ATTRIBUTINGEMOTIONTOSTATICBODYPOSTURES:RECOGNITIONACCURACY,CONFUSIONS, - PDF document

ATTRIBUTINGEMOTIONTOSTATICBODYPOSTURES:RECOGNITIONACCURACY,CONFUSIONS,ANDVIEWPOINTDEPENDENCEMarkCoulsontotalof176computer-generatedmannequinfigureswereproducedfromdescriptionsofposturalexpressionsofemotioninordertoinvestigatetheattri-butionofemotiontostaticbodypostures.Eachposturewasrenderedfrom3view-inganglesandpresentedtoparticipantsinaforced-decisiontask.Concordanceratesforattributionsof6emotions(anger,disgust,fear,happiness,sadness,andsurprise)rangedfromzeroformanydisgustposturestoover90percentforsomeangerandsadnesspostures.Anatomicalvariablesandviewinganglewereshowntopredictparticipants’responses.Analysisoftheconfusionmatrixsuggestedacir-cumplexsolutionwithhappinessandsurprisesharingasimilarposition,andfewconfusionsbetweentheotherfouremotions.Themeansbywhichemotionsmaybeattributedtostaticbodyposturesarediscussed,asareavenuesforfurtherresearch.KEYWORDS:emotion;expression;posture.Theroleofbodypostureintheexpressionandcommunicationofemo-tionremainsthepoorrelationofresearchintofacialandvocalexpression.SincepublicationofDarwin’sTheExpressionoftheEmotionsinManand(Darwin,1872/1965),agreatdealofattentionhasfocusedonhowemotionsarecommunicatedthroughfacialexpression,andmuchhasbeenlearnedrangingfromknowledgeofthewaysinwhichindividualmusclescontributetoeachexpression(e.g.Ekman&Friesen,1984;Izard,1979;Rinn,1984)tothesimilaritiesanddifferencesbetweenculturesinemo-tionalperception(e.g.Boucher&Carlson,1980;Ekmanet.al.,1987;HaidtKeltner,1999;Mesquita&Frijda,1992;Russell,1991).Asimilaral-thoughsmallerliteratureexistsontheperceptionofemotionfromthevoice(e.g.Banse&Scherer,1996;Murray&Arnott,1993;Scherer,1986,1995;vanBezooijen&Boves,1986;Wallbott&Scherer,1986).Thegeneralcon- MarkCoulsonisaffiliatedwiththeSchoolofHealthandSocialSciences,MiddlesexUni-AddresscorrespondencetoMarkCoulson,SchoolofHealthandSocialSciences,Middle-sexUniversity,Queensway,EnfieldEN34SF,UnitedKingdom.JournalofNonverbalBehavior,28(2),Summer20042004HumanSciencesPress,Inc. JOURNALOFNONVERBALBEHAVIORclusionisthatemotion,insofarasthetermreferstothesix‘basic’emotions(anger,disgust,fear,happiness,sadness,andsurprise)isaccuratelyper-ceivedthroughthefaceandvoice,andfurthermorethatthisprocessissimilaracrosscultures.Althoughmuchhasbeenwrittenaboutbodypostureandemotion,littlesystematicresearchhasbeencarriedout,andthatwhichexistshastendedtofocusonemotionalexpressionthroughmovementratherthanstaticposture.Thislackofresearchissurprisingwhenoneconsidersthatbodilyexpressioniswidelyrecognizedasanimportantdiagnostictoolintherapeu-ticpractice(e.g.Berger,1994;Dosamontes-Beaudry,1997;Flack,Laird,&Cavallaro,1999;Fuchs,1996;Hirsch,1994),andisacrucialfeatureofartis-ticformssuchasactinganddance.Furthermore,humanbodiesarelargeobjectspossessingmultipledegreesoffreedomwhichwouldappeartobeidealchannelsforemotionalcommunication(Montepare,Koff,Zaitchik,&Albert,1999),especiallyatdistanceswheretherecognitionofemotionfromfacialexpressionisdifficult(Walk&Walters,1988).However,therelation-shipbetweenpostureandemotionmaybeweakerthanforotherchannelsofnonverbalcommunication,andassuchpresentsspecialchallenges.Theresearchreportedhereinvestigatedtherelationshipbetweenpreciselyde-finedstaticposturesandattributionsofemotioninanattempttobeginspeci-fyinghowemotionmaybecommunicatedthroughthebody.HowWellDoBodiesCommunicateEmotion?Thereissomeevidencethatviewerscanreademotionsfromstaticbodypostureswithreasonablelevelsofaccuracy,andthatcertaincuesplayanimportantroleinthisprocess.Usingavarietyofmethodologiesincludingphotographs,anddynamicandstaticpoint-lightdisplays,Walkandcol-leagueshavedemonstratedthatemotioncanberecognizedfromstaticpos-turewithvaryinglevelsofaccuracy.WaltersandWalk(1986)claimthatrecognitionofemotionfromphotographsofposedexpressionswithfaceandhandsobscurediscomparableinaccuracytorecognitionoffacialex-pressions.Forpoint-lightstaticdisplays,althoughrecognitionratesweresignificantlyabovechanceforfiveofthesixbasicemotions(theexceptionbeingdisgust),recognitionratesonlyexceeded30%forhappiness(WaltersWalk,1988).Withdynamicdisplaysandusingasixalternativeforcedchoicemethodology,agreementratesincreasedtobetween71%forfearto96%forhappiness(Walk&Homan,1984).Aconsistentfindinginthesestudiesisthatanger,happinessandsadnessaremostaccuratelyrecog-nized,withsurpriseandfearsomewhatlower,anddisgustfrequentlyfailingtoreachstatisticallysignificantlevels. MARKCOULSONDespitedemonstratingthatpostureiscapableofcommunicatingemo-tion,fewstudieshavespecifiedtheanatomicalfeaturesresponsibleforattri-butionofparticularemotions.Aronoff,Woike,andHyman(1992)analyzedexpressiveposturesinballetdancersandfoundthatangularposturestendedtobeadoptedwhenplayingtheroleofathreateningcharacter,androundedposturesweremorecommonwhenplayingawarmorfriendlycharacter.Theseresultssuggestthatabstractgeometricalpropertiesofpos-turesareimportantcuestounderlyingemotion.SchouwstraandHoogstra-ten