadvantageinthestandardcondition[t(7)539p 01]Wealsoconductedananalysisofcovarianceontheresultswithconditionasacategoricalvariableandcycleasacontinuousvariable(covariate)Neitherthemaineffectofcycle[F(17)1]northecycle3conditioninteraction[F(17)5249p516]wassignificant
5 7 9 11 13 150
1
2
3
4
Cycles
k (it
ems)
Figure 3 Capacity as a function of cycle Squares denote the standard and triangles the specific target condition data from experiment 3 are plotted in black and those from experiment 2 in gray error bars indicate standard errors of the means
178 HOrOwiTz eT al
WhencomparingthetwoexperimentsitseemsclearthatincreasingthetemporalvariabilityinExperiment3didnotchangetheresultsnotablyWeconductedacross-experimentANOVAusingdataonlyfromCycles10ndash15inExperiment3Standardversusspecifictargetconditionandcyclewerewithin-subjectsfactorsandexperimentwasabetween-subjectsfactorWefoundnomaineffectofexperiment[F(114)511p 5 31]Thecondition3ex-perimentinteractionapproachedsignificance[F(114)541p506]probablybecausethespecifictargetcapac-itywassomewhathigherinExperiment3Addingtempo-raluncertaintyclearlydidnotdisruptperformanceinthisexperimentinrelationtoExperiment2
ThereisahintinFigure3thatperformanceespeciallyin thespecific targetconditiondroppedas trialdura-tionincreasedThiswouldbecompatiblewithdatafromOksamaandHyoumlnauml(2004)possiblyreflectingavigilancedecrement(Parasuraman1986)IfthedropovertimeweregreaterinthespecificconditionthiswouldsupporttheideathatlocationndashidentitybindingsweredecayingHowevertherewasnoevidenceforthisinanANOVAconductedondatafromalloftheconditionsofExperiment3inwhichneitherthemaineffectofcycle[F(1070)511p538]northecycle3conditioninteraction[F(1070)1]wassignificantIftherewasdecayithadapproachedasymp-totewithinthefirst6ndash7secoftracking
exPeriMeNT 4
OneobvioushypothesistoexplainthecontentdeficitisthatobserversencodeinformationdifferentlyinthetwoconditionsInthestandardconditiononlyspatiotemporalpropertiesareencodedInthespecifictargetconditionobserversknowthattheywillbeaskedaboutfeaturaloridentityinformationsotheyencodethataswellFeaturalinformation(orperhapsthebindingofidentityandlo-cation)takesupmoreresources(Bahrami2003Saiki2002) reducing tracking capacityThus informationaboutfeaturesoridentitiesrequiresmorecapacity
InExperiment4wetestedthisstrategicencodinghy-pothesisbymixingthequestionswithinablockoftri-alsObserversdidnotknowonagiventrialwhethertheywouldbeaskedtolocateallofthetargetsorjustaspecifictargetThereforetheyhadtoadoptthesamestrategiesonbothtypesoftrialsIfthestrategicencodinghypothesisiscorrectthenthecontentdeficitshouldbereducedinthisexperimentPerformanceshouldalsobereducedoverallsinceobserverswouldnotbeoptimallyencodinginfor-mationoneachtrialInotherrespectsthemethodwasidenticaltothatofExperiment2
Pilotworkindicatedthatwhenquestionsweremixedwithinablockof trialshavingto lookawayfromthecactusarraytoreadthequestionatthetopofthescreenimpairedperformanceThereforeinthisexperimentweusedauditoryratherthanvisualprobesInsteadofprintingthequestionatthetopofthescreenobserversheardtheprobequestionsoverheadphones
Meancapacityforthestandardconditionwas34items(SEM501)asopposedto22items(SEM502)forthe
specificconditionrepresentingasignificantadvantageforthestandardcondition[t(7)599p00005]Per-formanceinthisexperimentwithquestionsmixedwithinblockswasnearlyidenticaltothatinExperiment2withblockedquestionsEvenwhenobserverscouldnothaveadopteddifferentencodingstrategies for thedifferentconditionswestillobservedasizable(12-item)contentdeficitThissuggeststhatthecontentdeficitisnotaresultofthestrategicdemandsoftheconditionswedevised
exPeriMeNT 5
ThepreviousexperimentsestablishedaparadigmforstudyingMOTwithunique stimuliOnequestion thatwehadnotyetaddressediswhetheruniqueobjectshelporhinderMOTperformanceincomparisonwithidenticalobjectsForexampleiftheobjectsareidenticalanob-servermightaccidentallyswapatargetandadistractoriftheyapproachtoocloselyWithuniqueobjectsthisislesslikelytohappenSimilarlyifatsomepointtheobserverrealizedthathewasonlytrackingthreetargetsandhadlostonehecouldtheoreticallyrecoverthelosttargetifitwereuniqueOfcoursesuchadvantagesassumeasortofldquoidealobserverrdquowhoalwaysknowsexactlywhatheistrackingIfthetrackingsystemwascompletelyinsensitivetotargetfeaturesoridentityuniqueobjectswouldprovidenoadvan-tageSimilarlytheobservercouldonlyrecoveralosttargetifinformationwereavailableabouthowmanytargetswerecurrentlybeingtrackedandwhichonewasmissing
Inthisexperimentwecomparedtrackinguniqueob-jectswithtrackingidenticalobjectsSinceldquoWhereisthezebrardquoisnotadiagnosticquestionwhenallobjectsarezebrasweonlyusedthestandardresponsemodeTherewerethreeconditionsTheuniqueconditionwasidenti-caltothestandardresponseconditioninExperiment1Intheidenticalconditioneightidenticalanimalswerepresented animal identitywas selectedat randomoneachtrialFinallyinthepairedconditiontherewerefouruniquetargetsagainselectedatrandomForeachtargettherewasanidenticaldistractorProbeswerepresentedauditorilyasinExperiment4Atotalof10observerspar-ticipatedinthisexperiment
Wecancompareperformanceintheuniqueandidenti-calobjectconditionstodeterminewhetherthereisanyadvantageforuniqueobjectsIftheadvantageforuniqueobjectsderivesfromtheabilitytorecoverlosttargetsthisadvantageshouldbeabolishedinthepairedcondition
CapacityvaluesforthethreeconditionsareplottedinFigure4Observerstracked31items(SEM502)intheuniquecondition23(SEM502)inthepairedconditionand25(SEM501)intheidenticalconditionPlannedttestsshowedthatthepairedandidenticalconditionsdidnotdiffer[t(9)513p 10]butthatperformancewassignificantlyworseinboththanintheuniquecondition[forpairedt(9)527p05foridenticalt(9)527p01]
TheseresultsindicatethatobserverscantakeadvantageoftheadditionalinformationprovidedbyuniqueobjectsTheuniqueobjectadvantage(approximately06items
Tracking UniqUe ObjecTs 179
undertheseconditions)isentirelyeliminatediftargetsareidenticaltodistractorssuggestingthatsomesortoferrorrecoveryprocessisresponsible
exPeriMeNT 6
In the previous experiment pairing targets anddistractorsreducedperformanceinthestandardcondi-tionrelativetothatinthefullyuniqueconditionWouldthismanipulationaffectthespecificconditionaswellInExperiment6wemeasuredthecontentdeficitwithbothpairedanduniquestimuliBycomparingthesetwostimulusconditionswecandeterminewhethereliminat-ing featuraldistinctionsbetween target andnontargetsetsimpairstheabilitytotrackspecificobjectsaswellastheabilitytoholdontotargetlocationsProbeswerepresentedauditorilyasinExperiment5Therewere6ob-serversinthisexperiment
AsshowninFigure5weonceagainreplicated thecontentdeficitinthisexperimentsinceobserverscouldtrack11moreitemsinthestandardconditionthaninthe
specificcondition[F(19)51190p001]Pairingtar-getsandnontargetsreducedcapacitybyroughlythesameamountas inExperiment5 [09 itemsF(19)5373p001]Criticallythepairingmanipulationappearedtohavelessofadeleteriouseffectinthespecificconditionthaninthestandardcondition[interactionF(19)5185p005]Inthestandardconditionthepairingmanipu-lationcostobservers12itemsversusalossofonly07itemsinthespecificconditionThisinteractionsuggeststhatdifferentrepresentationsmayunderlieperformanceinthetwoconditions
Ontheotherhandcapacitywasalreadylowerinthespe-cificconditionIftheeffectofpairingstimuliismultipli-cativeratherthanadditivewewouldthenexpectthattheimpactofpairedstimuliwouldbeproportionallysmallerinthespecificconditionWhenwelog-transformedthecapacityvaluesconvertingproportionaldifferencesintoadditive ones the question3 pairing interaction waseliminated[F(19)1]
Sincewecannotrejectthehypothesisthattheeffectofpairingtargetswithnontargetsisproportionaltocapacitythisexperimentprovidesatbestweakevidencefordiffer-entrepresentations
exPeriMeNT 7
OneadvantageofourmethodisthatwecantestMOTperformanceinacasewithonlytargetsWhywouldwewanttodothisTherearereasonstobelievethatsuppress-ingnontargetsconsumessomeresourcesthatwouldother-wisebeavailablefortrackingtargetsForexamplethereisevidencethatobserverstrackbetterontrialsinwhichnontarget trajectories are repeated from earlier trials(OgawaampYagi2003)IfnontargetsareprocessedtheymayhavetobesuppressedinordertoreduceconfusionwithtargetsandPylyshynandLeonard(2003)showedevidenceforsuchinhibitionMoreoverunpublisheddatafromourlaboratoryindicatethatMOTperformancede-creasesasnontargetsareaddedtothedisplayevenwhenthenumberoftargetsremainsconstant
Accordingly we designed Experiment7 to test thehypothesisthattrackingcapacitywouldincreaseiftherewerenonontargetsThisexperimentwasidenticaltoEx-periment2exceptthatallitemsweretargetsCapacityinthestandardconditionwas58items(SEM502)sig-nificantlygreaterthanthe26items(SEM501)obtainedinthespecificcondition[t(7)5143p000005]
InagreementwithourhypothesiscapacityimprovedmarkedlyinthestandardconditionwhentherewerenodistractorsStandardcapacityinExperiment7wassig-nificantlygreaterthaninExperiments23and4whichweremethodologicallycomparableexceptforthenumberoftargetsandnontargets(allps00005)
Oneexplanationisthatpreviousexperimentsunderes-timatedcapacitybecauseofaceilingeffectTheseexperi-mentswerebasedonthehypothesisthatcapacityvariesacrosstrialswithnounderlyingdifferencesacrossexperi-mentsHoweverthemeasuredcapacityhasbeenlimitedbythenumberoftargetsForexampleifthereareonlyfour
Unique Paired Identical0
1
2
3
4
Condition
k (it
ems)
Figure 4 Capacity as a function of condition in experiment 5
Specific Standard0
1
2
3
4Unique
Paired
Condition
k (it
ems)
Figure 5 effect of pairing target and nontarget objects on ca-pacity in experiment 6 data from the paired conditions are plot-ted as gray bars and those from the unique conditions as open bars
180 HOrOwiTz eT al
targets(egExperiments2ndash4)observerscouldtrackallofthemontrialsinwhichtheircapacitywas4orgreateranytrialswithanactualcapacitygreaterthan4wouldthusappeartohaveacapacityof4InthiscasemeasuredmeancapacitywouldunderestimatethetruemeanInExperi-ment7whichincludedeighttargetsmeasuredcapacitycouldreflectactualcapacityuptoeightitemsleadingtoagreater(andmoreaccurate)estimateofmeancapacity
Wecantestthisceilinghypothesisagainstourhypoth-esisofincreasedmeancapacitybycomparingthecapacitydistributionsinExperiment7withthoseinExperiments2ndash4Accordingtotheceilinghypothesistheproportionoftrialswithacapacitylessthan4shouldbeidenticalacrossexperimentsFurthermoretheproportionoftrialswithacapacityof4orgreaterinExperiment7shouldbethesameastheproportionoftrialswithacapacityequalto4inExperiments2ndash4Incontrastthehypothesisthattrackingcapacityincreaseswhennonontargetsarepres-entpredictsthattheentiredistributionshouldshifttotherightleadingtofewertrialswithcapacitylessthan4inExperiment7thanintheearlierexperiments
We tested these hypotheses by computing capacityforeachtrialusingEquation2andgeneratingadistri-butionacrosstrialsforeachobserverFigure6plotsthedistributionofkforExperiment7andthecorrespondingdistributionforExperiments23and4combined(dis-tributionsfromeachoftheseexperimentswereaveragedacrossobservers)Clearlytheproportionoftrialswithanobservedcapacitylessthan4waslowerwitheighttargetsthanwithfourtargetsThisisconsistentwiththecapacitydistributionshiftingtotherightinExperiment7ratherthanmerespreadingofthek54trialsoverawiderrangeWesuggestthatthesuperiorperformanceinthisexperi-mentmayreflectadditionalcapacityfreedupwhenthereisnoneedtosuppressnontargetsHowevertheshapeofthedistributionforfourtargetsdoessuggestthatcapacitywasunderestimatedinthepreviousexperiments
TheconclusionsarequitedifferentifwelookatthespecifictargetconditionSpecifictargetcapacityinthis
experimentwasbetterthaninExperiment2(uncorrectedttestp05)butnotsignificantlydifferentfromtheca-pacitiesobservedinExperiments3and4( p05)Thusthetrackingoftargetidentitiesinthisconditiondoesnotseem tobe limitedby theneed to suppressnontargetidentities
diSCuSSioN
WehavedevelopedanewMOTparadigmthatallowsustotestthetrackingofuniqueobjectsFourmainfind-ingshaveemergedfromtheseexperimentsFirstthereisacontent-addressablerepresentationoftargetsduringMOT(specificcapacitywas1inallcases)SecondweobservedacontentdeficitThecontent-addressablerep-resentationhasalowercapacity(measuredinitems)thanthelocation-addressablerepresentation(iespecificca-pacitywaslowerthanstandardinallcases)Thirdthevi-sualsystemiscapableoftakingadvantageofdifferencesbetweenuniqueobjectstoimprovetrackingperformanceoverthelevelseenwithidenticalobjects(Experiment5)Thisadvantageappearstobedueprimarilytotheabilitytorecoverlosttargets(Experiment5)Fourthtrackingcapacityappearstobeconstrainedbytheneedtosup-pressnontargetswhentherearenonontargetscapacityincreasesmarkedly(Experiment7)
A Content-Addressable representation in MoTOurdata represent anexistenceproofof a content-
addressable representation inMOTSimilar data havebeenreportedbyOksamaandHyoumlnauml(2004)Intheirmul-tiple-identity-tracking(MIT)taskobserverstrackedmov-inglinedrawings(eitherobjectsorldquopseudo-objectsrdquo)forseveralsecondsAttheendofthetrialthestimuliweremaskedandasingleitemwasmarkedwithablackframeInasubsequentresponsescreenobservershadtoselectthepicturethatcorrespondedtothemarkedobjectOksamaandHyoumlnaumlobtainedamediancapacityoffouritems
Althoughitisdifficulttocomparecapacitymeasuresdirectlyacrossexperimentaltasks3itisinterestingthatbothweandOksamaandHyoumlnauml(2004)demonstratedthatobserversdoknowtosomeextentwhichtargetiswhichThisfindingisincontrasttothoseofPylyshyn(2004)whoseobservershaddifficultyidentifyingtargetsTwokeydifferencesbetweenourparadigmandMITontheonehandandPylyshynrsquos(2004)paradigmontheotherseemrelevanttothisdiscrepancyFirstinourstudyandthatofOksamaandHyoumlnaumlobjectidentitiesweredefinedbyintrinsicfeaturessuchasshapeand(inourexperi-ment)colorInPylyshynrsquos(2004)experimentldquoidentityrdquowasbasedonatangentialfeatureofanobjectsuchasinwhichcornerofthedisplayitbeganthetrialItseemslikelythattheshapeandcolorofanobjectmaybemoretightlyboundmarkersofldquoidentityrdquothaninitialpositionorabrieflypresentedletter
SecondinourexperimentsandthoseofOksamaandHyoumlnauml (2004) thevisualdifferencesbetweenobjectswerecontinuallyavailablethroughoutthetrackingphaseofthetrialonlymaskedduringtheresponsephaseIn
4 targets8 targets
0 2 4 6 80
1
2
3
4
5
6
Capacity
Pro
po
rtio
n o
f Tri
als
Figure 6 Capacity distributions for the standard condition in experiment 7 (gray lines striped area) and in experiments 2 3 and 4 (black lines stippled area) dashed lines indicate standard errors of the means
Tracking UniqUe ObjecTs 181
Pylyshynrsquos(2004)experimentstheobjectidentitieswerepresentedonlybrieflyatthestartofthetrialduringthemajorityofthetrackingphasetheobjectswerephysicallyidenticalItmaybethatthelinkbetweenobjectidentityandspatiotemporalpositiondecayedduringPylyshynrsquos(2004)taskbutwasconstantlyrefreshedinbothoursandOksamaandHyoumlnaumlrsquos
Ourexperimentsalsoprovidedsomeindirectevidenceforacontent-addressablerepresentationForexampleperformancewasconsistentlysuperior inexperimentsinwhichthesameobjectswereusedinthesamerolesfromtrialtotrial(Experiments23and4)thaninex-perimentsinwhichthestimulussetchangedfromtrialtotrial(Experiments15and6)Thisoccurrednotonlyinthespecifictargetconditionsinwhichobjectidentitywasrelevantbutalsointhestandardconditionsinwhichobjectidentitywasnot
one System or TwoIstrackingservedbyasinglerepresentationthatencodes
bothpositionandidentityoraretheretworepresentationsoneforpositionandoneforidentityThisquestionarisesbecauseofthestrikingcapacitydifferencesbetweenourresponseconditionsInallofourexperimentscapacitywassignificantlylowerforspecifictargetsthanfortargetsinthestandardMOTresponsemodeInExperiment4wedisconfirmedthehypothesisthatcapacitydifferencesresultedfromvariedencodingstrategies(seeegDavisWelchHolmesampShepherd2001)sinceweobtainedthesameresultwhenresponsemodesweremixedwithinablockaswedidwhentheywereruninseparateblocks
ThiscapacitydifferenceisnotconclusiveevidencethatseparaterepresentationsorsystemsareinvolvedThereareseveralwaysinwhichaunifiedaccountcouldex-plainthelowercapacityinthespecifictargetconditionForexampleonemightproposeasystemwithaslot-likestructure(VogelWoodmanampLuck2001)inwhichallslotsarenotcreatedequalIfthereweretwoslotswithsuf-ficientresolutiontoencodebothpositionandidentityandtwoslotsthatcouldonlyencodepositionwewouldalsoobtaindifferentcapacityestimatesinourtworesponseconditionsThisrepresentationcouldalsobedescribedintermsofKahnemanandTreismanrsquos(1984)objectfilesifweassumethatsomeoftheobjectfilesareemptyTheimpletionprocessmightinthatcasereturntheinforma-tionthatagivenobjectwasspatiotemporallycontinuouswithatargetobjectbutfailtoretrievethesemanticorperceptualcontentassociatedwiththatobjectfile
IntheMOTliteraturecapacityisgenerallyassumedtorefertoafixednumberofavailablerepresentationsinthevisualsystemeithervisualindexesorobjectfilesHow-everthereissomeevidencethatcapacityinMOTmightbebetterthoughtofasanundifferentiatedresource(AlvarezampCavanagh2005)InsteadofasetofslotsorpointerswemightassumethatthereisafixedamountofinformationthatcanbeencodedObserversmightencodepositionandidentityforalltargetsbutthebindingbetweenpositionsandidentitymaybenoisierthanthepositioninformationitself(ormaydecaymorequicklyExperiment3indicatedthatanysuchdecaywouldhavetoapproachitsasymptote
valuewithin6ndash7sec)leadingtoapparentlylowercapacitywhenresponsesdependonbinding
InsteadofonesystemtherealsomightbetwoAlthoughwetendtoassumethatasingleexperimentalparadigmprobestheperformanceofasinglecorrespondingfunc-tionalsystemitmaybethatMOTnaturallyrecruitsmulti-pleoverlappingsystemsThesimplesttwo-systemaccountmightbeaproposalthattrackingisservedbyaposition-trackingsystemsuchasvisualindexes(FINSTs)butthataccessingidentityndashpositionbindingsrequiresfocalatten-tion(WheelerampTreisman2002)Howeversincespecificcapacityisreliablygreaterthanoneitemwewouldhavetoassumethatmultipleitemscansimultaneouslybethefocusofattention(AwhampPashler2000Davisetal2001)
IfwewishtopreserveasinglefocusofattentionwemightinsteadproposethatbothvisualindexesandobjectfilescanbeusedVisualindexeswouldprovideonlypo-sitioninformationandobjectfilescouldsupportmorecomplexqueries
Itwillobviouslybedifficulttodefinitivelydecidebe-tweenoneandtwosystemsHowevertheevidenceofourfinalthreeexperimentspointstotwoindependentsystemsInExperiments5and6pairingtargetsandnontargetsre-ducedcapacityinthestandardbutnotinthespecifictar-getcondition(althoughwecouldnotrejectthehypothesisofaproportionaldecrease)InExperiment7standardca-pacityincreasedwhennonontargetswerepresentedbutspecificcapacitywasunaffected
MOThasbeenlikenedtoavisualworkingmemorytaskattimezero(CavanaghampAlvarez2005)soitisnaturaltolooktostudiesofworkingmemoryforanalogsofthecontent-addressableandlocation-onlysystemsTheclassictwo-pathwayschemeinvisionisthedistinctionbetweenventralanddorsalstreamsassociatedwithobjectrecogni-tionandspatialprocessingrespectively(UngerleiderampMishkin1982)ThisdistinctionhasbeenextendedintothedomainofworkingmemorybyWilsonOacuteScalaidheandGoldman-Rakic(1993)whoprovidedevidencethatobjectinformationandspatialinformationareneurallysegregatedinmonkeyprefrontalcortexwidelybelievedtobetheneuralsubstrateofworkingmemoryAsimilardissociationinhumanswasreportedusingPETimagingbySmithetal(1995)
Thedivisionofworkingmemoryintomultiplesubsys-temsgoesbackatleasttothemultiple-componentmodelofBaddeleyandHitch(1974)whichproposedseparatecomponents for different modalities a ldquophonologicallooprdquoforauditoryinformationandaldquovisuospatialsketchpadrdquoforvisualinformationInmorerecentversionsofthemodel(BaddeleyampLogie1999Logie1995)thesketchpadhasbeendividedintovisualandspatialsubcompo-nents roughlycorrespondingto theobjectandspatialsystemsdescribedbyWilsonetal(1993)Interestinglythespatialsubcomponent(theldquoinnerscriberdquo)isspecifi-callytaskedwithholdingsequenceorpathinformation(seeParmentierElfordampMayberry2005)makingitalikelysubstrateforMOTperformance
Thewidespreadinsistenceonaseparationbetweenvi-sualorobjectinformationandspatialinformationwouldseem problematic for our findings Standard MOT is
182 HOrOwiTz eT al
clearlyajobforthespatialsubsystem(PostleDrsquoEspositoampCorkin2005forexampleusedaversionofstandardMOTtotieupdorsalprocessinginamemoryinterferenceparadigm)Howevertheredoesnotseemtoberoominthataccountforacontent-addressablerepresentationOurspecifictargettask(aswellastheMITtaskofOksamaampHyoumlnauml2004)isnotapureobjectrecognitiontaskitrequireslocationndashidentitybindingswhichwouldseemtoinvolveintegratinginformationacrossstreamsTheproblemdisappears ifwe take intoaccount recentbe-havioral(OlsonampMarshuetz2005)andfMRI(PostleampDrsquoEsposito1999)investigationsdemonstratingthatobjectworkingmemoryrepresentationsalsocarrysomelocationinformation
NeuroimagingstudiesmightproveusefulhereStan-dardMOThasbeenshowntoactivateparietallobeareasinthedorsalstream(Culhametal1998Jovicichetal2001Seiffert2003)Itwouldbeinformativetoknowwhetherthepatternofactivationchangeswhenuniqueob-jectsareintroducedorwhenthetaskpotentiallyinvolvesobjectidentities
unique object Features in MoTInadditiontoourfindingsaboutcontentaddressability
