teleological and referential understanding of action in ...cynthiab/readings/csibra2003.pdfpublished...

Published online 18 February 2003

Teleological and referential understandingof action in infancy

Gergely CsibraCentre for Brain and Cognitive Development School of Psychology Birkbeck College Malet Street

London WC1E 7HX UK (gcsibrabbkacuk)

There are two fundamentally different ways to attribute intentional mental states to others upon observingtheir actions Actions can be interpreted as goal-directed which warrants ascribing intentions desires andbeliefs appropriate to the observed actions to the agents Recent studies suggest that young infants alsotend to interpret certain actions in terms of goals and their reasoning about these actions is based on asophisticated teleological representation Several theorists proposed that infants rely on motion cues suchas self-initiated movement in selecting goal-directed agents Our experiments revealed that althoughinfants are more likely to attribute goals to self-propelled than to non-self-propelled agents they do notneed direct evidence about the source of motion for interpreting actions in teleological terms The secondmode of action-based mental state attribution interprets actions as referential and allows ascription ofattentional states referential intents communicative messages etc to the agents Young infants alsodisplay evidence of interpreting actions in referential terms (for example when following othersrsquo gaze orpointing gesture) and are very sensitive to the communicative situations in which these actions occur Forexample young infants prefer faces with eye-contact and objects that react to them contingently andthese are the very situations that later elicit gaze following Whether or not these early abilities amountto a lsquotheory of mindrsquo is a matter of debate among infant researchers Nevertheless they represent skillsthat are vital for understanding social agents and engaging in social interactions

Keywords infancy referential action goal-directed action teleology theory of mind

1 INTRODUCTION

People cannot help but interpret each otherrsquos actions interms of hidden mental states like beliefs desires etc Theattribution of these intentional mental states helps themto explain observed and predict future behaviour ofothers and also enables them to influence what socialpartners will do According to philosophers (eg Dennett1987) the distinctive aspect of intentional mental statesis their lsquoaboutnessrsquomdashthey are lsquoaboutrsquo certain states of theworld There are two fundamentally different ways tointerpret actions as indicating mental states that are aboutsomething First we can interpret an observed behaviouras a goal-directed action A goal-directed action is lsquoaboutrsquothe end state of that action it is performed in order tomake the end state occur If you see someone drilling acorkscrew into the cork of a bottle of wine you will knowwhat this action is about an open bottle of wine Becausegoal-directed actions are seen to be determined by theirend state this kind of action interpretation is essentiallyteleological Teleological interpretation of an action enablesus to attach a goal to the action (an open bottle) but italso allows us to attribute to the agent desires (eg a desireto drink wine) beliefs (eg a belief that there is wine inthe bottle) and possibly further mental states as well Thesecond way to reveal what an action is about is to interpret

One contribution of 15 to a Theme Issue lsquoDecoding imitating andinfluencing the actions of others the mechanisms of social interactionrsquo

Phil Trans R Soc Lond B (2003) 358 447ndash458 447 Oacute 2003 The Royal SocietyDOI 101098rstb20021235

it as a referential action A referential action is about thestate of the world that it highlights If you see someonepointing to a car you will know what this action is aboutit is about the car Interpretation of an action as referentialemphasizes some aspect of the world as connected to theagent and allows attribution of referential intentions(drawing your attention to the car) communicative mess-ages (eg lsquothis is the car I was talking aboutrsquo) and othermental states to the agent

During the past decade several studies have been pub-lished that attempted to see whether young preverbalinfants are engaged in these kinds of action interpretationsand whether they attribute intentional mental states toothers The answer to these questions bears relevance tothe debates on the origins and the nature of the humanlsquotheory of mindrsquo and may also help us to understanddevelopmental disorders that are characterized by deficitsin social cognition This paper reviews some of the evi-dence relevant to these questions I shall argue that thisbody of evidence unambiguously shows that infantsyounger than 1 year of age employ both kinds of actioninterpretation and do this in a more and more sophisti-cated way as they approach their first birthday At thesame time I shall also argue that neither of these actioninterpretation systems is necessarily mentalistic both canfunction without attribution of representational mentalstates In other words goals can be attached to actionswithout understanding desires and referents can be ident-ified without figuring out the meaning of the underlyingcommunicative act I propose that these two action

448 G Csibra Action understanding in infancy

interpretation systems operate independently in earlyinfancy The independence of these cognitive mechanismsis supported by various facts they rely on different typesof computations they are triggered by different stimulusconditions and they can be dissociated in animal behav-iours and developmental disorders Crucially there is noevidence for transfer between these action interpretationsystems (for example attribution of goals on the basis ofreferential actions) in infants at or below 1 year of ageI suggest that the combining of these action interpretationsystems into a higher-order mentalistic representation ofactions takes place during the second year of life

2 TELEOLOGICAL UNDERSTANDING OF ACTIONS

Understanding goals requires connecting actions not totheir antecedents but to their consequents Some studiessuggest that infants as young as six months of age aresensitive to the end state of observed events Woodward(1998) for example repeatedly presented infants with anaction in which a hand reached towards and grasped oneof two toys When the infants had been habituated to thisevent she swapped the toys and the hand either graspedthe same toy in the new location or the other toy in thesame location Looking times for these two events weremarkedly different infants looked longer at the hand grasp-ing the lsquonewrsquo toy at the lsquooldrsquo location as if they expectedthe hand to reach towards the same toy This result showsthat infants are more likely to associate a grasping handwith the grasped object than with its location which is use-ful for interpreting the grasping action as directed to thegoal of acquiring a specific object Nothing in this experi-ment however demonstrates that infants do not simplydiscriminate between end states It does not show that theyrelate actions to end states in an lsquoin-order-torsquo clause thehand performed the reaching and grasping sequence inorder to acquire a certain object

An action is goal directed if it is performed not for itselfbut to achieve an end in other words if it is an instrumen-tal action Whether an action is instrumental in relationto an end state or not can be intuitively tested by consider-ing if we would expect it to be performed when it is notneeded for goal achievement Pulling out a cork from awine bottle for example can be an instrumental action toaccess the wine but only when the bottle is not emptyEven when the bottle is not empty we would not expectto see this action if all the glasses were filled with wineAs this example suggests interpreting something as aninstrumental action that has been performed to achieveand gets its meaning from a particular goal state requiresconsideration of many thingsmdashnot just the action and itsend state but also the environment in which it occurs Wehave demonstrated that nine-month-old infants arealready capable of doing this trick and depending on theenvironment interpret observed actions as goal directed