(1995)usedstickdrawingsofarmlessfigures,andvariedheadandspinepositions.Uprightpostureswerejudgedmorepositively,andforwards-leaningposturesmorenegatively.Anunpublishedstudywhichrequiredparticipantstoposeawoodenartist’sdollsuggestedthatthesixbasicemo-tionscouldbepartiallyrepresentedintermsofsagittalmovement,spinalflexion,open/closedandforwards/backwardsreaching,andfacialorienta-tiontowardsorawayfromtheelicitingstimulus(Inouye,1998).Inadditiontotheempiricalfindingsonstaticpostureoutlinedabove,numberofdescriptiveschemesforbodypostureandemotionhavebeenproduceddatingfromWilliamJames’originalstudy(James,1932).Al-thoughthesedescriptionspresentausefulframework,theytendtofocusongeneral,frequentlydynamicpropertiesofbodies,andfailtomakeclearpredictionsregardingthespecificpostureswhichmaybeassociatedwithdifferentemotionalstates.Dimensionssuchasapproach/withdrawaretoocoarse-grainedwhentheintentionistodeterminelikelyconfigurationsofhead,trunk,armsandlegs.Notwithstandingthis,thereexistavarietyofsourceswhichoffermoreorlessdetaileddescriptionsofemotionalpostures(e.g.Birdwhistell,1975;Boone&Cunningham,2001;Darwin,1872/1965;deMeijer,1989;Furnham,1999;Mehrabian,1981;Montepareet.al.,1999;Wallbott,1998).Forinstance,inthedescriptionsputforwardbytheseauthors,angerisvariouslydescribedasinvolvingajuttingchin,angu-larbodyshape,forwardweighttransfer,chestoutandangledforwards,andabowedhead.Bytranslatingthesedescriptionsintojointrotations,descriptiveschemesforselectedemotionscanbegenerated.Althoughemotioncanbeinferredfrombodyposture,themechanismsbywhichthisoccursareunclear.Posturesmayrepresentemotions,inten-tions,orrequests,adistinctionwhichistheoreticallyimportantbutdifficulttodisentangleanddemonstrateempirically.Also,emotionalexpressioncanbeconsideredemblematicandpropositional,orinvoluntaryandextempo-raneous(Buck,1991).Intermsofbodypostures,theformerincludebehav-iorssuchasraisedfists(emblematicofanger),whereasthelatteremphasizefunctionalactionsassociatedwiththebehavioralsignificanceoftheemo-tion(inthecaseofangerthesemightincludegeneraltensenessandafor-wardtransferofweight).Despitetheobviousimportanceofemblemsin JOURNALOFNONVERBALBEHAVIORthecommunicationofemotion,theresearchreportedherefocusedonthecontributionofanatomicalvariablestoexpression,andconsequentlylim-iteditselftotheinvestigationofnon-emblematicpostures.Furtherdifficultiesinstudyingposturerelatetoexpressiveidiosyncra-cies(deMeijer,1989;Wallbott,1998),controversiessurroundingthena-tureofposedversusspontaneousexpressions(Ekman,Hager,&Friesen,1981;Skinner&Mullin,1991),andtheproblemsinherentingeneratingthree-dimensionaldatafromtwo-dimensionalimages.Incombination,thesefactorsmaketheuseofvideoorphotographicrepresentationsofnaturalorperformedbehaviorproblematic,andconsequentlythestudyreportedherereliedoncomputer-generatedstimuli.Astheaimwastoexaminetheana-tomicalvariableswhichcommunicatedifferentemotions,completecontrolovertheexperimentalstimuliwasrequired.Computer-generatedfigures,unlikepeople,canbepreciselymanipulated,andoncedefinedcanbedis-playedfromanyangle.Finally,incontrastwithotherchannelsofnonverbalcommunication,bodypostureinvolvesanimportantthree-dimensionalpresencewhichof-fersdifferentperceptsdependingontheobserver’slocation,andchangesinviewinganglemayresultinocclusionofonebodypartbyanother.In-deed,thereissomeevidencewhichsuggeststhesamepostureviewedfromdifferentanglesdoesnotgiverisetothesamepercept,aspriorpresentationofaposturedoesnotprimeitslaterrecognitionfromadifferentviewpoint(Daems&Verfaillie,1999).Consequentlytheinclusionofviewpointasanadditionalvariablewasdeemedimportant.Withtheaboveinmindanumberofresearchquestionscanbeformu-lated.First,thenatureoftheposturesassociatedwithspecificemotionsneedtobedescribedanddefinedintermsofanatomicalfeatures.Second,thedegreetowhichhumanparticipantsagreeonwhichemotioneachpos-tureexpresses,andhowthischangesasaresultoftheanglefromwhichthepostureisviewedcanbeinvestigated.Finally,questionsaboutwhichanatomicalfeaturescontributetoattributions,andwhichemotionsarecon-fusedwitheachothercanbeaddressed.totalof61undergraduatestudents(36women,25men)tookpartintheexperimentforcoursecredit.Themeanagewas27years(range18–50). MARKCOULSONThestudylimiteditselftoexaminingposturesassociatedwithanger,dis-gust,fear,happiness,sadnessandsurprise.AlthoughthetheoreticalstatusofEkman’ssixbasicemotions,andthenotionofcategoricalemotionsitself,hasbeenchallengedfromanumberofdirections(Carroll&Russell,1996;Ortony&Turner,1990;Russell&Barrett,1999;Watson&Tellegen,1985,1999;Watson,Wiese,Vaidya,&Tellegen,1999),thereexistsawealthofdataregardingtheirabsoluteandrelativerecognitionaccuracywithinandacrosscultures.Inordertotranslategeneraldescriptionsofspontaneousfeaturesofbodilyexpressionofemotionintospecificpostures,aframeworkforde-scribingabodyisrequired.Thebodyistreatedasasystemofintercon-nectedrigidsegments,roughlycorrespondingtothebonesconnectingthemajorjoints,andtherelationshipsbetweenthesesegmentsdescribedintermsofrotationsaboutoneormoreaxesofthejointsconnectingthem.Ignoringfingersandtoes,andtreatingthespinalcolumnasthreesepa-ratejointsattheneck,chestandabdomen,therearefifteenmajorjointsinthehumanskeletonwithatotaloftwenty-ninedegreesoffreedom(ankles,knees,elbowsandthechestaremonaxial,wristsarebiaxial,andshoulders,hipsandthehead/neckandabdomen‘joints’aretriaxial).Certainsimplify-ingassumptionsarethereforerequiredifthenumberofposturesisnottobecomeunmanageablylarge.Consequently,amodelconsistingofthirteensegmentsandninedegreesoffreedomwaschosen.Theupperbodycon-sistedofsevensegments(head/neck,chest,abdomen,twoshoulders/upperarmsandtwoforearms),thelowerofsix(twothighs,twoshinsandtwofeet).Thedegreesoffreedomrelatetoaxesofjointrotation,andaredis-cussedfurtherbelow.Thismodelrepresentsacompromisedesignedtomaximizeflexibilityandminimizecomplexity.Theoverallmovementofthebody’scenterofmasscanbedescribedinanumberofmoreorlesscomplexways.Descriptionsofemotion,how-ever,arerarelymoreprecisethanindicatingwhetherabodyexperiencingspecificemotionislikelytomovetowardsorawayfromtheelicitingstimulus.Becauseofthis,movementofthemasscenterwascodedasoneofthreelevels—forwards,backwards,orneutral.Parametersspecifyingjointrotationsforthehips,kneesandanklesweregeneratedaccordinglyandpresentedtotenparticipants(whodidnottakepartinthemainstudy)whounanimouslyidentifiedthedirectionofweighttransferineachcase.Byintroducingafinalsimplifyingcriterion,thatofsymmetricalarms,itispossibletodescribeabodypostureintermsofoneparameterspecifyingmovementofthemasscenter,andsixjointrotations(headbend,chest JOURNALOFNONVERBALBEHAVIORbend,abdomentwist,shoulderadduct/abduct,shoulderswing,andelbowbend),foratotalofsevendegreesoffreedom.Thefinalstageoftheprocessinvolvesidentifyingthedegreetowhicheachjointcanrotate.Therearenoguidelinesastohowthismightbeachieved,andconsequentlythreeorfourlevelsofeachrotationwerese-lected,anumberwhichpermitsarangeofrealisticandperceptuallydis-tinctpositionstobeproducedwhilekeepingthetotalnumberofstimulimanageable.Bypermutingthequantitativejointrotationvalueswiththequalitativedescriptions,afamilyofdistinctconfigurationsforeachemotioncanbegenerated.Afulldescriptionofthejointrotationsdefiningeachemotion,andthenumberofdistinctposturesgeneratedbythisprocess,canbeseeninTable1.Eachofthe176descriptionswasusedtogenerateamannequinfigurewithinCuriousLabs’Poser4figureanimationpackage.Themannequinfig-urewaschosenbecauseitrepresentsaneasilyrecognizedhumanbodystrippedofage,sex,ethnicityandfacialfeatures.Eachposturewaspro-ducedataresolutionof400x400pixelsin24-bitcolour,andwasdepthshaded.Agroundplanewasintroducedtopreventthefigurefromappear-ingtofloatinspace.Eachimagewasfurtherrenderedfromthreedistinctviewingangles,fromthefront,thelefthandside,andaboveandbehindtheleftshoulder.Thisresultedinatotalstimulussetof528images(176postureseachrenderedfromthreeangles).Togiveanexampleoftherangeofposturesincludedinthestimuli,theresultsoftheparametersforangerareillustratedinFigure1wherethe32generatedposturesareshownfromthethreedifferentviewingangles.Inordertogainparitywithstudiesoffacialandvocalexpression,asix-alternativeforced-choicemethodologywasadopted.Directcomparisonsbetweenforcedchoiceandmoreopen-endedmethodssuggestthat‘thereisnotmuchwronwiththeforcedchoiceformat’(Izard,1994,p.291;seealsoFrank&Stennett,2001).The528imagesweresplitintothreeequalsets,witheachsetcontain-ingoneviewofeachposture,andathirdofposturesseenfromeachview-point.Theorderofpresentationswaspseudo-randomlyorderedsuchthatposturesdesignedtorepresentthesameemotionwereneverpresentedmorethanthreetimesinsequence,andthesameviewinganglewaspre-sentednomorethantwiceinsuccession.Fourseparatepseudo-randomordersweregeneratedforeachofthethreepresentationsets(thethirdandfourthbeingreversalsofthefirstandsecond). MARKCOULSONTABLE1JointRotationsforSixEmotions Shoulder AbdomenChestHeadElbowWeightNo.oftwist*bend**bend**ad/abduct***swing**bend**transferpostures Anger020,4020,258045,9050,110Forwards3220,025,450,50Backwards32Fear020,4025,50,6045,9050,110Backwards24Happiness00,200,20500,450,50Forwards32Sadness0,250,2025,508000Backwards32Surprise02025,505025,0,450,50Backwards24 Numberofposturesforeachemotioncategoryistheproductofthenumberofdistinctvaluesforeachjointrotationandweighttransfervaluesforthatcategory.*Theabdomenwasspecifiedastwistingtoonesideonlyornotatall.**Positiveandnegativevaluesrelatetoacoordinatesystemwithitsoriginatthejointcentre.Inthiscasearight-handcoordinatesystemisusedandpositiverotationsareforward.