intrackingwewereabletodemonstratethathavingvisu-allyuniqueobjectsmadetrackingeasierobserverswereabletotrackroughly06moreitemswhenobjectswereuniquethanwhenobjectswereidenticalInExperiments5and6weattributedthisresulttoanerrorrecoverystrat-egyIfIlosetrackofthezebraIcansearchforthezebraonthebasisofitsphysicalfeaturesandcontinuetotrackitThisstrategywouldobviouslybeuselessifallitemsarezebrasThisstrategyisoflimitedutilityifthereareazebratargetandazebradistractoronceIlosethezebraIthenhaveonlya50chanceofpickingupthetargetzebraasopposedtothedistractor
TheerrorrecoveryhypothesisisnottrivialIthassev-eralimportantimplicationsFirstitimpliesthatobserversknowwhattheyarenottrackingThephenomenologyofMOTissuchthatanobserverhasthefeelingofknow-inghowmanytargetsheorsheissuccessfullytrackingTheerrorrecoveryhypothesisimpliesthatwhenatargetislosttheobserverknowswhichone(givenuniquetargets)SecondtheerrorrecoveryhypothesisalsoassumesthatanobservercansearchforamissingtargetwhilecontinuingtotracktheremainingtargetsThisisinagreementwithrecentevidencethatobserverscansuccessfullyperformanMOTtaskandavisualsearchtaskonthesametrial(AlvarezHorowitzArsenioDiMaseampWolfe2005)Finallythehypothesisimpliestheabilitytoaddtothetrackingsetontheflywhiletrackingwithoutanyphysi-calcueTheseimplicationsneedtobeconfirmedthroughfurtherresearch
Howeverifobserversuseuniquephysicalfeaturesofobjectstorecoverlosttargetsinthisfashionitiscuriousthatthereisnosuchbenefitinthespecifictargetcondi-tionIfobserversdorecoverlosttargetsduringthecourse
ofatrialtheyappeartorecoveronlytheirlocationsnottheiridentities
AnalternativetotheerrorrecoveryhypothesisthoughnotamutuallyexclusiveoneisthattheresultsofEx-periment5canbeexplainedbytheperceptualsimilarityoftargetandnontargetitemsYantisrsquos(1992)approachtoMOThighlightstheimportanceofperceptualgroup-ingfactorsthatallowthetargetstobetreatedasaunitbythevisualsystemMostoftheresearchinthislinehasfocusedonspatiotemporalgroupingfactorssuchasthespatialrelationshipsbetweentheobjects(SearsampPylyshyn2000ViswanathanampMingolla2002Yantis1992)
Cangroupingby(nonspatial)featuralpropertiesim-provetrackingAnumberofstudieshavedemonstratedthatobservershave limitedexplicit access to featuralproperties (Bahrami 2003Saiki 2002Scholl etal1999)butthisfindingdoesnotaddressthedirectrolethatfeaturesmayplayWecouldexplaintheresultsofEx-periment5byproposingthatobserverswereabletouseaccidentaldifferencesintheshapeandcolorstatisticsoftargetsandnontargetstohelpsegregatethemvisuallyInthepairedconditiontargetsandnontargetshadidenticalfeaturestatisticssothatadvantagewaslostThishypoth-esiscouldalsoexplainwhyperformancewasbetterwithafixedstimulussetthanwithavariablestimulussetoveranumberoftrialswiththesamestimuliobserverslearnsubtledifferencesbetweenthetwosets
ConclusionsIntheintroductionwenotedthatreal-worlddynamic
attentiontasksdifferfromtraditionalMOTintwowaysFirstobserversintheeverydayworldusuallydealwithmultipleuniqueobjectsSecondtheyoftenwishtoknowwhereagiventargetmightberatherthanwhichobjectsarethetargetsTheexperimentspresentedhererepresentanattempttobridgelaboratoryMOTresearchandtheseeverydaycognitivetasksWehaveshownthatitiseasiertotrackuniqueobjectsthantotrackidenticalobjectsassum-ingthattheobjectsonewishestotrackarealsouniquelydifferentfromnontargetobjectsFurthermoreobserversdoknowwhattheyaretrackingTotakeoneofourex-amplesofecologicallyvalidtrackingtasksifyouwatchyourchildrenandtheirfriendsontheplaygroundyoucanbeawareofwhereyourkidsareatanygivenmomentyoumayalsobeawareofthelocationsoftheirfriendswithoutbeingabletotellonefriendfromtheotherFurthermoreifyoulosetrackofachildyouwillprobablybeawareofwhichoneyouhavelost
AuTHor NoTe
ThisresearchwassupportedbyNIMHGrantR0165576toTSHWethankKristinMichodandSkylerPlaceforassistancewithdatacollec-tionaswellasSteveLuckandananonymousreviewerforconstructivecriticismCorrespondencerelatingtothisarticlemaybesenttoTSHorowitzBrighamandWomenrsquosHospitalVisualAttentionLaboratory64SidneyStreetSuite170BostonMA02139(e-mailtoddhsearchbwhharvardedu)
Tracking UniqUe ObjecTs 183
reFereNCeS
Allen R McGeorge P Pearson D amp Milne A B (2004)AttentionandexpertiseinmultipletargettrackingApplied Cognitive Psychology18337-347
Alvarez G A amp Cavanagh P (2005)IndependentresourcesforattentionaltrackingintheleftandrightvisualhemifieldsPsychologi-cal Science16637-643
Alvarez G A amp Franconeri S (2004November)How many ob-jects can you trackPaperpresentedatthe12thAnnualWorkshoponObjectPerceptionampMemoryMinneapolisMN
Alvarez G A Horowitz T S Arsenio H C DiMase J S amp Wolfe J M (2005)DomultielementvisualtrackingandvisualsearchdrawcontinuouslyonthesamevisualattentionresourcesJournal of Experimental Psychology Human Perception amp Perfor-mance31643-667
Awh E Jonides J amp Reuter-Lorenz P A (1998)RehearsalinspatialworkingmemoryJournal of Experimental Psychology Human Perception amp Performance24780-790
Awh E amp Pashler H (2000)EvidenceforsplitattentionalfociJournal of Experimental Psychology Human Perception amp Perfor-mance26834-846
Baddeley A D amp Hitch G J (1974)WorkingmemoryInGHBower(Ed)The psychology of learning and motivation Advances in research and theory(Vol8pp47-90)NewYorkAcademicPress
Baddeley A D amp Logie R H (1999) Working memoryThemultiple-componentmodelInAMiyakeampPShah(Eds)Models of working memory Mechanisms of active maintenance and executive control(pp28-61)CambridgeCambridgeUniversityPress
Bahrami B (2003)ObjectpropertyencodingandchangeblindnessinmultipleobjecttrackingVisual Cognition10949-963
Brainard D H (1997)ThePsychophysicsToolboxSpatial Vision10433-436
Cavanagh P amp Alvarez G A (2005)TrackingmultipletargetswithmultifocalattentionTrends in Cognitive Sciences9349-354
Culham J C Brandt S A Cavanagh P Kanwisher N G Dale A M amp Tootell R B H (1998)CorticalfMRIactiva-tionproducedbyattentive trackingofmoving targetsJournal of Neurophysiology802657-2670
Davis G Welch V L Holmes A amp Shepherd A (2001)CanattentionselectonlyafixednumberofobjectsatatimePerception301227-1248
Henderson J M amp Hollingworth A (2003)EyemovementsandvisualmemoryDetectingchangestosaccadetargetsinscenesPer-ception amp Psychophysics6558-71
Horowitz T S Birnkrant R S Fencsik D E Tran L amp Wolfe J M (2006)HowdowetrackinvisibleobjectsPsychonomic Bulletin amp Review13516-523
Jovicich J Peters R J Koch C Braun J Chang L amp Ernst T (2001)Brainareasspecificforattentionalloadinamotion-trackingtaskJournal of Cognitive Neuroscience131048-1058
Kahneman D amp Treisman A (1984)ChangingviewsofattentionandautomaticityInRParasuramanampDRDavies(Eds)Varieties of attention(pp29-61)OrlandoAcademicPress
Kahneman D Treisman A amp Gibbs B J (1992)ThereviewingofobjectfilesObject-specificintegrationofinformationCognitive Psychology24175-219
Klieger S B Horowitz T S amp Wolfe J M (2004)Ismultipleobjecttrackingcolorblind[Abstract]Journal of Vision4(8)363a
Leonard C amp Pylyshyn Z W (2003)Measuringtheattentionaldemandofmultipleobjecttracking(MOT)[Abstract]Journal of Vi-sion3(9)582a
Logie R H (1995) Visuo-spatial working memory Hove UKErlbaum
Milliken B Tipper S P amp Weaver B (1994)NegativepriminginaspatiallocalizationtaskFeaturemismatchinganddistractorin-hibitionJournal of Experimental Psychology Human Perception amp Performance20624-646
Mitroff S R Scholl B J amp Wynn K (2004)DivideandconquerHowobjectfilesadaptwhenapersistingobjectsplitsintotwoPsy-chological Science15420-425
Ogawa H amp Yagi A (2003)Primingeffectsinmultipleobjecttrack-ingAnimplicitencodingbasedonglobalspatiotemporalinformation[Abstract]Journal of Vision3(9)339a
Oksama L amp Hyoumlnauml J (2004)Ismultipleobjecttrackingcarriedoutautomaticallybyanearlyvisionmechanismindependentofhigher-ordercognitionAnindividualdifferenceapproachVisual Cognition11631-671
Oliva A Torralba A Castelhano M S amp Henderson J M (2003)Top-downcontrolofvisualattentioninrealworldscenes[Ab-stract]Journal of Vision3(9)3a
Olson I R amp Marshuetz C (2005)RememberingldquowhatrdquobringsalongldquowhererdquoinvisualworkingmemoryPerception amp Psychophysics67185-194
Parasuraman R (1986)VigilancemonitoringandsearchInKRBoffLKaufmanampJPThomas(Eds)Handbook of perception and human performance Vol 2 Cognitive processes and performance(pp1-39)NewYorkWiley
Parmentier F B R Elford G amp Mayberry M (2005)Transi-tionalinformationinspatialserialmemoryPathcharacteristicsaffectrecallperformanceJournal of Experimental Psychology Learning Memory amp Cognition31412-427
Pelli D G (1997)TheVideoToolboxsoftwareforvisualpsychophysicsTransformingnumbersintomoviesSpatial Vision10437-442
Postle B R amp DrsquoEsposito M (1999)ldquoWhatrdquondashthenndashldquowhererdquoinvi-sualworkingmemoryAnevent-relatedfMRIstudyJournal of Cog-nitive Neuroscience11585-597
Postle B R DrsquoEsposito M amp Corkin S (2005)Effectsofver-balandnonverbalinterferenceonspatialandobjectvisualworkingmemoryMemory amp Cognition33203-212
Pylyshyn Z [W] (1989)Theroleoflocationindexesinspatialper-ceptionAsketchoftheFINSTspatial-indexmodelCognition3265-97
Pylyshyn Z W (2004)SomepuzzlingfindingsinmultipleobjecttrackingITrackingwithoutkeepingtrackofobjectidentitiesVisual Cognition11801-822
Pylyshyn Z W amp Leonard C (2003)Inhibitionofnontargetsdur-ingmultipleobjecttracking(MOT)[Abstract]Journal of Vision3(9)585a
Pylyshyn Z W amp Storm R W (1988)Trackingmultipleindepen-denttargetsEvidenceforaparalleltrackingmechanismSpatial Vi-sion3179-197
Saiki J (2002)Multiple-objectpermanencetrackingLimitationinmaintenanceandtransformationofperceptualobjectsProgress in Brain Research140133-148
Schneider W amp Shiffrin R M (1977)ControlledandautomatichumaninformationprocessingIDetectionsearchandattentionPsychological Review841-66
Scholl B J amp Pylyshyn Z W (1999)Trackingmultiple itemsthroughocclusionCluestovisualobjecthoodCognitive Psychology38259-290
Scholl B J Pylyshyn Z W amp Feldman J (2001)Whatisavi-sualobjectEvidencefromtargetmerginginmultipleobjecttrackingCognition80159-177
Scholl B J Pylyshyn Z W amp Franconeri S L (1999May)When are spatiotemporal and featural properties encoded as a result of attentional allocationPaperpresentedattheconferenceoftheAssociationforResearchinVisionandOphthalmologyFortLau-derdaleFL
Sears C R amp Pylyshyn Z W (2000)MultipleobjecttrackingandattentionalprocessingCanadian Journal of Experimental Psychol-ogy541-14
Seiffert A E (2003)Dissociatingneuralcorrelatesofattentionaltrackingandattentiontovisualmotion[Abstract]Journal of Vision3(9)868a
Smith E E Jonides J Koeppe R A Awh E Schumacher E H amp Minoshima S (1995)SpatialversusobjectworkingmemoryPETinvestigationsJournal of Cognitive Neuroscience7337-356
Tipper S P (1985)ThenegativeprimingeffectInhibitoryprimingbyignoredobjectsQuarterly Journal of Experimental Psychology37A571-590
Ungerleider L G amp Mishkin M (1982)Twocorticalvisualsys-
184 HOrOwiTz eT al
temsInDJIngleMAGoodaleampRJWMansfield(Eds)Anal-ysis of visual behavior(pp549-586)CambridgeMAMITPress
Viswanathan L amp Mingolla E (2002)DynamicsofattentionindepthEvidence frommulti-element trackingPerception311415-1437
Vogel E K Woodman G F amp Luck S J (2001)Storageoffea-turesconjunctionsandobjectsinvisualworkingmemoryJournal of Experimental Psychology Human Perception amp Performance2792-114
Wheeler M E amp Treisman A M (2002)Bindinginshort-termvisualmemoryJournal of Experimental Psychology General13148-64
Wilson F A Oacute Scalaidhe S P amp Goldman-Rakic P S (1993)Dissociationofobjectandspatialprocessingdomains inprimateprefrontalcortexScience2601955-1958
Wolfe J M Oliva A Horowitz T S Butcher S J amp Bompas A (2002)SegmentationofobjectsfrombackgroundsinvisualsearchtasksVision Research422985-3004
Yantis S (1992)MultielementvisualtrackingAttentionandpercep-tualorganizationCognitive Psychology24295-340
NoTeS
1Wearemissingageinformationfor1observer2Threeobserverswhoparticipatedinbothexperimentswereelimi-
natedfromthisanalysis3Measuredcapacitycanvarybasedontrackingdurationtherange
ofmovementspeedsandwhetherornotobjectscanoccludeonean-other(asinourexperiment)orbounceoffoneanother(asinOksamaampHyoumlnauml2004)
APPeNdix A Stimulus List
beaver fox horse ostrich rhino turtlecamel giraffe kangaroo pelican rooster wolfcrocodile goat lion rabbit squirrel zebraelephant gorilla moose tiger
APPeNdix b raw data
Table b1 raw Numbers of items reported in the Standard Condition
Total Targets Nontargets Filled
Experiment1 308Experiment2 344 014 359Experiment3 347 015 362Experiment4 339 015 354Experiment5 Unique 312 037 349
Paired 249 080 329Identical 261 086 347
Experiment6 248 078 326Experiment7 599 599
NotemdashTargetsindicatestheaveragenumberofcorrectlyreportedtargetspertrialNontargetstheaveragenumberoftimesobserversclickedonanontargetlocationpertrialandTotal Filledthesumofthesetwocat-egoriesNontargetclickcountswerenotcollectedforExperiment1andtherewerenonontargetsinExperiment7
Table b2 raw Proportions for the Specific Condition by experiment
Errors
Correct Empty Trialsper Responses Transposition Nontarget Space N Observer
Experiment1 44 14 16 26 8 5000Experiment2 56 18 10 16 8 5000Experiment3 68 12 10 10 8 5000Experiment4 58 25 09 08 8 5063Experiment6 41 14 21 23 7 5000Experiment7 33 39 00 28 8 5000
NotemdashTranspositionerrorsrefertotrialsinwhichtheobserverclickedonthewrongtargetlocationTrialsperobservervariedinExperiment4becauseconditionwasselectedrandomlyfromtrialtotrial
(ManuscriptreceivedMarch312005revisionacceptedforpublicationFebruary132006)
178 HOrOwiTz eT al
WhencomparingthetwoexperimentsitseemsclearthatincreasingthetemporalvariabilityinExperiment3didnotchangetheresultsnotablyWeconductedacross-experimentANOVAusingdataonlyfromCycles10ndash15inExperiment3Standardversusspecifictargetconditionandcyclewerewithin-subjectsfactorsandexperimentwasabetween-subjectsfactorWefoundnomaineffectofexperiment[F(114)511p 5 31]Thecondition3ex-perimentinteractionapproachedsignificance[F(114)541p506]probablybecausethespecifictargetcapac-itywassomewhathigherinExperiment3Addingtempo-raluncertaintyclearlydidnotdisruptperformanceinthisexperimentinrelationtoExperiment2
ThereisahintinFigure3thatperformanceespeciallyin thespecific targetconditiondroppedas trialdura-tionincreasedThiswouldbecompatiblewithdatafromOksamaandHyoumlnauml(2004)possiblyreflectingavigilancedecrement(Parasuraman1986)IfthedropovertimeweregreaterinthespecificconditionthiswouldsupporttheideathatlocationndashidentitybindingsweredecayingHowevertherewasnoevidenceforthisinanANOVAconductedondatafromalloftheconditionsofExperiment3inwhichneitherthemaineffectofcycle[F(1070)511p538]northecycle3conditioninteraction[F(1070)1]wassignificantIftherewasdecayithadapproachedasymp-totewithinthefirst6ndash7secoftracking
exPeriMeNT 4
OneobvioushypothesistoexplainthecontentdeficitisthatobserversencodeinformationdifferentlyinthetwoconditionsInthestandardconditiononlyspatiotemporalpropertiesareencodedInthespecifictargetconditionobserversknowthattheywillbeaskedaboutfeaturaloridentityinformationsotheyencodethataswellFeaturalinformation(orperhapsthebindingofidentityandlo-cation)takesupmoreresources(Bahrami2003Saiki2002) reducing tracking capacityThus informationaboutfeaturesoridentitiesrequiresmorecapacity
InExperiment4wetestedthisstrategicencodinghy-pothesisbymixingthequestionswithinablockoftri-alsObserversdidnotknowonagiventrialwhethertheywouldbeaskedtolocateallofthetargetsorjustaspecifictargetThereforetheyhadtoadoptthesamestrategiesonbothtypesoftrialsIfthestrategicencodinghypothesisiscorrectthenthecontentdeficitshouldbereducedinthisexperimentPerformanceshouldalsobereducedoverallsinceobserverswouldnotbeoptimallyencodinginfor-mationoneachtrialInotherrespectsthemethodwasidenticaltothatofExperiment2
Pilotworkindicatedthatwhenquestionsweremixedwithinablockof trialshavingto lookawayfromthecactusarraytoreadthequestionatthetopofthescreenimpairedperformanceThereforeinthisexperimentweusedauditoryratherthanvisualprobesInsteadofprintingthequestionatthetopofthescreenobserversheardtheprobequestionsoverheadphones
Meancapacityforthestandardconditionwas34items(SEM501)asopposedto22items(SEM502)forthe
specificconditionrepresentingasignificantadvantageforthestandardcondition[t(7)599p00005]Per-formanceinthisexperimentwithquestionsmixedwithinblockswasnearlyidenticaltothatinExperiment2withblockedquestionsEvenwhenobserverscouldnothaveadopteddifferentencodingstrategies for thedifferentconditionswestillobservedasizable(12-item)contentdeficitThissuggeststhatthecontentdeficitisnotaresultofthestrategicdemandsoftheconditionswedevised
exPeriMeNT 5
ThepreviousexperimentsestablishedaparadigmforstudyingMOTwithunique stimuliOnequestion thatwehadnotyetaddressediswhetheruniqueobjectshelporhinderMOTperformanceincomparisonwithidenticalobjectsForexampleiftheobjectsareidenticalanob-servermightaccidentallyswapatargetandadistractoriftheyapproachtoocloselyWithuniqueobjectsthisislesslikelytohappenSimilarlyifatsomepointtheobserverrealizedthathewasonlytrackingthreetargetsandhadlostonehecouldtheoreticallyrecoverthelosttargetifitwereuniqueOfcoursesuchadvantagesassumeasortofldquoidealobserverrdquowhoalwaysknowsexactlywhatheistrackingIfthetrackingsystemwascompletelyinsensitivetotargetfeaturesoridentityuniqueobjectswouldprovidenoadvan-tageSimilarlytheobservercouldonlyrecoveralosttargetifinformationwereavailableabouthowmanytargetswerecurrentlybeingtrackedandwhichonewasmissing
Inthisexperimentwecomparedtrackinguniqueob-jectswithtrackingidenticalobjectsSinceldquoWhereisthezebrardquoisnotadiagnosticquestionwhenallobjectsarezebrasweonlyusedthestandardresponsemodeTherewerethreeconditionsTheuniqueconditionwasidenti-caltothestandardresponseconditioninExperiment1Intheidenticalconditioneightidenticalanimalswerepresented animal identitywas selectedat randomoneachtrialFinallyinthepairedconditiontherewerefouruniquetargetsagainselectedatrandomForeachtargettherewasanidenticaldistractorProbeswerepresentedauditorilyasinExperiment4Atotalof10observerspar-ticipatedinthisexperiment
Wecancompareperformanceintheuniqueandidenti-calobjectconditionstodeterminewhetherthereisanyadvantageforuniqueobjectsIftheadvantageforuniqueobjectsderivesfromtheabilitytorecoverlosttargetsthisadvantageshouldbeabolishedinthepairedcondition
CapacityvaluesforthethreeconditionsareplottedinFigure4Observerstracked31items(SEM502)intheuniquecondition23(SEM502)inthepairedconditionand25(SEM501)intheidenticalconditionPlannedttestsshowedthatthepairedandidenticalconditionsdidnotdiffer[t(9)513p 10]butthatperformancewassignificantlyworseinboththanintheuniquecondition[forpairedt(9)527p05foridenticalt(9)527p01]
TheseresultsindicatethatobserverscantakeadvantageoftheadditionalinformationprovidedbyuniqueobjectsTheuniqueobjectadvantage(approximately06items
Tracking UniqUe ObjecTs 179
undertheseconditions)isentirelyeliminatediftargetsareidenticaltodistractorssuggestingthatsomesortoferrorrecoveryprocessisresponsible
exPeriMeNT 6
In the previous experiment pairing targets anddistractorsreducedperformanceinthestandardcondi-tionrelativetothatinthefullyuniqueconditionWouldthismanipulationaffectthespecificconditionaswellInExperiment6wemeasuredthecontentdeficitwithbothpairedanduniquestimuliBycomparingthesetwostimulusconditionswecandeterminewhethereliminat-ing featuraldistinctionsbetween target andnontargetsetsimpairstheabilitytotrackspecificobjectsaswellastheabilitytoholdontotargetlocationsProbeswerepresentedauditorilyasinExperiment5Therewere6ob-serversinthisexperiment
AsshowninFigure5weonceagainreplicated thecontentdeficitinthisexperimentsinceobserverscouldtrack11moreitemsinthestandardconditionthaninthe
specificcondition[F(19)51190p001]Pairingtar-getsandnontargetsreducedcapacitybyroughlythesameamountas inExperiment5 [09 itemsF(19)5373p001]Criticallythepairingmanipulationappearedtohavelessofadeleteriouseffectinthespecificconditionthaninthestandardcondition[interactionF(19)5185p005]Inthestandardconditionthepairingmanipu-lationcostobservers12itemsversusalossofonly07itemsinthespecificconditionThisinteractionsuggeststhatdifferentrepresentationsmayunderlieperformanceinthetwoconditions
Ontheotherhandcapacitywasalreadylowerinthespe-cificconditionIftheeffectofpairingstimuliismultipli-cativeratherthanadditivewewouldthenexpectthattheimpactofpairedstimuliwouldbeproportionallysmallerinthespecificconditionWhenwelog-transformedthecapacityvaluesconvertingproportionaldifferencesintoadditive ones the question3 pairing interaction waseliminated[F(19)1]
Sincewecannotrejectthehypothesisthattheeffectofpairingtargetswithnontargetsisproportionaltocapacitythisexperimentprovidesatbestweakevidencefordiffer-entrepresentations
exPeriMeNT 7
OneadvantageofourmethodisthatwecantestMOTperformanceinacasewithonlytargetsWhywouldwewanttodothisTherearereasonstobelievethatsuppress-ingnontargetsconsumessomeresourcesthatwouldother-wisebeavailablefortrackingtargetsForexamplethereisevidencethatobserverstrackbetterontrialsinwhichnontarget trajectories are repeated from earlier trials(OgawaampYagi2003)IfnontargetsareprocessedtheymayhavetobesuppressedinordertoreduceconfusionwithtargetsandPylyshynandLeonard(2003)showedevidenceforsuchinhibitionMoreoverunpublisheddatafromourlaboratoryindicatethatMOTperformancede-creasesasnontargetsareaddedtothedisplayevenwhenthenumberoftargetsremainsconstant