These studies (Gergely et al 1995 Csibra et al 1999)presented infants with computer animations Since theseminal work by Heider amp Simmel (1944) several experi-ments provided evidence that high-level sophisticatedsocial interpretation of actions does not require featuralidentification of agents (for a review see Scholl amp Tremoulet2000) Triangles and circles moving on a two-dimensionalsurface are readily interpreted as if they were people

Phil Trans R Soc Lond B (2003)

engaged in various kinds of social interactions One expla-nation of this phenomenon is that during development themotion patterns of human behaviour are abstracted awayfrom their usual context and from human features andwhen these patterns are recognized in artificially createdanimations we tend to project further human attributesto those abstract figures An alternative explanation whichI defend here suggests that certain motion patterns thatallow interpretation of the behaviour of the abstract figuresas goal-directed actions will always attract such aninterpretation and this tendency does not depend onextensive experience with human behaviour This expla-nation predicts that young infants will as readily attributegoals to animated figures as they do to human agents

Our experiments repeatedly presented infants with asimple animation (figure 1) in which a ball approachedand contacted another ball by jumping over an obstacleIn this event the jumping action can be interpreted as aninstrumental action because it is necessary to achieve theend state (the spatial position beside the other ball)Whether or not the infants arrived at the same interpret-ation was tested by showing them two modified versionsof the event and measuring their looking time In the testevents we removed the obstacle which changed therelevant aspects of the environment in a way that madethe jumping action unnecessary for goal achievement Oneof these events (old action) displayed the same jumpingapproach which however was no longer necessary andwas therefore inefficient while the other event (new action)displayed a straight-line approach to the same positionwhich was an efficient action in the new situation Ifinfants interpreted the original event as a goal-directedaction they should find the new instrumental action morecompatible with their interpretation and should respondwith longer looking at the old inefficient action

This is exactly what we found in 9- and 12-month-oldinfants (see figure 1) Further control studies were per-formed to verify whether these looking time differences wereindeed attributable to the interpretation of the habituationevent The control studies involved the same test eventsbut they were preceded by a habituation event which dif-fered only slightly from the original habituation event butdid not show a proper instrumental action In this eventthe lsquoobstaclersquo was positioned not in between the two ballsbut behind the moving ball (see figure 2) In this environ-ment the jumping action cannot be considered to be aninstrumental action and the infants were not expected toattribute the end position as the goal of the moving ballrsquosaction Their looking times confirmed this prediction

How can we explain infantsrsquo early emerging ability toattribute goals to animated shapes We have proposed(Gergely amp Csibra 1997 Csibra et al 2003) that watchingthese animations on a computer screen infants adopt ateleological stance The teleological stance is akin to theintentional stance of Dennett (1987) in that it representsan interpretational strategy that seeks to construe an eventin terms of goals (see Keil (1995) for a different versionof the teleological stance) It is however different fromthe intentional stance in that it does not attribute mentalstates to the agents (I will return to the relation betweenthese two stances later) Note that the teleological stance(and the intentional stance) is not an explicit inferentialsystem but a bias a tendency to construe events in accord

Action understanding in infancy G Csibra 449

habituation90

80

70

60

50

30

20

10

1 2 3 ndash3 ndash2 ndash1

p lt 005

p lt 001

old action new actiontest

seco

nds

Figure 1 The experimental events and the looking time results in Gergely et al (1995) and Csibra et al (1999) Light greybars six-month-old infants medium grey bars nine-month-old infants dark grey bars 12-month-old infants

habituation90

80

70

40

30

20

10



seco

nds

Figure 2 The control events and the the looking time results in Gergely et al (1995) and Csibra et al (1999) Medium greybars nine-month-old infants dark grey bars 12-month-old infants

with a certain formal structure Construing an action asgoal-directed implies as we have seen relating at leastthree different aspects of the observed event to each otherthe behaviour its physical context and the end state (seefigure 3) These three elements will create a well-formedteleological representation of the event if and only if thebehaviour is an efficient action towards the end state inthe given physical environment The habituation event inour study (figure 1) met this criterion (jumping over theobstacle was the most efficient action towards the end pos-ition in that environment) while the habituation event inthe control study (figure 2) did not (jumping over nothingwas not an efficient action towards the end position in thatenvironment) Thus the habituation event represented infigure 1 allowed goal attribution while the habituation


event represented in figure 2 did not and this differencewas reflected in the differential patterns of looking timeby infants

As figure 3 suggests goal attribution and efficiencyevaluation are inseparable in the teleological stance Butperhaps this is only true for behaviours of abstract figuresand infants may be willing to attribute goals to inefficientactions as well if they are performed by real humanbeings Recent studies suggest that this is not the caseWoodward amp Sommerville (2000) for example presentedinfants with two transparent boxes that contained two dif-ferent toys They habituated 12-month-old infants to anaction in which a hand first touched one of the boxes thenopened it and grasped the toy inside After habituationthe toys were swapped between the boxes During the test


observed action

behaviour action

satisfying well-formedness criterion

(efficiency)

physicalcontext

endstate constraints goal

interpreted action

Figure 3 The teleological representation for actions

event the hand either touched the same box as before(which however now contained the other toy) or ittouched the other box (which contained the same toy thathad been previously grasped) Infants looked longer at theformer action indicating that they did not expect the handto perform the familiar action seen before as that was nolonger necessary and in fact would have been an inef-ficient way to obtain the goal object (ie the toy that hadbeen grasped during habituation) In a control studyWoodward and Sommerville habituated the infants to thesame hand actions (first touching the box then graspingthe toy) with the exception that the toys were not insidebut in front of the boxes hence opening the box couldnot have been considered as an efficient instrumentalaction for grasping the toy In this condition the infantsdid not develop any specific expectation about which boxthe hand should touch when the toys were swapped andin fact they even looked slightly longer when the handtouched the other box Note that just as in our studydescribed above the only difference between the two con-ditions was the efficiency of the action opening the boxis an efficient instrumental action if and only if the targetobject is in the box