***Rotationisrelativetoaneutralpositionwhichisthearmraisedouttotheside,levelwiththeshoulder.Negativevaluesrelatetoarmsaboveshoulderlevel(abduction),positivevaluestoarmstowardsthesideofthetrunk(adduction). JOURNALOFNONVERBALBEHAVIOR Figure1.Thirty-twoposturesrepresentingangerviewedfromthreeangles. MARKCOULSONParticipantswerepresentedwitheachpostureinturnonacomputermonitor,andwererequiredtoselectwhichofthesixbasicemotionlabels(anger,disgust,fear,happiness,sadness,surprise)bestdescribedapersonadoptingthatparticularposture.Clickingthecomputermouseonabuttonrelatingtothechosenemotionlabeladvancedtheexperimenttothenexttrial.Theemotionlabelsappearedonthescreenatalltimesthroughouttheexperiment,andeachposturewaspresentedforaslongastheparticipantwished.Therewasaninter-trialintervalof500milliseconds,andpartici-pantswereaskedtotakeashortbreakafterevery30trials.Thetestingsessionlasted40–50minutes.Todeterminewhetherfatiguewasaffectingparticipants’responses,anitemsanalysiswasperformedonthefourpresentationsets.Foreachstimu-lus,themodalresponse(i.e.,thatemotionlabelwhichwaschosenbythegreatestnumberofparticipants)wasusedasthecriterionforaccuracy,andthenumberofattributionsforstimuliappearinginthefirsthalfofeachsetwerecomparedwiththoseforthesamestimulusappearinginthesecondhalfofthecomplementaryset.Arepeatedmeasurest-testfailedtoreachsignificance(n.s.),suggestingparticipants’accuracyofidentificationwasnotdeterioratingasthestudyprogressed.Inordertoexaminethedegreetowhichparticipantsagreedontheemotionbeingexpressedbyeachposture,binomialprobabilitieswereusedtocalculatesignificancelevels(chanceis16.7%).StandardlevelsofalphawerenotusedduetothepotentiallylargenumberofTypeIerrorsandthefactthatthenullhypothesis(thatthedistributionofresponsestoeachstim-ulusisrandom)isalmostalwayswrong,andrejectingitprovideslittleuse-fulinformation(Russell,1995).Militatingagainstthesefactorsistheun-knownnumberofstimuliforwhichthenullhypothesisisinfactcorrect.No‘pruning’ofthestimulitookplacepost-generation,withtheresultthatsomemayhaveelicitedrandomresponsesfromparticipants(followingtheposturegenerationprocedureoutlinedabove,theonlycheckperformedonthestimulussetwastoensurenotwopostureswereidentical).Despiteensuringthatalljointrotationswererealisticindegree,thecomplexityofthestimulimeantthatsomepostureslookedratherunusual.Ratherthanremovethesepriortothestudy,aprocesswhichmayinadvertentlyintro-ducebias,allgeneratedpostureswereincludedinthestudy.Withtheseconstraintsinmind,increasinglystrictconcordancere-quirementswereadoptedandthenumberofstimuliattributedtoasingle JOURNALOFNONVERBALBEHAVIORemotionateachlevelexamined.Table2summarizesthefindingsforcon-cordancelevelsof50%,60%,70%,80%and90%agreementbrokendownbyviewpoint(allconcordanceratesrepresentalphalevelsseveralordersofmagnitudebelowthosegenerallyadopted).Theposturereceivingthehighestconcordanceforeachemotionisshown,anditsdefiningpa-rameterslisted,inFigure2.Ingeneral,anger,happiness,andsadnessarebeingattributedtolargenumbersofpostures,withsomeidentifiedby90%ormoreofthesample.Fearandsurprisearelessfrequently,andlesscon-sensuallyidentified,andnoposturewasidentifiedasdisgustby50%ormoreofthesample.Toassessthedegreetowhichanatomicalvariablesandviewpointde-termineattributionofemotion,theeightvariablesdefiningeachposture(sixjointrotations,theweighttransferparameter,andviewpoint)wereen-teredaspredictorvariablesinamultinomiallogisticregression(MLR)usingresponsecategoryasthedependentvariable.MLRwaschoseninprefer-encetodiscriminantfunctionanalysisbecausethepredictorvariableswerenotnormallydistributedduetothesmallnumberoflevelsofeachrotation.TheresultsofthisanalysisarepresentedinTables3and4.Thefinalmodelwassignificantlybetterthantheintercept-onlymodel(0.001),andlikelihoodratiotestsrevealedthatallpredictorvariablesmadesignificantcontributionstothemodel(all0.001,Cox&Snellpseudo.581).Acrossallsixcategories,theregressioncorrectlypredicted48.5%ofresponses.Therewasconsiderablevariationinaccuracyfordif-ferentresponsecategories,withratesforhappinessandsadnessgreaterthan75%,andveryfewpredictionsfordisgust.Asallresponsestoallstimuliwereenteredintotheregression,theresultsaresomewhatconfoundedbythedegreeofconsensustoeachstimulus.However,asecondanalysisnotreportedherewhichremovedallstimulithatfailedtoreach40%consensusforatleastoneemotionattributionrevealedaverysimilarpatternofresults.Thebetacoefficientsindicatethatallemotionsarepredictedbyavari-etyofvariables.IncreatingtheregressionmodelinMLRanalysis,a‘refer-enceclass’forthedependentvariablemustbespecified.Thechoiceofreferenceclassisarbitraryinthatitdoesnotaffecttheoverallresults,butitdoesaffecttheinterpretationofthecoefficients.Disgustwaschosenasthereferenceclassinorderthatthefiveemotionsreceivingthehighestconcordanceswererepresentedintheanalysis.Asignificantcoefficientthusmeanstheparticularlevelofthevariablepredictscategorymember-ship(emotionattribution)relativetothereferenceclass.Forjointrotations,thereferenceclasswaschosenaszerodegreesofrotation(forshoulderadduct/abductthereisnozerorotation,and50wasselected),forweight MARKCOULSONTABLE2NumberofStimuliReachingConsensusLevelsofBetween50%and90%forSixEmotionsAcrossThreeDifferentViewpoints AngerDisgustFearHappinessSadnessSurprise FrontSideRearFrontSideRearFrontSideRearFrontSideRearFrontSideRearFrontSideRear 50%18121000004249282425433644160%168400002035252020392901070%103100000024161212261801080%210000000778512800090%100000000122031000 