Accordingly we designed Experiment7 to test thehypothesisthattrackingcapacitywouldincreaseiftherewerenonontargetsThisexperimentwasidenticaltoEx-periment2exceptthatallitemsweretargetsCapacityinthestandardconditionwas58items(SEM502)sig-nificantlygreaterthanthe26items(SEM501)obtainedinthespecificcondition[t(7)5143p000005]
InagreementwithourhypothesiscapacityimprovedmarkedlyinthestandardconditionwhentherewerenodistractorsStandardcapacityinExperiment7wassig-nificantlygreaterthaninExperiments23and4whichweremethodologicallycomparableexceptforthenumberoftargetsandnontargets(allps00005)
Oneexplanationisthatpreviousexperimentsunderes-timatedcapacitybecauseofaceilingeffectTheseexperi-mentswerebasedonthehypothesisthatcapacityvariesacrosstrialswithnounderlyingdifferencesacrossexperi-mentsHoweverthemeasuredcapacityhasbeenlimitedbythenumberoftargetsForexampleifthereareonlyfour
Unique Paired Identical0
1
2
3
4
Condition
k (it
ems)
Figure 4 Capacity as a function of condition in experiment 5
Specific Standard0
1
2
3
4Unique
Paired
Condition
k (it
ems)
Figure 5 effect of pairing target and nontarget objects on ca-pacity in experiment 6 data from the paired conditions are plot-ted as gray bars and those from the unique conditions as open bars
180 HOrOwiTz eT al
targets(egExperiments2ndash4)observerscouldtrackallofthemontrialsinwhichtheircapacitywas4orgreateranytrialswithanactualcapacitygreaterthan4wouldthusappeartohaveacapacityof4InthiscasemeasuredmeancapacitywouldunderestimatethetruemeanInExperi-ment7whichincludedeighttargetsmeasuredcapacitycouldreflectactualcapacityuptoeightitemsleadingtoagreater(andmoreaccurate)estimateofmeancapacity
Wecantestthisceilinghypothesisagainstourhypoth-esisofincreasedmeancapacitybycomparingthecapacitydistributionsinExperiment7withthoseinExperiments2ndash4Accordingtotheceilinghypothesistheproportionoftrialswithacapacitylessthan4shouldbeidenticalacrossexperimentsFurthermoretheproportionoftrialswithacapacityof4orgreaterinExperiment7shouldbethesameastheproportionoftrialswithacapacityequalto4inExperiments2ndash4Incontrastthehypothesisthattrackingcapacityincreaseswhennonontargetsarepres-entpredictsthattheentiredistributionshouldshifttotherightleadingtofewertrialswithcapacitylessthan4inExperiment7thanintheearlierexperiments
We tested these hypotheses by computing capacityforeachtrialusingEquation2andgeneratingadistri-butionacrosstrialsforeachobserverFigure6plotsthedistributionofkforExperiment7andthecorrespondingdistributionforExperiments23and4combined(dis-tributionsfromeachoftheseexperimentswereaveragedacrossobservers)Clearlytheproportionoftrialswithanobservedcapacitylessthan4waslowerwitheighttargetsthanwithfourtargetsThisisconsistentwiththecapacitydistributionshiftingtotherightinExperiment7ratherthanmerespreadingofthek54trialsoverawiderrangeWesuggestthatthesuperiorperformanceinthisexperi-mentmayreflectadditionalcapacityfreedupwhenthereisnoneedtosuppressnontargetsHowevertheshapeofthedistributionforfourtargetsdoessuggestthatcapacitywasunderestimatedinthepreviousexperiments
TheconclusionsarequitedifferentifwelookatthespecifictargetconditionSpecifictargetcapacityinthis
experimentwasbetterthaninExperiment2(uncorrectedttestp05)butnotsignificantlydifferentfromtheca-pacitiesobservedinExperiments3and4( p05)Thusthetrackingoftargetidentitiesinthisconditiondoesnotseem tobe limitedby theneed to suppressnontargetidentities
diSCuSSioN
WehavedevelopedanewMOTparadigmthatallowsustotestthetrackingofuniqueobjectsFourmainfind-ingshaveemergedfromtheseexperimentsFirstthereisacontent-addressablerepresentationoftargetsduringMOT(specificcapacitywas1inallcases)SecondweobservedacontentdeficitThecontent-addressablerep-resentationhasalowercapacity(measuredinitems)thanthelocation-addressablerepresentation(iespecificca-pacitywaslowerthanstandardinallcases)Thirdthevi-sualsystemiscapableoftakingadvantageofdifferencesbetweenuniqueobjectstoimprovetrackingperformanceoverthelevelseenwithidenticalobjects(Experiment5)Thisadvantageappearstobedueprimarilytotheabilitytorecoverlosttargets(Experiment5)Fourthtrackingcapacityappearstobeconstrainedbytheneedtosup-pressnontargetswhentherearenonontargetscapacityincreasesmarkedly(Experiment7)
A Content-Addressable representation in MoTOurdata represent anexistenceproofof a content-
addressable representation inMOTSimilar data havebeenreportedbyOksamaandHyoumlnauml(2004)Intheirmul-tiple-identity-tracking(MIT)taskobserverstrackedmov-inglinedrawings(eitherobjectsorldquopseudo-objectsrdquo)forseveralsecondsAttheendofthetrialthestimuliweremaskedandasingleitemwasmarkedwithablackframeInasubsequentresponsescreenobservershadtoselectthepicturethatcorrespondedtothemarkedobjectOksamaandHyoumlnaumlobtainedamediancapacityoffouritems
Althoughitisdifficulttocomparecapacitymeasuresdirectlyacrossexperimentaltasks3itisinterestingthatbothweandOksamaandHyoumlnauml(2004)demonstratedthatobserversdoknowtosomeextentwhichtargetiswhichThisfindingisincontrasttothoseofPylyshyn(2004)whoseobservershaddifficultyidentifyingtargetsTwokeydifferencesbetweenourparadigmandMITontheonehandandPylyshynrsquos(2004)paradigmontheotherseemrelevanttothisdiscrepancyFirstinourstudyandthatofOksamaandHyoumlnaumlobjectidentitiesweredefinedbyintrinsicfeaturessuchasshapeand(inourexperi-ment)colorInPylyshynrsquos(2004)experimentldquoidentityrdquowasbasedonatangentialfeatureofanobjectsuchasinwhichcornerofthedisplayitbeganthetrialItseemslikelythattheshapeandcolorofanobjectmaybemoretightlyboundmarkersofldquoidentityrdquothaninitialpositionorabrieflypresentedletter
SecondinourexperimentsandthoseofOksamaandHyoumlnauml (2004) thevisualdifferencesbetweenobjectswerecontinuallyavailablethroughoutthetrackingphaseofthetrialonlymaskedduringtheresponsephaseIn
4 targets8 targets
0 2 4 6 80
1
2
3
4
5
6
Capacity
Pro
po
rtio
n o
f Tri
als
Figure 6 Capacity distributions for the standard condition in experiment 7 (gray lines striped area) and in experiments 2 3 and 4 (black lines stippled area) dashed lines indicate standard errors of the means
Tracking UniqUe ObjecTs 181
Pylyshynrsquos(2004)experimentstheobjectidentitieswerepresentedonlybrieflyatthestartofthetrialduringthemajorityofthetrackingphasetheobjectswerephysicallyidenticalItmaybethatthelinkbetweenobjectidentityandspatiotemporalpositiondecayedduringPylyshynrsquos(2004)taskbutwasconstantlyrefreshedinbothoursandOksamaandHyoumlnaumlrsquos
Ourexperimentsalsoprovidedsomeindirectevidenceforacontent-addressablerepresentationForexampleperformancewasconsistentlysuperior inexperimentsinwhichthesameobjectswereusedinthesamerolesfromtrialtotrial(Experiments23and4)thaninex-perimentsinwhichthestimulussetchangedfromtrialtotrial(Experiments15and6)Thisoccurrednotonlyinthespecifictargetconditionsinwhichobjectidentitywasrelevantbutalsointhestandardconditionsinwhichobjectidentitywasnot
one System or TwoIstrackingservedbyasinglerepresentationthatencodes
bothpositionandidentityoraretheretworepresentationsoneforpositionandoneforidentityThisquestionarisesbecauseofthestrikingcapacitydifferencesbetweenourresponseconditionsInallofourexperimentscapacitywassignificantlylowerforspecifictargetsthanfortargetsinthestandardMOTresponsemodeInExperiment4wedisconfirmedthehypothesisthatcapacitydifferencesresultedfromvariedencodingstrategies(seeegDavisWelchHolmesampShepherd2001)sinceweobtainedthesameresultwhenresponsemodesweremixedwithinablockaswedidwhentheywereruninseparateblocks
ThiscapacitydifferenceisnotconclusiveevidencethatseparaterepresentationsorsystemsareinvolvedThereareseveralwaysinwhichaunifiedaccountcouldex-plainthelowercapacityinthespecifictargetconditionForexampleonemightproposeasystemwithaslot-likestructure(VogelWoodmanampLuck2001)inwhichallslotsarenotcreatedequalIfthereweretwoslotswithsuf-ficientresolutiontoencodebothpositionandidentityandtwoslotsthatcouldonlyencodepositionwewouldalsoobtaindifferentcapacityestimatesinourtworesponseconditionsThisrepresentationcouldalsobedescribedintermsofKahnemanandTreismanrsquos(1984)objectfilesifweassumethatsomeoftheobjectfilesareemptyTheimpletionprocessmightinthatcasereturntheinforma-tionthatagivenobjectwasspatiotemporallycontinuouswithatargetobjectbutfailtoretrievethesemanticorperceptualcontentassociatedwiththatobjectfile
IntheMOTliteraturecapacityisgenerallyassumedtorefertoafixednumberofavailablerepresentationsinthevisualsystemeithervisualindexesorobjectfilesHow-everthereissomeevidencethatcapacityinMOTmightbebetterthoughtofasanundifferentiatedresource(AlvarezampCavanagh2005)InsteadofasetofslotsorpointerswemightassumethatthereisafixedamountofinformationthatcanbeencodedObserversmightencodepositionandidentityforalltargetsbutthebindingbetweenpositionsandidentitymaybenoisierthanthepositioninformationitself(ormaydecaymorequicklyExperiment3indicatedthatanysuchdecaywouldhavetoapproachitsasymptote
valuewithin6ndash7sec)leadingtoapparentlylowercapacitywhenresponsesdependonbinding
InsteadofonesystemtherealsomightbetwoAlthoughwetendtoassumethatasingleexperimentalparadigmprobestheperformanceofasinglecorrespondingfunc-tionalsystemitmaybethatMOTnaturallyrecruitsmulti-pleoverlappingsystemsThesimplesttwo-systemaccountmightbeaproposalthattrackingisservedbyaposition-trackingsystemsuchasvisualindexes(FINSTs)butthataccessingidentityndashpositionbindingsrequiresfocalatten-tion(WheelerampTreisman2002)Howeversincespecificcapacityisreliablygreaterthanoneitemwewouldhavetoassumethatmultipleitemscansimultaneouslybethefocusofattention(AwhampPashler2000Davisetal2001)
IfwewishtopreserveasinglefocusofattentionwemightinsteadproposethatbothvisualindexesandobjectfilescanbeusedVisualindexeswouldprovideonlypo-sitioninformationandobjectfilescouldsupportmorecomplexqueries
Itwillobviouslybedifficulttodefinitivelydecidebe-tweenoneandtwosystemsHowevertheevidenceofourfinalthreeexperimentspointstotwoindependentsystemsInExperiments5and6pairingtargetsandnontargetsre-ducedcapacityinthestandardbutnotinthespecifictar-getcondition(althoughwecouldnotrejectthehypothesisofaproportionaldecrease)InExperiment7standardca-pacityincreasedwhennonontargetswerepresentedbutspecificcapacitywasunaffected
MOThasbeenlikenedtoavisualworkingmemorytaskattimezero(CavanaghampAlvarez2005)soitisnaturaltolooktostudiesofworkingmemoryforanalogsofthecontent-addressableandlocation-onlysystemsTheclassictwo-pathwayschemeinvisionisthedistinctionbetweenventralanddorsalstreamsassociatedwithobjectrecogni-tionandspatialprocessingrespectively(UngerleiderampMishkin1982)ThisdistinctionhasbeenextendedintothedomainofworkingmemorybyWilsonOacuteScalaidheandGoldman-Rakic(1993)whoprovidedevidencethatobjectinformationandspatialinformationareneurallysegregatedinmonkeyprefrontalcortexwidelybelievedtobetheneuralsubstrateofworkingmemoryAsimilardissociationinhumanswasreportedusingPETimagingbySmithetal(1995)
Thedivisionofworkingmemoryintomultiplesubsys-temsgoesbackatleasttothemultiple-componentmodelofBaddeleyandHitch(1974)whichproposedseparatecomponents for different modalities a ldquophonologicallooprdquoforauditoryinformationandaldquovisuospatialsketchpadrdquoforvisualinformationInmorerecentversionsofthemodel(BaddeleyampLogie1999Logie1995)thesketchpadhasbeendividedintovisualandspatialsubcompo-nents roughlycorrespondingto theobjectandspatialsystemsdescribedbyWilsonetal(1993)Interestinglythespatialsubcomponent(theldquoinnerscriberdquo)isspecifi-callytaskedwithholdingsequenceorpathinformation(seeParmentierElfordampMayberry2005)makingitalikelysubstrateforMOTperformance
Thewidespreadinsistenceonaseparationbetweenvi-sualorobjectinformationandspatialinformationwouldseem problematic for our findings Standard MOT is
182 HOrOwiTz eT al
clearlyajobforthespatialsubsystem(PostleDrsquoEspositoampCorkin2005forexampleusedaversionofstandardMOTtotieupdorsalprocessinginamemoryinterferenceparadigm)Howevertheredoesnotseemtoberoominthataccountforacontent-addressablerepresentationOurspecifictargettask(aswellastheMITtaskofOksamaampHyoumlnauml2004)isnotapureobjectrecognitiontaskitrequireslocationndashidentitybindingswhichwouldseemtoinvolveintegratinginformationacrossstreamsTheproblemdisappears ifwe take intoaccount recentbe-havioral(OlsonampMarshuetz2005)andfMRI(PostleampDrsquoEsposito1999)investigationsdemonstratingthatobjectworkingmemoryrepresentationsalsocarrysomelocationinformation
NeuroimagingstudiesmightproveusefulhereStan-dardMOThasbeenshowntoactivateparietallobeareasinthedorsalstream(Culhametal1998Jovicichetal2001Seiffert2003)Itwouldbeinformativetoknowwhetherthepatternofactivationchangeswhenuniqueob-jectsareintroducedorwhenthetaskpotentiallyinvolvesobjectidentities
unique object Features in MoTInadditiontoourfindingsaboutcontentaddressability
intrackingwewereabletodemonstratethathavingvisu-allyuniqueobjectsmadetrackingeasierobserverswereabletotrackroughly06moreitemswhenobjectswereuniquethanwhenobjectswereidenticalInExperiments5and6weattributedthisresulttoanerrorrecoverystrat-egyIfIlosetrackofthezebraIcansearchforthezebraonthebasisofitsphysicalfeaturesandcontinuetotrackitThisstrategywouldobviouslybeuselessifallitemsarezebrasThisstrategyisoflimitedutilityifthereareazebratargetandazebradistractoronceIlosethezebraIthenhaveonlya50chanceofpickingupthetargetzebraasopposedtothedistractor
TheerrorrecoveryhypothesisisnottrivialIthassev-eralimportantimplicationsFirstitimpliesthatobserversknowwhattheyarenottrackingThephenomenologyofMOTissuchthatanobserverhasthefeelingofknow-inghowmanytargetsheorsheissuccessfullytrackingTheerrorrecoveryhypothesisimpliesthatwhenatargetislosttheobserverknowswhichone(givenuniquetargets)SecondtheerrorrecoveryhypothesisalsoassumesthatanobservercansearchforamissingtargetwhilecontinuingtotracktheremainingtargetsThisisinagreementwithrecentevidencethatobserverscansuccessfullyperformanMOTtaskandavisualsearchtaskonthesametrial(AlvarezHorowitzArsenioDiMaseampWolfe2005)Finallythehypothesisimpliestheabilitytoaddtothetrackingsetontheflywhiletrackingwithoutanyphysi-calcueTheseimplicationsneedtobeconfirmedthroughfurtherresearch
Howeverifobserversuseuniquephysicalfeaturesofobjectstorecoverlosttargetsinthisfashionitiscuriousthatthereisnosuchbenefitinthespecifictargetcondi-tionIfobserversdorecoverlosttargetsduringthecourse
ofatrialtheyappeartorecoveronlytheirlocationsnottheiridentities
AnalternativetotheerrorrecoveryhypothesisthoughnotamutuallyexclusiveoneisthattheresultsofEx-periment5canbeexplainedbytheperceptualsimilarityoftargetandnontargetitemsYantisrsquos(1992)approachtoMOThighlightstheimportanceofperceptualgroup-ingfactorsthatallowthetargetstobetreatedasaunitbythevisualsystemMostoftheresearchinthislinehasfocusedonspatiotemporalgroupingfactorssuchasthespatialrelationshipsbetweentheobjects(SearsampPylyshyn2000ViswanathanampMingolla2002Yantis1992)
Cangroupingby(nonspatial)featuralpropertiesim-provetrackingAnumberofstudieshavedemonstratedthatobservershave limitedexplicit access to featuralproperties (Bahrami 2003Saiki 2002Scholl etal1999)butthisfindingdoesnotaddressthedirectrolethatfeaturesmayplayWecouldexplaintheresultsofEx-periment5byproposingthatobserverswereabletouseaccidentaldifferencesintheshapeandcolorstatisticsoftargetsandnontargetstohelpsegregatethemvisuallyInthepairedconditiontargetsandnontargetshadidenticalfeaturestatisticssothatadvantagewaslostThishypoth-esiscouldalsoexplainwhyperformancewasbetterwithafixedstimulussetthanwithavariablestimulussetoveranumberoftrialswiththesamestimuliobserverslearnsubtledifferencesbetweenthetwosets
ConclusionsIntheintroductionwenotedthatreal-worlddynamic
attentiontasksdifferfromtraditionalMOTintwowaysFirstobserversintheeverydayworldusuallydealwithmultipleuniqueobjectsSecondtheyoftenwishtoknowwhereagiventargetmightberatherthanwhichobjectsarethetargetsTheexperimentspresentedhererepresentanattempttobridgelaboratoryMOTresearchandtheseeverydaycognitivetasksWehaveshownthatitiseasiertotrackuniqueobjectsthantotrackidenticalobjectsassum-ingthattheobjectsonewishestotrackarealsouniquelydifferentfromnontargetobjectsFurthermoreobserversdoknowwhattheyaretrackingTotakeoneofourex-amplesofecologicallyvalidtrackingtasksifyouwatchyourchildrenandtheirfriendsontheplaygroundyoucanbeawareofwhereyourkidsareatanygivenmomentyoumayalsobeawareofthelocationsoftheirfriendswithoutbeingabletotellonefriendfromtheotherFurthermoreifyoulosetrackofachildyouwillprobablybeawareofwhichoneyouhavelost
AuTHor NoTe
ThisresearchwassupportedbyNIMHGrantR0165576toTSHWethankKristinMichodandSkylerPlaceforassistancewithdatacollec-tionaswellasSteveLuckandananonymousreviewerforconstructivecriticismCorrespondencerelatingtothisarticlemaybesenttoTSHorowitzBrighamandWomenrsquosHospitalVisualAttentionLaboratory64SidneyStreetSuite170BostonMA02139(e-mailtoddhsearchbwhharvardedu)
Tracking UniqUe ObjecTs 183
reFereNCeS
Allen R McGeorge P Pearson D amp Milne A B (2004)AttentionandexpertiseinmultipletargettrackingApplied Cognitive Psychology18337-347
Alvarez G A amp Cavanagh P (2005)IndependentresourcesforattentionaltrackingintheleftandrightvisualhemifieldsPsychologi-cal Science16637-643
Alvarez G A amp Franconeri S (2004November)How many ob-jects can you trackPaperpresentedatthe12thAnnualWorkshoponObjectPerceptionampMemoryMinneapolisMN
Alvarez G A Horowitz T S Arsenio H C DiMase J S amp Wolfe J M (2005)DomultielementvisualtrackingandvisualsearchdrawcontinuouslyonthesamevisualattentionresourcesJournal of Experimental Psychology Human Perception amp Perfor-mance31643-667
Awh E Jonides J amp Reuter-Lorenz P A (1998)RehearsalinspatialworkingmemoryJournal of Experimental Psychology Human Perception amp Performance24780-790
Awh E amp Pashler H (2000)EvidenceforsplitattentionalfociJournal of Experimental Psychology Human Perception amp Perfor-mance26834-846
Baddeley A D amp Hitch G J (1974)WorkingmemoryInGHBower(Ed)The psychology of learning and motivation Advances in research and theory(Vol8pp47-90)NewYorkAcademicPress
Baddeley A D amp Logie R H (1999) Working memoryThemultiple-componentmodelInAMiyakeampPShah(Eds)Models of working memory Mechanisms of active maintenance and executive control(pp28-61)CambridgeCambridgeUniversityPress
Bahrami B (2003)ObjectpropertyencodingandchangeblindnessinmultipleobjecttrackingVisual Cognition10949-963
Brainard D H (1997)ThePsychophysicsToolboxSpatial Vision10433-436
Cavanagh P amp Alvarez G A (2005)TrackingmultipletargetswithmultifocalattentionTrends in Cognitive Sciences9349-354
Culham J C Brandt S A Cavanagh P Kanwisher N G Dale A M amp Tootell R B H (1998)CorticalfMRIactiva-tionproducedbyattentive trackingofmoving targetsJournal of Neurophysiology802657-2670
Davis G Welch V L Holmes A amp Shepherd A (2001)CanattentionselectonlyafixednumberofobjectsatatimePerception301227-1248
Henderson J M amp Hollingworth A (2003)EyemovementsandvisualmemoryDetectingchangestosaccadetargetsinscenesPer-ception amp Psychophysics6558-71
Horowitz T S Birnkrant R S Fencsik D E Tran L amp Wolfe J M (2006)HowdowetrackinvisibleobjectsPsychonomic Bulletin amp Review13516-523
Jovicich J Peters R J Koch C Braun J Chang L amp Ernst T (2001)Brainareasspecificforattentionalloadinamotion-trackingtaskJournal of Cognitive Neuroscience131048-1058
Kahneman D amp Treisman A (1984)ChangingviewsofattentionandautomaticityInRParasuramanampDRDavies(Eds)Varieties of attention(pp29-61)OrlandoAcademicPress
Kahneman D Treisman A amp Gibbs B J (1992)ThereviewingofobjectfilesObject-specificintegrationofinformationCognitive Psychology24175-219
Klieger S B Horowitz T S amp Wolfe J M (2004)Ismultipleobjecttrackingcolorblind[Abstract]Journal of Vision4(8)363a
Leonard C amp Pylyshyn Z W (2003)Measuringtheattentionaldemandofmultipleobjecttracking(MOT)[Abstract]Journal of Vi-sion3(9)582a
Logie R H (1995) Visuo-spatial working memory Hove UKErlbaum
Milliken B Tipper S P amp Weaver B (1994)NegativepriminginaspatiallocalizationtaskFeaturemismatchinganddistractorin-hibitionJournal of Experimental Psychology Human Perception amp Performance20624-646
Mitroff S R Scholl B J amp Wynn K (2004)DivideandconquerHowobjectfilesadaptwhenapersistingobjectsplitsintotwoPsy-chological Science15420-425
Ogawa H amp Yagi A (2003)Primingeffectsinmultipleobjecttrack-ingAnimplicitencodingbasedonglobalspatiotemporalinformation[Abstract]Journal of Vision3(9)339a
Oksama L amp Hyoumlnauml J (2004)Ismultipleobjecttrackingcarriedoutautomaticallybyanearlyvisionmechanismindependentofhigher-ordercognitionAnindividualdifferenceapproachVisual Cognition11631-671
Oliva A Torralba A Castelhano M S amp Henderson J M (2003)Top-downcontrolofvisualattentioninrealworldscenes[Ab-stract]Journal of Vision3(9)3a
Olson I R amp Marshuetz C (2005)RememberingldquowhatrdquobringsalongldquowhererdquoinvisualworkingmemoryPerception amp Psychophysics67185-194
Parasuraman R (1986)VigilancemonitoringandsearchInKRBoffLKaufmanampJPThomas(Eds)Handbook of perception and human performance Vol 2 Cognitive processes and performance(pp1-39)NewYorkWiley