A more recent study has also demonstrated that infantsdo care about efficiency when they interpret actions interms of goals Onishirsquos (2001) studies also confirmed thatactions (removing obstacles) that make a target objectaccessible to a hand are interpreted as goal-directed by10-month-old infants but the same actions do not leadto goal attribution if they are not justified by the relativepositions of the obstacles and the target object Further-more evaluating the efficiency of actions is not restrictedto looking-time context either Gergely et al (2002) pro-vided evidence that infants modulate their imitative behav-iour according to the justifiability of the goal-directedactions performed by a model These authors replicatedthe well-known demonstration of Meltzoff (1988) that 14-month-old infants tend to imitate a behaviour that theyhave never seen before and would not perform spon-taneously In Meltzoffrsquos study infants watch a model lean-ing forward and touching an object on a table with hishead causing it to light up A week later when they arebrought back to the laboratory the majority of the infantsperform the same action In an additional condition how-ever Gergely et al slightly modified the modelrsquos behav-iour Before touching the object with her forehead theexperimenter covered her shoulder with a blanket whichthen she held onto tightly with her handsmdashthen she per-formed the same action Or was it the same actionTouching the object with her head was a perfectly reason-able action in a situation where the actorrsquos hands wereunavailable while in the other situation where she kept


her hands free it just seemed unjustifiable If infants arenot sensitive to this difference they should imitate thehead-action equally In fact only a minority of them imi-tated the head-action in the lsquohands occupiedrsquo versionwhich suggests that they interpreted this action as aninstrumental action that one does not have to copy if heor she is free from the constraints that affected the modelIn other words they could attempt to achieve the sameeffect in the most efficient manner that was available tothem by touching the object with their hands Infants docare about efficiency of perceived actions

We have seen that nine-month-old infants will attributethe end state as the goal of the action if it is an efficientinstrumental action but people do not have to observe acomplete action to attribute a goal to it We can figure outwhat the goal of an action could be by considering whatend state that action would be instrumental to Pulling outthe cork from a bottle is most likely to be carried out inorder for the agent be able to access the content of thebottle When we make these kinds of inferences we haveinformation about the agentrsquos behaviour and the physicalenvironment in which it takes place We then adopt theteleological stance and try to fill out the missing thirdelement (the end state) of the teleological representation inorder to satisfy the well-formedness criterion of efficiency(see figure 3) Can infants do the same trick

Meltzoff (1995) let 18-month-old infants watch a modelwho performed apparently failed actions on novel objectsWhen these babies had a chance to imitate the modelthey did not copy the failure but performed the completeintended action that led to the goal state that one couldhave inferred from the modelrsquos behaviour This is a cleardemonstration that 18-month-olds do not have to see thegoal realized in order to be able to attribute it to an actionTwelve-month-old infants however fail in this task(Bellagamba amp Tomasello 1999) suggesting that they maynot be able to extract the goal from the observation offailed attempts It is possible though that this testrequires too much from babies they not only have toattribute an unseen goal to the agent they also have toignore the observed end state of the action and replace itwith an inferred one In other words Meltzoffrsquos taskrequires a kind of counterfactual reasoning that mayexceed 12-month-old infantsrsquo capabilities even if they wereable to figure out goals for unfinished actions

To avoid the complications inherent in the Meltzoff(1995) study we created a computer animation thatallows teleological interpretation of an action even thoughthe goal is never seen achieved (Csibra et al 2003) Theanimation shows a simple chase event in which a biggerball follows a smaller one (see figure 4 habituation event)When the small one passes through a narrow gap betweentwo barriers the big ball takes a detour around the barriersand then continues its path in the direction where thesmall one left the screen The goal of the big ball can easilybe identified catching the small ball Note however thatthis goal is never seen achieved and can only be inferredfrom the evaluation of the big ballrsquos behaviour We perfor-med two different tests on two different groups of 12-month-old infants to check whether they interpret theunfinished chase event as a goal-directed action In thefirst test (figure 4) we changed the physical environmentby enlarging the gap between the barriers and presented


habituation90

80

70

60

50

30

20

10


p lt 005

incongruent action congruent actiontest

seco

nds

Figure 4 Evidence for attribution of an unseen goal in 12-month-old infants Medium grey bars nine-month-old infants darkgrey bars 12-month-old infants (Data from Csibra et al (2003) experiment 1)

habituation90

80

70

60

50

30

20

10


p lt 005

incongruent goal congruent goaltest

seco

nds

Figure 5 Evidence for attribution of an unseen goal in 12-month-old infants (Data from Csibra et al (2003) experiment 1A)

two actions the big ball either adjusted its path to the newconstraints and followed the small ball through the gap(congruent action) or took the same detour as beforewhich however was no longer an efficient action towardsthe same goal (incongruent action) In the second test(figure 5) we opened up the previously hidden part of thescene where the balls had left and confronted the infantswith two different endings for the story the small ballhalted and the big ball either stopped next to it (congruentgoal) or changed its path travelled past the small one andleft the scene (incongruent goal) Seeing an incongruentaction or an incongruent goal resulted in longer lookingtimes than seeing the corresponding congruent action orcongruent goal events indicating that 12-month-oldinfants took the teleological stance and were able to figure


out the unseen goal of an agent This result was recentlyreplicated by Wagner amp Carey (2002)

The teleological representational schema for actions(figure 3) allows a further type of inference as well Sofar we have seen that knowing the goal and the physicalconstraints infants can predict new actions and knowingthe physical constraints and the actions they can attributegoals A third type of inference that one can logicallyderive from this representational format would allow figur-ing out some invisible physical constraints on the basis ofthe observed action and its end state This inferencewould be drawn on the same basis as the previous infer-ences filling in the missing element (in this case thephysical constraints) of the schema with something thatmakes it a well-formed representation ie with something


habituation90

80

70

60

50

30

20

10


p lt 001

no obstacle obstacletest

seco

nds

Figure 6 Evidence for the inference of an invisible action constraint in 12-month-old infants Medium grey bars nine-month-old infants dark grey bars 12-month-old infants (Data from Csibra et al (2003) experiment 2)

that satisfies the efficiency principle To elaborate ourexample if someone takes a corkscrew and starts drillingit into a bottle we will spontaneously assume that thereis a cork in the bottle because otherwise this action wouldnot make any sense ie it could not be interpreted as anefficient action towards the known goal (accessing thewine in the bottle)

