how does practice reduce dual-task interference ...ruthruff/ruthruff single_task_prac.pdf · how...

Click here to load reader

Upload: truongdang

Post on 27-Jul-2018

216 views

Category:

Documents


0 download

TRANSCRIPT

  • PsychologicalResearchPsychologischeForschungSpringer-Verlag2004DOI10.1007/s00426-004-0192-7

    Original Article

    How does practice reduce dual-task interference:Integration, automatization, or juststage-shortening?EricRuthruff()MarkVanSelstJamesC.JohnstonRogerRemington

    E.RuthruffJ.C.JohnstonR.RemingtonMS2624,NASAAmesResearchCenter,MoffettField,CA94035,USA

    M.VanSelstDepartmentofPsychology,SanJosStateUniversity,SanJos,CA95192,USA

    E.RuthruffPhone:+1-650-6040343Fax:+1-650-6040801E-mail:[email protected]

    Received:29December2003/Accepted:31August2004

    AbstractThepresentstudyassessedthreehypothesesofhowpracticereducesdual-task

    interference:Practiceteachesparticipantstoefficientlyintegrateperformanceofataskpair;practice

    promotesautomatizationofindividualtasks,allowingthecentralbottlenecktobebypassed;

    practiceleavesthebottleneckintactbutshorterinduration.Thesehypothesesweretestedintwo

    transfer-of-trainingexperiments.Participantsreceivedoneofthreetrainingtypes(Task1only,or

    Task2only,ordual-task),followedbydual-tasktestsessions.PracticeeffectsinExperiment1

    (Task1:auditoryvocal;Task2:visualmanual)werefullyexplainedbytheintactbottleneck

    hypothesis,withouttaskintegrationorautomatization.Thishypothesisalsoaccountedwellfor

    themajorityofparticipantswhenthetaskorderwasreversed(Experiment2).Inthiscase,however,

    thereweremultipleindicatorsthatseveralparticipantshadsucceededineliminatingthebottleneck

    byautomatizingoneorbothtasks.Neitherexperimentprovidedanyevidencethatpracticepromotes

    efficienttaskintegration.

    1

  • IntroductionAlthoughpeopletypicallyhavedifficultyperformingtwochoice-responsetasksatthesametime,

    thisinterferencecanbedramaticallyreducedwithpractice.Isthisreductionduetoaqualitative

    changeintaskperformance?Specifically,doespracticeeliminatethecentralbottleneckwidely

    believedtounderliedual-taskinterference?Thereareseveralwaysinwhichthismighthappen.

    Ifthebottleneckoccursbecausecentraloperationsarecarriedoutbyscarcegeneral-purpose

    resources,practicemightallowpeopletoperformtaskswithoutthoseresources(task

    automatization).Forinstance,practicemightproducejumpercablepathwaysdirectlylinking

    stimuliandresponses(Johnston&Delgado,1993).Alternatively,practicemightallowanefficient

    integrationofthetwotasks.Forinstance,practicemightallowparticipantstore-organizetwo

    tasksintoasinglesuper-task,thuseliminatingresourcecompetition.

    Theplausibilityoftaskautomatizationandtaskintegrationprovidesmotivationtolookfor

    qualitativechangesinperformancewithpractice,butitisbynomeansobviousthatsuchchanges

    actuallyoccur.Itispossiblethatpracticeinsteadleavesthebottleneckintactbutshorterinduration

    duetostage-shortening.Todeterminehowpracticereducesdual-taskinterference,thepresent

    studyusedatransfer-of-trainingparadigm,comparingtheeffectsofsingle-taskpracticeand

    dual-taskpracticeonsubsequentdual-taskperformance.Lookingahead,theresultsobtainedwith

    thetaskorderadoptedbymanypreviousdual-taskstudies(Task1:auditoryvocal;Task2:

    visualmanual)canbeexplainedentirelybythehypothesisthatpracticeleavesthebottleneck

    intactbutshorterinduration.Withtheopposite(rarelystudied)taskorder,asubstantialmajority

    ofparticipantsagainshowedanintactbottleneck.However,aminorityofparticipantssucceeded

    ineliminatingthebottleneck,automatizingoneorbothtasks.

    Background

    The psychological refractory period

    Manyearlydual-taskstudiesusedaccuracyastheprimarydependentmeasure(e.g.,Allport,

    Antonis,&Reynolds,1972;Hirst,Spelke,Reaves,Caharack,&Neisser,1980).Accuracy,however,

    maybeinsensitivetocapacitylimitations.Evenastrictinabilitytoperformcentralstagesofthe

    twotasksinparallelwouldnotnecessarilycausemanyerrors.Peoplemightbeabletobuffer

    informationononetask,switchtoasecondtaskforaperiodoftime,andlaterretrieveenoughof

    2

  • thebufferedinformationfromthefirsttasktorespondaccurately(seeShaffer,1975).Although

    thisbuffer-and-switchstrategymightallowparticipantstoavoidmakingoverterrors,itwould

    increaseresponsetime(RT)tooneorbothtasks(Pashler&Johnston,1989).Forthisreason,

    recentdual-taskresearchhasemphasizedparadigmsthatallowprecisemeasurementofRT.

    ByfarthemostwidelyusedRTparadigmindual-taskresearchisthepsychologicalrefractory

    period(PRP)paradigm,inwhichthekeyindependentvariableisthestimulusonsetasynchrony

    (SOA)betweenthetwotasks.LongSOAs,wheretemporaloverlapbetweentasksisminimal,

    provideabaselinemeasurementofRTforeachtaskperformedseparately.ShortSOAs,where

    temporaloverlapishigh,provideanopportunitytoobservewhetherparallelcentralprocessing

    hasoccurred.ThetypicalPRPfindingisadramaticincreaseinRTtothesecondtask(RT2)at

    shortSOAsrelativeto(baseline)longSOAs.ThisPRPeffectisremarkablyrobust,occurringeven

    withverysimplepairsoftasksintheabsenceofinputoroutputmodalityconflicts(forreviews

    seeLien&Proctor,2002;Pashler&Johnston,1998).

    ThereissubstantialevidencethattheprimarycauseofthePRPeffectisacentralbottleneck,

    requiringcentralstagessuchasresponseselectiontoproceedononlyonetaskatatime(Davis,

    1957;Welford,1952).Figure1showsageneralizedbottleneckmodeloftheshortSOAcondition,

    undertheassumptionthatTask1bottleneckoperationsareperformedbeforethoseofTask2.

    Eachtaskisdividedintothreeprocessingstages:Pre-bottleneck(A),bottleneck(B),and

    post-bottleneck(C).Byhypothesis,stagesAandCcanproceedinparallelwithanystageon

    anothertask,butstageB(thebottleneckstage)proceedsononlyonetaskatatime.Thus,atshort

    SOAs,theTask2bottleneckstage(2B)mustwaitfortheTask1bottleneckstage(1B)tofinish.

    ThiswaitingtimeproducesaPRPeffectequalto1A+1B2ASOAshort(seePashler&Johnston,

    1989).Becausethisequationdoesnotcontainatermforstage2B,thePRPeffectshouldnot

    dependonthedurationoftheTask2bottleneckstage.Thisbottleneckmodelpredictionandmany

    othershavebeenconfirmedinseveralstudies(e.g.,McCann&Johnston,1992;Pashler&Johnston,

    1989).

    Fig.1Stage-timediagramsforthe0-msstimulusonsetasynchrony(SOA),accordingtoageneralized

    bottleneckmodel.Processingisdividedintothreestages,arbitrarilylabeledA,B,andC.StagesAandCof

    3

  • onetaskcanproceedinparallelwithanystageoftheothertask.StageB(thebottleneckstage),however,

    canproceedononlyonetaskatatime.

    AlthoughitisclearthatacentralbottleneckoccursinthePRPparadigm,MeyerandKieras

    (1997a,1997b)havearguedthatthebottleneckmayreflectstrategicchoicesratherthanstructural

    limitations(seealsoSchumacheretal.,2001).Ifthebottleneckisstrategic,thenthereshouldbe

    conditionsunderwhichpeoplechoosetoperformtasksinparallel.Tolookforsuchconditions,

    Ruthruff,Pashler,&Klaassen(2001)explicitlyencouragingprocessingoverlap,placedequal

    emphasisoneachtask,askedparticipantstogrouptheirresponses(i.e.,emitthematthesame

    time),minimizedinputandoutputconflicts,andpresentedbothstimuliatthesametime.Despite

    theseefforts,dual-taskcostswereaboutaslargeasthosetypicallyobservedinthePRPparadigm.

    Severalotherrecentstudieshavealsoarguedthatthecentralbottleneckisstructuralratherthan

    strategic,atleastatrelativelylowpracticelevels(Levy&Pashler,2001;Pashler,1994;Ruthruff,

    Miller,&Lachmann,1995,Experiment3;Ruthruffetal.,2001;Ruthruff,Pashler,&Hazeltine,

    2003;Tombu&Jolicoeur,2000).

    Can practice eliminate the central bottleneck?

    Althoughthecentralbottleneckappearstobestructural,itmightneverthelessbepossibleto

    overcomeitwithpractice.Untilrecently,onlyafewPRPstudieshadexaminedtheeffectsof

    extensivepractice.ThesestudiesreportedsurprisinglylittleeffectofpracticeonPRPinterference

    (e.g.,Bertelson&Tisseyre,1969;Davis,1956,1957;Dutta&Walker,1995;Hick,1948;Karlin

    &Kestenbaum,1968;VanSelst&Jolicoeur,1997).KarlinandKestenbaum,forinstance,reported

    residualPRPeffectsgreaterthan200ms,evenafterseventoeightpracticesessions.VanSelst

    andJolicoeur(1997)laterreplicatedthisaspectofKarlinandKestenbaumsresults.

    TheresultsoftheseearlystudiessuggestedthatPRPinterferenceisremarkablyresistantto

    practice.VanSelst,Ruthruff,andJohnston(1999)arguedthatthisconclusionwaspremature,

    however,becausethesepreviousPRPstudiesrequiredmanualresponsestobothtasks.Giventhe

    likelihoodofstrongoutputinterferencebetweentwomanualresponses(eitherinprogrammingor

    execution;seeDeJong,1993;Keele,1973),thesestudiesmighthaveprovidedlessthanideal

    conditionsforobservingeffectsofpracticeoncentralinterference.Toaddressthisconcern,Van

    Selstetal.(1999)conductedaPRPpracticestudyusingtaskpairsthatminimizedoutput(and

    input)interference.Onetaskmappedauditorystimulitovocalresponses(high-andlow-pitched

    tonestothespokenwordshighandlow)andtheothertaskmappedvisualstimulitomanual

    4

  • responses(alphanumericcharacterstokeypresses).Contrarytopreviousstudies,practice

    dramaticallyreduceddual-taskinterference,from353msinthefirstsessiontoonly40msinthe

    18thsession.Thismorerecentfindingdemonstratesthatpracticecanindeedsubstantiallyreduce

    PRPinterference,providedthatinputandoutputmodalityconflictshavebeeneliminated(seealso

    Hazeltine,Teague,&Ivry,2002;Schumacheretal.,2001).