Parmentier F B R Elford G amp Mayberry M (2005)Transi-tionalinformationinspatialserialmemoryPathcharacteristicsaffectrecallperformanceJournal of Experimental Psychology Learning Memory amp Cognition31412-427
Pelli D G (1997)TheVideoToolboxsoftwareforvisualpsychophysicsTransformingnumbersintomoviesSpatial Vision10437-442
Postle B R amp DrsquoEsposito M (1999)ldquoWhatrdquondashthenndashldquowhererdquoinvi-sualworkingmemoryAnevent-relatedfMRIstudyJournal of Cog-nitive Neuroscience11585-597
Postle B R DrsquoEsposito M amp Corkin S (2005)Effectsofver-balandnonverbalinterferenceonspatialandobjectvisualworkingmemoryMemory amp Cognition33203-212
Pylyshyn Z [W] (1989)Theroleoflocationindexesinspatialper-ceptionAsketchoftheFINSTspatial-indexmodelCognition3265-97
Pylyshyn Z W (2004)SomepuzzlingfindingsinmultipleobjecttrackingITrackingwithoutkeepingtrackofobjectidentitiesVisual Cognition11801-822
Pylyshyn Z W amp Leonard C (2003)Inhibitionofnontargetsdur-ingmultipleobjecttracking(MOT)[Abstract]Journal of Vision3(9)585a
Pylyshyn Z W amp Storm R W (1988)Trackingmultipleindepen-denttargetsEvidenceforaparalleltrackingmechanismSpatial Vi-sion3179-197
Saiki J (2002)Multiple-objectpermanencetrackingLimitationinmaintenanceandtransformationofperceptualobjectsProgress in Brain Research140133-148
Schneider W amp Shiffrin R M (1977)ControlledandautomatichumaninformationprocessingIDetectionsearchandattentionPsychological Review841-66
Scholl B J amp Pylyshyn Z W (1999)Trackingmultiple itemsthroughocclusionCluestovisualobjecthoodCognitive Psychology38259-290
Scholl B J Pylyshyn Z W amp Feldman J (2001)Whatisavi-sualobjectEvidencefromtargetmerginginmultipleobjecttrackingCognition80159-177
Scholl B J Pylyshyn Z W amp Franconeri S L (1999May)When are spatiotemporal and featural properties encoded as a result of attentional allocationPaperpresentedattheconferenceoftheAssociationforResearchinVisionandOphthalmologyFortLau-derdaleFL
Sears C R amp Pylyshyn Z W (2000)MultipleobjecttrackingandattentionalprocessingCanadian Journal of Experimental Psychol-ogy541-14
Seiffert A E (2003)Dissociatingneuralcorrelatesofattentionaltrackingandattentiontovisualmotion[Abstract]Journal of Vision3(9)868a
Smith E E Jonides J Koeppe R A Awh E Schumacher E H amp Minoshima S (1995)SpatialversusobjectworkingmemoryPETinvestigationsJournal of Cognitive Neuroscience7337-356
Tipper S P (1985)ThenegativeprimingeffectInhibitoryprimingbyignoredobjectsQuarterly Journal of Experimental Psychology37A571-590
Ungerleider L G amp Mishkin M (1982)Twocorticalvisualsys-
184 HOrOwiTz eT al
temsInDJIngleMAGoodaleampRJWMansfield(Eds)Anal-ysis of visual behavior(pp549-586)CambridgeMAMITPress
Viswanathan L amp Mingolla E (2002)DynamicsofattentionindepthEvidence frommulti-element trackingPerception311415-1437
Vogel E K Woodman G F amp Luck S J (2001)Storageoffea-turesconjunctionsandobjectsinvisualworkingmemoryJournal of Experimental Psychology Human Perception amp Performance2792-114
Wheeler M E amp Treisman A M (2002)Bindinginshort-termvisualmemoryJournal of Experimental Psychology General13148-64
Wilson F A Oacute Scalaidhe S P amp Goldman-Rakic P S (1993)Dissociationofobjectandspatialprocessingdomains inprimateprefrontalcortexScience2601955-1958
Wolfe J M Oliva A Horowitz T S Butcher S J amp Bompas A (2002)SegmentationofobjectsfrombackgroundsinvisualsearchtasksVision Research422985-3004
Yantis S (1992)MultielementvisualtrackingAttentionandpercep-tualorganizationCognitive Psychology24295-340
NoTeS
1Wearemissingageinformationfor1observer2Threeobserverswhoparticipatedinbothexperimentswereelimi-
natedfromthisanalysis3Measuredcapacitycanvarybasedontrackingdurationtherange
ofmovementspeedsandwhetherornotobjectscanoccludeonean-other(asinourexperiment)orbounceoffoneanother(asinOksamaampHyoumlnauml2004)
APPeNdix A Stimulus List
beaver fox horse ostrich rhino turtlecamel giraffe kangaroo pelican rooster wolfcrocodile goat lion rabbit squirrel zebraelephant gorilla moose tiger
APPeNdix b raw data
Table b1 raw Numbers of items reported in the Standard Condition
Total Targets Nontargets Filled
Experiment1 308Experiment2 344 014 359Experiment3 347 015 362Experiment4 339 015 354Experiment5 Unique 312 037 349
Paired 249 080 329Identical 261 086 347
Experiment6 248 078 326Experiment7 599 599
NotemdashTargetsindicatestheaveragenumberofcorrectlyreportedtargetspertrialNontargetstheaveragenumberoftimesobserversclickedonanontargetlocationpertrialandTotal Filledthesumofthesetwocat-egoriesNontargetclickcountswerenotcollectedforExperiment1andtherewerenonontargetsinExperiment7
Table b2 raw Proportions for the Specific Condition by experiment
Errors
Correct Empty Trialsper Responses Transposition Nontarget Space N Observer
Experiment1 44 14 16 26 8 5000Experiment2 56 18 10 16 8 5000Experiment3 68 12 10 10 8 5000Experiment4 58 25 09 08 8 5063Experiment6 41 14 21 23 7 5000Experiment7 33 39 00 28 8 5000
NotemdashTranspositionerrorsrefertotrialsinwhichtheobserverclickedonthewrongtargetlocationTrialsperobservervariedinExperiment4becauseconditionwasselectedrandomlyfromtrialtotrial
(ManuscriptreceivedMarch312005revisionacceptedforpublicationFebruary132006)
Tracking UniqUe ObjecTs 179
undertheseconditions)isentirelyeliminatediftargetsareidenticaltodistractorssuggestingthatsomesortoferrorrecoveryprocessisresponsible
exPeriMeNT 6
In the previous experiment pairing targets anddistractorsreducedperformanceinthestandardcondi-tionrelativetothatinthefullyuniqueconditionWouldthismanipulationaffectthespecificconditionaswellInExperiment6wemeasuredthecontentdeficitwithbothpairedanduniquestimuliBycomparingthesetwostimulusconditionswecandeterminewhethereliminat-ing featuraldistinctionsbetween target andnontargetsetsimpairstheabilitytotrackspecificobjectsaswellastheabilitytoholdontotargetlocationsProbeswerepresentedauditorilyasinExperiment5Therewere6ob-serversinthisexperiment
AsshowninFigure5weonceagainreplicated thecontentdeficitinthisexperimentsinceobserverscouldtrack11moreitemsinthestandardconditionthaninthe
specificcondition[F(19)51190p001]Pairingtar-getsandnontargetsreducedcapacitybyroughlythesameamountas inExperiment5 [09 itemsF(19)5373p001]Criticallythepairingmanipulationappearedtohavelessofadeleteriouseffectinthespecificconditionthaninthestandardcondition[interactionF(19)5185p005]Inthestandardconditionthepairingmanipu-lationcostobservers12itemsversusalossofonly07itemsinthespecificconditionThisinteractionsuggeststhatdifferentrepresentationsmayunderlieperformanceinthetwoconditions
Ontheotherhandcapacitywasalreadylowerinthespe-cificconditionIftheeffectofpairingstimuliismultipli-cativeratherthanadditivewewouldthenexpectthattheimpactofpairedstimuliwouldbeproportionallysmallerinthespecificconditionWhenwelog-transformedthecapacityvaluesconvertingproportionaldifferencesintoadditive ones the question3 pairing interaction waseliminated[F(19)1]
Sincewecannotrejectthehypothesisthattheeffectofpairingtargetswithnontargetsisproportionaltocapacitythisexperimentprovidesatbestweakevidencefordiffer-entrepresentations
exPeriMeNT 7
OneadvantageofourmethodisthatwecantestMOTperformanceinacasewithonlytargetsWhywouldwewanttodothisTherearereasonstobelievethatsuppress-ingnontargetsconsumessomeresourcesthatwouldother-wisebeavailablefortrackingtargetsForexamplethereisevidencethatobserverstrackbetterontrialsinwhichnontarget trajectories are repeated from earlier trials(OgawaampYagi2003)IfnontargetsareprocessedtheymayhavetobesuppressedinordertoreduceconfusionwithtargetsandPylyshynandLeonard(2003)showedevidenceforsuchinhibitionMoreoverunpublisheddatafromourlaboratoryindicatethatMOTperformancede-creasesasnontargetsareaddedtothedisplayevenwhenthenumberoftargetsremainsconstant
Accordingly we designed Experiment7 to test thehypothesisthattrackingcapacitywouldincreaseiftherewerenonontargetsThisexperimentwasidenticaltoEx-periment2exceptthatallitemsweretargetsCapacityinthestandardconditionwas58items(SEM502)sig-nificantlygreaterthanthe26items(SEM501)obtainedinthespecificcondition[t(7)5143p000005]
InagreementwithourhypothesiscapacityimprovedmarkedlyinthestandardconditionwhentherewerenodistractorsStandardcapacityinExperiment7wassig-nificantlygreaterthaninExperiments23and4whichweremethodologicallycomparableexceptforthenumberoftargetsandnontargets(allps00005)
Oneexplanationisthatpreviousexperimentsunderes-timatedcapacitybecauseofaceilingeffectTheseexperi-mentswerebasedonthehypothesisthatcapacityvariesacrosstrialswithnounderlyingdifferencesacrossexperi-mentsHoweverthemeasuredcapacityhasbeenlimitedbythenumberoftargetsForexampleifthereareonlyfour
Unique Paired Identical0
1
2
3
4
Condition
k (it
ems)
Figure 4 Capacity as a function of condition in experiment 5
Specific Standard0
1
2
3
4Unique
Paired
Condition
k (it
ems)
Figure 5 effect of pairing target and nontarget objects on ca-pacity in experiment 6 data from the paired conditions are plot-ted as gray bars and those from the unique conditions as open bars
180 HOrOwiTz eT al
targets(egExperiments2ndash4)observerscouldtrackallofthemontrialsinwhichtheircapacitywas4orgreateranytrialswithanactualcapacitygreaterthan4wouldthusappeartohaveacapacityof4InthiscasemeasuredmeancapacitywouldunderestimatethetruemeanInExperi-ment7whichincludedeighttargetsmeasuredcapacitycouldreflectactualcapacityuptoeightitemsleadingtoagreater(andmoreaccurate)estimateofmeancapacity
Wecantestthisceilinghypothesisagainstourhypoth-esisofincreasedmeancapacitybycomparingthecapacitydistributionsinExperiment7withthoseinExperiments2ndash4Accordingtotheceilinghypothesistheproportionoftrialswithacapacitylessthan4shouldbeidenticalacrossexperimentsFurthermoretheproportionoftrialswithacapacityof4orgreaterinExperiment7shouldbethesameastheproportionoftrialswithacapacityequalto4inExperiments2ndash4Incontrastthehypothesisthattrackingcapacityincreaseswhennonontargetsarepres-entpredictsthattheentiredistributionshouldshifttotherightleadingtofewertrialswithcapacitylessthan4inExperiment7thanintheearlierexperiments
We tested these hypotheses by computing capacityforeachtrialusingEquation2andgeneratingadistri-butionacrosstrialsforeachobserverFigure6plotsthedistributionofkforExperiment7andthecorrespondingdistributionforExperiments23and4combined(dis-tributionsfromeachoftheseexperimentswereaveragedacrossobservers)Clearlytheproportionoftrialswithanobservedcapacitylessthan4waslowerwitheighttargetsthanwithfourtargetsThisisconsistentwiththecapacitydistributionshiftingtotherightinExperiment7ratherthanmerespreadingofthek54trialsoverawiderrangeWesuggestthatthesuperiorperformanceinthisexperi-mentmayreflectadditionalcapacityfreedupwhenthereisnoneedtosuppressnontargetsHowevertheshapeofthedistributionforfourtargetsdoessuggestthatcapacitywasunderestimatedinthepreviousexperiments
TheconclusionsarequitedifferentifwelookatthespecifictargetconditionSpecifictargetcapacityinthis
experimentwasbetterthaninExperiment2(uncorrectedttestp05)butnotsignificantlydifferentfromtheca-pacitiesobservedinExperiments3and4( p05)Thusthetrackingoftargetidentitiesinthisconditiondoesnotseem tobe limitedby theneed to suppressnontargetidentities
diSCuSSioN
WehavedevelopedanewMOTparadigmthatallowsustotestthetrackingofuniqueobjectsFourmainfind-ingshaveemergedfromtheseexperimentsFirstthereisacontent-addressablerepresentationoftargetsduringMOT(specificcapacitywas1inallcases)SecondweobservedacontentdeficitThecontent-addressablerep-resentationhasalowercapacity(measuredinitems)thanthelocation-addressablerepresentation(iespecificca-pacitywaslowerthanstandardinallcases)Thirdthevi-sualsystemiscapableoftakingadvantageofdifferencesbetweenuniqueobjectstoimprovetrackingperformanceoverthelevelseenwithidenticalobjects(Experiment5)Thisadvantageappearstobedueprimarilytotheabilitytorecoverlosttargets(Experiment5)Fourthtrackingcapacityappearstobeconstrainedbytheneedtosup-pressnontargetswhentherearenonontargetscapacityincreasesmarkedly(Experiment7)
A Content-Addressable representation in MoTOurdata represent anexistenceproofof a content-
addressable representation inMOTSimilar data havebeenreportedbyOksamaandHyoumlnauml(2004)Intheirmul-tiple-identity-tracking(MIT)taskobserverstrackedmov-inglinedrawings(eitherobjectsorldquopseudo-objectsrdquo)forseveralsecondsAttheendofthetrialthestimuliweremaskedandasingleitemwasmarkedwithablackframeInasubsequentresponsescreenobservershadtoselectthepicturethatcorrespondedtothemarkedobjectOksamaandHyoumlnaumlobtainedamediancapacityoffouritems
Althoughitisdifficulttocomparecapacitymeasuresdirectlyacrossexperimentaltasks3itisinterestingthatbothweandOksamaandHyoumlnauml(2004)demonstratedthatobserversdoknowtosomeextentwhichtargetiswhichThisfindingisincontrasttothoseofPylyshyn(2004)whoseobservershaddifficultyidentifyingtargetsTwokeydifferencesbetweenourparadigmandMITontheonehandandPylyshynrsquos(2004)paradigmontheotherseemrelevanttothisdiscrepancyFirstinourstudyandthatofOksamaandHyoumlnaumlobjectidentitiesweredefinedbyintrinsicfeaturessuchasshapeand(inourexperi-ment)colorInPylyshynrsquos(2004)experimentldquoidentityrdquowasbasedonatangentialfeatureofanobjectsuchasinwhichcornerofthedisplayitbeganthetrialItseemslikelythattheshapeandcolorofanobjectmaybemoretightlyboundmarkersofldquoidentityrdquothaninitialpositionorabrieflypresentedletter
SecondinourexperimentsandthoseofOksamaandHyoumlnauml (2004) thevisualdifferencesbetweenobjectswerecontinuallyavailablethroughoutthetrackingphaseofthetrialonlymaskedduringtheresponsephaseIn
4 targets8 targets
0 2 4 6 80
1
2
3
4
5
6
Capacity
Pro
po
rtio
n o
f Tri
als
Figure 6 Capacity distributions for the standard condition in experiment 7 (gray lines striped area) and in experiments 2 3 and 4 (black lines stippled area) dashed lines indicate standard errors of the means
Tracking UniqUe ObjecTs 181
Pylyshynrsquos(2004)experimentstheobjectidentitieswerepresentedonlybrieflyatthestartofthetrialduringthemajorityofthetrackingphasetheobjectswerephysicallyidenticalItmaybethatthelinkbetweenobjectidentityandspatiotemporalpositiondecayedduringPylyshynrsquos(2004)taskbutwasconstantlyrefreshedinbothoursandOksamaandHyoumlnaumlrsquos
Ourexperimentsalsoprovidedsomeindirectevidenceforacontent-addressablerepresentationForexampleperformancewasconsistentlysuperior inexperimentsinwhichthesameobjectswereusedinthesamerolesfromtrialtotrial(Experiments23and4)thaninex-perimentsinwhichthestimulussetchangedfromtrialtotrial(Experiments15and6)Thisoccurrednotonlyinthespecifictargetconditionsinwhichobjectidentitywasrelevantbutalsointhestandardconditionsinwhichobjectidentitywasnot
one System or TwoIstrackingservedbyasinglerepresentationthatencodes
bothpositionandidentityoraretheretworepresentationsoneforpositionandoneforidentityThisquestionarisesbecauseofthestrikingcapacitydifferencesbetweenourresponseconditionsInallofourexperimentscapacitywassignificantlylowerforspecifictargetsthanfortargetsinthestandardMOTresponsemodeInExperiment4wedisconfirmedthehypothesisthatcapacitydifferencesresultedfromvariedencodingstrategies(seeegDavisWelchHolmesampShepherd2001)sinceweobtainedthesameresultwhenresponsemodesweremixedwithinablockaswedidwhentheywereruninseparateblocks
ThiscapacitydifferenceisnotconclusiveevidencethatseparaterepresentationsorsystemsareinvolvedThereareseveralwaysinwhichaunifiedaccountcouldex-plainthelowercapacityinthespecifictargetconditionForexampleonemightproposeasystemwithaslot-likestructure(VogelWoodmanampLuck2001)inwhichallslotsarenotcreatedequalIfthereweretwoslotswithsuf-ficientresolutiontoencodebothpositionandidentityandtwoslotsthatcouldonlyencodepositionwewouldalsoobtaindifferentcapacityestimatesinourtworesponseconditionsThisrepresentationcouldalsobedescribedintermsofKahnemanandTreismanrsquos(1984)objectfilesifweassumethatsomeoftheobjectfilesareemptyTheimpletionprocessmightinthatcasereturntheinforma-tionthatagivenobjectwasspatiotemporallycontinuouswithatargetobjectbutfailtoretrievethesemanticorperceptualcontentassociatedwiththatobjectfile
IntheMOTliteraturecapacityisgenerallyassumedtorefertoafixednumberofavailablerepresentationsinthevisualsystemeithervisualindexesorobjectfilesHow-everthereissomeevidencethatcapacityinMOTmightbebetterthoughtofasanundifferentiatedresource(AlvarezampCavanagh2005)InsteadofasetofslotsorpointerswemightassumethatthereisafixedamountofinformationthatcanbeencodedObserversmightencodepositionandidentityforalltargetsbutthebindingbetweenpositionsandidentitymaybenoisierthanthepositioninformationitself(ormaydecaymorequicklyExperiment3indicatedthatanysuchdecaywouldhavetoapproachitsasymptote
valuewithin6ndash7sec)leadingtoapparentlylowercapacitywhenresponsesdependonbinding
InsteadofonesystemtherealsomightbetwoAlthoughwetendtoassumethatasingleexperimentalparadigmprobestheperformanceofasinglecorrespondingfunc-tionalsystemitmaybethatMOTnaturallyrecruitsmulti-pleoverlappingsystemsThesimplesttwo-systemaccountmightbeaproposalthattrackingisservedbyaposition-trackingsystemsuchasvisualindexes(FINSTs)butthataccessingidentityndashpositionbindingsrequiresfocalatten-tion(WheelerampTreisman2002)Howeversincespecificcapacityisreliablygreaterthanoneitemwewouldhavetoassumethatmultipleitemscansimultaneouslybethefocusofattention(AwhampPashler2000Davisetal2001)
IfwewishtopreserveasinglefocusofattentionwemightinsteadproposethatbothvisualindexesandobjectfilescanbeusedVisualindexeswouldprovideonlypo-sitioninformationandobjectfilescouldsupportmorecomplexqueries
Itwillobviouslybedifficulttodefinitivelydecidebe-tweenoneandtwosystemsHowevertheevidenceofourfinalthreeexperimentspointstotwoindependentsystemsInExperiments5and6pairingtargetsandnontargetsre-ducedcapacityinthestandardbutnotinthespecifictar-getcondition(althoughwecouldnotrejectthehypothesisofaproportionaldecrease)InExperiment7standardca-pacityincreasedwhennonontargetswerepresentedbutspecificcapacitywasunaffected
MOThasbeenlikenedtoavisualworkingmemorytaskattimezero(CavanaghampAlvarez2005)soitisnaturaltolooktostudiesofworkingmemoryforanalogsofthecontent-addressableandlocation-onlysystemsTheclassictwo-pathwayschemeinvisionisthedistinctionbetweenventralanddorsalstreamsassociatedwithobjectrecogni-tionandspatialprocessingrespectively(UngerleiderampMishkin1982)ThisdistinctionhasbeenextendedintothedomainofworkingmemorybyWilsonOacuteScalaidheandGoldman-Rakic(1993)whoprovidedevidencethatobjectinformationandspatialinformationareneurallysegregatedinmonkeyprefrontalcortexwidelybelievedtobetheneuralsubstrateofworkingmemoryAsimilardissociationinhumanswasreportedusingPETimagingbySmithetal(1995)
Thedivisionofworkingmemoryintomultiplesubsys-temsgoesbackatleasttothemultiple-componentmodelofBaddeleyandHitch(1974)whichproposedseparatecomponents for different modalities a ldquophonologicallooprdquoforauditoryinformationandaldquovisuospatialsketchpadrdquoforvisualinformationInmorerecentversionsofthemodel(BaddeleyampLogie1999Logie1995)thesketchpadhasbeendividedintovisualandspatialsubcompo-nents roughlycorrespondingto theobjectandspatialsystemsdescribedbyWilsonetal(1993)Interestinglythespatialsubcomponent(theldquoinnerscriberdquo)isspecifi-callytaskedwithholdingsequenceorpathinformation(seeParmentierElfordampMayberry2005)makingitalikelysubstrateforMOTperformance
Thewidespreadinsistenceonaseparationbetweenvi-sualorobjectinformationandspatialinformationwouldseem problematic for our findings Standard MOT is
182 HOrOwiTz eT al
clearlyajobforthespatialsubsystem(PostleDrsquoEspositoampCorkin2005forexampleusedaversionofstandardMOTtotieupdorsalprocessinginamemoryinterferenceparadigm)Howevertheredoesnotseemtoberoominthataccountforacontent-addressablerepresentationOurspecifictargettask(aswellastheMITtaskofOksamaampHyoumlnauml2004)isnotapureobjectrecognitiontaskitrequireslocationndashidentitybindingswhichwouldseemtoinvolveintegratinginformationacrossstreamsTheproblemdisappears ifwe take intoaccount recentbe-havioral(OlsonampMarshuetz2005)andfMRI(PostleampDrsquoEsposito1999)investigationsdemonstratingthatobjectworkingmemoryrepresentationsalsocarrysomelocationinformation
NeuroimagingstudiesmightproveusefulhereStan-dardMOThasbeenshowntoactivateparietallobeareasinthedorsalstream(Culhametal1998Jovicichetal2001Seiffert2003)Itwouldbeinformativetoknowwhetherthepatternofactivationchangeswhenuniqueob-jectsareintroducedorwhenthetaskpotentiallyinvolvesobjectidentities
unique object Features in MoTInadditiontoourfindingsaboutcontentaddressability
intrackingwewereabletodemonstratethathavingvisu-allyuniqueobjectsmadetrackingeasierobserverswereabletotrackroughly06moreitemswhenobjectswereuniquethanwhenobjectswereidenticalInExperiments5and6weattributedthisresulttoanerrorrecoverystrat-egyIfIlosetrackofthezebraIcansearchforthezebraonthebasisofitsphysicalfeaturesandcontinuetotrackitThisstrategywouldobviouslybeuselessifallitemsarezebrasThisstrategyisoflimitedutilityifthereareazebratargetandazebradistractoronceIlosethezebraIthenhaveonlya50chanceofpickingupthetargetzebraasopposedtothedistractor