To test whether infants make similar inferences wepresented them with a computer-animated event (Csibraet al 2003) which was similar to those we used in ourearlier studies (see figure 1) a ball approached anotherball by a jumping action The event here however dif-fered from the original studies in two respects we madethe animation three dimensional and occluded the partof the space that the ball jumped over (figure 6) In thetest phase the occluder was removed and it either revealedan object or an empty space If infants justify the observedjumping action by inferring the presence of an obstaclebehind the occluder seeing the obstacle would confirmwhile seeing the empty space would violate their expec-tation which should be reflected by longer looking timein the latter case This is exactly what we found Twelve-month-old infants inferred the presence of an obstacle onthe sole basis of the behaviour of the ball Note that theabsence of the obstacle does not violate any physicalknowledge it does not have to be there But its absenceviolates our expectation that the object approaches its goalefficiently and questions the interpretation that the actionis performed in order to achieve an end ie that it is agoal-directed action

All these results indicate that at least by their first birth-day if not earlier infants rely on a quite sophisticatedteleological representational system when they interpretbehaviours and they use this system productively to figureout invisible aspects of actions (Csibra et al 2003) Doesthis conclusion entail that they use a lsquotheory of mindrsquo iedo they attribute representational mental states such asintentions desires and beliefs to the agents Infants mayindeed interpret the observed behaviours in mentalistic


teleological representation

constraints

action

goal

mentalistic representation

beliefs

intentions

desires

Figure 7 The teleological and mentalistic actionexplanations

terms but this is not a necessary implication of theseresults No doubt one could account for these findings byassuming that the infants interpreted the observed event aslsquothe ball wanted to touch the other ball believed that theobstacle was impenetrable and decided to jump over itrsquoIn this interpretation the elements of the teleologicalschema are projected into the agentrsquos mind as contentsof hisher mental states goals become desires constraintsbecome beliefs and actions become intentions (figure 7)Note however that within the particular context of actioninterpretation there is no benefit gained from the compu-tationally more demanding mental state attributionmdashonecan predict exactly the same actions from goals and physi-cal constraints as from contents of desires (ie goals) andcontents of beliefs (ie physical constraints) The additionalbenefit of relying on mental states comes from situationswhere the mental states are attributed independently of theactual action and are used in action predictions (lsquohe opensthe bottle but he does not like wine so he wants to offerit to someone elsersquo) But none of the studies I have reviewedabove required such inferences and therefore they did notprovide conclusive evidence for mental state attributionIn fact we have argued earlier (Csibra amp Gergely 1998)that a non-mentalistic purely teleological action interpret-ation system (the teleological stance) developmentally pre-cedes the later emerging mature theory of mind (theintentional stance)


A further question arises from the fact that interpretingan action as goal-directed is not a causal inference buta result of a specific stance (whether it is teleological orintentional) What makes infants decide that an observedbehaviour is to be interpreted from this stance ie it isto be evaluated in terms of its efficiency We all take theintentional stance when dealing with fellow human beingsbut we also apply mentalistic terms to animals to naturalphenomena and even to machines (see Dennett 1987)and neither Heider amp Simmelrsquos (1944) triangles nor ourjumping balls had any animate let alone human featuresA plausible assumption is therefore that we (and infants)rely on behavioural rather than featural cues to identifyagents that are possibly engaged in goal-directed actionsThe most direct hypothesis about these cues was put for-ward by David Premack (1990) According to him infantswill treat as intentional (hence goal-directed) any agent thatappears to be self-propelled This idea has been incorpor-ated into several theories of infant development (Mandler1992 Baron-Cohen 1994 Carey amp Spelke 1994 Leslie1994) and was originally one of the hypotheses thatinspired our studies (Gergely et al 1995) Surprisinglyhowever not much research was devoted to verifying thishypothesis Some studies (eg Kaufman 1998) suggest thatinfants make a distinction between self-propelled and exter-nally driven objects from an early age and develop differentkinds of expectations towards the two classes of objects(Luo amp Baillargeon 2002) Our recent studies indicate that12-month-old infants are more likely to attribute goals toanimated agents that appear to move by themselves thanto agents that are launched by other objects (Gergely ampCsibra 2003) This indicates that self-motion does indeedwork as a cue for goal-directedness

But this is not the whole story Infants in our study weremore reluctant to attribute goals to externally propelledobjects but still tended to take the teleological stancetowards them as long as their behaviour appearedefficient Even when we removed all cues of self-motionand animacy but left enough information to evaluate theefficiency of goal approach (Csibra et al 1999) 9- and12-month-old infants were willing to attribute a goal tothe observed action These results suggest that althoughself-propulsion works as cue it is not obligatory and theremay also be other cues for goal-directedness Such cuescan be derived from the efficiency principle itself Theprinciple requires for example that behaviours directedto the same goal be adjusted in relation to the relevantaspects of the environment in which they occur Conse-quently the perception of behavioural adjustment that isa function of situational constraints may serve as the trig-gering condition for analysing the behaviour as goaldirected

The left side of table 1 summarizes the proposed spe-cifics of the cognitive system that allows infants to inter-pret actions as goal-directed

3 REFERENTIAL UNDERSTANDING OF ACTIONS

Understanding an action as referential requires linkingthe actorrsquos behaviour to specific objects or to specificaspects of the environment These actions such as point-ing to or looking at an object normally occur in communi-cative contexts direct the observerrsquos attention to that


object and may help to secure a referent for other com-municative signals such as verbal utterances

Studies on early language acquisition indicate that thereis a special context where referential understanding ofactions is indispensable Young children have been shownto be specifically sensitive to where someone is looking orpointing at when uttering a new word and interpret theword and the observed action as referring to the samething (Baldwin 1993 Tomasello 1999 Bloom 2000) Inother words they use referential interpretation of non-ver-bal actions to disambiguate the new wordrsquos referent Thisis a clear demonstration that 18-month-old infants canunderstand actions in referential terms Unfortunatelylinguistic tests are difficult to administer before 18 monthsof age so they are not feasible for testing the understandingof referential actions in younger infants