JOURNALOFNONVERBALBEHAVIOR Figure2.Posturereceivinghighestconcordanceforeachemotion(postureshownfromallthreeview-points—viewpointreceivinghighestconcordancewasthehighestforthatemotion). MARKCOULSON Figure2. JOURNALOFNONVERBALBEHAVIORTABLE3ResultsofMultinomialLogisticRegression—ClassificationTable Predicted ObservedAngerDisgustFearHappinessSadnessSurprise%Correct Anger617154434447110434.9%Disgust289272145505218818.9%Fear1021151919174196776.4%Happiness23981711966138678.6%Sadness31115777209429627.4%Surprise80822363382079.9%Overall%12.6%7.6%33.4%33.9%12.1%.4%48.5% MARKCOULSONTABLE4ResultsofMultinomialLogisticRegression—BetaCoefficients Betacoefficients AngerFearHappinessSadnessSurprise Headbend0.42*0.62*0.58**0.86*0.65**0.130.340.070.310.010.320.290.72*Chestbend0.020.39**0.58**0.080.370.101.12**0.85**Abdomentwist0.66**0.070.42**0.35Shoulderswing0.140.220.200.210.32*0.02Shoulderad/abduct0.88**0.61*2.23**2.51**1.01**0.76**1.89**2.40**Elbowbend1.62**0.71**0.83**0.130.71**0.31*WeighttransferBackwards0.29*0.92**0.130.150.25Forwards0.49**0.82**0.090.050.16Front0.42**0.37**0.87**0.170.49**0.100.04 eferenceclassisdisgust.Referenceclassforjointrotationsiszerorotationexceptforshoulderadduct/abduct,whichis50 JOURNALOFNONVERBALBEHAVIORtransfertheneutralpositionwaschosen,andforviewpointthesideviewwasselected.Consideringeachemotioninturn,angerispredictedbybackwardsheadbendandtheabsenceofabackwardschestbend,noabdominaltwist,andarmsraisedforwardsandupwards.Weighttransferiseitherforwardsorbackwards,andattributionsaremorelikelywhenposturesareseenfromthefront.Forfear,headbackwardsandnoabdominaltwistarepredictive,andthereisnoeffectofchestbendorupperarmposition.Forearmsareraisedandweighttransferiseitherbackwardsorforwards,andattributionsarelesslikelywhenviewedfromthefront.Happinessischaracterisedbyheadbackwardsandnoforwardsmovementofthechest.Armsareraisedaboveshoulderlevelandstraightattheelbow,weighttransferisnotpredic-tive,andattributionsaremorelikelywhenviewedfromthefront.Sadnessistheonlyemotioncharacterisedbyaforwardsheadbend,andinadditionincludesforwardschestbend,notwisting,andarmsatthesideofthetrunk.Weighttransferisnotpredictiveandattributionsarelesslikelywhenviewedfrombehind.Finally,surpriseinvolvesbackwardsheadandchestbends,anydegreeofabdominaltwisting,andarmsraisedwithforearmsstraight.Weighttransferisnotpredictive,andattributionsareagainmorelikelywhenviewedfromthefront.Inadditiontoexaminingwhetherchangesinviewpointaffectrecogni-tionaccuracy,itisimportanttoinvestigateconsistencyofattributionacrossviewpoints.Inordertoaccomplishthis,aseriesofnon-parametriccorrela-tionswasperformed.Foreachpostureseenfromeachviewpoint,thesixemotionswererankedintermsofhowoftentheywereselectedbypartici-pants.Correlationswerethencalculatedacrossthethreeviewpointsforeachemotion.Totheextentthatattributionsareconsistentacrossview-points,thecorrelationsforsameemotionpairsshouldbehigherthanfordifferentemotionpairs.TheresultsaresummarizedinTable5,andrevealapatternofcorrela-tionswhereallemotionsexceptdisgustshowedthehighestcorrelationsforsameemotionpairings.Thatis,forfiveoutofsixemotions,apostureseenasrepresentingaspecificemotionfromoneviewpointwasmostlikelytobeattributedtothesameemotionwhenviewedfromadifferentviewpoint.Fordisgust,althoughtherewassomeevidenceforconsistency,posturesseenasdisgustwereequallylikelytobeseenasdisgustorfearwhenviewedfromadifferentviewpoint.Tofurtherinvestigatethepatternsofconfusionbetweenemotions,amultidimensionalscaling(MDS)procedurewasappliedtothematrixofstimuluscategoriesandparticipants’responses.Confusionisnotcommuta-tive;confusingexpressionAwithexpressionBdoesnotimplyexpression MARKCOULSONTABLE5SummaryofViewpointShowingSameEmotionCorrelations,andRangeofOtherEmotionCorrelations Front-sideFront-rearSide-rear (range)r(range)r(range) Anger.36(.35–.18).46(.34–.29).55(.29–.15)Disgust.15*(.20–.25).23**(.21–.23).08*(.21–.11)Fear.42(.43–.29).47(.40–.25).62(.43–.28)Happiness.60(.51–.38).63(.56–.41).71(.60–.53)Sadness.64(.64–.36).71(.60–.29).79(.70–.26)Surprise.60(.48–.48).58(.46–.50).62(.49–.51) *Sameemotioncorrelationwasnotthehighestcorrelationobserved.**Sameemotioncorrelationwasjointhighestcorrelationobserved. willbeconfusedwithexpressionA,andthefullmatrixasopposedtoaveraged(upperorlowertriangle)valueswasused.two-dimensionalcircumplexfitstheconfusiondatawell(Kruskal0.08,seeFigure3).Happinessandsurpriseoccupysimilarloca-tionsonthesurfacewiththeotherfouremotionswellspreadoutaroundthecircumplex.Theresultsofthisstudyshedlightonanumberofquestionsregardingtheperceptionofemotionfrombodyposture.Theseconcernhowdifferentemotionsareattributedtobodypostures,theeffectofanatomicalvariablesandviewpointonattribution,andthenatureofconfusionsacrossemotions.Eachoftheseisaddressedbelow.Theresultspaintacomplexpictureofhowwellbodyposturecommu-nicatesemotion.Disgustwasnotattributedtoanyposturebymorethan50%ofthesample,andfearandsurprisewereconsistentlyattributedtoonlyasmallnumberofpostures.Anger,sadness,andhappiness,bycon-trast,wereattributedtolargenumbersofpostures,andsomestimuliwereattributedtothesameemotionby90%ormoreofthesample.Forthesethreeemotions,agreementratesarecomparabletothoseobtainedfrom