TheerrorrecoveryhypothesisisnottrivialIthassev-eralimportantimplicationsFirstitimpliesthatobserversknowwhattheyarenottrackingThephenomenologyofMOTissuchthatanobserverhasthefeelingofknow-inghowmanytargetsheorsheissuccessfullytrackingTheerrorrecoveryhypothesisimpliesthatwhenatargetislosttheobserverknowswhichone(givenuniquetargets)SecondtheerrorrecoveryhypothesisalsoassumesthatanobservercansearchforamissingtargetwhilecontinuingtotracktheremainingtargetsThisisinagreementwithrecentevidencethatobserverscansuccessfullyperformanMOTtaskandavisualsearchtaskonthesametrial(AlvarezHorowitzArsenioDiMaseampWolfe2005)Finallythehypothesisimpliestheabilitytoaddtothetrackingsetontheflywhiletrackingwithoutanyphysi-calcueTheseimplicationsneedtobeconfirmedthroughfurtherresearch
Howeverifobserversuseuniquephysicalfeaturesofobjectstorecoverlosttargetsinthisfashionitiscuriousthatthereisnosuchbenefitinthespecifictargetcondi-tionIfobserversdorecoverlosttargetsduringthecourse
ofatrialtheyappeartorecoveronlytheirlocationsnottheiridentities
AnalternativetotheerrorrecoveryhypothesisthoughnotamutuallyexclusiveoneisthattheresultsofEx-periment5canbeexplainedbytheperceptualsimilarityoftargetandnontargetitemsYantisrsquos(1992)approachtoMOThighlightstheimportanceofperceptualgroup-ingfactorsthatallowthetargetstobetreatedasaunitbythevisualsystemMostoftheresearchinthislinehasfocusedonspatiotemporalgroupingfactorssuchasthespatialrelationshipsbetweentheobjects(SearsampPylyshyn2000ViswanathanampMingolla2002Yantis1992)
Cangroupingby(nonspatial)featuralpropertiesim-provetrackingAnumberofstudieshavedemonstratedthatobservershave limitedexplicit access to featuralproperties (Bahrami 2003Saiki 2002Scholl etal1999)butthisfindingdoesnotaddressthedirectrolethatfeaturesmayplayWecouldexplaintheresultsofEx-periment5byproposingthatobserverswereabletouseaccidentaldifferencesintheshapeandcolorstatisticsoftargetsandnontargetstohelpsegregatethemvisuallyInthepairedconditiontargetsandnontargetshadidenticalfeaturestatisticssothatadvantagewaslostThishypoth-esiscouldalsoexplainwhyperformancewasbetterwithafixedstimulussetthanwithavariablestimulussetoveranumberoftrialswiththesamestimuliobserverslearnsubtledifferencesbetweenthetwosets
ConclusionsIntheintroductionwenotedthatreal-worlddynamic
attentiontasksdifferfromtraditionalMOTintwowaysFirstobserversintheeverydayworldusuallydealwithmultipleuniqueobjectsSecondtheyoftenwishtoknowwhereagiventargetmightberatherthanwhichobjectsarethetargetsTheexperimentspresentedhererepresentanattempttobridgelaboratoryMOTresearchandtheseeverydaycognitivetasksWehaveshownthatitiseasiertotrackuniqueobjectsthantotrackidenticalobjectsassum-ingthattheobjectsonewishestotrackarealsouniquelydifferentfromnontargetobjectsFurthermoreobserversdoknowwhattheyaretrackingTotakeoneofourex-amplesofecologicallyvalidtrackingtasksifyouwatchyourchildrenandtheirfriendsontheplaygroundyoucanbeawareofwhereyourkidsareatanygivenmomentyoumayalsobeawareofthelocationsoftheirfriendswithoutbeingabletotellonefriendfromtheotherFurthermoreifyoulosetrackofachildyouwillprobablybeawareofwhichoneyouhavelost
AuTHor NoTe
ThisresearchwassupportedbyNIMHGrantR0165576toTSHWethankKristinMichodandSkylerPlaceforassistancewithdatacollec-tionaswellasSteveLuckandananonymousreviewerforconstructivecriticismCorrespondencerelatingtothisarticlemaybesenttoTSHorowitzBrighamandWomenrsquosHospitalVisualAttentionLaboratory64SidneyStreetSuite170BostonMA02139(e-mailtoddhsearchbwhharvardedu)
Tracking UniqUe ObjecTs 183
reFereNCeS
Allen R McGeorge P Pearson D amp Milne A B (2004)AttentionandexpertiseinmultipletargettrackingApplied Cognitive Psychology18337-347
Alvarez G A amp Cavanagh P (2005)IndependentresourcesforattentionaltrackingintheleftandrightvisualhemifieldsPsychologi-cal Science16637-643
Alvarez G A amp Franconeri S (2004November)How many ob-jects can you trackPaperpresentedatthe12thAnnualWorkshoponObjectPerceptionampMemoryMinneapolisMN
Alvarez G A Horowitz T S Arsenio H C DiMase J S amp Wolfe J M (2005)DomultielementvisualtrackingandvisualsearchdrawcontinuouslyonthesamevisualattentionresourcesJournal of Experimental Psychology Human Perception amp Perfor-mance31643-667
Awh E Jonides J amp Reuter-Lorenz P A (1998)RehearsalinspatialworkingmemoryJournal of Experimental Psychology Human Perception amp Performance24780-790
Awh E amp Pashler H (2000)EvidenceforsplitattentionalfociJournal of Experimental Psychology Human Perception amp Perfor-mance26834-846
Baddeley A D amp Hitch G J (1974)WorkingmemoryInGHBower(Ed)The psychology of learning and motivation Advances in research and theory(Vol8pp47-90)NewYorkAcademicPress
Baddeley A D amp Logie R H (1999) Working memoryThemultiple-componentmodelInAMiyakeampPShah(Eds)Models of working memory Mechanisms of active maintenance and executive control(pp28-61)CambridgeCambridgeUniversityPress
Bahrami B (2003)ObjectpropertyencodingandchangeblindnessinmultipleobjecttrackingVisual Cognition10949-963
Brainard D H (1997)ThePsychophysicsToolboxSpatial Vision10433-436
Cavanagh P amp Alvarez G A (2005)TrackingmultipletargetswithmultifocalattentionTrends in Cognitive Sciences9349-354
Culham J C Brandt S A Cavanagh P Kanwisher N G Dale A M amp Tootell R B H (1998)CorticalfMRIactiva-tionproducedbyattentive trackingofmoving targetsJournal of Neurophysiology802657-2670
Davis G Welch V L Holmes A amp Shepherd A (2001)CanattentionselectonlyafixednumberofobjectsatatimePerception301227-1248
Henderson J M amp Hollingworth A (2003)EyemovementsandvisualmemoryDetectingchangestosaccadetargetsinscenesPer-ception amp Psychophysics6558-71
Horowitz T S Birnkrant R S Fencsik D E Tran L amp Wolfe J M (2006)HowdowetrackinvisibleobjectsPsychonomic Bulletin amp Review13516-523
Jovicich J Peters R J Koch C Braun J Chang L amp Ernst T (2001)Brainareasspecificforattentionalloadinamotion-trackingtaskJournal of Cognitive Neuroscience131048-1058
Kahneman D amp Treisman A (1984)ChangingviewsofattentionandautomaticityInRParasuramanampDRDavies(Eds)Varieties of attention(pp29-61)OrlandoAcademicPress
Kahneman D Treisman A amp Gibbs B J (1992)ThereviewingofobjectfilesObject-specificintegrationofinformationCognitive Psychology24175-219
Klieger S B Horowitz T S amp Wolfe J M (2004)Ismultipleobjecttrackingcolorblind[Abstract]Journal of Vision4(8)363a
Leonard C amp Pylyshyn Z W (2003)Measuringtheattentionaldemandofmultipleobjecttracking(MOT)[Abstract]Journal of Vi-sion3(9)582a
Logie R H (1995) Visuo-spatial working memory Hove UKErlbaum
Milliken B Tipper S P amp Weaver B (1994)NegativepriminginaspatiallocalizationtaskFeaturemismatchinganddistractorin-hibitionJournal of Experimental Psychology Human Perception amp Performance20624-646
Mitroff S R Scholl B J amp Wynn K (2004)DivideandconquerHowobjectfilesadaptwhenapersistingobjectsplitsintotwoPsy-chological Science15420-425
Ogawa H amp Yagi A (2003)Primingeffectsinmultipleobjecttrack-ingAnimplicitencodingbasedonglobalspatiotemporalinformation[Abstract]Journal of Vision3(9)339a
Oksama L amp Hyoumlnauml J (2004)Ismultipleobjecttrackingcarriedoutautomaticallybyanearlyvisionmechanismindependentofhigher-ordercognitionAnindividualdifferenceapproachVisual Cognition11631-671
Oliva A Torralba A Castelhano M S amp Henderson J M (2003)Top-downcontrolofvisualattentioninrealworldscenes[Ab-stract]Journal of Vision3(9)3a
Olson I R amp Marshuetz C (2005)RememberingldquowhatrdquobringsalongldquowhererdquoinvisualworkingmemoryPerception amp Psychophysics67185-194
Parasuraman R (1986)VigilancemonitoringandsearchInKRBoffLKaufmanampJPThomas(Eds)Handbook of perception and human performance Vol 2 Cognitive processes and performance(pp1-39)NewYorkWiley
Parmentier F B R Elford G amp Mayberry M (2005)Transi-tionalinformationinspatialserialmemoryPathcharacteristicsaffectrecallperformanceJournal of Experimental Psychology Learning Memory amp Cognition31412-427
Pelli D G (1997)TheVideoToolboxsoftwareforvisualpsychophysicsTransformingnumbersintomoviesSpatial Vision10437-442
Postle B R amp DrsquoEsposito M (1999)ldquoWhatrdquondashthenndashldquowhererdquoinvi-sualworkingmemoryAnevent-relatedfMRIstudyJournal of Cog-nitive Neuroscience11585-597
Postle B R DrsquoEsposito M amp Corkin S (2005)Effectsofver-balandnonverbalinterferenceonspatialandobjectvisualworkingmemoryMemory amp Cognition33203-212
Pylyshyn Z [W] (1989)Theroleoflocationindexesinspatialper-ceptionAsketchoftheFINSTspatial-indexmodelCognition3265-97
Pylyshyn Z W (2004)SomepuzzlingfindingsinmultipleobjecttrackingITrackingwithoutkeepingtrackofobjectidentitiesVisual Cognition11801-822
Pylyshyn Z W amp Leonard C (2003)Inhibitionofnontargetsdur-ingmultipleobjecttracking(MOT)[Abstract]Journal of Vision3(9)585a
Pylyshyn Z W amp Storm R W (1988)Trackingmultipleindepen-denttargetsEvidenceforaparalleltrackingmechanismSpatial Vi-sion3179-197
Saiki J (2002)Multiple-objectpermanencetrackingLimitationinmaintenanceandtransformationofperceptualobjectsProgress in Brain Research140133-148
Schneider W amp Shiffrin R M (1977)ControlledandautomatichumaninformationprocessingIDetectionsearchandattentionPsychological Review841-66
Scholl B J amp Pylyshyn Z W (1999)Trackingmultiple itemsthroughocclusionCluestovisualobjecthoodCognitive Psychology38259-290
Scholl B J Pylyshyn Z W amp Feldman J (2001)Whatisavi-sualobjectEvidencefromtargetmerginginmultipleobjecttrackingCognition80159-177
Scholl B J Pylyshyn Z W amp Franconeri S L (1999May)When are spatiotemporal and featural properties encoded as a result of attentional allocationPaperpresentedattheconferenceoftheAssociationforResearchinVisionandOphthalmologyFortLau-derdaleFL
Sears C R amp Pylyshyn Z W (2000)MultipleobjecttrackingandattentionalprocessingCanadian Journal of Experimental Psychol-ogy541-14
Seiffert A E (2003)Dissociatingneuralcorrelatesofattentionaltrackingandattentiontovisualmotion[Abstract]Journal of Vision3(9)868a
Smith E E Jonides J Koeppe R A Awh E Schumacher E H amp Minoshima S (1995)SpatialversusobjectworkingmemoryPETinvestigationsJournal of Cognitive Neuroscience7337-356
Tipper S P (1985)ThenegativeprimingeffectInhibitoryprimingbyignoredobjectsQuarterly Journal of Experimental Psychology37A571-590
Ungerleider L G amp Mishkin M (1982)Twocorticalvisualsys-
184 HOrOwiTz eT al
temsInDJIngleMAGoodaleampRJWMansfield(Eds)Anal-ysis of visual behavior(pp549-586)CambridgeMAMITPress
Viswanathan L amp Mingolla E (2002)DynamicsofattentionindepthEvidence frommulti-element trackingPerception311415-1437
Vogel E K Woodman G F amp Luck S J (2001)Storageoffea-turesconjunctionsandobjectsinvisualworkingmemoryJournal of Experimental Psychology Human Perception amp Performance2792-114
Wheeler M E amp Treisman A M (2002)Bindinginshort-termvisualmemoryJournal of Experimental Psychology General13148-64
Wilson F A Oacute Scalaidhe S P amp Goldman-Rakic P S (1993)Dissociationofobjectandspatialprocessingdomains inprimateprefrontalcortexScience2601955-1958
Wolfe J M Oliva A Horowitz T S Butcher S J amp Bompas A (2002)SegmentationofobjectsfrombackgroundsinvisualsearchtasksVision Research422985-3004
Yantis S (1992)MultielementvisualtrackingAttentionandpercep-tualorganizationCognitive Psychology24295-340
NoTeS
1Wearemissingageinformationfor1observer2Threeobserverswhoparticipatedinbothexperimentswereelimi-
natedfromthisanalysis3Measuredcapacitycanvarybasedontrackingdurationtherange
ofmovementspeedsandwhetherornotobjectscanoccludeonean-other(asinourexperiment)orbounceoffoneanother(asinOksamaampHyoumlnauml2004)
APPeNdix A Stimulus List
beaver fox horse ostrich rhino turtlecamel giraffe kangaroo pelican rooster wolfcrocodile goat lion rabbit squirrel zebraelephant gorilla moose tiger
APPeNdix b raw data
Table b1 raw Numbers of items reported in the Standard Condition
Total Targets Nontargets Filled
Experiment1 308Experiment2 344 014 359Experiment3 347 015 362Experiment4 339 015 354Experiment5 Unique 312 037 349
Paired 249 080 329Identical 261 086 347
Experiment6 248 078 326Experiment7 599 599
NotemdashTargetsindicatestheaveragenumberofcorrectlyreportedtargetspertrialNontargetstheaveragenumberoftimesobserversclickedonanontargetlocationpertrialandTotal Filledthesumofthesetwocat-egoriesNontargetclickcountswerenotcollectedforExperiment1andtherewerenonontargetsinExperiment7
Table b2 raw Proportions for the Specific Condition by experiment
Errors
Correct Empty Trialsper Responses Transposition Nontarget Space N Observer
Experiment1 44 14 16 26 8 5000Experiment2 56 18 10 16 8 5000Experiment3 68 12 10 10 8 5000Experiment4 58 25 09 08 8 5063Experiment6 41 14 21 23 7 5000Experiment7 33 39 00 28 8 5000
NotemdashTranspositionerrorsrefertotrialsinwhichtheobserverclickedonthewrongtargetlocationTrialsperobservervariedinExperiment4becauseconditionwasselectedrandomlyfromtrialtotrial
(ManuscriptreceivedMarch312005revisionacceptedforpublicationFebruary132006)
180 HOrOwiTz eT al
targets(egExperiments2ndash4)observerscouldtrackallofthemontrialsinwhichtheircapacitywas4orgreateranytrialswithanactualcapacitygreaterthan4wouldthusappeartohaveacapacityof4InthiscasemeasuredmeancapacitywouldunderestimatethetruemeanInExperi-ment7whichincludedeighttargetsmeasuredcapacitycouldreflectactualcapacityuptoeightitemsleadingtoagreater(andmoreaccurate)estimateofmeancapacity
Wecantestthisceilinghypothesisagainstourhypoth-esisofincreasedmeancapacitybycomparingthecapacitydistributionsinExperiment7withthoseinExperiments2ndash4Accordingtotheceilinghypothesistheproportionoftrialswithacapacitylessthan4shouldbeidenticalacrossexperimentsFurthermoretheproportionoftrialswithacapacityof4orgreaterinExperiment7shouldbethesameastheproportionoftrialswithacapacityequalto4inExperiments2ndash4Incontrastthehypothesisthattrackingcapacityincreaseswhennonontargetsarepres-entpredictsthattheentiredistributionshouldshifttotherightleadingtofewertrialswithcapacitylessthan4inExperiment7thanintheearlierexperiments
We tested these hypotheses by computing capacityforeachtrialusingEquation2andgeneratingadistri-butionacrosstrialsforeachobserverFigure6plotsthedistributionofkforExperiment7andthecorrespondingdistributionforExperiments23and4combined(dis-tributionsfromeachoftheseexperimentswereaveragedacrossobservers)Clearlytheproportionoftrialswithanobservedcapacitylessthan4waslowerwitheighttargetsthanwithfourtargetsThisisconsistentwiththecapacitydistributionshiftingtotherightinExperiment7ratherthanmerespreadingofthek54trialsoverawiderrangeWesuggestthatthesuperiorperformanceinthisexperi-mentmayreflectadditionalcapacityfreedupwhenthereisnoneedtosuppressnontargetsHowevertheshapeofthedistributionforfourtargetsdoessuggestthatcapacitywasunderestimatedinthepreviousexperiments
TheconclusionsarequitedifferentifwelookatthespecifictargetconditionSpecifictargetcapacityinthis
experimentwasbetterthaninExperiment2(uncorrectedttestp05)butnotsignificantlydifferentfromtheca-pacitiesobservedinExperiments3and4( p05)Thusthetrackingoftargetidentitiesinthisconditiondoesnotseem tobe limitedby theneed to suppressnontargetidentities
diSCuSSioN
WehavedevelopedanewMOTparadigmthatallowsustotestthetrackingofuniqueobjectsFourmainfind-ingshaveemergedfromtheseexperimentsFirstthereisacontent-addressablerepresentationoftargetsduringMOT(specificcapacitywas1inallcases)SecondweobservedacontentdeficitThecontent-addressablerep-resentationhasalowercapacity(measuredinitems)thanthelocation-addressablerepresentation(iespecificca-pacitywaslowerthanstandardinallcases)Thirdthevi-sualsystemiscapableoftakingadvantageofdifferencesbetweenuniqueobjectstoimprovetrackingperformanceoverthelevelseenwithidenticalobjects(Experiment5)Thisadvantageappearstobedueprimarilytotheabilitytorecoverlosttargets(Experiment5)Fourthtrackingcapacityappearstobeconstrainedbytheneedtosup-pressnontargetswhentherearenonontargetscapacityincreasesmarkedly(Experiment7)
A Content-Addressable representation in MoTOurdata represent anexistenceproofof a content-
addressable representation inMOTSimilar data havebeenreportedbyOksamaandHyoumlnauml(2004)Intheirmul-tiple-identity-tracking(MIT)taskobserverstrackedmov-inglinedrawings(eitherobjectsorldquopseudo-objectsrdquo)forseveralsecondsAttheendofthetrialthestimuliweremaskedandasingleitemwasmarkedwithablackframeInasubsequentresponsescreenobservershadtoselectthepicturethatcorrespondedtothemarkedobjectOksamaandHyoumlnaumlobtainedamediancapacityoffouritems
Althoughitisdifficulttocomparecapacitymeasuresdirectlyacrossexperimentaltasks3itisinterestingthatbothweandOksamaandHyoumlnauml(2004)demonstratedthatobserversdoknowtosomeextentwhichtargetiswhichThisfindingisincontrasttothoseofPylyshyn(2004)whoseobservershaddifficultyidentifyingtargetsTwokeydifferencesbetweenourparadigmandMITontheonehandandPylyshynrsquos(2004)paradigmontheotherseemrelevanttothisdiscrepancyFirstinourstudyandthatofOksamaandHyoumlnaumlobjectidentitiesweredefinedbyintrinsicfeaturessuchasshapeand(inourexperi-ment)colorInPylyshynrsquos(2004)experimentldquoidentityrdquowasbasedonatangentialfeatureofanobjectsuchasinwhichcornerofthedisplayitbeganthetrialItseemslikelythattheshapeandcolorofanobjectmaybemoretightlyboundmarkersofldquoidentityrdquothaninitialpositionorabrieflypresentedletter
SecondinourexperimentsandthoseofOksamaandHyoumlnauml (2004) thevisualdifferencesbetweenobjectswerecontinuallyavailablethroughoutthetrackingphaseofthetrialonlymaskedduringtheresponsephaseIn
4 targets8 targets
0 2 4 6 80
1
2
3
4
5
6
Capacity
Pro
po
rtio
n o
f Tri
als
Figure 6 Capacity distributions for the standard condition in experiment 7 (gray lines striped area) and in experiments 2 3 and 4 (black lines stippled area) dashed lines indicate standard errors of the means
Tracking UniqUe ObjecTs 181
Pylyshynrsquos(2004)experimentstheobjectidentitieswerepresentedonlybrieflyatthestartofthetrialduringthemajorityofthetrackingphasetheobjectswerephysicallyidenticalItmaybethatthelinkbetweenobjectidentityandspatiotemporalpositiondecayedduringPylyshynrsquos(2004)taskbutwasconstantlyrefreshedinbothoursandOksamaandHyoumlnaumlrsquos
Ourexperimentsalsoprovidedsomeindirectevidenceforacontent-addressablerepresentationForexampleperformancewasconsistentlysuperior inexperimentsinwhichthesameobjectswereusedinthesamerolesfromtrialtotrial(Experiments23and4)thaninex-perimentsinwhichthestimulussetchangedfromtrialtotrial(Experiments15and6)Thisoccurrednotonlyinthespecifictargetconditionsinwhichobjectidentitywasrelevantbutalsointhestandardconditionsinwhichobjectidentitywasnot
one System or TwoIstrackingservedbyasinglerepresentationthatencodes
bothpositionandidentityoraretheretworepresentationsoneforpositionandoneforidentityThisquestionarisesbecauseofthestrikingcapacitydifferencesbetweenourresponseconditionsInallofourexperimentscapacitywassignificantlylowerforspecifictargetsthanfortargetsinthestandardMOTresponsemodeInExperiment4wedisconfirmedthehypothesisthatcapacitydifferencesresultedfromvariedencodingstrategies(seeegDavisWelchHolmesampShepherd2001)sinceweobtainedthesameresultwhenresponsemodesweremixedwithinablockaswedidwhentheywereruninseparateblocks
ThiscapacitydifferenceisnotconclusiveevidencethatseparaterepresentationsorsystemsareinvolvedThereareseveralwaysinwhichaunifiedaccountcouldex-plainthelowercapacityinthespecifictargetconditionForexampleonemightproposeasystemwithaslot-likestructure(VogelWoodmanampLuck2001)inwhichallslotsarenotcreatedequalIfthereweretwoslotswithsuf-ficientresolutiontoencodebothpositionandidentityandtwoslotsthatcouldonlyencodepositionwewouldalsoobtaindifferentcapacityestimatesinourtworesponseconditionsThisrepresentationcouldalsobedescribedintermsofKahnemanandTreismanrsquos(1984)objectfilesifweassumethatsomeoftheobjectfilesareemptyTheimpletionprocessmightinthatcasereturntheinforma-tionthatagivenobjectwasspatiotemporallycontinuouswithatargetobjectbutfailtoretrievethesemanticorperceptualcontentassociatedwiththatobjectfile
IntheMOTliteraturecapacityisgenerallyassumedtorefertoafixednumberofavailablerepresentationsinthevisualsystemeithervisualindexesorobjectfilesHow-everthereissomeevidencethatcapacityinMOTmightbebetterthoughtofasanundifferentiatedresource(AlvarezampCavanagh2005)InsteadofasetofslotsorpointerswemightassumethatthereisafixedamountofinformationthatcanbeencodedObserversmightencodepositionandidentityforalltargetsbutthebindingbetweenpositionsandidentitymaybenoisierthanthepositioninformationitself(ormaydecaymorequicklyExperiment3indicatedthatanysuchdecaywouldhavetoapproachitsasymptote
valuewithin6ndash7sec)leadingtoapparentlylowercapacitywhenresponsesdependonbinding
InsteadofonesystemtherealsomightbetwoAlthoughwetendtoassumethatasingleexperimentalparadigmprobestheperformanceofasinglecorrespondingfunc-tionalsystemitmaybethatMOTnaturallyrecruitsmulti-pleoverlappingsystemsThesimplesttwo-systemaccountmightbeaproposalthattrackingisservedbyaposition-trackingsystemsuchasvisualindexes(FINSTs)butthataccessingidentityndashpositionbindingsrequiresfocalatten-tion(WheelerampTreisman2002)Howeversincespecificcapacityisreliablygreaterthanoneitemwewouldhavetoassumethatmultipleitemscansimultaneouslybethefocusofattention(AwhampPashler2000Davisetal2001)
IfwewishtopreserveasinglefocusofattentionwemightinsteadproposethatbothvisualindexesandobjectfilescanbeusedVisualindexeswouldprovideonlypo-sitioninformationandobjectfilescouldsupportmorecomplexqueries
Itwillobviouslybedifficulttodefinitivelydecidebe-tweenoneandtwosystemsHowevertheevidenceofourfinalthreeexperimentspointstotwoindependentsystemsInExperiments5and6pairingtargetsandnontargetsre-ducedcapacityinthestandardbutnotinthespecifictar-getcondition(althoughwecouldnotrejectthehypothesisofaproportionaldecrease)InExperiment7standardca-pacityincreasedwhennonontargetswerepresentedbutspecificcapacitywasunaffected
MOThasbeenlikenedtoavisualworkingmemorytaskattimezero(CavanaghampAlvarez2005)soitisnaturaltolooktostudiesofworkingmemoryforanalogsofthecontent-addressableandlocation-onlysystemsTheclassictwo-pathwayschemeinvisionisthedistinctionbetweenventralanddorsalstreamsassociatedwithobjectrecogni-tionandspatialprocessingrespectively(UngerleiderampMishkin1982)ThisdistinctionhasbeenextendedintothedomainofworkingmemorybyWilsonOacuteScalaidheandGoldman-Rakic(1993)whoprovidedevidencethatobjectinformationandspatialinformationareneurallysegregatedinmonkeyprefrontalcortexwidelybelievedtobetheneuralsubstrateofworkingmemoryAsimilardissociationinhumanswasreportedusingPETimagingbySmithetal(1995)
Thedivisionofworkingmemoryintomultiplesubsys-temsgoesbackatleasttothemultiple-componentmodelofBaddeleyandHitch(1974)whichproposedseparatecomponents for different modalities a ldquophonologicallooprdquoforauditoryinformationandaldquovisuospatialsketchpadrdquoforvisualinformationInmorerecentversionsofthemodel(BaddeleyampLogie1999Logie1995)thesketchpadhasbeendividedintovisualandspatialsubcompo-nents roughlycorrespondingto theobjectandspatialsystemsdescribedbyWilsonetal(1993)Interestinglythespatialsubcomponent(theldquoinnerscriberdquo)isspecifi-callytaskedwithholdingsequenceorpathinformation(seeParmentierElfordampMayberry2005)makingitalikelysubstrateforMOTperformance
Thewidespreadinsistenceonaseparationbetweenvi-sualorobjectinformationandspatialinformationwouldseem problematic for our findings Standard MOT is
182 HOrOwiTz eT al
clearlyajobforthespatialsubsystem(PostleDrsquoEspositoampCorkin2005forexampleusedaversionofstandardMOTtotieupdorsalprocessinginamemoryinterferenceparadigm)Howevertheredoesnotseemtoberoominthataccountforacontent-addressablerepresentationOurspecifictargettask(aswellastheMITtaskofOksamaampHyoumlnauml2004)isnotapureobjectrecognitiontaskitrequireslocationndashidentitybindingswhichwouldseemtoinvolveintegratinginformationacrossstreamsTheproblemdisappears ifwe take intoaccount recentbe-havioral(OlsonampMarshuetz2005)andfMRI(PostleampDrsquoEsposito1999)investigationsdemonstratingthatobjectworkingmemoryrepresentationsalsocarrysomelocationinformation
NeuroimagingstudiesmightproveusefulhereStan-dardMOThasbeenshowntoactivateparietallobeareasinthedorsalstream(Culhametal1998Jovicichetal2001Seiffert2003)Itwouldbeinformativetoknowwhetherthepatternofactivationchangeswhenuniqueob-jectsareintroducedorwhenthetaskpotentiallyinvolvesobjectidentities
unique object Features in MoTInadditiontoourfindingsaboutcontentaddressability
intrackingwewereabletodemonstratethathavingvisu-allyuniqueobjectsmadetrackingeasierobserverswereabletotrackroughly06moreitemswhenobjectswereuniquethanwhenobjectswereidenticalInExperiments5and6weattributedthisresulttoanerrorrecoverystrat-egyIfIlosetrackofthezebraIcansearchforthezebraonthebasisofitsphysicalfeaturesandcontinuetotrackitThisstrategywouldobviouslybeuselessifallitemsarezebrasThisstrategyisoflimitedutilityifthereareazebratargetandazebradistractoronceIlosethezebraIthenhaveonlya50chanceofpickingupthetargetzebraasopposedtothedistractor
TheerrorrecoveryhypothesisisnottrivialIthassev-eralimportantimplicationsFirstitimpliesthatobserversknowwhattheyarenottrackingThephenomenologyofMOTissuchthatanobserverhasthefeelingofknow-inghowmanytargetsheorsheissuccessfullytrackingTheerrorrecoveryhypothesisimpliesthatwhenatargetislosttheobserverknowswhichone(givenuniquetargets)SecondtheerrorrecoveryhypothesisalsoassumesthatanobservercansearchforamissingtargetwhilecontinuingtotracktheremainingtargetsThisisinagreementwithrecentevidencethatobserverscansuccessfullyperformanMOTtaskandavisualsearchtaskonthesametrial(AlvarezHorowitzArsenioDiMaseampWolfe2005)Finallythehypothesisimpliestheabilitytoaddtothetrackingsetontheflywhiletrackingwithoutanyphysi-calcueTheseimplicationsneedtobeconfirmedthroughfurtherresearch
Howeverifobserversuseuniquephysicalfeaturesofobjectstorecoverlosttargetsinthisfashionitiscuriousthatthereisnosuchbenefitinthespecifictargetcondi-tionIfobserversdorecoverlosttargetsduringthecourse
ofatrialtheyappeartorecoveronlytheirlocationsnottheiridentities
AnalternativetotheerrorrecoveryhypothesisthoughnotamutuallyexclusiveoneisthattheresultsofEx-periment5canbeexplainedbytheperceptualsimilarityoftargetandnontargetitemsYantisrsquos(1992)approachtoMOThighlightstheimportanceofperceptualgroup-ingfactorsthatallowthetargetstobetreatedasaunitbythevisualsystemMostoftheresearchinthislinehasfocusedonspatiotemporalgroupingfactorssuchasthespatialrelationshipsbetweentheobjects(SearsampPylyshyn2000ViswanathanampMingolla2002Yantis1992)
Cangroupingby(nonspatial)featuralpropertiesim-provetrackingAnumberofstudieshavedemonstratedthatobservershave limitedexplicit access to featuralproperties (Bahrami 2003Saiki 2002Scholl etal1999)butthisfindingdoesnotaddressthedirectrolethatfeaturesmayplayWecouldexplaintheresultsofEx-periment5byproposingthatobserverswereabletouseaccidentaldifferencesintheshapeandcolorstatisticsoftargetsandnontargetstohelpsegregatethemvisuallyInthepairedconditiontargetsandnontargetshadidenticalfeaturestatisticssothatadvantagewaslostThishypoth-esiscouldalsoexplainwhyperformancewasbetterwithafixedstimulussetthanwithavariablestimulussetoveranumberoftrialswiththesamestimuliobserverslearnsubtledifferencesbetweenthetwosets
ConclusionsIntheintroductionwenotedthatreal-worlddynamic
attentiontasksdifferfromtraditionalMOTintwowaysFirstobserversintheeverydayworldusuallydealwithmultipleuniqueobjectsSecondtheyoftenwishtoknowwhereagiventargetmightberatherthanwhichobjectsarethetargetsTheexperimentspresentedhererepresentanattempttobridgelaboratoryMOTresearchandtheseeverydaycognitivetasksWehaveshownthatitiseasiertotrackuniqueobjectsthantotrackidenticalobjectsassum-ingthattheobjectsonewishestotrackarealsouniquelydifferentfromnontargetobjectsFurthermoreobserversdoknowwhattheyaretrackingTotakeoneofourex-amplesofecologicallyvalidtrackingtasksifyouwatchyourchildrenandtheirfriendsontheplaygroundyoucanbeawareofwhereyourkidsareatanygivenmomentyoumayalsobeawareofthelocationsoftheirfriendswithoutbeingabletotellonefriendfromtheotherFurthermoreifyoulosetrackofachildyouwillprobablybeawareofwhichoneyouhavelost
AuTHor NoTe
ThisresearchwassupportedbyNIMHGrantR0165576toTSHWethankKristinMichodandSkylerPlaceforassistancewithdatacollec-tionaswellasSteveLuckandananonymousreviewerforconstructivecriticismCorrespondencerelatingtothisarticlemaybesenttoTSHorowitzBrighamandWomenrsquosHospitalVisualAttentionLaboratory64SidneyStreetSuite170BostonMA02139(e-mailtoddhsearchbwhharvardedu)
Tracking UniqUe ObjecTs 183
reFereNCeS
Allen R McGeorge P Pearson D amp Milne A B (2004)AttentionandexpertiseinmultipletargettrackingApplied Cognitive Psychology18337-347
Alvarez G A amp Cavanagh P (2005)IndependentresourcesforattentionaltrackingintheleftandrightvisualhemifieldsPsychologi-cal Science16637-643
Alvarez G A amp Franconeri S (2004November)How many ob-jects can you trackPaperpresentedatthe12thAnnualWorkshoponObjectPerceptionampMemoryMinneapolisMN
Alvarez G A Horowitz T S Arsenio H C DiMase J S amp Wolfe J M (2005)DomultielementvisualtrackingandvisualsearchdrawcontinuouslyonthesamevisualattentionresourcesJournal of Experimental Psychology Human Perception amp Perfor-mance31643-667
Awh E Jonides J amp Reuter-Lorenz P A (1998)RehearsalinspatialworkingmemoryJournal of Experimental Psychology Human Perception amp Performance24780-790
Awh E amp Pashler H (2000)EvidenceforsplitattentionalfociJournal of Experimental Psychology Human Perception amp Perfor-mance26834-846
Baddeley A D amp Hitch G J (1974)WorkingmemoryInGHBower(Ed)The psychology of learning and motivation Advances in research and theory(Vol8pp47-90)NewYorkAcademicPress
Baddeley A D amp Logie R H (1999) Working memoryThemultiple-componentmodelInAMiyakeampPShah(Eds)Models of working memory Mechanisms of active maintenance and executive control(pp28-61)CambridgeCambridgeUniversityPress
Bahrami B (2003)ObjectpropertyencodingandchangeblindnessinmultipleobjecttrackingVisual Cognition10949-963
Brainard D H (1997)ThePsychophysicsToolboxSpatial Vision10433-436
Cavanagh P amp Alvarez G A (2005)TrackingmultipletargetswithmultifocalattentionTrends in Cognitive Sciences9349-354
Culham J C Brandt S A Cavanagh P Kanwisher N G Dale A M amp Tootell R B H (1998)CorticalfMRIactiva-tionproducedbyattentive trackingofmoving targetsJournal of Neurophysiology802657-2670
Davis G Welch V L Holmes A amp Shepherd A (2001)CanattentionselectonlyafixednumberofobjectsatatimePerception301227-1248
Henderson J M amp Hollingworth A (2003)EyemovementsandvisualmemoryDetectingchangestosaccadetargetsinscenesPer-ception amp Psychophysics6558-71
Horowitz T S Birnkrant R S Fencsik D E Tran L amp Wolfe J M (2006)HowdowetrackinvisibleobjectsPsychonomic Bulletin amp Review13516-523
Jovicich J Peters R J Koch C Braun J Chang L amp Ernst T (2001)Brainareasspecificforattentionalloadinamotion-trackingtaskJournal of Cognitive Neuroscience131048-1058
Kahneman D amp Treisman A (1984)ChangingviewsofattentionandautomaticityInRParasuramanampDRDavies(Eds)Varieties of attention(pp29-61)OrlandoAcademicPress
Kahneman D Treisman A amp Gibbs B J (1992)ThereviewingofobjectfilesObject-specificintegrationofinformationCognitive Psychology24175-219
Klieger S B Horowitz T S amp Wolfe J M (2004)Ismultipleobjecttrackingcolorblind[Abstract]Journal of Vision4(8)363a
Leonard C amp Pylyshyn Z W (2003)Measuringtheattentionaldemandofmultipleobjecttracking(MOT)[Abstract]Journal of Vi-sion3(9)582a
Logie R H (1995) Visuo-spatial working memory Hove UKErlbaum
Milliken B Tipper S P amp Weaver B (1994)NegativepriminginaspatiallocalizationtaskFeaturemismatchinganddistractorin-hibitionJournal of Experimental Psychology Human Perception amp Performance20624-646
Mitroff S R Scholl B J amp Wynn K (2004)DivideandconquerHowobjectfilesadaptwhenapersistingobjectsplitsintotwoPsy-chological Science15420-425
Ogawa H amp Yagi A (2003)Primingeffectsinmultipleobjecttrack-ingAnimplicitencodingbasedonglobalspatiotemporalinformation[Abstract]Journal of Vision3(9)339a
Oksama L amp Hyoumlnauml J (2004)Ismultipleobjecttrackingcarriedoutautomaticallybyanearlyvisionmechanismindependentofhigher-ordercognitionAnindividualdifferenceapproachVisual Cognition11631-671
Oliva A Torralba A Castelhano M S amp Henderson J M (2003)Top-downcontrolofvisualattentioninrealworldscenes[Ab-stract]Journal of Vision3(9)3a
Olson I R amp Marshuetz C (2005)RememberingldquowhatrdquobringsalongldquowhererdquoinvisualworkingmemoryPerception amp Psychophysics67185-194
Parasuraman R (1986)VigilancemonitoringandsearchInKRBoffLKaufmanampJPThomas(Eds)Handbook of perception and human performance Vol 2 Cognitive processes and performance(pp1-39)NewYorkWiley
Parmentier F B R Elford G amp Mayberry M (2005)Transi-tionalinformationinspatialserialmemoryPathcharacteristicsaffectrecallperformanceJournal of Experimental Psychology Learning Memory amp Cognition31412-427
Pelli D G (1997)TheVideoToolboxsoftwareforvisualpsychophysicsTransformingnumbersintomoviesSpatial Vision10437-442
Postle B R amp DrsquoEsposito M (1999)ldquoWhatrdquondashthenndashldquowhererdquoinvi-sualworkingmemoryAnevent-relatedfMRIstudyJournal of Cog-nitive Neuroscience11585-597
Postle B R DrsquoEsposito M amp Corkin S (2005)Effectsofver-balandnonverbalinterferenceonspatialandobjectvisualworkingmemoryMemory amp Cognition33203-212
Pylyshyn Z [W] (1989)Theroleoflocationindexesinspatialper-ceptionAsketchoftheFINSTspatial-indexmodelCognition3265-97
Pylyshyn Z W (2004)SomepuzzlingfindingsinmultipleobjecttrackingITrackingwithoutkeepingtrackofobjectidentitiesVisual Cognition11801-822
Pylyshyn Z W amp Leonard C (2003)Inhibitionofnontargetsdur-ingmultipleobjecttracking(MOT)[Abstract]Journal of Vision3(9)585a
Pylyshyn Z W amp Storm R W (1988)Trackingmultipleindepen-denttargetsEvidenceforaparalleltrackingmechanismSpatial Vi-sion3179-197
Saiki J (2002)Multiple-objectpermanencetrackingLimitationinmaintenanceandtransformationofperceptualobjectsProgress in Brain Research140133-148
Schneider W amp Shiffrin R M (1977)ControlledandautomatichumaninformationprocessingIDetectionsearchandattentionPsychological Review841-66
Scholl B J amp Pylyshyn Z W (1999)Trackingmultiple itemsthroughocclusionCluestovisualobjecthoodCognitive Psychology38259-290
Scholl B J Pylyshyn Z W amp Feldman J (2001)Whatisavi-sualobjectEvidencefromtargetmerginginmultipleobjecttrackingCognition80159-177
Scholl B J Pylyshyn Z W amp Franconeri S L (1999May)When are spatiotemporal and featural properties encoded as a result of attentional allocationPaperpresentedattheconferenceoftheAssociationforResearchinVisionandOphthalmologyFortLau-derdaleFL
Sears C R amp Pylyshyn Z W (2000)MultipleobjecttrackingandattentionalprocessingCanadian Journal of Experimental Psychol-ogy541-14
Seiffert A E (2003)Dissociatingneuralcorrelatesofattentionaltrackingandattentiontovisualmotion[Abstract]Journal of Vision3(9)868a
Smith E E Jonides J Koeppe R A Awh E Schumacher E H amp Minoshima S (1995)SpatialversusobjectworkingmemoryPETinvestigationsJournal of Cognitive Neuroscience7337-356
Tipper S P (1985)ThenegativeprimingeffectInhibitoryprimingbyignoredobjectsQuarterly Journal of Experimental Psychology37A571-590
Ungerleider L G amp Mishkin M (1982)Twocorticalvisualsys-
184 HOrOwiTz eT al
temsInDJIngleMAGoodaleampRJWMansfield(Eds)Anal-ysis of visual behavior(pp549-586)CambridgeMAMITPress
Viswanathan L amp Mingolla E (2002)DynamicsofattentionindepthEvidence frommulti-element trackingPerception311415-1437
Vogel E K Woodman G F amp Luck S J (2001)Storageoffea-turesconjunctionsandobjectsinvisualworkingmemoryJournal of Experimental Psychology Human Perception amp Performance2792-114
Wheeler M E amp Treisman A M (2002)Bindinginshort-termvisualmemoryJournal of Experimental Psychology General13148-64
Wilson F A Oacute Scalaidhe S P amp Goldman-Rakic P S (1993)Dissociationofobjectandspatialprocessingdomains inprimateprefrontalcortexScience2601955-1958
Wolfe J M Oliva A Horowitz T S Butcher S J amp Bompas A (2002)SegmentationofobjectsfrombackgroundsinvisualsearchtasksVision Research422985-3004
Yantis S (1992)MultielementvisualtrackingAttentionandpercep-tualorganizationCognitive Psychology24295-340
NoTeS
1Wearemissingageinformationfor1observer2Threeobserverswhoparticipatedinbothexperimentswereelimi-
natedfromthisanalysis3Measuredcapacitycanvarybasedontrackingdurationtherange
ofmovementspeedsandwhetherornotobjectscanoccludeonean-other(asinourexperiment)orbounceoffoneanother(asinOksamaampHyoumlnauml2004)
APPeNdix A Stimulus List
beaver fox horse ostrich rhino turtlecamel giraffe kangaroo pelican rooster wolfcrocodile goat lion rabbit squirrel zebraelephant gorilla moose tiger
APPeNdix b raw data
Table b1 raw Numbers of items reported in the Standard Condition
Total Targets Nontargets Filled
Experiment1 308Experiment2 344 014 359Experiment3 347 015 362Experiment4 339 015 354Experiment5 Unique 312 037 349
Paired 249 080 329Identical 261 086 347
Experiment6 248 078 326Experiment7 599 599
NotemdashTargetsindicatestheaveragenumberofcorrectlyreportedtargetspertrialNontargetstheaveragenumberoftimesobserversclickedonanontargetlocationpertrialandTotal Filledthesumofthesetwocat-egoriesNontargetclickcountswerenotcollectedforExperiment1andtherewerenonontargetsinExperiment7
Table b2 raw Proportions for the Specific Condition by experiment
Errors
Correct Empty Trialsper Responses Transposition Nontarget Space N Observer
Experiment1 44 14 16 26 8 5000Experiment2 56 18 10 16 8 5000Experiment3 68 12 10 10 8 5000Experiment4 58 25 09 08 8 5063Experiment6 41 14 21 23 7 5000Experiment7 33 39 00 28 8 5000
NotemdashTranspositionerrorsrefertotrialsinwhichtheobserverclickedonthewrongtargetlocationTrialsperobservervariedinExperiment4becauseconditionwasselectedrandomlyfromtrialtotrial
(ManuscriptreceivedMarch312005revisionacceptedforpublicationFebruary132006)
Tracking UniqUe ObjecTs 181
Pylyshynrsquos(2004)experimentstheobjectidentitieswerepresentedonlybrieflyatthestartofthetrialduringthemajorityofthetrackingphasetheobjectswerephysicallyidenticalItmaybethatthelinkbetweenobjectidentityandspatiotemporalpositiondecayedduringPylyshynrsquos(2004)taskbutwasconstantlyrefreshedinbothoursandOksamaandHyoumlnaumlrsquos
Ourexperimentsalsoprovidedsomeindirectevidenceforacontent-addressablerepresentationForexampleperformancewasconsistentlysuperior inexperimentsinwhichthesameobjectswereusedinthesamerolesfromtrialtotrial(Experiments23and4)thaninex-perimentsinwhichthestimulussetchangedfromtrialtotrial(Experiments15and6)Thisoccurrednotonlyinthespecifictargetconditionsinwhichobjectidentitywasrelevantbutalsointhestandardconditionsinwhichobjectidentitywasnot
one System or TwoIstrackingservedbyasinglerepresentationthatencodes
bothpositionandidentityoraretheretworepresentationsoneforpositionandoneforidentityThisquestionarisesbecauseofthestrikingcapacitydifferencesbetweenourresponseconditionsInallofourexperimentscapacitywassignificantlylowerforspecifictargetsthanfortargetsinthestandardMOTresponsemodeInExperiment4wedisconfirmedthehypothesisthatcapacitydifferencesresultedfromvariedencodingstrategies(seeegDavisWelchHolmesampShepherd2001)sinceweobtainedthesameresultwhenresponsemodesweremixedwithinablockaswedidwhentheywereruninseparateblocks
ThiscapacitydifferenceisnotconclusiveevidencethatseparaterepresentationsorsystemsareinvolvedThereareseveralwaysinwhichaunifiedaccountcouldex-plainthelowercapacityinthespecifictargetconditionForexampleonemightproposeasystemwithaslot-likestructure(VogelWoodmanampLuck2001)inwhichallslotsarenotcreatedequalIfthereweretwoslotswithsuf-ficientresolutiontoencodebothpositionandidentityandtwoslotsthatcouldonlyencodepositionwewouldalsoobtaindifferentcapacityestimatesinourtworesponseconditionsThisrepresentationcouldalsobedescribedintermsofKahnemanandTreismanrsquos(1984)objectfilesifweassumethatsomeoftheobjectfilesareemptyTheimpletionprocessmightinthatcasereturntheinforma-tionthatagivenobjectwasspatiotemporallycontinuouswithatargetobjectbutfailtoretrievethesemanticorperceptualcontentassociatedwiththatobjectfile
IntheMOTliteraturecapacityisgenerallyassumedtorefertoafixednumberofavailablerepresentationsinthevisualsystemeithervisualindexesorobjectfilesHow-everthereissomeevidencethatcapacityinMOTmightbebetterthoughtofasanundifferentiatedresource(AlvarezampCavanagh2005)InsteadofasetofslotsorpointerswemightassumethatthereisafixedamountofinformationthatcanbeencodedObserversmightencodepositionandidentityforalltargetsbutthebindingbetweenpositionsandidentitymaybenoisierthanthepositioninformationitself(ormaydecaymorequicklyExperiment3indicatedthatanysuchdecaywouldhavetoapproachitsasymptote
valuewithin6ndash7sec)leadingtoapparentlylowercapacitywhenresponsesdependonbinding
InsteadofonesystemtherealsomightbetwoAlthoughwetendtoassumethatasingleexperimentalparadigmprobestheperformanceofasinglecorrespondingfunc-tionalsystemitmaybethatMOTnaturallyrecruitsmulti-pleoverlappingsystemsThesimplesttwo-systemaccountmightbeaproposalthattrackingisservedbyaposition-trackingsystemsuchasvisualindexes(FINSTs)butthataccessingidentityndashpositionbindingsrequiresfocalatten-tion(WheelerampTreisman2002)Howeversincespecificcapacityisreliablygreaterthanoneitemwewouldhavetoassumethatmultipleitemscansimultaneouslybethefocusofattention(AwhampPashler2000Davisetal2001)
IfwewishtopreserveasinglefocusofattentionwemightinsteadproposethatbothvisualindexesandobjectfilescanbeusedVisualindexeswouldprovideonlypo-sitioninformationandobjectfilescouldsupportmorecomplexqueries
Itwillobviouslybedifficulttodefinitivelydecidebe-tweenoneandtwosystemsHowevertheevidenceofourfinalthreeexperimentspointstotwoindependentsystemsInExperiments5and6pairingtargetsandnontargetsre-ducedcapacityinthestandardbutnotinthespecifictar-getcondition(althoughwecouldnotrejectthehypothesisofaproportionaldecrease)InExperiment7standardca-pacityincreasedwhennonontargetswerepresentedbutspecificcapacitywasunaffected
MOThasbeenlikenedtoavisualworkingmemorytaskattimezero(CavanaghampAlvarez2005)soitisnaturaltolooktostudiesofworkingmemoryforanalogsofthecontent-addressableandlocation-onlysystemsTheclassictwo-pathwayschemeinvisionisthedistinctionbetweenventralanddorsalstreamsassociatedwithobjectrecogni-tionandspatialprocessingrespectively(UngerleiderampMishkin1982)ThisdistinctionhasbeenextendedintothedomainofworkingmemorybyWilsonOacuteScalaidheandGoldman-Rakic(1993)whoprovidedevidencethatobjectinformationandspatialinformationareneurallysegregatedinmonkeyprefrontalcortexwidelybelievedtobetheneuralsubstrateofworkingmemoryAsimilardissociationinhumanswasreportedusingPETimagingbySmithetal(1995)