To find earlier referential contexts that do not rely onword learning Louis Moses and his colleagues turned toanother phenomenon known as lsquosocial referencingrsquo(Moses et al 2001) When infants are confronted with anew situation or with a novel object they tend to checktheir parentsrsquo or other adultsrsquo face before approaching itand modulate their behaviour accordingly (Campos ampStenberg 1981) For example they cross a visual cliff iftheir mother is smiling at the other side but refrain fromcrawling over if she looks worried (Sorce et al 1985)Moses and his colleagues tested whether this modulationof behaviour is specific to the object that the adult waslooking at when she expressed a certain emotion Theyarranged situations where an experimenter and the infantwere focusing on either the same or different objects andthen the experimenter expressed either a positive or anegative affect both verbally and by facial expressions Inresponse to this infants always looked at the adultrsquos faceand checked their line of regard If infants understand theadultrsquos emotional expressions in referential terms theyshould modulate their behaviour towards the object thatthe experimenter was looking at with that emotion evenwhen their own attention was engaged by another objectAnd this is precisely what they found Twelve-month-oldinfants explored the target object longer when it had beenassociated with positive affect even when it was not intheir own attentional focus This is a clear evidence forreferential understanding of looking

This referential understanding of looking behaviour isassisted by infantsrsquo tendency to follow the gaze of otherhuman beings If you make eye contact with a 12-month-old infant and then conspicuously look at some otherobject in the environment she will follow your gaze andmany times she will rest her gaze on the same object Thisbehaviour which is often called lsquojoint attentionrsquo emergesduring the second half of the first year and its accuracydevelops rapidly (Butterworth amp Jarrett 1991)

But sensitivity to the gaze direction of others can bedemonstrated even earlier in laboratory situations Ifthree- to six-month-old infants are presented with a targetobject on one side of a computer screen they are moreinclined to orient towards it if they perceive a gaze-shifton a face to the same direction just milliseconds earlier(Hood et al 1998) This phenomenon has been shown tobe partly explainable by sensitivity to motion cues pro-vided by the perceived shift of pupil position (Farroni etal 2000) but this does not account for all aspects of the


Table 1 The main characteristics of the two-action interpretation systems

goal-directed action referential action

about

criteria of application

initial principlemature principlerepresentation

a future stateself-propelled agent contingent adjustment biological motion

efficiencyrationality

constraints goal

action

a current statecommunicative situations (eye contact contingent

directionality

relevance agent

action

referent

reactivity)

results Recent studies revealed that pupil motion cues areeffective in eliciting shifts of attention in infants if andonly if they are preceded by a period of eye contact(Farroni et al 2003) Motion cues do not elicit attentionalshifts if the perceived gaze of the face on the computerscreen is moving from the side to the centre (ie from anaverted to an eye-contact position) or when the face ispresented upside down This makes sense because at leastin humans making eye contact is the simplest way ofestablishing a communicative situation and referentialactions usually occur in communicative contexts

An interesting hypothesis that one can draw from theseresults is that referential understanding of actions isassisted by infantsrsquo sensitivity to two kinds of cues thosethat indicate a communicative situation and those thatindicate spatial directions The combination of these twotendencies (lsquolook for communicative situations and if youfind one follow the directionrsquo) may ensure that infantswill find the referent of a communicative act in most casesAgain the sensitivity to communicative situations doesnot imply an understanding of communication Rather itrepresents a bias in processing the information availablein the infantrsquos environment This hypothesis provided uswith the prediction that even the youngest infants mustbe sensitive to the best cue for a communicative situationie eye contact We tested this hypothesis with newbornsin a simple preferential looking paradigm (Farroni et al2002) Seventeen 1ndash5-day-old newborns were shown twofaces one that looked straight at them (direct gaze) andone that looked away (averted gaze) All but two of themlooked longer at and all of them looked more timestowards the face with the direct gaze (figure 8) This earlyand most probably innate preference for eye contact canbe interpreted in various theoretical frameworks that weredeveloped to explain early sensitivity to social cues It fitswell into the eye-direction detection mechanism hypothes-ized by Baron-Cohen (1994) and it can also be adaptedinto Morton amp Johnsonrsquos (1991) lsquoCONSPECrsquo mech-anism which orients babies to faces The fact that faceswith direct gaze engage four-month-old infantsrsquo brain cir-cuits that are associated with face perception stronger thando faces with averted gaze (Farroni et al 2002) is moreconsistent with the latter theory But whatever the exactmechanism is the fact that newborns are sensitive not justto faces (Johnson et al 1991) and eyes (Batki et al 2000)but also to eye contact gives them a kick-start towardsunderstanding referential actions


However eye contact is not the only method for estab-lishing a communicative situation Contingent responsesfrom a source may also indicate that the source is com-municating with you John Watson argued 30 years agothat very young infantsrsquo well-known sensitivity for objectsthat respond to their own actions with high but imperfectcontingency is not simply a preparedness for operant con-ditioning but a way to find social partners in the world(Watson 1972) Suzan Johnson and colleagues have dem-onstrated that 12-month-old infants will follow the lsquogazersquoof a non-human object if it reacts to the childrenrsquos actionsand vocalizations contingently (Johnson et al 1998 seealso Johnson 2003) This is a very clear example of inter-preting a behaviour as a referential action when the onlycue for treating the objectrsquos behaviour meaningfully is thecommunicative situation established by contingent reac-tivity And contingent reactivity can indicate a communi-cative context for 10-month-old infants even when itcomes from a clearly mechanical object like a robot(Movellan amp Watson 2002)

This example suggests that early understanding of refer-ential actions evident in lsquogaze followingrsquo phenomena orig-inates not from a rich comprehension of the link betweena mental state of an agent and its referent but from a blindtracking of motion cues in communicative contexts Inother words referential interpretation of actions just liketeleological interpretation of actions represents not aknowledge but a lsquostancersquo Taking this lsquoreferential stancersquotriggers a search for referents on the basis of directionalcues and should be initially restricted to well-defined com-municative contexts This interpretation of early capacitiesexplains several aspects of the results in this fieldAlthough infants tend to follow the gaze of others initiallythey will not find the object looked at by the other person(Butterworth amp Jarrett 1991) Even when nine-month-oldinfants follow pointing gestures they would not necessar-ily associate the pointing action with the pointed object(Woodward amp Guajardo 2002) It is not before they are16 months of age that they tie directional motion cues tothe line of regard of others correctly for example whenthey follow head turns that they observe from behind(Muir amp Lee 2002 see also Johnson 2003)

What purpose does this rudimentary understanding ofreferential actions serve if it does not specify what theactorrsquos intentions have been behind hisher actions ie ifit does not provide mental state attribution The answerseems obvious referential understanding of actions might


(a)

(c)

(b)

100

mea

n lo

okin

g ti

me

(s)

mea

n nu

mbe

r of

ori

enta

tion

s

50

looking time

orientations

ndash10 ndash05 05 10

20

10

Figure 8 Newborn babies look longer at (a) and more times towards (b) faces with direct eye gaze (c) Individual preferencescores also show a bias towards direct eye gaze both in looking times and number of orientations (Data from Farroni et al(2002))

have evolved to support childrenrsquos word learning Ifinfants interpret linguistic utterances and non-verbal com-municative actions like looking or pointing in referentialterms then they can assume that simultaneous referencesby the same person will refer to the same object If non-arbitrary referential relationships like the spatial relation-ships inherent in looking and pointing help them find areferent it may also help them establish arbitrary referen-tial relationships like the one between words and objects(The study by Moses et al (2001) that is mentioned aboveshows how this works in a non-linguistic context) Thisrole of action interpretation in word learning is well docu-mented (Baldwin 1993 Tomasello 1999 Bloom 2000)but it is usually assumed that this interpretation must yielda mental state (ie the referential intention of the speaker)to which both the verbal and non-verbal actions can bemapped This is however not required direct mappingbetween words and non-verbally referred objects can alsofunction as a bootstrapping mechanism into word learn-ing Recently Sperber amp Wilson (2002) have also pro-posed that the early processing of communicative signalsmay not be based on general-purpose mind-readingmechanisms but relies on a sub-module which evolved tosupport fast comprehension of ostensive stimuli Under-standing of referential intentions does indeed play animportant role in language acquisition from 18 to 24months of age Nevertheless the development of thisessential cognitive skill suggests that an understanding ofreferential intentions is not a precondition for but the pro-


duct of understanding referential actions This hypothesisis parallel with the one I proposed for teleological actionunderstanding Just as the notion of pre-existing goals (iedesires) is derived developmentally from teleologicalunderstanding of actions and not the other way aroundI propose that the notion of pre-existing lsquomeaningrsquo (iethe communicative message) is derived developmentallyfrom referential understanding of actions and not theother way around

It is not yet clear how this development is achievedHowever early understanding of referential actions atleast in one respect seems to be very similar to the prin-ciple that governs mature human communication therelevance principle (Sperber amp Wilson 1986) The appli-cation of both the relevance principle and the referentialinterpretation of actions depends only on the recognitionof a communicative context and both work as a presump-tion that cannot be violated The specifics of the referen-tial action interpretation system are summarized in table 1

4 TELEOLOGICAL VERSUS REFERENTIALUNDERSTANDING OF ACTIONS

The evidence reviewed so far suggests that 12-month-old or even younger infants readily interpret action asgoal-directed or referential What is the relationshipbetween these two types of action understanding Someresearchers of infants tend to treat these two kinds ofaction interpretation as equivalent (eg Phillips et al


2002 Woodward amp Guajardo 2002) calling them bothlsquoobject-directed actionsrsquo This is because in the majorityof experiments with infants the agentrsquos goal is to seize anobject or the referred state of affairs is an object I thinkhowever that lsquoobject-directednessrsquo is a misleading termbecause it relies on a surface similarity between theseparticular cases of action interpretation (namely that bothrefer to a relationship between an agent and a distalobject) and conceals the fact that they rely on differentkinds of action understanding Indeed a comparison ofthe specifics of these two systems (see table 1) suggeststhat there is not much common in them they are triggeredby different cues apply different representations and com-putations serve different functions and are likely to beimplemented in separate mechanisms

One may argue however that the distinction betweenthese systems is artificial after all both represent inten-tional actions and the difference between them is simplydetermined by the content of the actual intention In otherwords they may be subsystems of a single lsquotheory of mindrsquomechanism (Leslie 1994) Nevertheless it is not only for-mal arguments that support the claim that these two kindsof action interpretation reflect two distinct cognitivemechanisms Various kinds of dissociations confirm thatthese action interpretation systems can operate indepen-dently without the help of a higher-level theory of mindThe first of these dissociations occurs in the animal king-dom Chimpanzees for example seem to be able to attri-bute goals to observed actions (Uller amp Nichols 2000)and can be trained to follow gaze alterations (Tomaselloet al 2001 Okamoto et al 2002) They also understandwhat other individuals can see (Call 2001) Howeverthere is no evidence that they conceive representationalmental states at all (Call amp Tomasello 1999) and indeedtheir understanding of seeing does not reflect an under-standing of reference (Povinelli et al 1999) A second dis-sociation can be seen in autism a developmental disordercharacterized by severe difficulties in attributing mentalstates to others (Frith 2001) However people with autismcan attribute goals to animated shapes the same way astypical children (Abell et al 2000 Castelli et al 2002)suggesting that their teleological action interpretation sys-tem is intact At the same time their main difficulties seemto be rooted in a non-functioning referential interpret-ational system they do not make eye contact (Phillips etal 1992) fail to understand eye gaze (Baron-Cohen1994) and as a result their acquisition of language is seri-ously delayed and impaired These dissociations couldhardly occur if the two action interpretation systems dis-cussed in this paper were simply different manifestationsof the same mechanism (for a more detailed discussion ofthese dissociations see Gergely (2002))

A third type of dissociation is provided by the studiesof human infants themselves If a single mentalistic actioninterpretation system existed which could attribute inten-tions by both teleological and referential interpretation ofactions it would provide a link between these mech-anisms as it does in older children and adults We canuse a referential act (eg pointing) to figure out the likelygoal of another person (eg obtaining the referred object)and can use a goal-directed act (eg searching) to figureout the referent of a word (Tomasello amp Barton 1994)There is no evidence however that young infants would


be able to make such inferences Thus I hypothesize thatthese two action interpretation systems initially representseparate mechanisms which will be integrated into ahigher-order mentalistic action interpretation during thesecond year of life