JOURNALOFNONVERBALBEHAVIOR Figure3.Multidimensionalscalinganalysisofconfusionmatrix. staticfacialexpressions,andreplicateotherfindingssuggestingtheyareaccuratelyperceivedfromposture(Boone&Cunningham,1998;Walk&Homan,1984).Forthelesswellrecognizedemotions,anumberofexplanationscanbeputforward.Fordisgust,accuracyratesfromotherstudieshavebeensimilarlylow,andtheremaybenostaticbodyposturefordisgustotherthantheactofretching.Disgustmaythereforebeprimarilycommunicatedthroughtheface,althoughitremainstobeseenwhethercertaindynamicfeaturesofthebodyalsocontribute.Surpriseandhappinessgeneratedperceptuallysimilarpostures,andthelowfiguresforsurprisemaybeduetoparticipantsoptingforanattribu-tionofhappinessintheabsenceofsituationalcuesforsurprise.Thisobser-vationissupportedbytheanalysisoftheconfusionmatrix,whichidentifiedsurpriseandhappinessastheonlysystematicallyconfusedpostures,andisdevelopedfurtherinthediscussionofconfusionsbelow.Theresultsforfearareperhapsmoresurprisingwhenoneconsiders MARKCOULSONtheeasewithwhichposturalexpressionsoffearcanbebroughttomind.Thecoweringaspectoffeariswellrepresentedinthestimulusset,yetparticipantswerereluctanttoattributefear.Onepossibleexplanationforthisconcernstherelativecontributionsofpostureandmovementtotheattributionofemotion.Althoughstaticimagescanimplymovement,andhavebeenshowntoactivateneuralsystemsassociatedwithmotiondetec-tion,attributionsofcertainemotionsmaybeespeciallysensitivetomotioncues,andstaticimagesmaynotrepresentanappropriatemediumwithinwhichdynamicallyexpressedemotionscanbecommunicated.Inadynam-icallyunfoldingresponse,snapshotsmaysometimesprovideagreatdealofusefulinformation,butatothertimesbeuninformativeorevenmislead-ing.Fearandsurprise(andperhapstoalesserextentdisgust)areemergencyresponseswhosebodilyexpressionmayhavemoretodowithvelocityandformofmovementthanposturalconfiguration.Clearly,futureresearchshouldinvestigatehowpostureandmovementcontributecollectivelyorindependentlytotheattributionofemotion.Thejointrotationsandweighttransfervariablesthatdefinedthepos-tureswereshowntopredictparticipants’attributionswithvaryingdegreesofaccuracy.Eachemotionwasassociatedwithauniquesetofjointrota-tionsandnoneoftheanatomicalvariableswasredundant.Whilethesedatapointtoameansbywhichstructuralandpsychologicaldescriptionsofbodyposturemaybeintegrated,theyarefarfromdefinitive.Therangeofjointrotationsincludedwassmall,andlittlecanbesaidaboutthedegreetowhicheachrotationisabletoaffecttheoverallattributionofemotion.Futureresearchshouldaddressthesequestionsinmoredetail,andmapoutthecomplexrelationshipsbetweenpatternsofrotationandresultingattributions.Unlikefacialexpression,wheremuscleactionsarequantifiedinfairlygrossways,thestudyofposturalexpressionshouldfocusonthedegreeofchange,withtheaimofformallydescribingtherelationshipsbe-tweenanatomicalvariablesandperceptualgestalts.Turningtotheeffectsofviewpoint,therewasageneraltendencyforfrontalviewstoleadtomoreconsensualattributions,althoughthispatterndidnotapplytofearorsadness.Itispossiblethatthesedifferencesarisefromtheoccludingeffectsofparticularviewpointsonsomeclassesofstim-ulus.Forinstance,themoreclosedanddownwardlookingposturesfordisgust,fearandsadnessappearsmallerfromthefront,andpresentlessinformationtotheviewerthansideandrearviews.Aninspectionofthestimulithemselvesleadstoasimilarobservationforangerandhappinessposturesseenfromtheside.Theoverallpreferenceforfrontalviewssuggeststhatattributingemo-tiontoabodypostureisagreatdealeasierwhenthepersonadoptingthe JOURNALOFNONVERBALBEHAVIORpostureisfacingtheperceiver.Suchanorientation,whilenotnecessarilyidealforperceivingthethreedimensionalrelationshipsbetweenbodyseg-ments,maynonethelessenhancerecognitionduetoitsinterpersonalsignif-icance.Ifapersonisfacingaway,decodingtheiremotionalstatemaybeusefulbutnotcritical.Whenweareconfronted,rapidandaccuratedecod-ingofemotionalstateoffersimportantinformationthatcanbeusedtoguideAnalternativeexplanationfortheeffectsofviewpointisthatpartici-pantsarenotbasingtheirdecisionsontheinferredthree-dimensionalprop-ertiesofthestimulus,butonsomeotherunknownperceptualvariableorvariables.Whilethisexplanationremainsapossibility,itisrenderedlessplausiblebytheresultsofthecorrelationalanalysesofviewpoint,whichshowedthatforallemotionsbardisgusttheattributionmadetooneview-pointwastheonemostlikelytobeselectedfortheothertwo.Thissuggestsparticipantsareextractingthree-dimensionalinformationfromthestimuliandusingthisasthebasisfortheirdecisionsratherthanviewpoint-depen-dentcues.Withregardtoconfusionsbetweenpostures,thepatternobtainedherewasconsistentwithatwo-dimensionalcircumplexofemotion.Fouroutofsixemotions(anger,disgust,fear,andsadness)werenotconfusedwithoneanother,buttheremainingtwo(happinessandsurprise)demonstratedconsiderableoverlap.Thiscouldbeduetothephysicalsimilarityoftheposturesproducedforhappinessandsurprise,butmayalsoreflectconcep-tualsimilarities.CarrollandRussell(1996)arguethattheattributionofemotiontodisplaysoccursinacontextthattheylabellimitedsituationalThatis,knowledgeoftheelicitingcontexthasadirecteffectonperceptionsofemotion.Inthecontext-freestimuliusedinthisstudy,participantsmayusehappinessasadefaultcategoryforstimuliwhicharegenerallyuprightwiththearmsraisedtothesides.Intheabsenceofanexplicitemotion-elicitingevent,itmaybemoreparsimonioustoattributeaninternalcausetotheemotion,andselecthappinessoversurprise.Ananalogouseffectwouldnotbepredictedinattributionstofacialexpressionsduetothelargedifferencesbetweenhappyandsurprisedfaces.Severalquestionsremainunanswered.First,therelativelylowagree-mentratesforfearsuggesteitherthatfearispoorlyrecognizedfromstaticposture(andthatrecognitionmayrelymoreondynamiccues),orthatthefearposturesusedinthisstudywereunrepresentative.Thelatteraccountisdifficulttoreconcilewiththelevelsofagreementfoundforthemajorityoftheotheremotions,butthisrequiresfurtherinvestigation.Themoregeneralissueoftherelativerolesofpostureandmovement,andhowthesemayvaryacrossdifferentemotions,alsorequiresfurtherinvestigation,andmay MARKCOULSONcastlightonthegenerallysmalleffectsofweighttransferobservedinthisstudy.Researchersarebecomingincreasinglyawareoftheimportanceofthetemporalsignatureoffacialexpressions(Wehrle,Kaiser,Schmidt,&Scherer,2000),anditislikelythatbodilymovementwillprovesimilarlyimportant.Investigatingphysicalvariablesdescribingdynamicasopposedtostaticfeaturesofthebody(velocityandaccelerationofsegments,timecourseofonsetandtermination,etc.)shouldcontinuetheintegrationofpsychologicalandanatomicalaccountsattemptedhere.Second,therangeofemotionswaslimitedtothosesixwhichhaveattractedthegreatestattentioninpreviousworkonrecognition.Thereisanentireemotionalrepertoiretobeinvestigated,anditislikelythatotheremotionswillberecognizedfrompostureatlevelscomparabletothosestudiedhere.Asimilarexercisewhichconsideredsomeoftheself-con-sciousemotionssuchasshame,guilt,andembarrassmentwhichtypicallyincludehidingofthefaceandmightthereforebeeffectivelycommunicatedthroughthebodycouldenhanceunderstandingoftheseemotions,andoftheroleofpostureincommunicatingemotiongenerally.Thesefindingsasawholesupportthecontentionthatstaticbodypos-tureoffersareliablesourceofinformationconcerningemotion,andcon-tributetoourunderstandingofhowemotionisexpressedthroughthebody.Recognitionofemotionfrompostureiscomparabletorecognitionfromthevoice,andsomeposturesarerecognizedaswellasfacialexpressions.Anaccountofbodilyexpressionofemotioninthelanguageofanatomicalvariablesistractable,andwouldrepresentatoolwithconsiderablere-searchandpracticalsignificance.Aronoff,J.,Woike,B.A.,&Hyman,L.M.(1992).Whicharethestimuliinfacialdisplaysofangerandhappiness?Configurationalbasesofemotionrecognition.JournalofPersonalityandSocialPsychology(6),1050–1066.Banse,R.,&Scherer,K.R.(1996).Acousticprofilesinvocalemotionexpression.JournalofPersonalityandSocialPsychology(3),614–636.Berger,M.(1994).Psychologicalstudiesconcerninghuman-structuraldance.(1–2),50–66.Birdwhistell,R.L.(1975).Backgroundconsiderationstothestudyofthebodyasamediumof‘expression’.InJ.Benthall&T.Polhemud(Eds.),Thebodyasamediumofexpression(pp.36–58).Bungay,Suffolk:Allen.Boone,R.T.,&Cunningham,J.G.(1998).Children’sdecodingofemotioninexpressivebodymovement:Thedevelopmentofcueattunement.DevelopmentalPsychology1007–1016.Boone,R.T.,&Cunningham,J.G.(2001).Children’sexpressionofemotionalmeaninginmusicthroughexpressivebodymovement.JournalofNonverbalBehavior(1),21–41. JOURNALOFNONVERBALBEHAVIORBoucher,J.D.,&Carlson,G.E.(1980).Recognitionoffacialexpressioninthreecultures.JournalofCrossCulturalPsychology263–280.Buck,R.(1991).Motivation,emotion,andcognition:Adevelopmental-interactionalistview.InK.T.Strongman(Ed.),InternationalreviewofstudiesonemotionewYork:Wiley.Carroll,J.M.,&Russell,J.A.(1996).Dofacialexpressionssignalspecificemotions?Judgingemotionfromthefaceincontext.JournalofPersonalityandSocialPsychology205–Daems,A.,&Verfaillie,K.(1999).Viewpoint-dependentprimingeffectsintheperceptionofhumanactionsandbodypostures.VisualCognition(6),665–693.Darwin,C.(1872;1965).TheexpressionoftheemotionsinmanandanimalsChicago:Uni-versityofChicagoPress.deMeijer,M.(1989).Thecontributionofgeneralfeaturesofbodymovementtotheattributionofemotions.JournalofNonverbalBehavior247–268.Dittrich,W.H.,Troscianko,T.,Lea,S.E.G.,&Morgan,D.(1996).Perceptionofemotionfromdynamicpoint-lightdisplaysrepresentedindance.727–738.Dosamantes-Beaudry,I.(1997).Somaticexperienceinpsychoanalysis.PsychoanalyticPsy-(4),517–530.Ekman,P.,&Friesen,W.V.(1984).Emotionfacialactioncodingsystem(EM-FACS)Francisco:UniversityofCalifornia,SanFrancisco.Ekman,P.,Friesen,W.V.,O’Sullivan,M.,Diacoyanni-Tarlatzis,I.,Krause,R.,Pitcairn,T.,etal.