    Whatcausesthedramaticreductionindual-taskinterferencewithpractice?Inparticular,does

    practiceactuallyeliminatethebottleneck?Belowweoutlinethreeplausiblehypotheses,eachof

    whichfocusesonadifferentconsequencethatpracticemighthave.Aswillbeseen,thesemodels

    makedistinctpredictionsaboutwhattypesofpractice(e.g.,single-taskversusdual-task)should

    bemosteffectiveinreducingdual-taskinterference.

    First,dual-taskpracticemightproduceanefficientintegrationofthetwotasks(see,e.g.,Hirst

    etal.,1980;Spelke,Hirst,&Neisser,1976),increasingthedegreetowhichcentralstagescan

    proceedinparallel.Asanextremeexample,participantsmightlearntocombinethetwotasksinto

    asinglesuper-task(Kahneman,1973),sothatasinglecognitiveoperationjointlyselectsresponses

    tobothtasksatthesametime.Accordingtothistaskintegrationhypothesis,integrationrequires

    dual-taskpractice.Thus,dual-taskpracticeshouldproducePRPreductionaboveandbeyondwhat

    ispossiblewithsingle-taskpractice.

    Second,practicemightautomatizetheperformanceofindividualtasks.Forthepresentpurposes,

    wedefinetaskautomatizationastheeliminationofdemandsonthelimitedcentralresources

    responsibleforthecentralbottleneck.Forinstance,practicemightallowthedevelopmentof

    jumpercablepathwaysdirectlylinkingstimuliandresponses(Johnston&Delgado,1993).If

    eithertaskisautomatized,thenthereshouldbenoconflictforlimitedcentralresourcesandhence

    nobottleneck.Accordingtothistaskautomatizationhypothesis,therefore,completebypassof

    thebottleneck(i.e.,parallelprocessingofcentraloperations)shouldoccurfollowingeither

    single-taskordual-taskpractice.

    Third,practicemightcausenoqualitativechangeinperformancenotaskintegrationandno

    taskautomatizationleavingthebottleneckintact.Eveninthiscase,practicewouldreduce

    dual-taskinterferencebyshorteningstagedurations.Thisintactbottleneck(withstage-shortening1)

    1Stage-shorteningisaninevitableconsequenceoftaskpractice.Itplaysacriticalroleintheintact

    bottleneckhypothesisinfact,itisthesolecauseofPRPreduction.Althoughthetaskintegration

    andtaskautomatizationhypothesesalsoallowforstage-shortening,itisnotassumedtobethe

    5

  • hypothesismakesthesamepredictionsforthePRPeffectthatallbottleneckmodels(seeFig.1)

    make:PRPeffect=1A+1B2ASOA(Pashler&Johnston,1989).Thisequationindicates

    thatthePRPeffectcanbereducedonlybyshorteningstages1Aor1B,whichcanoccureither

    withdual-taskpracticeorwithsingle-taskpracticeonTask1.Single-taskpracticeonTask2should

    notreducethePRPeffect(notethatreductionsin2AwouldactuallyincreasethePRPeffect).

    Althoughthepredictionsofourthreemainhypotheseshaveneverbeentestedinanappropriate

    transfer-of-trainingdesign,previouspracticestudiesprovidesomeusefulclues.VanSelstetal.

    (1999)confirmedseveralbottleneckmodelpredictionsregardingfactorinteractionswithSOA

    (seebelowformoredetails)afterpractice.Theirresultsprovidesomesupportfortheintact

    bottleneckhypothesis.However,VanSelstetal.didnotincludeasingle-taskpracticecondition,

    soadirecttestbetweenthecompetingmodels(taskintegration,taskautomatization,intact

    bottleneck)wasnotpossible.

    Ruthruff,Johnston,andVanSelst(2001)transferredthehighly-practicedVanSelstetal.

    participantstonewdual-taskconditionsinvolvingeithertheoldTask2pairedwithanewTask1

    (Experiment1),oranewTask2pairedwiththeoldTask1(Experiment2).Althoughtheirfinding

    ofbettertransferwiththeoldTask1thanwiththeoldTask2providessomesupportfortheintact

    bottleneckhypothesis,theirdesignhasseriouslimitations.Specifically,thetwotransferconditions

    hadneitherTask1norTask2incommon.Hence,theresultsmightbeduetospecialproperties

    ofthetesttasks(e.g.,onetaskpairmighthavebeenmoredifficultthantheother).Thisproblem

    isanunfortunatebyproductofusingthesamecohortofparticipantsthroughoutalltheexperiments

    (takingadvantageoftheirheroiclevelsofcumulativepractice).Asexplainedinthenextsection,

    thepresentstudyavoidedtheseproblemsbyusingdifferentcohortsofparticipantsfordifferent

    trainingconditions.

    The present studyThegoalofthepresentstudywastoevaluatethetaskintegration,taskautomatization,andintact

    bottleneckhypothesesofhowpracticereducesdual-taskinterference.Wetestedpredictionsof

    thesethreehypothesesusingthetransfer-of-trainingparadigmsummarizedinTable1.Participants

    receivedoneofthreetypesoftraining:

    [Table1willappearhere.Seeendofdocument.]

    principlecauseofPRPreduction(e.g.,ifthebottleneckisbypassed,theninterferencewillbesmall

    regardlessofstagedurations).

    6

  • a)Single-taskpracticeonTask1

    b)Single-taskpracticeonTask2

    c)Dual-task(PRP)practice

    Followingeighttrainingsessions,allthreegroupsthencompletedfourdual-task(PRP)test

    sessions.Importantly,thedual-tasktestsessionswereidenticalforthethreetraininggroups.

    Assessing the consequences of practice

    Weanalyzedtwopossibleconsequencesofpractice:ReductionsinthesizeofthePRPeffect,and

    eliminationofthebottleneck.AssessmentofPRPreductionisstraightforward,butassessmentof

    bottleneckeliminationisnot.Somepreviousdual-taskstudieshaveusedtheamountofdual-task

    costasaproxyforbottleneckelimination,assumingthatsmallPRPeffectsdirectlyindicatethe

    eliminationofthebottleneckandlargecostsdirectlyindicatethecontinuedexistenceofthe

    bottleneck.Unfortunately,theabsoluteamountofdual-taskinterferenceisnot,byitself,a

    completelyreliablediagnostic.EventhebottleneckmodelpredictsthatPRPeffectsshoulddecline

    withpractice.Infact,PRPeffectscouldapproachzero,ifTask1centraloperationsarefinished

    veryquickly,beforeTask2centraloperationsaresettobegin.Insuchcases,thebottleneckissaid

    tobelatent,producingnoobservableRTdelays(seeRuthruff,Johnston,VanSelst,Whitsell,

    &Remington,2003).

    Weusedseveraldifferentmeasurestoassesswhetherthebottleneckwasinfacteliminated.

    First,asexplainedinmoredetailbelow,weexaminedwhetherthePRPeffectwasproportional

    toRT1(accordingtothebottleneckmodel,PRP=RT11C2ASOA).Wealsoexamined

    severalspecificbottleneckmodelpredictionsforhowtheeffectsofvarioustaskdifficultyfactors

    shouldinteractwiththeeffectsofSOA(seebelow).Asanadditionalindicatorofwhethera

    bottleneckoccurred,weexaminedtheorderofresponseswithinatrial.Accordingtothecentral

    bottleneckmodel,theovertTask1responseshouldprecedetheTask2responseonalmostevery

    trial.Butifparticipantsperformcentraloperationsinparallel(asallowedbythetaskintegration

    andtaskautomatizationhypotheses),responsesmightoccurineitherorder.

    Summary of key predictions

    Thetaskintegrationhypothesispredictsthatdual-taskpracticeshouldbemuchmoreeffective

    thansingle-taskpractice,reducingandpossiblyeveneliminatingthebottleneck.Thetask

    automatizationhypothesispredictseliminationofthebottleneck,withlittleornoresidualPRP

    7

  • effect,followingTask1,Task2,ordual-taskpractice.Theintactbottleneckhypothesispredicts

    thatthebottleneckshouldpersistwithpractice(albeitshorterindurationduetostage-shortening),

    andthatPRPeffectsshouldbemuchsmallerfollowingeitherdual-taskorTask1practicethan

    followingTask2practice.

    Experiment 1AsinVanSelstetal.(1999),Task1requiredparticipantstosayloworhighinresponseto

    low-pitchedorhigh-pitchedtonesandTask2requiredparticipantstopressakeycorresponding

    totheidentityofanalphanumericcharacter(1,2,3,4,A,B,C,orD).Forhalfoftheparticipants,

    thelettersweremappedcompatiblyontothefourresponsekeysandthedigitsweremapped

    incompatibly.Fortheotherhalfofparticipants,theassignmentswerereversed.

    FollowingVanSelstetal.(1999),wemanipulatedthreetaskdifficultyfactors:

    a)Tonetaskdifficulty(whetherthetonepitchwasnearorfarfromthehigh/lowboundary)

    b)Charactertaskcontrast

    c)Charactertaskstimulusresponse(SR)compatibility

    Ifthebottleneckremainsintactwithpractice:

    a)Task1difficultyeffectsshouldcarryoverfullyontoRT2atshortSOAs

    b)Task2characterintensityeffectsonRT2shouldbesmalleratshortSOAs(duetoabsorption

    intocognitiveslack)thanatlongSOAs

    c)Task2SRcompatibilityeffectsonRT2shouldberoughlythesameatallSOAs

    Formoredetaileddiscussionofthesepredictions,seeVanSelstetal.(1999;seealsoPashler

    &Johnston,1989,1998;Schweickert,1978).

    Method

    Participants

    Theparticipantswere18studentsrecruitedfromcollegesanduniversitiesintheMountainView,

    California,areaandfromwork-studyprogramsatNASAAmesResearchCenter.Eachofthe18

    participantsperformed12sessions,foragrandtotalof216sessions.Theywerepaidfortheir

    participation.