Indeed a suitable test for whether and when infantsattribute mental states to others would be a demonstrationof a transfer between these systems At what age do infantspass this test A recent study claimed to demonstrate sucha transfer ie goal prediction on the basis of lookingbehaviour at 12 and 14 months of age Phillips et al(2002) habituated infants to a person looking at one oftwo objects with a positive emotion In the test phaseinfants displayed longer looking time when the personheld in her hands the other object than when she held thesame object Phillips et al concluded that infants were ableto predict a goal-directed action (grabbing an object) fromthe referential relation (looking) between the person andone of the objects Note however that no instrumentalaction was presented to the infants in these studies Theymay have inferred that a certain action (grabbing theobject) must have taken place between looking and hold-ing behind the closed curtain but they did not have to dothat All they needed to do was match two referentialactions (looking and holding plus looking) with theirreferent and notice the change of the referred object Inother words they could have detected the incompatibilitybetween habituation and test events without goal attri-bution and action prediction ie entirely within the refer-ential action interpretation system

Other laboratories (eg Sodian amp Thormer 2000) havealso been trying to establish the age when a link betweenteleological and referential understanding of actions canbe demonstrated Further research is needed to establishnot just the timing but also the mechanism of the inte-gration between these systems as this step represents amajor milestone in the development of a mature theoryof mind

I thank Teresa Farroni Gyorgy Gergely Mark Johnson and ananonymous reviewer for their helpful comments on an earlierversion of this paper The author was supported by the UKMedical Research Council (programme grant G9715587)

REFERENCES

Abell F Happe F amp Frith U 2000 Do triangles play tricksAttribution of mental states to animated shapes in normaland abnormal development Cogn Dev 15 1ndash16

Baldwin D A 1993 Early referential understanding infantsrsquoability to recognize referential acts for what they are DevlPsychol 29 832ndash843

Baron-Cohen S 1994 How to build a baby that can readminds cognitive mechanisms in mind reading Cah PsycholCognCurr Psychol Cogn 13 513ndash552

Batki A Baron-Cohen S Wheelwright S Connellan J ampAhluwalia J 2000 Is there an innate gaze module Evidencefrom human neonates Infant Behav Dev 23 223ndash229

Bellagamba F amp Tomasello M 1999 Re-enacting intendedacts comparing 12- and 18-month-olds Infant Behav Dev22 277ndash282

Bloom P 2000 How children learn the meanings of words Cam-bridge MA MIT Press

Butterworth G amp Jarrett N 1991 What minds have in com-mon is space spatial mechanisms serving joint visual atten-tion in infancy Br J Devl Psychol 9 55ndash72


Call J 2001 Chimpanzee social cognition Trends Cogn Sci5 388ndash393

Call J amp Tomasello M 1999 A non-verbal false belief taskthe performance of children and great apes Child Dev 70381ndash395

Campos J J amp Stenberg C R 1981 Perception appraisaland emotion the onset of social referencing In Infant socialcognition (ed M Lamb amp L Sherrod) pp 273ndash314Hillsdale NJ Lawrence Erlbaum Associates

Carey S amp Spelke E 1994 Domain-specific knowledge andconceptual change In Mapping the mind domain specificityin cognition and culture (ed L Hirschfeld amp S Gelman) pp169ndash200 New York Cambridge University Press

Castelli F Frith C Happe F amp Frith U 2002 AutismAsperger syndrome and brain mechanisms for the attri-bution of mental states to animated shapes Brain 1251839ndash1849

Csibra G amp Gergely G 1998 The teleological origins of men-talistic action explanations a developmental hypothesisDevl Sci 1 255ndash259

Csibra G Gergely G Bgro S Koos O amp Brockbank M1999 Goal attribution without agency cues the perceptionof lsquopure reasonrsquo in infancy Cognition 72 237ndash267

Csibra G Bgro S Koos O amp Gergely G 2003 One-year-old infants use teleological representations of actions pro-ductively Cogn Sci 27 111ndash133

Dennett D C 1987 The intentional stance Cambridge MAMIT Press

Farroni T Johnson M H Brockbank M amp Simion F2000 Infantsrsquo use of gaze direction to cue attention theimportance of perceived motion Visual Cogn 7 705ndash718

Farroni T Csibra G Simion F amp Johnson M F 2002Eye-contact detection in humans from birth Proc NatlAcad Sci USA 99 9602ndash9605

Farroni T Mansfield E M Lai C amp Johnson M H 2003Infants perceiving and acting on the eyes tests of an evol-utionary hypothesis J Exp Child Psychol (In the press)

Frith U 2001 Mind blindness and the brain in autism Neuron32 969ndash979

Gergely G 2002 The development of understanding self andagency In Blackwellrsquos handbook of childhood cognitive develop-ment (ed U Goswami) pp 26ndash46 Oxford Blackwell

Gergely G amp Csibra G 1997 Teleological reasoning ininfancy the infantrsquos naive theory of rational action A replyto Premack and Premack Cognition 63 227ndash233

Gergely G amp Csibra G 2003 Perception of autonomousmotion and goal attribution by one-year-old infants (Inpreparation)

Gergely G Nadasdy Z Csibra G amp Bgro S 1995 Takingthe intentional stance at 12 months of age Cognition 56165ndash193

Gergely G Bekkering H amp Kiraly I 2002 Rational imitationin preverbal infants Nature 415 755

Heider F amp Simmel S 1944 An experimental study of appar-ent behavior Am J Psychol 57 243ndash259

Hood B M Willen J D amp Driver J 1998 Adultrsquos eyes trig-ger shifts of visual attention in infants Psychol Sci 9131ndash134

Johnson M H Dziurawiec S Ellis H amp Morton J 1991Newbornsrsquo preferential tracking of face-like stimuli and itssubsequent decline Cognition 40 1ndash19

Johnson S C 2003 Detecting agents Phil Trans R SocLond B 358 549ndash559 (DOI 101098rstb20021237)