(1987).UniveralsandculturaldifferencesinthejudgmentsoffacialexpressionsofJournalofPersonalityandSocialPsychology712–717.Ekman,P.,Hager,J.C.,&Friesen,W.V.(1981).Thesymmetryofemotionalanddeliberatefacialactions.101–106.Flack,W.F.,Laird,J.D.,&Cavallaro,L.A.(1999).Emotionalexpressionandfeelinginschizo-phrenia:Effectsofspecificexpressivebehaviorsonemotionalexperiences.JournalofClinicalPsychology(1),1–20.Frank,M.G.,&Stennett,J.(2001).Theforced-choiceparadigmandtheperceptionoffacialexpressionofemotion.JournalofPersonalityandSocialPsychology(1),75–85.Fuchs,T.(1996).Bodilycommunicationandrelateddisorders.ZeitschriftFurKlinischePsy-chologiePsychiatrieUndPsychotherapie(4),415–428.Furnham,A.(1999).BodylanguageatworkhannelIslands:TheGuernseyPress.Haidt,J.,&Keltner,D.(1999).Cultureandfacialexpression:Open-endedmethodsfindmoreexpressionsandagradientofrecognition.CognitionandEmotion(3),225–266.Hirsch,M.(1994).Thepatient’sbodyinpsychoanalyticpsychotherapy.(3),153–157.Horstmann,G.(2003).Whatdofacialexpressionconvey:Feelingstates,behavioralintentions,oractionrequests?(2),150–166.Inouye,K.S.(1998).Featurebasedcodingofemotivebodyposturenpublishedmanuscript,EarlhamCollege,IN.Izard,C.(1994).Innateanduniversalfacialexpression:Evidencefromdevelopmentalandcross-culturalresearch.PsychologicalBulletin288–299.Izard,C.E.(1979).MAXtrainingmanualnpublishedmanuscript,UniversityofDelaware,Newark,Delaware.James,W.T.(1932).Astudyoftheexpressionofbodyposture.JournalofGeneralPsychology405–437.Kourtzi,Z.,&Kanwisher,N.(2000).ActivationinhumanMT;MSTbystaticimageswithim-pliedmotion.JournalofCognitiveNeuroscience(1),48–55.Mehrabian,A.(1981).Silentmessages:Implicitcommunicationofemotionsandattitudes(2ndBelmont,CA:Wadsworth,Inc.Mesquita,B.,&Frijda,N.H.(1992).Culturalvariationsinemotion:Areview.179–204.Montepare,J.,Koff,E.,Zaitchik,D.,&Albert,M.(1999).Theuseofbodymovementsandgesturesascuestoemotionsinyoungerandolderadults.JournalofNonverbalBehavior(2),133–152. MARKCOULSONMurray,I.R.,&Arnott,J.L.(1993).Towardthesimulationofemotioninsyntheticspeech:Areviewoftheliteratureonhumanvocalemotion.JournaloftheAcousticalSocietyof(2),1097–1108.Ortony,A.,&Turner,T.J.(1990).What’sbasicaboutbasicemotions?PsychologicalReview315–331.Pollick,F.E.,Paterson,H.M.,Bruderlin,A.,&Sanford,A.J.(2001).Perceivingaffectfromarmmovement.B51–B61.Rinn,W.E.(1984).Theneuropsychologyoffacialexpression:Areviewoftheneurologicalandpsychologicalmechanismsforproducingfacialexpression.PsychologicalBulletin,52–Russell,J.A.(1991).Cultureandthecategorizationofemotions.PsychologicalBulletin426–450.Russell,J.A.(1995).Facialexpressionsofemotion:Whatliesbeyondminimaluniversality?PsychologicalBulletin379–391.Russell,J.A.,&Barrett,L.F.(1999).Coreaffect,prototypicalemotionalepisodes,andotherthingscalledEmotion:Dissectingtheelephant.JournalofPersonalityandSocialPsychol-(5),805–819.Scherer,K.R.(1986).Vocalaffectexpression:Areviewandamodelforfutureresearch.PsychologicalBulletin(2),143–165.Scherer,K.R.(1995).Expressionofemotioninvoiceandmusic.JournalofVoice(3),235–Schouwstra,S.J.,&Hoogstraten,J.(1995).Headpositionandspinalpositionasdeterminantsofperceivedemotionalstate.PerceptualandMotorSkills673–674.Skinner,M.,&Mullen,B.(1991).Facialasymmetryinemotionalexpression:Ameta-analysisofresearch.BritishJournalofSocialPsychology113–124.vanBezooijen,R.,&Boves,L.(1986).Theeffectsoflow-passfilteringandrandomsplicingontheperceptionofspeech.JournalofPsycholinguisticResearch(5),403–417.Walk,R.D.,&Homan,C.P.(1984).Emotionanddanceindynamiclightdisplays.ofthePsychonomicSociety(5),437–440.Walk,R.D.,&Walters,K.L.(1988).Perceptionofthesmileandotheremotionsofthebodyandfaceatdifferentdistances.BulletinofthePsychonomicSociety(6),510–510.Wallbott,H.G.(1998).Bodilyexpressionofemotion.EuropeanJournalofSocialPsychology(6),879–896.Wallbott,H.G.,&Scherer,K.R.(1986).Cuesandchannelsinemotionrecognition.ofPersonalityandSocialPsychology(4),690–699.Walters,K.L.,&Walk,R.D.(1986).Perceptionofemotionfrombodyposture.BulletinofthePsychonomicSociety(5),329–329.Walters,K.L.,&Walk,R.D.(1988).Perceptionofemotionfrommovingbodycuesinphoto-BulletinofthePsychonomicSociety(2),112–114.Watson,D.,&Tellegen,A.(1985).Towardaconsensualstructureofmood.219–235.Watson,D.,&Tellegen,A.(1999).Issuesinthedimensionalstructureofaffect—Effectsofdescriptors,measurementerror,andresponseformats:CommentonRussellandCarrollPsychologicalBulletin(5),601–610.Watson,D.,Wiese,D.,Vaidya,J.,&Tellegen,A.(1999).Thetwogeneralactivationsystemsofaffect:Structuralfindings,evolutionaryconsiderations,andpsychobiologicalevidence.JournalofPersonalityandSocialPsychology(5),820–830.Wehrle,T.,Kaiser,S.,Schmidt,S.,&Scherer,K.R.(2000).Studyingthedynamicsofemotionalexpressionusingsynthesizedfacialmusclemovements.JournalofPersonalityandSocial(1),105–119.Yik,M.S.M.,&Russell,J.A.(1999).Interpretationoffaces:Across-culturalstudyofapredic-tionfromFridlund’stheory.CognitionandEmotion(1),93–104. Reproducedwithpermissionofthecopyrightowner.Furtherreproductionprohibitedwithoutpermission.