    8

  • Stimuli

    ThestimulusforTask1wasoneoffourtonespresentedfor150ms.Thetwotoneshighestin

    pitch(3,125and625Hz)wereclassifiedashightones,andthetwotoneslowestinpitch(400

    and80Hz)wereclassifiedaslowtones.ThestimulusforTask2wasasinglealphanumeric

    characterdrawnatrandomfromtheset{1,2,3,4,A,B,C,D}.Thecharacterswerepresentedin

    TimesRomanfontataviewingdistanceofabout65cm.Allcharactersfittedintoarectangular

    areaof1.41.94visualangle.Thebackgroundwaswhite;thecharacterswereblack(high

    contrastcondition)onhalfofthetrialsandgray(lowcontrastcondition)ontheotherhalfofthe

    trials.

    Apparatus

    StimuluspresentationandtimingwereperformedbyIBMPC-compatiblecomputersequipped

    withaSchmitt-triggervoicekeyfordetectingspeechonsetandaVoiceConnexionsystem(external)

    forrecognizingspeech.

    Procedure

    Participantswereinstructedtorespondtohightonesbysayinghighandtolowtonesbysaying

    low.Theywereinstructedtorespondtotheidentityofthealphanumericcharacterbypressing

    theh,j,korlkeysonastandardkeyboard,usingthefingersoftheirrighthand.Forhalf

    oftheparticipantsineachtraininggroup,thelettersA,B,C,andDweremappedinalphabetical

    orderontothefourresponsekeysfromlefttoright(i.e.,compatibly).Thenumbers,meanwhile,

    weremappedinanincompatible,non-sequentialorder(3,1,4,2)ontothesamefourresponse

    keys.Fortheremainingparticipants,numbersweremappedcompatibly(1,2,3,4)butletterswere

    mappedincompatibly(C,A,D,B).Theinstructionsemphasizedtheimportanceofresponding

    quicklyandaccurately.

    Eachparticipantwasrandomlyassignedtooneofthethreetraininggroups.TheTask1training

    groupperformedonlyTask1(thetonetask)duringtheeighttrainingsessionsandtheTask2

    traininggroupperformedonlyTask2(thecharactertask).Thedual-tasktraininggroup,meanwhile,

    performedbothTask1andTask2inaPRPdesign.Allthreegroupsperformedthesamenumber

    oftrialspertrainingsession(560),althoughthedual-taskgroupmadetworesponsespertrialand

    thesingle-taskgroupsmadeonlyoneresponsepertrial.Thus,thedual-taskgroupperformed

    Task1asoftenastheTask1traininggroupandperformedTask2asoftenastheTask2training

    group.

    9

  • Aftertheeighttrainingsessions,allparticipantsthenperformedfourdual-task(PRP)test

    sessions(seeTable1).Thesedual-tasktestsessionswereidenticalforallthreegroups.Because

    twoofthetraininggroupshadtolearnanewtaskandanewparadigm,thefirsttestsessionwas

    consideredpractice.Participantswereinstructedtorespondtobothtasksquicklyandaccurately,

    buttoplaceparticularemphasisonthespeedofTask1responses.

    Eachsessionconsistedof35warm-uptrialsfollowedby525experimentaltrials.Thesession

    wasbrokenintosevenblocksof80trials,separatedbyshortbreaks.Duringeachbreak,the

    computerprovidedfeedbackontheaveragespeedofTask1andtheaccuracyofbothTask1and

    Task2.Eachtrialbeganwiththepresentationofafixationcrossfor500ms,followedbyarandom

    fore-period(100,150,200,or250ms).Inthedual-taskcondition,theTask1tonesoundedfor

    150msandtheTask2alphanumericcharacterappearedafteravariableSOA(17,67,150,250,

    450,or850ms).IntheTask1trainingcondition,onlythetonewaspresented.IntheTask2training

    condition,onlythecharacterwaspresented.ThetimingoftheTask2characterinthissingle-task

    conditionwasyokedtothatofthedual-taskcondition:Followingtherandomfore-periodthere

    wasanadditionalSOAdelay.Ifeitherresponsewasincorrect,amessageindicatedthetaskon

    whichtheerrorhadbeenmade.Also,iftheparticipantrespondedwithin100msofstimulusonset,

    aTOOEARLYmessagewasdisplayed.Iftheparticipantfailedtorespondwithin2,500msof

    stimulusonset,aTOOSLOWmessagewasdisplayed.Thefixationcrossforthenexttrialwas

    presented750mslater.

    Analyses

    WeconductedseparateANOVAsonmeanRT1,RT2,Task1errorrate,andTask2errorratein

    thetestsessions(sessions1012),usingthefactorsoftraininggroup(Task1only,Task2only,

    Dual-Task),SOA,Task1difficulty,Task2SRcompatibility,andTask2contrast.Whenassessing

    interactionsonRT2betweentask-difficultymanipulationsandSOA,wereliedonaseparate,and

    moresensitive,ANOVAthatincludedjusttheshortestandlongestSOAsratherthanallsixSOAs

    (whereadjacentSOAstendtoproduceverysimilarperformance).BeforeconductingRTanalyses,

    wefirsteliminatedtrialscontaininganerroroneithertask.Then,forboththeerrorandRTanalyses,

    weremovedalltrialswithanRToflessthan100ms,followedbytrialsinwhicheitherRT1or

    RT2wasidentifiedasanoutlier(lessthan5.6%oftrialsineachexperiment)usingamodified

    recursiveoutliereliminationprocedurewithmovingcriterion(describedinVanSelst&Jolicoeur,

    1994).Allanalysesusedanalphalevelof.05.

    10

  • Results

    Dual-task group only (training and test)

    Wefirstexamineseparatelytheresultsfromthedual-tasktraininggroup,whichrepresentsa

    replicationoftheVanSelstetal.(1999)experiment,albeitwithfewersessions(12vs.36)anda

    differentsetofsixparticipants.TheresultsforthisgroupareshowninFig.2(unfilleddiamonds)

    acrossthe12sessions,alongwiththeresultsofVanSelstetal.(1999;filledsquares).Eachpoint

    representsthemeanPRPeffect(definedasRT2attheshortestSOAminusRT2atthelongest

    SOA)forasessionplottedagainstthemeanRT1forthatsession.

    Fig.2Meanpsychologicalrefractoryperiod(PRP)effectforeachsessionasafunctionofmeanTask1

    responsetime(RT1)acrosssessionsinVanSelstetal.(1999;filledsquares),thepresentExperiment1

    (unfilleddiamonds),andthepresentExperiment2(asterisks).Thebestfittingregressionlinewithaslope

    of1.0isshownforeachexperiment.

    Ifpracticeprimarilyreducesthedurationofcentralstages(forsupportingevidenceseeFletcher

    &Rabbitt,1978;Mowbray&Rhoades,1959;Pashler&Baylis,1991;Welford,1976),theintact

    bottleneckhypothesismakesanelegantpredictionregardingthesizeofthePRPeffectandRT1

    acrosssessions.NotethatRT1equals1A+1B+1CandthePRPeffectequals

    1A+1BSOAshort2A(assumingthatabottleneckdelayalwaysoccursattheshortestSOAbut

    neveratthelongestSOA;Pashler&Johnston,1989).Theonlytermineachequationthat,by

    hypothesis,canvarysubstantiallywithpracticeisstage1B.Hence,practiceshouldreducethe

    PRPeffectandRT1byroughlythesameamountacrosssessions.Accordingly,agraphofthePRP

    effectagainstRT1acrosssessionsshouldhaveaslopeofabout1.0.AsshowninFig.2,theVan

    Selstetal.(1999;filledsquares)dataconfirmedthissimpleprediction(thebest-fittingslopewas

    1.02),supportingabottleneckmodelwithcentralstageshortening.

    11

  • Infittingthedatafromthedual-tasktraininggroupofthepresentExperiment1,session1was

    excludedbecausethemeanRT1wasverylong(1,091ms),indicatingaclearviolationofthe

    assumptionthatabottleneckdelayneveroccurredatthelongestSOA(850ms).AsshowninFig.2

    (unfilleddiamonds),declinesinthePRPeffectacrosssessionstrackeddeclinesinRT1

    approximatelyoneforone,replicatingthefindingofVanSelstetal.(1999).Thebest-fitting

    regressionlinewithaslopeof1.0,showninFig.2,wasPRP=RT1321.Allowingtheslopeto

    takeonanyvalue,thebest-fittingregressionlinewasPRP=.86*RT1249,buttheadditional

    varianceaccountedfor(beyondthemodelwiththefixed1.0slope)wasnotsignificant,

    F(1,9)=2.43,MSE=387.6,p>.05.

    AlthoughtherelationbetweenPRPandRT1isqualitativelysimilarbetweenthetwostudies,

    therearetwonotablequantitativedifferences.First,meanRT1foragivenlevelofpracticewas

    muchshorterinVanSelstetal.thaninthepresentstudy(e.g.,308vs.515msinsession12).Some

    ofthisdifferencecanbeattributedtogreatermotivationand/orexperienceoftheparticipantsin

    theVanSelstetal.study(1999),twoofwhomwereco-authorsofthepaper.Second,theVan

    Selstetal.participantsshowedslightlymorePRPeffectforagivenRT1,asreflectedinthePRP

    axisintercepts(243vs.321).Oneplausibleexplanationforthisdifferenceisthatthehighly

    motivatedVanSelstetal.participantsreducedRT1inpartbyspeakingtheresponsesveryabruptly.

    Theconsequentreductioninthedurationofstage1CwouldreduceRT1withoutreducingthePRP

    effect,ineffectproducingmorePRPeffectforagivenRT1.

    Training sessions

    Table2showsmeanRT1andRT2acrosstheeighttrainingsessionsforthethreegroups.To

    facilitatecomparisonswiththesingle-taskgroups,thedatafromthedual-tasktraininggroupare

    basedononlythelongestSOA(theleastinfluencedbydual-taskinterference).Notsurprisingly,

    RTdecreasedsteadilyacrosssessionsforeachtraininggroup.MeanRT1wasshorterforthe

    single-tasktraininggroupthanforthedual-tasktraininggroup,butmeanRT2wasnot;aswillbe

    seen,asimilareffectwasobservedinExperiment2aswell.Onepossibleexplanationisthat

    dual-taskparticipantsinitiallyprepareforTask1aswellasthesubsequentswitchtoTask2(De

    Jong,1995;Lien&Ruthruff,2004).Thisinitialspreadingofpreparationbetweenthetwotasks

    couldslowdual-taskRT1,butwouldnotslowdual-taskRT2atthelongestSOA(atthisSOA,

    participantscanprepareexclusivelyforTask2).