Thedivisionofworkingmemoryintomultiplesubsys-temsgoesbackatleasttothemultiple-componentmodelofBaddeleyandHitch(1974)whichproposedseparatecomponents for different modalities a ldquophonologicallooprdquoforauditoryinformationandaldquovisuospatialsketchpadrdquoforvisualinformationInmorerecentversionsofthemodel(BaddeleyampLogie1999Logie1995)thesketchpadhasbeendividedintovisualandspatialsubcompo-nents roughlycorrespondingto theobjectandspatialsystemsdescribedbyWilsonetal(1993)Interestinglythespatialsubcomponent(theldquoinnerscriberdquo)isspecifi-callytaskedwithholdingsequenceorpathinformation(seeParmentierElfordampMayberry2005)makingitalikelysubstrateforMOTperformance
Thewidespreadinsistenceonaseparationbetweenvi-sualorobjectinformationandspatialinformationwouldseem problematic for our findings Standard MOT is
182 HOrOwiTz eT al
clearlyajobforthespatialsubsystem(PostleDrsquoEspositoampCorkin2005forexampleusedaversionofstandardMOTtotieupdorsalprocessinginamemoryinterferenceparadigm)Howevertheredoesnotseemtoberoominthataccountforacontent-addressablerepresentationOurspecifictargettask(aswellastheMITtaskofOksamaampHyoumlnauml2004)isnotapureobjectrecognitiontaskitrequireslocationndashidentitybindingswhichwouldseemtoinvolveintegratinginformationacrossstreamsTheproblemdisappears ifwe take intoaccount recentbe-havioral(OlsonampMarshuetz2005)andfMRI(PostleampDrsquoEsposito1999)investigationsdemonstratingthatobjectworkingmemoryrepresentationsalsocarrysomelocationinformation
NeuroimagingstudiesmightproveusefulhereStan-dardMOThasbeenshowntoactivateparietallobeareasinthedorsalstream(Culhametal1998Jovicichetal2001Seiffert2003)Itwouldbeinformativetoknowwhetherthepatternofactivationchangeswhenuniqueob-jectsareintroducedorwhenthetaskpotentiallyinvolvesobjectidentities
unique object Features in MoTInadditiontoourfindingsaboutcontentaddressability
intrackingwewereabletodemonstratethathavingvisu-allyuniqueobjectsmadetrackingeasierobserverswereabletotrackroughly06moreitemswhenobjectswereuniquethanwhenobjectswereidenticalInExperiments5and6weattributedthisresulttoanerrorrecoverystrat-egyIfIlosetrackofthezebraIcansearchforthezebraonthebasisofitsphysicalfeaturesandcontinuetotrackitThisstrategywouldobviouslybeuselessifallitemsarezebrasThisstrategyisoflimitedutilityifthereareazebratargetandazebradistractoronceIlosethezebraIthenhaveonlya50chanceofpickingupthetargetzebraasopposedtothedistractor
TheerrorrecoveryhypothesisisnottrivialIthassev-eralimportantimplicationsFirstitimpliesthatobserversknowwhattheyarenottrackingThephenomenologyofMOTissuchthatanobserverhasthefeelingofknow-inghowmanytargetsheorsheissuccessfullytrackingTheerrorrecoveryhypothesisimpliesthatwhenatargetislosttheobserverknowswhichone(givenuniquetargets)SecondtheerrorrecoveryhypothesisalsoassumesthatanobservercansearchforamissingtargetwhilecontinuingtotracktheremainingtargetsThisisinagreementwithrecentevidencethatobserverscansuccessfullyperformanMOTtaskandavisualsearchtaskonthesametrial(AlvarezHorowitzArsenioDiMaseampWolfe2005)Finallythehypothesisimpliestheabilitytoaddtothetrackingsetontheflywhiletrackingwithoutanyphysi-calcueTheseimplicationsneedtobeconfirmedthroughfurtherresearch
Howeverifobserversuseuniquephysicalfeaturesofobjectstorecoverlosttargetsinthisfashionitiscuriousthatthereisnosuchbenefitinthespecifictargetcondi-tionIfobserversdorecoverlosttargetsduringthecourse
ofatrialtheyappeartorecoveronlytheirlocationsnottheiridentities
AnalternativetotheerrorrecoveryhypothesisthoughnotamutuallyexclusiveoneisthattheresultsofEx-periment5canbeexplainedbytheperceptualsimilarityoftargetandnontargetitemsYantisrsquos(1992)approachtoMOThighlightstheimportanceofperceptualgroup-ingfactorsthatallowthetargetstobetreatedasaunitbythevisualsystemMostoftheresearchinthislinehasfocusedonspatiotemporalgroupingfactorssuchasthespatialrelationshipsbetweentheobjects(SearsampPylyshyn2000ViswanathanampMingolla2002Yantis1992)
Cangroupingby(nonspatial)featuralpropertiesim-provetrackingAnumberofstudieshavedemonstratedthatobservershave limitedexplicit access to featuralproperties (Bahrami 2003Saiki 2002Scholl etal1999)butthisfindingdoesnotaddressthedirectrolethatfeaturesmayplayWecouldexplaintheresultsofEx-periment5byproposingthatobserverswereabletouseaccidentaldifferencesintheshapeandcolorstatisticsoftargetsandnontargetstohelpsegregatethemvisuallyInthepairedconditiontargetsandnontargetshadidenticalfeaturestatisticssothatadvantagewaslostThishypoth-esiscouldalsoexplainwhyperformancewasbetterwithafixedstimulussetthanwithavariablestimulussetoveranumberoftrialswiththesamestimuliobserverslearnsubtledifferencesbetweenthetwosets
ConclusionsIntheintroductionwenotedthatreal-worlddynamic
attentiontasksdifferfromtraditionalMOTintwowaysFirstobserversintheeverydayworldusuallydealwithmultipleuniqueobjectsSecondtheyoftenwishtoknowwhereagiventargetmightberatherthanwhichobjectsarethetargetsTheexperimentspresentedhererepresentanattempttobridgelaboratoryMOTresearchandtheseeverydaycognitivetasksWehaveshownthatitiseasiertotrackuniqueobjectsthantotrackidenticalobjectsassum-ingthattheobjectsonewishestotrackarealsouniquelydifferentfromnontargetobjectsFurthermoreobserversdoknowwhattheyaretrackingTotakeoneofourex-amplesofecologicallyvalidtrackingtasksifyouwatchyourchildrenandtheirfriendsontheplaygroundyoucanbeawareofwhereyourkidsareatanygivenmomentyoumayalsobeawareofthelocationsoftheirfriendswithoutbeingabletotellonefriendfromtheotherFurthermoreifyoulosetrackofachildyouwillprobablybeawareofwhichoneyouhavelost
AuTHor NoTe
ThisresearchwassupportedbyNIMHGrantR0165576toTSHWethankKristinMichodandSkylerPlaceforassistancewithdatacollec-tionaswellasSteveLuckandananonymousreviewerforconstructivecriticismCorrespondencerelatingtothisarticlemaybesenttoTSHorowitzBrighamandWomenrsquosHospitalVisualAttentionLaboratory64SidneyStreetSuite170BostonMA02139(e-mailtoddhsearchbwhharvardedu)
Tracking UniqUe ObjecTs 183
reFereNCeS
Allen R McGeorge P Pearson D amp Milne A B (2004)AttentionandexpertiseinmultipletargettrackingApplied Cognitive Psychology18337-347
Alvarez G A amp Cavanagh P (2005)IndependentresourcesforattentionaltrackingintheleftandrightvisualhemifieldsPsychologi-cal Science16637-643
Alvarez G A amp Franconeri S (2004November)How many ob-jects can you trackPaperpresentedatthe12thAnnualWorkshoponObjectPerceptionampMemoryMinneapolisMN
Alvarez G A Horowitz T S Arsenio H C DiMase J S amp Wolfe J M (2005)DomultielementvisualtrackingandvisualsearchdrawcontinuouslyonthesamevisualattentionresourcesJournal of Experimental Psychology Human Perception amp Perfor-mance31643-667
Awh E Jonides J amp Reuter-Lorenz P A (1998)RehearsalinspatialworkingmemoryJournal of Experimental Psychology Human Perception amp Performance24780-790
Awh E amp Pashler H (2000)EvidenceforsplitattentionalfociJournal of Experimental Psychology Human Perception amp Perfor-mance26834-846
Baddeley A D amp Hitch G J (1974)WorkingmemoryInGHBower(Ed)The psychology of learning and motivation Advances in research and theory(Vol8pp47-90)NewYorkAcademicPress
Baddeley A D amp Logie R H (1999) Working memoryThemultiple-componentmodelInAMiyakeampPShah(Eds)Models of working memory Mechanisms of active maintenance and executive control(pp28-61)CambridgeCambridgeUniversityPress
Bahrami B (2003)ObjectpropertyencodingandchangeblindnessinmultipleobjecttrackingVisual Cognition10949-963
Brainard D H (1997)ThePsychophysicsToolboxSpatial Vision10433-436
Cavanagh P amp Alvarez G A (2005)TrackingmultipletargetswithmultifocalattentionTrends in Cognitive Sciences9349-354
Culham J C Brandt S A Cavanagh P Kanwisher N G Dale A M amp Tootell R B H (1998)CorticalfMRIactiva-tionproducedbyattentive trackingofmoving targetsJournal of Neurophysiology802657-2670
Davis G Welch V L Holmes A amp Shepherd A (2001)CanattentionselectonlyafixednumberofobjectsatatimePerception301227-1248
Henderson J M amp Hollingworth A (2003)EyemovementsandvisualmemoryDetectingchangestosaccadetargetsinscenesPer-ception amp Psychophysics6558-71
Horowitz T S Birnkrant R S Fencsik D E Tran L amp Wolfe J M (2006)HowdowetrackinvisibleobjectsPsychonomic Bulletin amp Review13516-523
Jovicich J Peters R J Koch C Braun J Chang L amp Ernst T (2001)Brainareasspecificforattentionalloadinamotion-trackingtaskJournal of Cognitive Neuroscience131048-1058
Kahneman D amp Treisman A (1984)ChangingviewsofattentionandautomaticityInRParasuramanampDRDavies(Eds)Varieties of attention(pp29-61)OrlandoAcademicPress
Kahneman D Treisman A amp Gibbs B J (1992)ThereviewingofobjectfilesObject-specificintegrationofinformationCognitive Psychology24175-219
Klieger S B Horowitz T S amp Wolfe J M (2004)Ismultipleobjecttrackingcolorblind[Abstract]Journal of Vision4(8)363a
Leonard C amp Pylyshyn Z W (2003)Measuringtheattentionaldemandofmultipleobjecttracking(MOT)[Abstract]Journal of Vi-sion3(9)582a
Logie R H (1995) Visuo-spatial working memory Hove UKErlbaum
Milliken B Tipper S P amp Weaver B (1994)NegativepriminginaspatiallocalizationtaskFeaturemismatchinganddistractorin-hibitionJournal of Experimental Psychology Human Perception amp Performance20624-646
Mitroff S R Scholl B J amp Wynn K (2004)DivideandconquerHowobjectfilesadaptwhenapersistingobjectsplitsintotwoPsy-chological Science15420-425
Ogawa H amp Yagi A (2003)Primingeffectsinmultipleobjecttrack-ingAnimplicitencodingbasedonglobalspatiotemporalinformation[Abstract]Journal of Vision3(9)339a
Oksama L amp Hyoumlnauml J (2004)Ismultipleobjecttrackingcarriedoutautomaticallybyanearlyvisionmechanismindependentofhigher-ordercognitionAnindividualdifferenceapproachVisual Cognition11631-671
Oliva A Torralba A Castelhano M S amp Henderson J M (2003)Top-downcontrolofvisualattentioninrealworldscenes[Ab-stract]Journal of Vision3(9)3a
Olson I R amp Marshuetz C (2005)RememberingldquowhatrdquobringsalongldquowhererdquoinvisualworkingmemoryPerception amp Psychophysics67185-194
Parasuraman R (1986)VigilancemonitoringandsearchInKRBoffLKaufmanampJPThomas(Eds)Handbook of perception and human performance Vol 2 Cognitive processes and performance(pp1-39)NewYorkWiley
Parmentier F B R Elford G amp Mayberry M (2005)Transi-tionalinformationinspatialserialmemoryPathcharacteristicsaffectrecallperformanceJournal of Experimental Psychology Learning Memory amp Cognition31412-427
Pelli D G (1997)TheVideoToolboxsoftwareforvisualpsychophysicsTransformingnumbersintomoviesSpatial Vision10437-442
Postle B R amp DrsquoEsposito M (1999)ldquoWhatrdquondashthenndashldquowhererdquoinvi-sualworkingmemoryAnevent-relatedfMRIstudyJournal of Cog-nitive Neuroscience11585-597
Postle B R DrsquoEsposito M amp Corkin S (2005)Effectsofver-balandnonverbalinterferenceonspatialandobjectvisualworkingmemoryMemory amp Cognition33203-212
Pylyshyn Z [W] (1989)Theroleoflocationindexesinspatialper-ceptionAsketchoftheFINSTspatial-indexmodelCognition3265-97
Pylyshyn Z W (2004)SomepuzzlingfindingsinmultipleobjecttrackingITrackingwithoutkeepingtrackofobjectidentitiesVisual Cognition11801-822
Pylyshyn Z W amp Leonard C (2003)Inhibitionofnontargetsdur-ingmultipleobjecttracking(MOT)[Abstract]Journal of Vision3(9)585a
Pylyshyn Z W amp Storm R W (1988)Trackingmultipleindepen-denttargetsEvidenceforaparalleltrackingmechanismSpatial Vi-sion3179-197
Saiki J (2002)Multiple-objectpermanencetrackingLimitationinmaintenanceandtransformationofperceptualobjectsProgress in Brain Research140133-148
Schneider W amp Shiffrin R M (1977)ControlledandautomatichumaninformationprocessingIDetectionsearchandattentionPsychological Review841-66
Scholl B J amp Pylyshyn Z W (1999)Trackingmultiple itemsthroughocclusionCluestovisualobjecthoodCognitive Psychology38259-290
Scholl B J Pylyshyn Z W amp Feldman J (2001)Whatisavi-sualobjectEvidencefromtargetmerginginmultipleobjecttrackingCognition80159-177
Scholl B J Pylyshyn Z W amp Franconeri S L (1999May)When are spatiotemporal and featural properties encoded as a result of attentional allocationPaperpresentedattheconferenceoftheAssociationforResearchinVisionandOphthalmologyFortLau-derdaleFL
Sears C R amp Pylyshyn Z W (2000)MultipleobjecttrackingandattentionalprocessingCanadian Journal of Experimental Psychol-ogy541-14
Seiffert A E (2003)Dissociatingneuralcorrelatesofattentionaltrackingandattentiontovisualmotion[Abstract]Journal of Vision3(9)868a
Smith E E Jonides J Koeppe R A Awh E Schumacher E H amp Minoshima S (1995)SpatialversusobjectworkingmemoryPETinvestigationsJournal of Cognitive Neuroscience7337-356
Tipper S P (1985)ThenegativeprimingeffectInhibitoryprimingbyignoredobjectsQuarterly Journal of Experimental Psychology37A571-590
Ungerleider L G amp Mishkin M (1982)Twocorticalvisualsys-
184 HOrOwiTz eT al
temsInDJIngleMAGoodaleampRJWMansfield(Eds)Anal-ysis of visual behavior(pp549-586)CambridgeMAMITPress
Viswanathan L amp Mingolla E (2002)DynamicsofattentionindepthEvidence frommulti-element trackingPerception311415-1437
Vogel E K Woodman G F amp Luck S J (2001)Storageoffea-turesconjunctionsandobjectsinvisualworkingmemoryJournal of Experimental Psychology Human Perception amp Performance2792-114
Wheeler M E amp Treisman A M (2002)Bindinginshort-termvisualmemoryJournal of Experimental Psychology General13148-64
Wilson F A Oacute Scalaidhe S P amp Goldman-Rakic P S (1993)Dissociationofobjectandspatialprocessingdomains inprimateprefrontalcortexScience2601955-1958
Wolfe J M Oliva A Horowitz T S Butcher S J amp Bompas A (2002)SegmentationofobjectsfrombackgroundsinvisualsearchtasksVision Research422985-3004
Yantis S (1992)MultielementvisualtrackingAttentionandpercep-tualorganizationCognitive Psychology24295-340
NoTeS
1Wearemissingageinformationfor1observer2Threeobserverswhoparticipatedinbothexperimentswereelimi-
natedfromthisanalysis3Measuredcapacitycanvarybasedontrackingdurationtherange
ofmovementspeedsandwhetherornotobjectscanoccludeonean-other(asinourexperiment)orbounceoffoneanother(asinOksamaampHyoumlnauml2004)
APPeNdix A Stimulus List
beaver fox horse ostrich rhino turtlecamel giraffe kangaroo pelican rooster wolfcrocodile goat lion rabbit squirrel zebraelephant gorilla moose tiger
APPeNdix b raw data
Table b1 raw Numbers of items reported in the Standard Condition
Total Targets Nontargets Filled
Experiment1 308Experiment2 344 014 359Experiment3 347 015 362Experiment4 339 015 354Experiment5 Unique 312 037 349
Paired 249 080 329Identical 261 086 347
Experiment6 248 078 326Experiment7 599 599
NotemdashTargetsindicatestheaveragenumberofcorrectlyreportedtargetspertrialNontargetstheaveragenumberoftimesobserversclickedonanontargetlocationpertrialandTotal Filledthesumofthesetwocat-egoriesNontargetclickcountswerenotcollectedforExperiment1andtherewerenonontargetsinExperiment7
Table b2 raw Proportions for the Specific Condition by experiment
Errors
Correct Empty Trialsper Responses Transposition Nontarget Space N Observer
Experiment1 44 14 16 26 8 5000Experiment2 56 18 10 16 8 5000Experiment3 68 12 10 10 8 5000Experiment4 58 25 09 08 8 5063Experiment6 41 14 21 23 7 5000Experiment7 33 39 00 28 8 5000
NotemdashTranspositionerrorsrefertotrialsinwhichtheobserverclickedonthewrongtargetlocationTrialsperobservervariedinExperiment4becauseconditionwasselectedrandomlyfromtrialtotrial
(ManuscriptreceivedMarch312005revisionacceptedforpublicationFebruary132006)
182 HOrOwiTz eT al
clearlyajobforthespatialsubsystem(PostleDrsquoEspositoampCorkin2005forexampleusedaversionofstandardMOTtotieupdorsalprocessinginamemoryinterferenceparadigm)Howevertheredoesnotseemtoberoominthataccountforacontent-addressablerepresentationOurspecifictargettask(aswellastheMITtaskofOksamaampHyoumlnauml2004)isnotapureobjectrecognitiontaskitrequireslocationndashidentitybindingswhichwouldseemtoinvolveintegratinginformationacrossstreamsTheproblemdisappears ifwe take intoaccount recentbe-havioral(OlsonampMarshuetz2005)andfMRI(PostleampDrsquoEsposito1999)investigationsdemonstratingthatobjectworkingmemoryrepresentationsalsocarrysomelocationinformation
NeuroimagingstudiesmightproveusefulhereStan-dardMOThasbeenshowntoactivateparietallobeareasinthedorsalstream(Culhametal1998Jovicichetal2001Seiffert2003)Itwouldbeinformativetoknowwhetherthepatternofactivationchangeswhenuniqueob-jectsareintroducedorwhenthetaskpotentiallyinvolvesobjectidentities
unique object Features in MoTInadditiontoourfindingsaboutcontentaddressability
intrackingwewereabletodemonstratethathavingvisu-allyuniqueobjectsmadetrackingeasierobserverswereabletotrackroughly06moreitemswhenobjectswereuniquethanwhenobjectswereidenticalInExperiments5and6weattributedthisresulttoanerrorrecoverystrat-egyIfIlosetrackofthezebraIcansearchforthezebraonthebasisofitsphysicalfeaturesandcontinuetotrackitThisstrategywouldobviouslybeuselessifallitemsarezebrasThisstrategyisoflimitedutilityifthereareazebratargetandazebradistractoronceIlosethezebraIthenhaveonlya50chanceofpickingupthetargetzebraasopposedtothedistractor
TheerrorrecoveryhypothesisisnottrivialIthassev-eralimportantimplicationsFirstitimpliesthatobserversknowwhattheyarenottrackingThephenomenologyofMOTissuchthatanobserverhasthefeelingofknow-inghowmanytargetsheorsheissuccessfullytrackingTheerrorrecoveryhypothesisimpliesthatwhenatargetislosttheobserverknowswhichone(givenuniquetargets)SecondtheerrorrecoveryhypothesisalsoassumesthatanobservercansearchforamissingtargetwhilecontinuingtotracktheremainingtargetsThisisinagreementwithrecentevidencethatobserverscansuccessfullyperformanMOTtaskandavisualsearchtaskonthesametrial(AlvarezHorowitzArsenioDiMaseampWolfe2005)Finallythehypothesisimpliestheabilitytoaddtothetrackingsetontheflywhiletrackingwithoutanyphysi-calcueTheseimplicationsneedtobeconfirmedthroughfurtherresearch
Howeverifobserversuseuniquephysicalfeaturesofobjectstorecoverlosttargetsinthisfashionitiscuriousthatthereisnosuchbenefitinthespecifictargetcondi-tionIfobserversdorecoverlosttargetsduringthecourse
ofatrialtheyappeartorecoveronlytheirlocationsnottheiridentities
AnalternativetotheerrorrecoveryhypothesisthoughnotamutuallyexclusiveoneisthattheresultsofEx-periment5canbeexplainedbytheperceptualsimilarityoftargetandnontargetitemsYantisrsquos(1992)approachtoMOThighlightstheimportanceofperceptualgroup-ingfactorsthatallowthetargetstobetreatedasaunitbythevisualsystemMostoftheresearchinthislinehasfocusedonspatiotemporalgroupingfactorssuchasthespatialrelationshipsbetweentheobjects(SearsampPylyshyn2000ViswanathanampMingolla2002Yantis1992)
Cangroupingby(nonspatial)featuralpropertiesim-provetrackingAnumberofstudieshavedemonstratedthatobservershave limitedexplicit access to featuralproperties (Bahrami 2003Saiki 2002Scholl etal1999)butthisfindingdoesnotaddressthedirectrolethatfeaturesmayplayWecouldexplaintheresultsofEx-periment5byproposingthatobserverswereabletouseaccidentaldifferencesintheshapeandcolorstatisticsoftargetsandnontargetstohelpsegregatethemvisuallyInthepairedconditiontargetsandnontargetshadidenticalfeaturestatisticssothatadvantagewaslostThishypoth-esiscouldalsoexplainwhyperformancewasbetterwithafixedstimulussetthanwithavariablestimulussetoveranumberoftrialswiththesamestimuliobserverslearnsubtledifferencesbetweenthetwosets
ConclusionsIntheintroductionwenotedthatreal-worlddynamic
attentiontasksdifferfromtraditionalMOTintwowaysFirstobserversintheeverydayworldusuallydealwithmultipleuniqueobjectsSecondtheyoftenwishtoknowwhereagiventargetmightberatherthanwhichobjectsarethetargetsTheexperimentspresentedhererepresentanattempttobridgelaboratoryMOTresearchandtheseeverydaycognitivetasksWehaveshownthatitiseasiertotrackuniqueobjectsthantotrackidenticalobjectsassum-ingthattheobjectsonewishestotrackarealsouniquelydifferentfromnontargetobjectsFurthermoreobserversdoknowwhattheyaretrackingTotakeoneofourex-amplesofecologicallyvalidtrackingtasksifyouwatchyourchildrenandtheirfriendsontheplaygroundyoucanbeawareofwhereyourkidsareatanygivenmomentyoumayalsobeawareofthelocationsoftheirfriendswithoutbeingabletotellonefriendfromtheotherFurthermoreifyoulosetrackofachildyouwillprobablybeawareofwhichoneyouhavelost
AuTHor NoTe
ThisresearchwassupportedbyNIMHGrantR0165576toTSHWethankKristinMichodandSkylerPlaceforassistancewithdatacollec-tionaswellasSteveLuckandananonymousreviewerforconstructivecriticismCorrespondencerelatingtothisarticlemaybesenttoTSHorowitzBrighamandWomenrsquosHospitalVisualAttentionLaboratory64SidneyStreetSuite170BostonMA02139(e-mailtoddhsearchbwhharvardedu)
Tracking UniqUe ObjecTs 183
reFereNCeS
Allen R McGeorge P Pearson D amp Milne A B (2004)AttentionandexpertiseinmultipletargettrackingApplied Cognitive Psychology18337-347
Alvarez G A amp Cavanagh P (2005)IndependentresourcesforattentionaltrackingintheleftandrightvisualhemifieldsPsychologi-cal Science16637-643