    [Table2willappearhere.Seeendofdocument.]

    12

  • Task1 RT in test sessions

    Figure3showsmeanRT1andRT2,averagedacrossthedual-tasktestsessions,asafunctionof

    SOAandtraininggroup.MeanRT1wasshortestforthegroupthathadpracticedTask1only

    (365.4ms),nextfastestforthegroupthathadpracticedbothTask1andTask2(522ms),and

    slowestforthegroupthathadnotpracticedTask1(575ms).Thesedifferencesweremarginally

    significantoverall,F(2,15)=3.2,MSE=3,183,749.1,.05

  • fortheTask2practicegroup,and221msforthedual-taskpracticegroup.Theinteractionbetween

    traininggroupandSOAwasstatisticallysignificant,F(10,75)=2.22,MSE=74,977.8,p

  • bottleneck.Furthermore,thispersonhadaverylongmeanRT1(990ms),whichislongenough

    toclearlyviolatethepreviouslynotedassumptionthatnobottleneckdelayoccurredatthelong

    SOA(requiredforthelinearPRP/RT1relationshiptohold).Afurtherreasontodoubtthatany

    individualinExperiment1eliminatedthebottleneckistheverylowrateofresponsereversals

    (lowsingledigitsforeachparticipant).

    Task1 carry-over prediction

    ThedataconfirmedtheintactbottleneckpredictionthatTask1difficultyeffectsshouldcarry

    overfullyontoRT2atshortSOAs,butnotatlongSOAs.Averagedacrossthethreetraining

    conditions,theeffectonRT1was78ms.ThiseffectcarriedoverfullyontoRT2atshortSOAs

    (85ms),butnotatlongSOAs(14ms),F(5,75)=11.0,MSE=10,036.0,p

  • monotonictrendinthepredicteddirection(witha56%reductionincontrasteffectsattheshortest

    SOAcomparedwiththelongestSOA);however,thedecreasewasonlymarginallysignificantin

    theANOVAcomparingjustthelongestandshortestSOAs,F(1,15)=3.53,MSE=6,148.1,

    .05.2,orwhencomparingjusttheshortestandlongestSOAs,F(1,15)=2.2,MSE=14,089.3,

    p>.1.Therewasalsonosignificantthree-wayinteractionbetweentraininggroup,Task2SR

    compatibility,andSOA,F(2,15)

  • compatible(2.2%),F(1,15)=28.0,MSE=344.5,p

  • additiontothisspecificconcernitis,ofcourse,genericallyvaluabletoreplicateresultsunder

    differentconditions.

    Toaddresstheseconcerns,wereversedthetaskorderinExperiment2:Thecharactertask

    becameTask1andthetonetaskbecameTask2.Inallotherrespects,thedesignwasidenticalto

    thatofExperiment1.Notethatmostpreviousstudiesusingauditoryvocalandvisualmanual

    taskshaveutilizedthetaskorderofExperiment1;studiesusingtheoppositeorderarerelatively

    rare.Accordingtotheintactbottleneckhypothesis,thePRPeffectshouldstillbesmallerfollowing

    practiceonTask1alonethanfollowingpracticeonTask2alone,eventhoughtheparticular

    judgments(charactervs.tone)assignedtothesetaskshavebeenexchanged.Furthermore,we

    shouldstillfindnoevidenceoftheeliminationofthebottleneck.

    Withthetaskorderreversed,abottleneckwouldproduceadifferentpatternofinteractionswith

    SOA.ThecharacterintensityandSRcompatibilitymanipulationeffectsarenowplantedinTask1

    and,therefore,shouldcarryoverfullyontoRT2atshortSOAs.Thebottleneckmodelpredictions

    forthedifficultyofthetonetask(nowTask2)arelessclear,becausetheprocessinglocusofthis

    effectisnotknown.IfthisfactorprimarilyinfluencesthecentralstageofTask2(i.e.,stage2B),

    theeffectsonRT2shouldbeconstantacrossSOAs.Ifthisfactorinsteadinfluencestheinputstage

    ofTask2(i.e.,stage2A),theeffectsonRT2shouldbesmalleratshortSOAsthanatlongSOAs

    (duetoabsorptionintocognitiveslack).

    Method

    Exceptforreversingtherolesofthetonetask(nowTask2)andthecharactertask(nowTask1),

    themethodwasidenticaltothatofExperiment1.The18participantsinthisexperimentdidnot

    participateinExperiment1.

    Results

    Dual-task group only (training test)

    WefirstreporttheresultsfromthegroupinExperiment2thatperformeddual-tasksessionsduring

    boththetrainingandtestphases,showninFig.2(asterisksymbols).DeclinesinthePRPeffect

    trackeddeclinesinRT1approximatelyoneforoneacrosssessions,justasfoundinExperiment1

    andVanSelstetal.(1999).Thebest-fittingregressionlinewithaslopeof1.0was

    PRP=RT1218;thislineisshowninFig.2.Ifweinsteadallowtheslopetotakeonanyvalue,

    thebestfittingregressionlineisPRP=1.06*RT1252.8.Allowingtheslopetovarydoesnot

    18

  • accountforasignificantamountofvariancebeyondthatalreadyaccountedforwithaslopefixed

    at1.0,F(1,9).05.Thesefindingsareconsistentwiththeintactbottleneck

    hypothesis.

    ThepointsforExperiment2generallylieabout100msabovethoseforExperiment1.Why

    mightthishaveoccurred?Apriori,fasterinputprocessingforTask2(fasterstage2A)in

    Experiment2thaninExperiment1(thatis,fastertoneclassificationthancharacterclassification)

    wouldhavebeenexpectedbecauseauditoryinformationshouldgettothecortexfasterthanvisual

    informationandbecauseaneasyphysicaldiscrimination(tonepitch)shouldbefasterthanone

    basedonlearnedcategories(characteridentification).Suchadifferenceinthedurationofstage2A

    wouldincreasethesizeofthePRPeffectforagivenRT1;hence,itwouldproducetheobserved

    differencesinFig.2.

    Training phase

    Table3showsmeanRT1andRT2acrosstheeighttrainingsessionsforthethreegroups.Not

    surprisingly,RTsdecreasedsteadilyacrosssessions.Also,meanRT1againtendedtobelonger

    forthedual-tasktraininggroupthanforthesingle-tasktraininggroup,whichmightreflectacost

    ofhavingtoinitiallypreparetwotasksratherthanjustone.

    [Table3willappearhere.Seeendofdocument.]

    Task1 RT in test sessions

    Figure6showsmeanRT1andRT2asafunctionofSOAforeachtraininggroup.Averagedacross

    thefourtestsessions,meanRT1wasfastestforthegroupsthathadpracticedTask1inthetraining

    phase(550msfortheTask1traininggroupand528msforthedual-tasktraininggroup),and

    slowestforthegroupthatdidnot(609msfortheTask2traininggroup).Withonlysixparticipants

    pergroup,however,thisdifferencedidnotquitereachstatisticalsignificanceacrossparticipants,

    F(2,15)=1.672350,MSE=904,083.5,p=.22.Afollow-upanalysisrevealedthattheeffectof

    previoustrainingonTask1RTwaslargeduringtheinitialtestsession,butthenshrankoverthe

    nextfewtestsessions.AsdiscussedintheTask1carry-oversectionbelow,meanRT1depended

    onbothTask1contrast,F(1,15)=147.0,MSE=4,336.3,p

  • tointerpretthem.Thesesamethreefactorsalsointeractedsignificantlywiththetypeoftraining,

    F(2,15)=3.8,p

  • butseveralclearlydidnot.Thislattersubsetofparticipants(filledsymbols)showsamuchsmaller

    PRPeffectthanwouldbepredictedbasedontheirmeanRT1,andthereforearecandidatestohave

    eliminatedthebottleneck.Wereturntothisissuelater,afterpresentingadditionalevidenceregarding

    factorinteractionswithSOA.

    Fig.7MeanPRPeffectversusmeanRT1foreachparticipantinExperiment2.Trianglesdual-tasktraining

    group,squaresTask1traininggroup,circlesTask2traininggroup.Thepointsrepresentingthe4participants

    suspectedofhavingbypassedthebottleneckarefilled;thepointsrepresentingtheother14participantsare

    unfilled.

    Task1 carry-over prediction

    Thebottleneckmodelpredictsthatmanipulationsofbottleneckorpre-bottleneckstagesofTask1

    shouldcarryoverfullyontoRT2atshortSOAsbutnotatlongSOAs.ThereweretwoTask1

    variablesinExperiment2thatwecanusetotestthecarry-overprediction:ContrastandSR

    compatibility.

    TheeffectofTask1contrastonRTtoTask1itselfwas31.4msanditseffectonRT2was10.6,

    20.3,23.1,18.0,22.9,and3.0msatthe17,67,150,250,450,or850msSOAsrespectively.Thus,

    contrarytothecentralbottleneckmodelprediction,theeffectofcontrastonTask1didnotappear

    tocarryoverfullyontoRT2attheshortSOAs.Furthermore,theinteractiononRT2between

    contrastandSOA(shortvs.long)wasnotstatisticallysignificant,F(1,15).2.

    BecausetheeffectofTask1SRcompatibilityonRT1(96ms)wasmuchlargerthantheeffect

    ofTask1contrast(31ms),itprovidesanevenbetteropportunitytoevaluatetheTask1carry-over

    prediction.Figure8showsthecarry-overofTask1compatibilityeffectsontoRT2foreachtraining

    21

  • group(filledsymbols)aswellasthemeanTask1compatibilityeffectonRT1(averagedacross

    SOAs;unfilledsymbols).Averagedacrossthethreetraininggroups,theeffectofTask1

    compatibilitywas85.2,71.4,61.8,66.9,41.8,and10.3msatthe17,67,150,250,450,or850ms

    SOAsrespectively.ThisinteractionbetweenTask1compatibilityeffectsandSOA(shortvs.long)

    wassignificant,F(1,15)=14.9,MSE=20,326.6,p

  • compatibility,Task2tonedifficulty,andtrainingtypewasalsosignificant,F(2,15)=3.72,

    MSE=51.8,p

  • Evidence of two distinct subgroups

    Ourfirstindicationthatsomeparticipantshadeliminatedthebottleneckcamefromananalysisof

    responseorder.BecausethebottleneckmodelassumesthatTask1centraloperationsareperformed

    beforeTask2centraloperations,theresponsesshouldbeemittedinthissameorder;thisprediction

    wasconfirmedinExperiment1andinmanypreviousPRPstudies.Withoutabottleneck,however,

    itisentirelypossibleforTask2responsestobeemittedbeforeTask1responsesattheshortest

    SOA;infact,becauseTask2tooklesstimethanTask1(whenperformedalone),responsereversals

    shouldoccurmoreoftenthannot.Indeed,fourparticipantsinExperiment2reversedresponses

    onaveryhighproportionoftrials(6698%),oftenrespondingtoTask2afewhundredmilliseconds

    beforeTask1.Noneoftheother14participantsreversedresponsesonmorethan8%oftrials.

    Thisenormousgapinthepercentageofresponsereversalsbetweenmembersofthetwogroups

    stronglysupportsacategoricaldifferencebetweenthesegroups.

    Thefourparticipantswithhighfrequenciesofresponsereversalsalsohappenedtoshow

    considerablysmallerPRPeffects(range:3367ms)thananyoftheremaining14participants

    (range:125505ms).Evenmoreimportantly,theyproducedsmallerPRPeffectsthanthebottleneck

    modelwouldpredictbasedontheirmeanRT1,byroughly300ms.InFig.7,thesefourparticipants

    correspondtothefourdatapoints(filledsymbols)thatliethefurthestbelowtheregressionline.

    TherewereafewadditionalparticipantswithPRPeffectsbelowtheregressionline,butnoneof

    themfrequentlyreversedtheresponseorder.Thus,thereareonlyfourparticipantswithclear

    evidenceofhavingeliminatedthebottleneck.Forthesakeofbrevity,wecallthesefourindividuals

    bypassersandtheremaining14participantsbottleneckers.

    Asanadditionalindicatorofwhetherthebottleneckwasbypassed,wecomputedtheamount

    ofTask1carry-overforthebypassersandbottleneckers.Ifthebottleneckhasbeeneliminated,

    thereisnoobviousreasonforTask1difficultyeffectstostronglyinfluenceRT2.Butifthereis

    abottleneck,theeffectsofthemanipulationsshouldcarryoverfullyontoRT2(asdiscussedabove).

    Forthebypassers,theeffectofTask1SRcompatibilityeffectwas62msonTask1,butonly

    13msonTask2.Forthebottleneckers,however,theTask1SRcompatibilityeffectonRT1was

    103ms,andthiseffectcarriedoverfullyontoRT2(109ms).Thesefindingsfurthersupportour

    classificationofparticipantsasbypassersandbottleneckers.Theyalsoexplainwhytheoverall

    datashowedsignsofincompletecarry-over:Thedatarepresentedamixtureoffourparticipants

    whobypassedthebottleneck(producingverylittlecarry-over)and14participantswhodidnot

    (producingcompletecarry-over).Table4summarizesthedifferentialamountsofcarry-over

    24

  • betweenthetwogroupsofparticipants,aswellastheotherprominentempiricaldifferences

    mentionedabove.

    [Table4willappearhere.Seeendofdocument.]

    Attempts to reconcile the bottleneck model with small PRP effects

    Torescuethebottleneckmodel,itmightbearguedthatthebypassersperformedTask1central

    operationsveryquickly,beforeTask2centraloperationswereevenreadytobegin(i.e.,the

    bottleneckwaslatent;Ruthruff,Pashler,&Hazeltine,2003).However,thisaccountisimplausible

    giventherelativelylongmeanRT1(563ms).Anotherpossibilityisthatthebypassersgenerally

    reversedthecentralprocessingorder(Task2beforeTask1).Thishypothesiscorrectlypredicts

    thehighfrequencyofresponsereversalsandtheneareliminationofinterferenceonTask2.

    However,itthenmustpredictdual-taskinterferenceonTask1.Contrarytothisprediction,mean

    RT1was565,565,572,559,551,and564msatthe17,67,150,250,450,and850msSOAs

    respectively.

    Itisalsoconceivablethatthelimitationunderlyingthebottleneckstillheld,butparticipants

    circumventeditbyperformingcentraloperationsononlyonetaskpertrial(guessingrandomly

    ontheother).Thishypothesispredictsthatthebypassersshouldshowadramaticincreaseinerror

    rates.Inactuality,therewasonlyaslighttrendforthebypasserstomakemoreerrorsthanthe

    bottleneckers(5.3vs.4.0%onTask1;4.2vs.3.6%onTask2),withconsiderableoverlapinerror

    ratesbetweenmembersofthetwogroups.Becauseweseenowaytoreconcilethedatafromthese

    fourparticipantswiththebottleneckmodel,weinsteadconcludethattheyfoundawaytoselect

    andexecuteresponsestobothtasksatthesametime.

    Factors that promote bottleneck bypass

    ClearevidenceofbottleneckeliminationisextremelyrareinthePRPliterature(asdiscussedin

    moredetailbelow).Itisimportant,therefore,toconsiderwhatfactorsmayhavefacilitatedbypassing

    thebottleneckinExperiment2.Themostobviousenablingfactoristhehighlevelpractice(4,480

    trainingtrialsplus2,240dual-tasktesttrialsperparticipant).Anotherpotentiallyimportantfactor

    istheabsenceofinputmodalityconflictsandoutputmodalityconflicts.Athirdfactoristheuse

    ofparticularinput/outputmodalitypairings(auditoryvocalandvisualmanual)thatminimize

    dual-taskinterference(seeHazeltine,Ruthruff,&Remington,2004;Shaffer,1975;VanSelst&

    Johnston,1997).Althoughthesefactorsarelikelytobeimportant,evennecessary,foreliminating

    thebottleneck,theyclearlyarenotsufficient.Thesefactorswerepresentforallparticipantsinall

    25

  • conditionsofbothexperiments,yetbottleneckeliminationoccurredinonlyafewcases.Below

    weconsiderseveralotherfactorsthatmighthelpexplainwhenthebottleneckcanbebypassed.

    Trainingconditiondoesnotappeartobeacriticalfactor.InExperiment2,atleast1outofthe

    6participantsineachofthethreetraininggroupsappearedtobypassthebottleneck(1intheTask1

    onlygroup,2intheTask2onlygroup,and1inthedual-taskgroup).Althoughmoreparticipants

    eliminatedthebottleneckfollowingtrainingonTask2alone,thisdifferencewasnotstatistically

    significant.Giventhehighcostsofrunningtheseexperiments(12sessionsperparticipant),it

    wouldbeprohibitivetotestmuchlargersamplestodefinitivelyassessdifferentialeffectsoftraining

    condition.However,toimproveourestimateoftheextenttowhichTask2trainingpromotes

    bottleneckelimination,weinstructedanadditionalsixparticipantstoperformunderthiscondition.

    Only1ofthese6additionalparticipantsshowedevidenceofbottleneckelimination,asinthe

    originalTask1practiceanddual-taskpracticegroups.Thus,thereisnoevidencethatanyparticular

    trainingconditioniscriticalforeliminatingthebottleneck.

    Onefactorthatdidappeartostronglyinfluencethelikelihoodofeliminatingthebottleneckwas

    taskorder.WiththetaskorderofExperiment1(Task1:Auditoryvocal;Task2:Visualmanual)

    therewasnoevidencethatanyofthe18participantswereabletoeliminatethebottleneck.

    Furthermore,noneofthesixparticipantsinVanSelstetal.(1999)wereabletoeliminatethe

    bottleneckwiththistaskorder.Withtheoppositetaskorder(Task1:Visualmanual;Task2:

    Auditoryvocal;Experiment2),however,therewasstrongevidencethat4outof18participants

    wereabletoeliminatethebottleneck.Inthefollow-upexperimentdiscussedabove,usingthe

    Task2trainingconditionofExperiment2,anotherparticipant(1outof6)alsobypassedthe

    bottleneck.Combiningalloftheseresultstogether,0outof24participantseliminatedthebottleneck

    withthetypicaltaskorder,but5outof24participantseliminatedthebottleneckwiththereverse

    taskorder.2Thisfindingsuggeststhatfuturestudiestaketaskorderintoconsideration,although

    itisnotyetclearwhetherthisfactorwillbeimportantwithothertaskpairs.

    AnothercriticalfactormightbeanindividualsskilllevelinperformingTask1and/orTask2,

    whichshouldbereflectedinthemeanRTs.ThemeanRT1was563msforbypassersand564ms

    forbottleneckers.Thus,Task1skillappearsnottobeimportant.ThemeanRT2atthebaseline

    longSOA,however,wasmuchsmallerforbypassers(358ms)thanforbottleneckers(572ms),

    2Achi-squaredtestrevealedthatthisdifferenceisstatisticallysignificant,2(1)=5.581,p

  • t(16)=3.71,p

  • criticalempiricalquestionswerehowthedifferenttrainingtypeswouldinfluencethePRPeffect

    andwhethertheywouldallowparticipantstoeliminatetheprocessingbottleneck.

    Task integration hypothesis

    Accordingtothetaskintegrationhypothesis,dual-taskpractice(butnotsingle-taskpractice)allows

    participantstoefficientlyintegrateperformanceofthetwotasks,possiblyeliminatingthebottleneck.

    Accordingly,thedual-tasktrainingconditionshouldproducethegreatestreductioninthePRP

    effect.Contrarytothisprediction,dual-taskinterferenceinExperiments1and2wasnosmaller

    followingdual-taskpracticethanfollowingsingle-taskpracticeonTask1.Hence,thepresentdata

    providenoevidenceoftaskintegration.Itremainspossible,ofcourse,thatintegrationoccurs

    underdifferentcircumstances(i.e.,withevenhigherlevelsofpracticeorwithtaskscloselytied

    toacommongoal,suchasbrakingandsteeringacar).

    Intact bottleneck hypothesis

    Accordingtotheintactbottleneckhypothesis,practicedoesnotleadtotaskintegrationor

    automatization.Althoughthebottleneckremainsintactwithpractice,itsdurationcandecline

    (producingasmallerPRPeffect)asthedurationofcertainTask1stages(1Aand1B)become

    shorter.Accordingly,Task1practiceanddual-taskpracticeshouldgreatlyreducedual-task

    interference,butTask2practiceshouldnot.ThispredictionwasconfirmedinExperiment1

    (Task1:Auditoryvocal;Task2:Visualmanual).Thedatafurtherconfirmedthepredictionsthat

    Task1difficultyeffectsshouldcarryoverfullyontoTask2atshortSOAs,andthatthePRPeffect

    andRT1shouldbelinearlyrelatedacrosspracticesessions(seeFig.2)andacrossindividuals(see

    Fig.4).TherewasnoevidencethatanyindividualparticipantinExperiment1wasabletoeliminate

    thebottleneck.Thus,withthistaskorder,weconcludethatpracticereduceddual-taskinterference

    solelybyshorteningTask1stagedurations,leavingthebottleneckintact.Thisconclusionisthe

    sameonereachedbyVanSelstetal.(1999),whoalsostudieddual-taskpracticeeffectswiththe

    sametasksinthesameorder.