Alvarez G A amp Franconeri S (2004November)How many ob-jects can you trackPaperpresentedatthe12thAnnualWorkshoponObjectPerceptionampMemoryMinneapolisMN
Alvarez G A Horowitz T S Arsenio H C DiMase J S amp Wolfe J M (2005)DomultielementvisualtrackingandvisualsearchdrawcontinuouslyonthesamevisualattentionresourcesJournal of Experimental Psychology Human Perception amp Perfor-mance31643-667
Awh E Jonides J amp Reuter-Lorenz P A (1998)RehearsalinspatialworkingmemoryJournal of Experimental Psychology Human Perception amp Performance24780-790
Awh E amp Pashler H (2000)EvidenceforsplitattentionalfociJournal of Experimental Psychology Human Perception amp Perfor-mance26834-846
Baddeley A D amp Hitch G J (1974)WorkingmemoryInGHBower(Ed)The psychology of learning and motivation Advances in research and theory(Vol8pp47-90)NewYorkAcademicPress
Baddeley A D amp Logie R H (1999) Working memoryThemultiple-componentmodelInAMiyakeampPShah(Eds)Models of working memory Mechanisms of active maintenance and executive control(pp28-61)CambridgeCambridgeUniversityPress
Bahrami B (2003)ObjectpropertyencodingandchangeblindnessinmultipleobjecttrackingVisual Cognition10949-963
Brainard D H (1997)ThePsychophysicsToolboxSpatial Vision10433-436
Cavanagh P amp Alvarez G A (2005)TrackingmultipletargetswithmultifocalattentionTrends in Cognitive Sciences9349-354
Culham J C Brandt S A Cavanagh P Kanwisher N G Dale A M amp Tootell R B H (1998)CorticalfMRIactiva-tionproducedbyattentive trackingofmoving targetsJournal of Neurophysiology802657-2670
Davis G Welch V L Holmes A amp Shepherd A (2001)CanattentionselectonlyafixednumberofobjectsatatimePerception301227-1248
Henderson J M amp Hollingworth A (2003)EyemovementsandvisualmemoryDetectingchangestosaccadetargetsinscenesPer-ception amp Psychophysics6558-71
Horowitz T S Birnkrant R S Fencsik D E Tran L amp Wolfe J M (2006)HowdowetrackinvisibleobjectsPsychonomic Bulletin amp Review13516-523
Jovicich J Peters R J Koch C Braun J Chang L amp Ernst T (2001)Brainareasspecificforattentionalloadinamotion-trackingtaskJournal of Cognitive Neuroscience131048-1058
Kahneman D amp Treisman A (1984)ChangingviewsofattentionandautomaticityInRParasuramanampDRDavies(Eds)Varieties of attention(pp29-61)OrlandoAcademicPress
Kahneman D Treisman A amp Gibbs B J (1992)ThereviewingofobjectfilesObject-specificintegrationofinformationCognitive Psychology24175-219
Klieger S B Horowitz T S amp Wolfe J M (2004)Ismultipleobjecttrackingcolorblind[Abstract]Journal of Vision4(8)363a
Leonard C amp Pylyshyn Z W (2003)Measuringtheattentionaldemandofmultipleobjecttracking(MOT)[Abstract]Journal of Vi-sion3(9)582a
Logie R H (1995) Visuo-spatial working memory Hove UKErlbaum
Milliken B Tipper S P amp Weaver B (1994)NegativepriminginaspatiallocalizationtaskFeaturemismatchinganddistractorin-hibitionJournal of Experimental Psychology Human Perception amp Performance20624-646
Mitroff S R Scholl B J amp Wynn K (2004)DivideandconquerHowobjectfilesadaptwhenapersistingobjectsplitsintotwoPsy-chological Science15420-425
Ogawa H amp Yagi A (2003)Primingeffectsinmultipleobjecttrack-ingAnimplicitencodingbasedonglobalspatiotemporalinformation[Abstract]Journal of Vision3(9)339a
Oksama L amp Hyoumlnauml J (2004)Ismultipleobjecttrackingcarriedoutautomaticallybyanearlyvisionmechanismindependentofhigher-ordercognitionAnindividualdifferenceapproachVisual Cognition11631-671
Oliva A Torralba A Castelhano M S amp Henderson J M (2003)Top-downcontrolofvisualattentioninrealworldscenes[Ab-stract]Journal of Vision3(9)3a
Olson I R amp Marshuetz C (2005)RememberingldquowhatrdquobringsalongldquowhererdquoinvisualworkingmemoryPerception amp Psychophysics67185-194
Parasuraman R (1986)VigilancemonitoringandsearchInKRBoffLKaufmanampJPThomas(Eds)Handbook of perception and human performance Vol 2 Cognitive processes and performance(pp1-39)NewYorkWiley
Parmentier F B R Elford G amp Mayberry M (2005)Transi-tionalinformationinspatialserialmemoryPathcharacteristicsaffectrecallperformanceJournal of Experimental Psychology Learning Memory amp Cognition31412-427
Pelli D G (1997)TheVideoToolboxsoftwareforvisualpsychophysicsTransformingnumbersintomoviesSpatial Vision10437-442
Postle B R amp DrsquoEsposito M (1999)ldquoWhatrdquondashthenndashldquowhererdquoinvi-sualworkingmemoryAnevent-relatedfMRIstudyJournal of Cog-nitive Neuroscience11585-597
Postle B R DrsquoEsposito M amp Corkin S (2005)Effectsofver-balandnonverbalinterferenceonspatialandobjectvisualworkingmemoryMemory amp Cognition33203-212
Pylyshyn Z [W] (1989)Theroleoflocationindexesinspatialper-ceptionAsketchoftheFINSTspatial-indexmodelCognition3265-97
Pylyshyn Z W (2004)SomepuzzlingfindingsinmultipleobjecttrackingITrackingwithoutkeepingtrackofobjectidentitiesVisual Cognition11801-822
Pylyshyn Z W amp Leonard C (2003)Inhibitionofnontargetsdur-ingmultipleobjecttracking(MOT)[Abstract]Journal of Vision3(9)585a
Pylyshyn Z W amp Storm R W (1988)Trackingmultipleindepen-denttargetsEvidenceforaparalleltrackingmechanismSpatial Vi-sion3179-197
Saiki J (2002)Multiple-objectpermanencetrackingLimitationinmaintenanceandtransformationofperceptualobjectsProgress in Brain Research140133-148
Schneider W amp Shiffrin R M (1977)ControlledandautomatichumaninformationprocessingIDetectionsearchandattentionPsychological Review841-66
Scholl B J amp Pylyshyn Z W (1999)Trackingmultiple itemsthroughocclusionCluestovisualobjecthoodCognitive Psychology38259-290
Scholl B J Pylyshyn Z W amp Feldman J (2001)Whatisavi-sualobjectEvidencefromtargetmerginginmultipleobjecttrackingCognition80159-177
Scholl B J Pylyshyn Z W amp Franconeri S L (1999May)When are spatiotemporal and featural properties encoded as a result of attentional allocationPaperpresentedattheconferenceoftheAssociationforResearchinVisionandOphthalmologyFortLau-derdaleFL
Sears C R amp Pylyshyn Z W (2000)MultipleobjecttrackingandattentionalprocessingCanadian Journal of Experimental Psychol-ogy541-14
Seiffert A E (2003)Dissociatingneuralcorrelatesofattentionaltrackingandattentiontovisualmotion[Abstract]Journal of Vision3(9)868a
Smith E E Jonides J Koeppe R A Awh E Schumacher E H amp Minoshima S (1995)SpatialversusobjectworkingmemoryPETinvestigationsJournal of Cognitive Neuroscience7337-356
Tipper S P (1985)ThenegativeprimingeffectInhibitoryprimingbyignoredobjectsQuarterly Journal of Experimental Psychology37A571-590
Ungerleider L G amp Mishkin M (1982)Twocorticalvisualsys-
184 HOrOwiTz eT al
temsInDJIngleMAGoodaleampRJWMansfield(Eds)Anal-ysis of visual behavior(pp549-586)CambridgeMAMITPress
Viswanathan L amp Mingolla E (2002)DynamicsofattentionindepthEvidence frommulti-element trackingPerception311415-1437
Vogel E K Woodman G F amp Luck S J (2001)Storageoffea-turesconjunctionsandobjectsinvisualworkingmemoryJournal of Experimental Psychology Human Perception amp Performance2792-114
Wheeler M E amp Treisman A M (2002)Bindinginshort-termvisualmemoryJournal of Experimental Psychology General13148-64
Wilson F A Oacute Scalaidhe S P amp Goldman-Rakic P S (1993)Dissociationofobjectandspatialprocessingdomains inprimateprefrontalcortexScience2601955-1958
Wolfe J M Oliva A Horowitz T S Butcher S J amp Bompas A (2002)SegmentationofobjectsfrombackgroundsinvisualsearchtasksVision Research422985-3004
Yantis S (1992)MultielementvisualtrackingAttentionandpercep-tualorganizationCognitive Psychology24295-340
NoTeS
1Wearemissingageinformationfor1observer2Threeobserverswhoparticipatedinbothexperimentswereelimi-
natedfromthisanalysis3Measuredcapacitycanvarybasedontrackingdurationtherange
ofmovementspeedsandwhetherornotobjectscanoccludeonean-other(asinourexperiment)orbounceoffoneanother(asinOksamaampHyoumlnauml2004)
APPeNdix A Stimulus List
beaver fox horse ostrich rhino turtlecamel giraffe kangaroo pelican rooster wolfcrocodile goat lion rabbit squirrel zebraelephant gorilla moose tiger
APPeNdix b raw data
Table b1 raw Numbers of items reported in the Standard Condition
Total Targets Nontargets Filled
Experiment1 308Experiment2 344 014 359Experiment3 347 015 362Experiment4 339 015 354Experiment5 Unique 312 037 349
Paired 249 080 329Identical 261 086 347
Experiment6 248 078 326Experiment7 599 599
NotemdashTargetsindicatestheaveragenumberofcorrectlyreportedtargetspertrialNontargetstheaveragenumberoftimesobserversclickedonanontargetlocationpertrialandTotal Filledthesumofthesetwocat-egoriesNontargetclickcountswerenotcollectedforExperiment1andtherewerenonontargetsinExperiment7
Table b2 raw Proportions for the Specific Condition by experiment
Errors
Correct Empty Trialsper Responses Transposition Nontarget Space N Observer
Experiment1 44 14 16 26 8 5000Experiment2 56 18 10 16 8 5000Experiment3 68 12 10 10 8 5000Experiment4 58 25 09 08 8 5063Experiment6 41 14 21 23 7 5000Experiment7 33 39 00 28 8 5000
NotemdashTranspositionerrorsrefertotrialsinwhichtheobserverclickedonthewrongtargetlocationTrialsperobservervariedinExperiment4becauseconditionwasselectedrandomlyfromtrialtotrial
(ManuscriptreceivedMarch312005revisionacceptedforpublicationFebruary132006)
Tracking UniqUe ObjecTs 183
reFereNCeS
Allen R McGeorge P Pearson D amp Milne A B (2004)AttentionandexpertiseinmultipletargettrackingApplied Cognitive Psychology18337-347
Alvarez G A amp Cavanagh P (2005)IndependentresourcesforattentionaltrackingintheleftandrightvisualhemifieldsPsychologi-cal Science16637-643
Alvarez G A amp Franconeri S (2004November)How many ob-jects can you trackPaperpresentedatthe12thAnnualWorkshoponObjectPerceptionampMemoryMinneapolisMN
Alvarez G A Horowitz T S Arsenio H C DiMase J S amp Wolfe J M (2005)DomultielementvisualtrackingandvisualsearchdrawcontinuouslyonthesamevisualattentionresourcesJournal of Experimental Psychology Human Perception amp Perfor-mance31643-667
Awh E Jonides J amp Reuter-Lorenz P A (1998)RehearsalinspatialworkingmemoryJournal of Experimental Psychology Human Perception amp Performance24780-790
Awh E amp Pashler H (2000)EvidenceforsplitattentionalfociJournal of Experimental Psychology Human Perception amp Perfor-mance26834-846
Baddeley A D amp Hitch G J (1974)WorkingmemoryInGHBower(Ed)The psychology of learning and motivation Advances in research and theory(Vol8pp47-90)NewYorkAcademicPress
Baddeley A D amp Logie R H (1999) Working memoryThemultiple-componentmodelInAMiyakeampPShah(Eds)Models of working memory Mechanisms of active maintenance and executive control(pp28-61)CambridgeCambridgeUniversityPress
Bahrami B (2003)ObjectpropertyencodingandchangeblindnessinmultipleobjecttrackingVisual Cognition10949-963
Brainard D H (1997)ThePsychophysicsToolboxSpatial Vision10433-436
Cavanagh P amp Alvarez G A (2005)TrackingmultipletargetswithmultifocalattentionTrends in Cognitive Sciences9349-354
Culham J C Brandt S A Cavanagh P Kanwisher N G Dale A M amp Tootell R B H (1998)CorticalfMRIactiva-tionproducedbyattentive trackingofmoving targetsJournal of Neurophysiology802657-2670
Davis G Welch V L Holmes A amp Shepherd A (2001)CanattentionselectonlyafixednumberofobjectsatatimePerception301227-1248
Henderson J M amp Hollingworth A (2003)EyemovementsandvisualmemoryDetectingchangestosaccadetargetsinscenesPer-ception amp Psychophysics6558-71
Horowitz T S Birnkrant R S Fencsik D E Tran L amp Wolfe J M (2006)HowdowetrackinvisibleobjectsPsychonomic Bulletin amp Review13516-523
Jovicich J Peters R J Koch C Braun J Chang L amp Ernst T (2001)Brainareasspecificforattentionalloadinamotion-trackingtaskJournal of Cognitive Neuroscience131048-1058
Kahneman D amp Treisman A (1984)ChangingviewsofattentionandautomaticityInRParasuramanampDRDavies(Eds)Varieties of attention(pp29-61)OrlandoAcademicPress
Kahneman D Treisman A amp Gibbs B J (1992)ThereviewingofobjectfilesObject-specificintegrationofinformationCognitive Psychology24175-219
Klieger S B Horowitz T S amp Wolfe J M (2004)Ismultipleobjecttrackingcolorblind[Abstract]Journal of Vision4(8)363a
Leonard C amp Pylyshyn Z W (2003)Measuringtheattentionaldemandofmultipleobjecttracking(MOT)[Abstract]Journal of Vi-sion3(9)582a
Logie R H (1995) Visuo-spatial working memory Hove UKErlbaum
Milliken B Tipper S P amp Weaver B (1994)NegativepriminginaspatiallocalizationtaskFeaturemismatchinganddistractorin-hibitionJournal of Experimental Psychology Human Perception amp Performance20624-646
Mitroff S R Scholl B J amp Wynn K (2004)DivideandconquerHowobjectfilesadaptwhenapersistingobjectsplitsintotwoPsy-chological Science15420-425
Ogawa H amp Yagi A (2003)Primingeffectsinmultipleobjecttrack-ingAnimplicitencodingbasedonglobalspatiotemporalinformation[Abstract]Journal of Vision3(9)339a
Oksama L amp Hyoumlnauml J (2004)Ismultipleobjecttrackingcarriedoutautomaticallybyanearlyvisionmechanismindependentofhigher-ordercognitionAnindividualdifferenceapproachVisual Cognition11631-671
Oliva A Torralba A Castelhano M S amp Henderson J M (2003)Top-downcontrolofvisualattentioninrealworldscenes[Ab-stract]Journal of Vision3(9)3a
Olson I R amp Marshuetz C (2005)RememberingldquowhatrdquobringsalongldquowhererdquoinvisualworkingmemoryPerception amp Psychophysics67185-194
Parasuraman R (1986)VigilancemonitoringandsearchInKRBoffLKaufmanampJPThomas(Eds)Handbook of perception and human performance Vol 2 Cognitive processes and performance(pp1-39)NewYorkWiley
Parmentier F B R Elford G amp Mayberry M (2005)Transi-tionalinformationinspatialserialmemoryPathcharacteristicsaffectrecallperformanceJournal of Experimental Psychology Learning Memory amp Cognition31412-427
Pelli D G (1997)TheVideoToolboxsoftwareforvisualpsychophysicsTransformingnumbersintomoviesSpatial Vision10437-442
Postle B R amp DrsquoEsposito M (1999)ldquoWhatrdquondashthenndashldquowhererdquoinvi-sualworkingmemoryAnevent-relatedfMRIstudyJournal of Cog-nitive Neuroscience11585-597
Postle B R DrsquoEsposito M amp Corkin S (2005)Effectsofver-balandnonverbalinterferenceonspatialandobjectvisualworkingmemoryMemory amp Cognition33203-212
Pylyshyn Z [W] (1989)Theroleoflocationindexesinspatialper-ceptionAsketchoftheFINSTspatial-indexmodelCognition3265-97
Pylyshyn Z W (2004)SomepuzzlingfindingsinmultipleobjecttrackingITrackingwithoutkeepingtrackofobjectidentitiesVisual Cognition11801-822
Pylyshyn Z W amp Leonard C (2003)Inhibitionofnontargetsdur-ingmultipleobjecttracking(MOT)[Abstract]Journal of Vision3(9)585a
Pylyshyn Z W amp Storm R W (1988)Trackingmultipleindepen-denttargetsEvidenceforaparalleltrackingmechanismSpatial Vi-sion3179-197
Saiki J (2002)Multiple-objectpermanencetrackingLimitationinmaintenanceandtransformationofperceptualobjectsProgress in Brain Research140133-148
Schneider W amp Shiffrin R M (1977)ControlledandautomatichumaninformationprocessingIDetectionsearchandattentionPsychological Review841-66
Scholl B J amp Pylyshyn Z W (1999)Trackingmultiple itemsthroughocclusionCluestovisualobjecthoodCognitive Psychology38259-290
Scholl B J Pylyshyn Z W amp Feldman J (2001)Whatisavi-sualobjectEvidencefromtargetmerginginmultipleobjecttrackingCognition80159-177
Scholl B J Pylyshyn Z W amp Franconeri S L (1999May)When are spatiotemporal and featural properties encoded as a result of attentional allocationPaperpresentedattheconferenceoftheAssociationforResearchinVisionandOphthalmologyFortLau-derdaleFL
Sears C R amp Pylyshyn Z W (2000)MultipleobjecttrackingandattentionalprocessingCanadian Journal of Experimental Psychol-ogy541-14
Seiffert A E (2003)Dissociatingneuralcorrelatesofattentionaltrackingandattentiontovisualmotion[Abstract]Journal of Vision3(9)868a
Smith E E Jonides J Koeppe R A Awh E Schumacher E H amp Minoshima S (1995)SpatialversusobjectworkingmemoryPETinvestigationsJournal of Cognitive Neuroscience7337-356
Tipper S P (1985)ThenegativeprimingeffectInhibitoryprimingbyignoredobjectsQuarterly Journal of Experimental Psychology37A571-590
Ungerleider L G amp Mishkin M (1982)Twocorticalvisualsys-
184 HOrOwiTz eT al
temsInDJIngleMAGoodaleampRJWMansfield(Eds)Anal-ysis of visual behavior(pp549-586)CambridgeMAMITPress
Viswanathan L amp Mingolla E (2002)DynamicsofattentionindepthEvidence frommulti-element trackingPerception311415-1437
Vogel E K Woodman G F amp Luck S J (2001)Storageoffea-turesconjunctionsandobjectsinvisualworkingmemoryJournal of Experimental Psychology Human Perception amp Performance2792-114
Wheeler M E amp Treisman A M (2002)Bindinginshort-termvisualmemoryJournal of Experimental Psychology General13148-64
Wilson F A Oacute Scalaidhe S P amp Goldman-Rakic P S (1993)Dissociationofobjectandspatialprocessingdomains inprimateprefrontalcortexScience2601955-1958
Wolfe J M Oliva A Horowitz T S Butcher S J amp Bompas A (2002)SegmentationofobjectsfrombackgroundsinvisualsearchtasksVision Research422985-3004
Yantis S (1992)MultielementvisualtrackingAttentionandpercep-tualorganizationCognitive Psychology24295-340
NoTeS
1Wearemissingageinformationfor1observer2Threeobserverswhoparticipatedinbothexperimentswereelimi-
natedfromthisanalysis3Measuredcapacitycanvarybasedontrackingdurationtherange
ofmovementspeedsandwhetherornotobjectscanoccludeonean-other(asinourexperiment)orbounceoffoneanother(asinOksamaampHyoumlnauml2004)
APPeNdix A Stimulus List
beaver fox horse ostrich rhino turtlecamel giraffe kangaroo pelican rooster wolfcrocodile goat lion rabbit squirrel zebraelephant gorilla moose tiger
APPeNdix b raw data
Table b1 raw Numbers of items reported in the Standard Condition
Total Targets Nontargets Filled
Experiment1 308Experiment2 344 014 359Experiment3 347 015 362Experiment4 339 015 354Experiment5 Unique 312 037 349
Paired 249 080 329Identical 261 086 347
Experiment6 248 078 326Experiment7 599 599
NotemdashTargetsindicatestheaveragenumberofcorrectlyreportedtargetspertrialNontargetstheaveragenumberoftimesobserversclickedonanontargetlocationpertrialandTotal Filledthesumofthesetwocat-egoriesNontargetclickcountswerenotcollectedforExperiment1andtherewerenonontargetsinExperiment7
Table b2 raw Proportions for the Specific Condition by experiment
Errors
Correct Empty Trialsper Responses Transposition Nontarget Space N Observer
Experiment1 44 14 16 26 8 5000Experiment2 56 18 10 16 8 5000Experiment3 68 12 10 10 8 5000Experiment4 58 25 09 08 8 5063Experiment6 41 14 21 23 7 5000Experiment7 33 39 00 28 8 5000
NotemdashTranspositionerrorsrefertotrialsinwhichtheobserverclickedonthewrongtargetlocationTrialsperobservervariedinExperiment4becauseconditionwasselectedrandomlyfromtrialtotrial
(ManuscriptreceivedMarch312005revisionacceptedforpublicationFebruary132006)
184 HOrOwiTz eT al
temsInDJIngleMAGoodaleampRJWMansfield(Eds)Anal-ysis of visual behavior(pp549-586)CambridgeMAMITPress
Viswanathan L amp Mingolla E (2002)DynamicsofattentionindepthEvidence frommulti-element trackingPerception311415-1437
Vogel E K Woodman G F amp Luck S J (2001)Storageoffea-turesconjunctionsandobjectsinvisualworkingmemoryJournal of Experimental Psychology Human Perception amp Performance2792-114
Wheeler M E amp Treisman A M (2002)Bindinginshort-termvisualmemoryJournal of Experimental Psychology General13148-64
Wilson F A Oacute Scalaidhe S P amp Goldman-Rakic P S (1993)Dissociationofobjectandspatialprocessingdomains inprimateprefrontalcortexScience2601955-1958
Wolfe J M Oliva A Horowitz T S Butcher S J amp Bompas A (2002)SegmentationofobjectsfrombackgroundsinvisualsearchtasksVision Research422985-3004
Yantis S (1992)MultielementvisualtrackingAttentionandpercep-tualorganizationCognitive Psychology24295-340
NoTeS
1Wearemissingageinformationfor1observer2Threeobserverswhoparticipatedinbothexperimentswereelimi-
natedfromthisanalysis3Measuredcapacitycanvarybasedontrackingdurationtherange
ofmovementspeedsandwhetherornotobjectscanoccludeonean-other(asinourexperiment)orbounceoffoneanother(asinOksamaampHyoumlnauml2004)
APPeNdix A Stimulus List
beaver fox horse ostrich rhino turtlecamel giraffe kangaroo pelican rooster wolfcrocodile goat lion rabbit squirrel zebraelephant gorilla moose tiger
APPeNdix b raw data
Table b1 raw Numbers of items reported in the Standard Condition
Total Targets Nontargets Filled
Experiment1 308Experiment2 344 014 359Experiment3 347 015 362Experiment4 339 015 354Experiment5 Unique 312 037 349
Paired 249 080 329Identical 261 086 347
Experiment6 248 078 326Experiment7 599 599
NotemdashTargetsindicatestheaveragenumberofcorrectlyreportedtargetspertrialNontargetstheaveragenumberoftimesobserversclickedonanontargetlocationpertrialandTotal Filledthesumofthesetwocat-egoriesNontargetclickcountswerenotcollectedforExperiment1andtherewerenonontargetsinExperiment7
Table b2 raw Proportions for the Specific Condition by experiment
Errors
Correct Empty Trialsper Responses Transposition Nontarget Space N Observer
Experiment1 44 14 16 26 8 5000Experiment2 56 18 10 16 8 5000Experiment3 68 12 10 10 8 5000Experiment4 58 25 09 08 8 5063Experiment6 41 14 21 23 7 5000Experiment7 33 39 00 28 8 5000
NotemdashTranspositionerrorsrefertotrialsinwhichtheobserverclickedonthewrongtargetlocationTrialsperobservervariedinExperiment4becauseconditionwasselectedrandomlyfromtrialtotrial
(ManuscriptreceivedMarch312005revisionacceptedforpublicationFebruary132006)