    InExperiment2,weusedataskorderoppositetothatofExperiment1.Thisparticulartask

    order(Task1:Visualmanual;Task2:Auditoryvocal)hasrarelybeeninvestigatedindual-task

    studies,butweuseditheretounconfoundtheeffectsofTask1/Task2practicefromtheeffects

    oftone/charactertaskpractice.Ontheonehand,wefoundevidencethatforthemajorityof

    participants(14outof18)thebottleneckremainedintact.Ontheotherhand,wealsofoundclear

    evidencethataminorityofparticipants(4outof18)hadcompletelyeliminatedthebottleneck.

    28

  • AsdiscussedinExperiment2andsummarizedinTable4,thisdivisionofparticipantsinto

    bypassersandbottleneckersissupportedbyseveralconvergingmeasures.First,thebypassers

    showedverysmallPRPeffects(only55ms,comparedwith342msforthebottleneckers),which

    wereroughly200mssmallerthantheintactbottleneckhypothesiswouldhavepredictedbased

    ontheirmeanRT1(seethefourfilleddatapointsinFig.7).Second,eachbypasserrespondedto

    thetasksinanorderoppositetothepresentationorder(oftenbyafewhundredmilliseconds)on

    atleast66%oftrialsattheshortestSOA.Third,thebypassersshowedverylittlecarry-overof

    Task1difficultyeffectsontoRT2.Thesefindingsstronglysuggestthat,contrarytotheintact

    bottleneckhypothesis,thebypasserswereabletoselectresponsestothetwotasksvirtually

    independently.

    Task automatization hypothesis

    Accordingtothetaskautomatizationhypothesis,practice(eithersingle-taskordual-task)allows

    participantstolearnhowtoperformtaskswithoutlimitedcentralresourcesandthusbypassthe

    centralbottleneck.Thus,allthreetypesoftrainingshouldproduceverysmallPRPeffects.Contrary

    tothisprediction,weobservedlargemeanPRPeffectsforallthetrainingtypesinbothexperiments.

    Thus,thetaskautomatizationhypothesis,byitself,isinconsistentwiththeoverallresults.Looking

    atindividuals,thishypothesisisinconsistentwithresultsfromalloftheparticipantsinExperiment1

    and14of18participantsinExperiment2.However,itcanexplaintheresultsofthe4participants

    inExperiment2whobypassedthebottleneckdespitehavingpracticedonlyoneofthetwotasks.

    Which task was automatized?

    Wenotedabovethatautomatizationofeithertaskcould,inprinciple,allowparticipantstobypass

    thebottleneck.Therefore,itispossiblethatthebypassersinExperiment2hadautomatizedjust

    oneofthetwotasks.Consistentwiththispossibility,thebypassersrespondedmuchmorequickly

    thanthebottleneckerstoTask2(evenatthelongSOA)butnottoTask1.Thisobservationsuggests

    thatthebypassershadautomatizedTask2(thetonetask)butnotTask1(thecharactertask).

    Thetonetaskmighthavebeenmoresusceptibletoautomatizationbecauseithadlessneedof

    centralbottleneckresourcestobeginwith,eitherbecausethestimulusclassificationwasbasedon

    aphysicalproperty(tonepitch)ratherthananabstract,learnedproperty(characteridentity),or

    becausetheSRmappingwashighlycompatible(sayhightoahightone)ratherthanincompatible

    (thevisualtaskhadanincompatiblemappingforhalfthestimuli).Adifferentpossibilityisthat

    auditorystimuliarebetterthanvisualstimuliatgainingaccesstothebottleneckbypassroute(e.g.,

    29

  • ajumper-cableroute)becauseauditorystimulimorereadilyattractattentiontothemselves

    (cf.Posner,Nissen,&Klein,1976;Wickens&Liu,1988).Furtherresearchisneededtodetermine

    whichofthesefactors(easeofstimulusclassification,SRcompatibility,differenceinalerting

    ability)isthekeytofacilitatingtaskautomatization.Along-termgoalistodeterminehowtasks

    canbeautomatizedandwhysomepeoplecanaccomplishthisbetterthanothers.

    What caused the effects of task order?

    Whenwepresentedtheauditoryvocaltaskfirstandthevisualmanualtasksecond,noneofthe

    24participants(poolingtogetherExperiment1andVanSelstetal.,1999)producedclearevidence

    ofhavingbypassedthebottleneck.Usingthissametaskorderandasampleofsixyoungeradults

    andsixolderadults,Maquestiaux,Hartley,andBertsch(inpress)alsofoundthatparticipantshad

    bypassedthebottleneckafterpractice.Whenweusedtheoppositetaskorder,however,5outof

    24participants(poolingtogetherExperiment2andtheassociatedcontrolexperiment)produced

    clearevidenceofhavingbypassedthebottleneck.AsnotedintheDiscussionsectionof

    Experiment2,thiseffectoftaskorderwasstatisticallysignificant.Interestingly,asimilareffect

    oftaskorderwasobservedinJohnstonandDelgado(1993)whenanunpracticedbutexceptionally

    easytaskwasusedaseitherTask1orTask2(seebelowforfurtherdiscussionoftheirfindings).

    Howcanweexplaintheeffectsoftaskorderontheprobabilityofeliminatingthebottleneck?

    Aboveweproposedthatthetonetaskwasautomatizedforsomeparticipants.Because

    automatizationofeitherTask1orTask2shouldeliminateconflictforthecentralbottleneck

    resource,thishypothesisincorrectlypredictsbottleneckeliminationwithbothtaskorders(notjust

    whenthetonetaskservedasTask2).Clearly,someadditionalassumptionsareneededtoexplain

    taskordereffects.

    Oneattractivehypothesistoexplaintaskordereffectsisthattasksrecruitavailablecentral

    resourcesevenwhentheyarenotneededtoperformthattask.Toillustratethisgreedyresource

    usagehypothesis,considertheanalogyofabankteller(abottleneckmechanismthatcanserve

    onlyonecustomeratatime).Supposethattwocustomersarriveatthebank,onewhorequiresthe

    teller(e.g.,toperformawiretransfer)andonewhodoesnot(e.g.,towithdrawcash).Iftheperson

    whodoesnotrequirethetellerarrivesfirsthemight,forconvenience,engagethetellerevenif

    thereisanearbyautomatedtellermachine(ATM).Oncethiscustomerhasengagedtheteller,the

    othercustomer(whorequirestellerassistance)mustwait(abottleneckdelay).Nowconsiderthe

    oppositearrivalorder,wherethecustomerrequiringtellerassistancearrivesfirstandengagesthe

    teller.Whentheothercustomerarrives,hecannotimmediatelyusetheteller.Insteadofwaiting,

    30

  • however,hecouldsimplyutilizethenearbyATMtowithdrawcash,thusavoidingabottleneck

    delay.Followingthisanalogy,thegreedyresourceusagehypothesispredictsanasymmetryinthe

    PRPparadigm,suchthatabottleneckdelaycanbeavoidedonlywithonetaskorder(wherethe

    taskthatcanusethebypassrouteservesasTask2).

    Ifwenowcombinethegreedyresourceusagehypothesiswiththeaboveconclusionthatthe

    tonetaskismoreamenabletoautomatizationwithpracticethanthelettertask,itfollowsthat

    bottleneckbypasswouldbemorelikelytooccurwhenthetonetaskservesasTask2(asin

    Experiment2)thanwhenitservesasTask1(Experiment1).Thus,thishypothesiscanexplain

    thetaskordereffectsobservedinourstudy(andthoseofJohnston&Delgado,1993).Atthesame

    time,weacknowledgethatthishypothesisispost-hocandrequiresfurthertesting.

    Practice can (sometimes) eliminate the bottleneck

    Oneofthemostimportantfindingsofthepresentstudyisthatitisindeedpossibletoeliminate

    thecentralbottleneckwithpractice.Priortothisstudy,therewaslittlepublishedevidencethat

    thiswaspossible.ManyearlystudieshadfoundthatpracticehadverylittleinfluenceonthePRP

    effect.VanSelstetal.(1999)latershowedthatpracticecandramaticallyreducePRPeffectswhen

    inputandoutputconflictswereeliminated.Interestingly,oneparticipantshowednoPRPeffect

    all.ThisparticipantsmeanRT1wassoshort(~250ms),however,thateventhecentralbottleneck

    couldhavebeenlatent,producinglittleornoobservablePRPinterference.Ruthruff,Johnston,

    VanSelst,Whitsell,&Remington(2003)laterconfirmedthelatentbottleneckaccountofthis

    participantsdata.WhentheSOAbetweentaskswasadjustedtobringthecentraloperationsof

    thetwotasksmorecloselyintoalignment,PRPinterferencereturned.

    Smalldual-taskinterferenceeffectshavealsobeenobservedafterpracticeinanon-PRPparadigm

    whereparticipantsperformpureblocksthatcontainjustonetaskortheotherandmixedblocks

    thatcontainsomesingle-tasktrialsandsomedual-tasktrialswitha0-msSOA(Hazeltineetal.,

    2002;Levy&Pashler,2001;Schumacheretal.,2001).However,thesesmalldual-taskeffects

    wereobtainedwithanunusuallyeasytask:Participantssimplypressedabuttonthatcorresponded

    spatiallywiththelocationofavisualstimulus.Consequently,meanRTsweretypicallylessthan

    300ms.Itisquitepossiblethatthebottleneckwasnoteliminatedinthesestudies,butinsteadwas

    latent(seeHazeltineetal.,2002forrelevantdiscussion).Interestingly,afewstudiesusingthis

    dual-taskparadigmbutwithmoredifficulttaskshaveoftenfoundlongermeanRTsandlarger

    amountsofdual-taskinterference(Levy&Pashler,2001,Experiment2;Schumacheretal.,2001,

    Experiment3;butseealsoHazeltineetal.,2004,Experiment3).

    31

  • Inconsideringthelatentbottleneckhypothesis,itiscriticalthatthebypassersinthepresent

    Experiment2showedsolittlePRPeffectdespitehavingrelativelylongmeanRT1s(between492

    and631ms).LongRT1sareusuallytheresultoflongcentralstagedurations.Accordingly,itis

    implausiblethatinterferencewasavoidedsimplybecausethetwotasksneverdemandedcentral

    operationsatthesametime,especiallygiventhatweusedmultipleSOAsacrossawiderange(17

    to800ms).Itisalsoimportanttonotethat,unlikemanypreviouslypublishedstudies,wedidnot

    baseourinferencessolelyonthemagnitudeofPRPeffects.Instead,wealsoexaminedthePRP/RT1

    relationship,severalfactoreffects,andthefrequencyofresponsereversals.Futurestudiesdesigned

    tolookforbottleneckbypassshouldalsousetheseconvergingindicators.

    Previous evidence of the elimination of the bottleneck

    HavingfoundclearevidenceofbottleneckeliminationinExperiment2,wenowtakeastepback

    andconsiderhowthisfindingfitsinwiththepreviousevidence(albeitscarce)ofbottleneck

    eliminationinRTparadigms.Indoingso,itishelpfultoclassifychoiceRTtasksaccordingtothe

    compatibilityoftheSRtranslationand,hence,theirdemandoncentraloperations.Theeasiest

    tasksareideomotorcompatibletasks,suchasshadowing(repeatingaloudwhateverwordyou

    hear)wherethestimuluscodescloselyresembletheresponsecodes(Greenwald,1972).Intermediate

    indifficultyarecompatibletasks,wherethestimuluscodesandresponsecodesdonotresemble

    eachotherbuthavesomenaturalrelationship(e.g.,inourtonetaskparticipantssaidlowto

    low-pitchedtonesandhightohigh-pitchedtones).Thenexthighesttierofdifficultywould

    includearbitrarytasks,wherethereisnopre-learnedassociationbetweenstimuliandresponses.

    Thehighesttierofdifficultyincludesincompatibletasks,wherethemappingopposessome

    pre-learnedassociation.

    Fortheeasiesttaskcategory(ideomotorcompatible),thebottleneckmightbeexpectedto

    generallybebypassed.Interestingly,severalstudieshavereachedtheoppositeconclusion(e.g.,

    Lien,McCann,Ruthruff,&Proctor,inpress;Lien,Proctor,&Allen,2002).Otherstudieshave

    reportedsmallPRPeffects(e.g.,Greenwald,1972;Greenwald&Shulman,1973;Greenwald,

    2003),butbecauseRTsweresoshortitisdifficulttoruleoutalatentbottleneckaccount(see

    Lien,Proctor,&Ruthruff,2003).Pashler,Carrier,andHoffman(1993;Experiments1and2),

    however,didfindstrongevidenceforbottleneckbypasswhenTask1wastheusualtonetaskand

    Task2requiredaneyemovementtoavisualstimulus(anideomotorcompatibletask).Interestingly,

    thepatternofresultswassimilartothatofthebypassersinExperiment2:AlthoughthePRPeffects

    weregreaterthanzero,participantsgenerallyrespondedtoTask2wellbeforeTask1atshort

    32

  • SOAs.JohnstonandDelgado(1993)foundsimilarevidenceofbottleneckbypasswithacompatible

    Task1andanideomotorTask2thatrequiredcontinuoustrackingofavisualstimulususinga

    joystick.Atthesametime,theyfailedtofindclearevidenceofbypasswithseveralothertracking

    tasks,soideomotorcompatibilityisapparentlynotasufficientconditionforbottleneckbypass.

    Interestingly,JohnstonandDelgadoalsofailedtofindclearevidenceofbottleneckbypasswhen

    thecontinuoustrackingtaskservedasTask1.Thisfindingprovidesfurthersupportforourgreedy

    resourceusagehypothesis,whichpredictsthatbottleneckbypassshouldoccuronlywhenthe

    automatizedtaskservesasTask2.

    Weareawareofnopreviousstudieswithclearevidenceoftheeliminationofthebottleneck

    witheithercompatible,arbitrary,orincompatibletasks.Thepresentstudy,however,didfindsuch

    evidence.Ourtonetaskusedacompatiblestimulusresponsemapping,whereasourcharactertask

    usedacompatible/incompatiblemapping(dependingonwhetherthecharacterwasadigitora

    letter).Itshouldbenoted,however,thatbottleneckeliminationoccurredonlyforasubsetof

    participants,onlywithonetaskorder,andonlyafterextensivepractice.

    SummaryThepresentstudyprovidednoevidencethatdual-taskpracticeallowsparticipantstoefficiently

    integratetaskpairs.Instead,theprimaryeffectofpracticewastoshortenstages,leavingthe

    bottleneckintactbutshorterinduration.However,practice(single-taskanddual-task)alsoallowed

    afewparticipantstoautomatizeoneorbothtasksandavoidthecentralbottleneck.

    Thesedatarepresentthestrongestevidenceyetthatpracticecaneliminatethebottleneck.This

    findingisanimportantleadforfurtherresearchaimedatuncoveringthenatureoftheprocessing

    bottleneckandhowtocircumventit.Atthesametime,weshouldnotlosesightofthefactthat

    thisintriguingfindingrequiredthecombinationofseveralfavorableconditions.Theseconditions

    includeextensivepractice,relativelyeasytasks(especiallycomparedwithmanyreal-worldtasks),

    anabsenceofinputconflictsandoutputconflicts,andpreferredmodalitypairings(auditoryvocal

    andvisualmanual).Evenwiththesefavorableconditions,wefoundevidenceofbottleneck

    eliminationonlywithoneofthetwopossibletaskorders.Evenwiththemorefavorabletaskorder,

    itoccurredonlyforaminorityofparticipants(4outof18).Thus,atthispoint,eliminationofthe

    bottleneckisstillarareexception.

    33

  • AcknowledgementsThisresearchwasfundedbytheAirspaceOperationsSystemsProjectof

    NASAsAirspaceSystemsProgram.WethankMei-ChingLien,HalPashler,andRobertProctor

    forhelpfulcommentsonanearlierdraftofthepaper.

    ReferencesAllport,D.A.,Antonis,B.,&Reynolds,P.(1972).Onthedivisionofattention:Adisproofofthesinglechannelhypothesis.QuarterlyJournalofExperimentalPsychology,24,225235.

    Bertelson,P.,&Tisseyre,F.(1969).Refractoryperiodofc-reactions.JournalofExperimentalPsychology,79,122128.

    Davis,R.(1956).ThelimitsofthePsychologicalRefractoryPeriod.QuarterlyJournalofExperimentalPsychology,8,2438.

    Davis,R.(1957).Thehumanoperatorasasinglechannelinformationsystem.QuarterlyJournalofExperimentalPsychology,9,119129.

    DeJong,R.(1993).Multiplebottlenecksinoverlappingtaskperformance.JournalofExperimentalPsychology:HumanPerceptionandPerformance,19,965980.

    DeJong,R.(1995).Theroleofpreparationinoverlapping-taskperformance.QuarterlyJournalofExperimentalPsychology,48,225.

    Dutta,A.,&Walker,B.N.(1995,November).PersistenceofthePRPeffect:Evaluatingtheresponse-selectionbottleneck.Posterpresentedatthe36thAnnualMeetingofthePsychonomicSociety,LosAngeles.

    Fletcher,B.C.,&Rabbitt,P.M.A.(1978).Thechangingpatternofperceptualanalyticstrategiesandresponseselectionwithpracticeinatwo-choicereactiontimetask.QuarterlyJournalofExperimentalPsychology,30,417427.

    Greenwald,A.,&Shulman,H.G.(1973).Ondoingtwothingsatonce.II.Eliminationofthepsychologicalrefractoryperiodeffect.JournalofExperimentalPsychology,101,7076.

    Greenwald,A.G.(1972).Ondoingtwothingsatonce.I.Time-sharingasafunctionofideomotorcompatibility.JournalofExperimentalPsychology,94,5257.

    Greenwald,A.G.(2003).Ondoingtwothingsatonce.III.Confirmationofperfecttimesharingwhensimultaneoustasksareideomotorcompatible.JournalofExperimentalPsychology:HumanPerceptionandPerformance,29,859868.

    Hazeltine,E.,Teague,D.,&Ivry,R.B.(2002).Simultaneousdual-taskperformancerevealsparallelresponseselection.JournalofExperimentalPsychology:HumanPerceptionandPerformance,28,527545.

    Hazeltine,E.,Ruthruff,E.,&Remington,R.W.(2004).Thepairingsofstimulusandresponsesmodalitiesaffectdual-taskcosts.Manuscriptsubmittedforpublication.

    Hick,W.E.(1948).Thediscontinuousfunctionofthehumanoperatorinpursuittasks.QuarterlyJournalofExperimentalPsychology,1,3651.

    Hirst,W.,Spelke,E.S.,Reaves,C.C.,Caharack,G.,&Neisser,U.(1980).Dividingattentionwithoutalternationorautomaticity.JournalofExperimentalPsychology:General,109,98117.

    Johnston,J.C.,&Delgado,D.F.(1993,November).Bypassingthesingle-channelbottleneckindual-taskperformance.Paperpresentedtothe34thannualmeetingofthePsychonomicSociety,WashingtonD.C.

    Kahneman,D.(1973).Attentionandeffort.EnglewoodCliffs,NJ:Prentice-Hall.

    Karlin,L.,&Kestenbaum,R.(1968).Effectsofnumberofalternativesonthepsychologicalrefractoryperiod.QuarterlyJournalofExperimentalPsychology,20,167178.

    Keele,S.(1973).Attentionandhumanperformance.PacificPalisades,CA:Goodyear.

    Levy,J.,&Pashler,H.(2001).Isdual-taskslowinginstructiondependent?JournalofExperimentalPsychology:HumanPerceptionandPerformance,27,862869.

    34

  • Lien,M.-C.,&Proctor,R.W.(2002).Stimulus-responsecompatibilityandpsychologicalrefractoryperiodeffects:Implicationsforresponseselection.PsychonomicBulletin&Review,9,212238.

    Lien,M.-C.,&Ruthruff,E.(2004).Taskswitchinginahierarchicaltaskstructure:Evidenceforthefragilityofthetaskrepetitionbenefit.JournalofExperimentalPsychology:Learning,Memory,andCognition,30,697713.

    Lien,M.-C.,Proctor,R.W.,&Allen,P.A.(2002).Ideomotorcompatibilityinthepsychologicalrefractoryperiodeffect:29yearsofoversimplification.JournalofExperimentalPsychology:HumanPerceptionandPerformance,28,396409.

    Lien,M.-C.,Proctor,R.W.,&Ruthruff,E.(2003).Stillnoevidenceforperfecttimesharingwithtwoideomotorcompatibletasks:AnobservationonGreenwald(2003).JournalofExperimentalPsychology:HumanPerceptionandPerformance,29,12671272.

    Lien,M.-C.,McCann,R.E.,Ruthruff,E.,&Proctor,R.W.(inpress).Processinglimitationsindual-taskperformance:Canthecentralbottleneckbebypassedwithideomotorcompatibletasks?JournalofExperimentalPsychology:HumanPerceptionandPerformance.

    Maquestiaux,F.,Hartley,A.A.,Bertsch,J.(inpress).Canpracticeovercomeage-relateddifferencesinthepsychologicalrefractoryperiodeffect?PsychologyandAging.

    McCann,R.S.,&Johnston,J.C.(1992).Locusofthesingle-channelbottleneckindual-taskinterference,JournalofExperimentalPsychology:HumanPerceptionandPerformance,18,471484.

    Meyer,D.E.,&Kieras,D.E.(1997a).Acomputationaltheoryofhumanmultiple-taskperformance:TheEPICinformation-processingarchitectureandstrategicresponsedefermentmodel.PsychologicalReview,104,165.

    Meyer,D.E.,&Kieras,D.E.(1997b).Acomputationaltheoryofhumanmultiple-taskperformance.II.Accountsofpsychologicalrefractoryphenomena.PsychologicalReview,107,749791.

    Mowbray,G.H.,&Rhoades,M.V.(1959).Onthereductionofchoicereactiontimeswithpractice.TheQuarterlyJournalofExperimentalPsychology,14,136.

    Pashler,H.(1994).Gradedcapacity-sharingindual-taskinterference?JournalofExperimentalPsychology:HumanPerceptionandPerformance,20,330342.

    Pashler,H.,&Baylis,G.(1991).Procedurallearning.I.Locusofpracticeeffectsinspeededchoicetasks.JournalofExperimentalPsychology:Learning,MemoryandCognition,17,2032.

    Pashler,H.,&Johnston,J.C.(1989).Chronometricevidenceforcentralpostponementintemporallyoverlappingtasks.QuarterlyJournalofExperimentalPsychology,41A,1945.

    Pashler,H.,&Johnston,J.C.(1998).Attentionallimitationsindual-taskperformance.InH.Pashler(Ed.),Attention(pp.155189).Hove,UK:Psychology.

    Pashler,H.,Carrier,M.,&Hoffman,J.(1993).Saccadiceyemovementsanddual-taskinterference.QuarterlyJournalofExperimentalPsychology,46A,5182.

    Posner,M.I.,Nissen,J.M.,&Klein,R.(1976).Visualdominance:Aninformationprocessingaccountofitsoriginsandsignificance.PsychologicalReview,83,157171.

    Ruthruff,E.,Miller,J.O.,&Lachmann,T.(1995).Doesmentalrotationrequirecentralmechanisms?JournalofExperimentalPsychology:HumanPerceptionandPerformance,21,552570.

    Ruthruff,E.,Pashler,H.E.,&Klaassen,A.(2001).Processingbottlenecksindual-taskperformance:Structurallimitationorvoluntarypostponement?PsychonomicBulletin&Review,8,7380.

    Ruthruff,E.,Johnston,J.C.,&VanSelst,M.(2001).Whypracticereducesdual-taskinterference.JournalofExperimentalPsychology:HumanPerceptionandPerformance,27,321.

    Ruthruff,E.,Pashler,H.,&Hazeltine,E.(2003).Dual-taskinterferencewithequaltaskemphasis:Gradedcapacity-sharingorcentralpostponement?PerceptionandPsychophysics,65,801816.

    Ruthruff,E.,Johnston,J.C.,VanSelst,M.A.,Whitsell,S.,&Remington,R.(2003).Vanishingdual-taskinterferenceafterpractice:Hasthebottleneckbeeneliminatedorisitmerelylatent?JournalofExperimentalPsychology:HumanPerceptionandPerformance,29,280289.

    Schumacher,E.H.,Seymour,T.L.,Glass,J.M.,Fencsik,D.,Lauber,E.,Kieras,D.E.,&Meyer,D.E.(2001).Virtuallyperfecttimesharingindual-taskperformance:Uncorkingthecentralattentionalbottleneck.PsychologicalScience,12,101108.

    35

  • Schweickert,R.(1978).Acriticalpathgeneralizationoftheadditivefactormethod:AnalysisofaStrooptask.JournalofMathematicalPsychology,18,105139.

    Shaffer,L.H.(1975).Multipleattentionincontinuousverbaltasks.InP.M.A.Rabbitt&S.Dornic(Eds.)AttentionandPerformanceV(pp.157167).SanDiego,CA:Academic.

    Spelke,E.S.,Hirst,W.C.,&Neisser,U.(1976).Skillsofdividedattention.Cognition,4,215230.

    Tombu,M.,&Jolicoeur,P.(2000,November).IsthePRPeffectduetoastrategicorstructuralbottleneck?Paperpresentedatthe41stannualmeetingofthePsychonomicSociety,NewOrleans,LA.

    VanSelst,M.,&Johnston,J.C.(1997,November).ModalitymodulatesPRPinterference.Presentedatthe38thannualmeetingofthePsychonomicSociety,Philadelphia.

    VanSelst,M.,&Jolicoeur,P.(1994).Asolutiontotheeffectofsamplesizeonoutlierelimination.QuarterlyJournalofExperimentalPsychology,47A,631650.

    VanSelst,M.,&Jolicoeur,P.(1997).Decisionandresponse.CognitivePsychology,33,266307.

    VanSelst,M.,Ruthruff,E.,&Johnston,J.C.(1999).CanpracticeeliminatethePsychologicalRefractoryPeriodeffect?JournalofExperimentalPsychology:HumanPerceptionandPerformance,25,12681283.

    Welford,A.T.(1952).Thepsychologicalrefractoryperiodandthetimingofhigh-speedperformanceAreviewandatheory.BritishJournalofPsychology,43,219.

    Welford,A.T.(1976).Skilledperformance:Perceptualandmotorskills.Glenview,IL:Scott,Foresman.

    Wickens,C.D.,&Liu,Y.(1988).Codesandmodalitiesinmultipleresources:Asuccessandaqualification.HumanFactors,30,599616.

    36

  • Tab

    le1Task(s)performedineachsessionofExperiments1and2asafunctionofthetraininggroup(Single/Task1,Single/Task2,Dual-task).Eachtrainingconditionusedadifferent

    groupofparticipants.InExperiment1,Task1wasthetonetask(auditoryvocal)andTask2wasthecharactertask(visualmanual)task.InExperiment2,thetaskorderwasreversed

    Trainingcondition

    Session

    Dual-task

    Single/Task2

    Single/Task1

    Task1/Task2

    Task2

    Task1

    1Training

    Task1/Task2

    Task2

    Task1

    2Task1/Task2

    Task2

    Task1

    3Task1/Task2

    Task2

    Task1

    4Task1/Task2

    Task2

    Task1

    5Task1/Task2

    Task2

    Task1

    6Task1/Task2

    Task2

    Task1

    7Task1/Task2

    Task2

    Task1

    8Task1/Task2

    Task1/Task2

    Task1/Task2

    1Test

    Task1/Task2

    Task1/Task2

    Task1/Task2

    2Task1/Task2

    Task1/Task2

    Task1/Task2

    3Task1/Task2

    Task1/Task2

    Task1/Task2

    4

    37

  • Tab

    le2MeanresponsetimeforTask1(tone)andTask2(character)duringtheeighttrainingsessionsforeachtraininggroupinExperiment1.Forthedual-taskpracticegroup,mean

    responsetimesarebasedonthelongstimulusonsetasynchrony(SOA;baseline)conditiononly

    Trainingcondition

    Session

    Dual-task(longSOA)

    Single/Task2

    Single/Task1

    Task2character

    Task1tone

    Task2character

    Task1tone

    Task2character

    Task1tone

    993

    1,090

    1,095

    440

    1722

    767

    718

    365

    2669

    697

    672

    336

    3629

    647

    644

    321

    4595

    602

    639

    311

    5569

    581

    600

    320

    6585

    617

    602

    308

    7566

    566

    624

    297

    8

    38

  • Tab

    le3MeanresponsetimeforTask1(character)andTask2(tone)duringtheeighttrainingsessionsforeachtraininggroupinExperiment2.Forthedual-taskpracticegroup,mean

    responsetimesarebasedonthelongSOA(baseline)conditiononly

    Trainingcondition

    Session

    Dual-task(longSOA)

    Single/Task2

    Single/Task1

    Task2tone

    Task1character

    Task2tone

    Task1character

    Task2tone

    Task1character

    726

    845

    597

    592

    1633

    664

    564

    554

    2556

    620

    549

    522

    3509

    588

    549

    501

    4500

    589

    595

    525

    5516

    567

    533

    459

    6509

    558

    522

    450

    7538

    558

    503

    424

    8

    39

  • Tab

    le4DifferencesinresultsbetweenthetwosubgroupsofparticipantsinExperiment2.PRPpsychologicalrefractoryperiod,RT2responsetimetoTask2

    Nobottleneckgroup(n=4)

    Bottleneckgroup(n=14)

    55ms

    342ms

    PRPeffect

    82%

    2%Responsereversals

    21%

    106%

    Carry-overofTask1compatibilityeffectsontoRT2

    40

    IntroductionBackgroundThe psychological refractory periodCan practice eliminate the central bottleneck?

    The present studyAssessing the consequences of practiceSummary of key predictions

    Experiment 1MethodParticipantsStimuliApparatusProcedureAnalyses

    ResultsDual-task group only (training and test)Training sessionsTask1 RT in test sessionsPRP effect on Task2 in test sessionsPRP effect vs. RT1 for individualsTask1 carry-over predictionTask2 absorption predictionTask 2 SR compatibilityTask1 error ratesTask2 error rates

    Discussion

    Experiment2 (reverse task order)MethodResultsDual-task group only (training test)Training phaseTask1 RT in test sessionsPRP effect on Task2 in test sessionsPRP effect vs. RT1 for individualsTask1 carry-over predictionTask2 tone difficulty effectsTask1 error ratesTask2 error rates

    DiscussionEvidence of two distinct subgroupsAttempts to reconcile the bottleneck model with small PRP effectsFactors that promote bottleneck bypassRe-examining PRP effects as a function of training group

    General discussionTask integration hypothesisIntact bottleneck hypothesisTask automatization hypothesisWhich task was automatized?What caused the effects of task order?Practice can (sometimes) eliminate the bottleneckPrevious evidence of the elimination of the bottleneck

    SummaryReferences