Johnson S C Slaughter V amp Carey S 1998 Whose gazewill infants follow The elicitation of gaze-following in 12-month-olds Devl Sci 1 233ndash238

Kaufman L E 1998 Five-month-old infants distinguishbetween inert and self-moving inanimate objects In 11thBiannual Int Conf on Infant Stud April 1998 Atlanta GA


Keil F C 1995 The growth of causal understanding of natu-ral kinds In Causal cognition a multidisciplinary debate (edD Sperber D Premack amp A J Premack) pp 234ndash262Oxford University Press

Leslie A M 1994 ToMM ToBy and Agency core architec-ture and domain specificity In Mapping the mind domainspecificity in cognition and culture (ed L Hirschfeld amp SGelman) pp 119ndash148 New York Cambridge UniversityPress

Luo Y amp Baillargeon R 2002 Does a self-moving box possessan lsquointernal forcersquo and lsquomindrsquo In 13th Biennial Int Conf onInfant Stud April 2002 Toronto Ontario Canada

Mandler J M 1992 How to build a baby II Conceptualprimitives Psychol Rev 99 587ndash604

Meltzoff A N 1988 Infant imitation after a 1-week delaylong-term memory for novel acts and multiple stimuli DevlPsychol 24 470ndash476

Meltzoff A N 1995 Understanding the intentions of othersre-enactment of intended acts by 18-month-old childrenDevl Psychol 31 838ndash850

Morton J amp Johnson M H 1991 CONSPEC and CON-LERN a two-process theory of infant face recognition Psy-chol Rev 98 164ndash181

Moses L J Baldwin D A Rosicky J G amp Tidball G 2001Evidence for referential understanding in the emotionsdomain at twelve and eighteen months Child Dev 72718ndash735

Movellan J R amp Watson J S 2002 The development of gazefollowing as a Bayesian systems identification problemUCSD Machine Perception Laboratory Technical Reports200201

Muir D amp Lee K 2002 Invisible gaze following perspectivetaking at 12 months of age In 13th Biennial Int Conf onInfant Stud April 2002 Toronto Ontario Canada

Okamoto S Tomonaga M Ishii K Kawai N TanakaM amp Matsuzawa T 2002 An infant chimpanzee (Pantroglodytes) follows human gaze Anim Cogn 5 107ndash114

Onishi K H 2001 Ten-month-old infants understand that agoal object can be obtained in multiple ways In The SocRes Child Dev Minneapolis April 2001

Phillips A T Wellman H M amp Spelke E S 2002 Infantsrsquoability to connect gaze and emotional expression to inten-tional action Cognition 85 53ndash78

Phillips W Baron-Cohen S amp Rutter M 1992 The role ofeye contact in goal detection evidence from normal infantsand children with autism or mental handicap Dev Psycho-pathol 4 375ndash383

Povinelli D J Bierschwale D T amp Cech C G 1999 Com-prehension of seeing as a referential act in young childrenbut not juvenile chimpanzees Br J Devl Psychol 17 37ndash60

Premack D 1990 The infantrsquos theory of self-propelled objectsCognition 36 1ndash16

Scholl B J amp Tremoulet P D 2000 Perceptual causality andanimacy Trends Cogn Sci 4 299ndash309

Sodian B amp Thormer C 2000 Understanding the epistemicaspects of seeing Precursors in infancy In 12th Biennial IntConf on Infant Stud July 2000 Brighton UK

Sorce J Emde R N Campos J J amp Klinnert M 1985Maternal emotional signaling its effect on the visual cliffbehavior in 1-year-olds Devl Psychol 21 195ndash200

Sperber D amp Wilson D 1986 Relevance communication andcognition Oxford Blackwell

Sperber D amp Wilson D 2002 Pragmatics modularity andmond-reading Mind Lang 17 3ndash23

Tomasello M 1999 The cultural origins of human cognitionCambridge MA Harvard University Press

Tomasello M amp Barton M E 1994 Learning words in non-ostensive contexts Devl Psychol 30 639ndash650


Tomasello M Hare B amp Fogleman T 2001 The ontogenyof gaze following in chimpanzees Pan troglodytes and rhesusmacaques Macaca mulatta Anim Behav 61 335ndash343

Uller C amp Nichols S 2000 Goal attribution in chimpanzeesCognition 76 B27ndashB34

Wagner L amp Carey S 2002 Coming to conclusions 12-month-old infants form expectations about probable endingsof motion events In 13th Biennial Int Conf on Infant StudApril 2002 Toronto Ontario Canada


Watson J S 1972 Smiling cooing and lsquothe gamersquo MerrilPalmer Q 18 323ndash339

Woodward A L 1998 Infants selectively encode the goalobject of an actorrsquos reach Cognition 69 1ndash34

Woodward A L amp Guajardo J J 2002 Infantsrsquo understand-ing of the point gesture as an object-directed action CognDev 17 1061ndash1084

Woodward A L amp Sommerville J A 2000 Twelve-month-oldinfants interpret action in context Psychol Sci 11 73ndash77


interpretation systems operate independently in earlyinfancy The independence of these cognitive mechanismsis supported by various facts they rely on different typesof computations they are triggered by different stimulusconditions and they can be dissociated in animal behav-iours and developmental disorders Crucially there is noevidence for transfer between these action interpretationsystems (for example attribution of goals on the basis ofreferential actions) in infants at or below 1 year of ageI suggest that the combining of these action interpretationsystems into a higher-order mentalistic representation ofactions takes place during the second year of life

2 TELEOLOGICAL UNDERSTANDING OF ACTIONS

Understanding goals requires connecting actions not totheir antecedents but to their consequents Some studiessuggest that infants as young as six months of age aresensitive to the end state of observed events Woodward(1998) for example repeatedly presented infants with anaction in which a hand reached towards and grasped oneof two toys When the infants had been habituated to thisevent she swapped the toys and the hand either graspedthe same toy in the new location or the other toy in thesame location Looking times for these two events weremarkedly different infants looked longer at the hand grasp-ing the lsquonewrsquo toy at the lsquooldrsquo location as if they expectedthe hand to reach towards the same toy This result showsthat infants are more likely to associate a grasping handwith the grasped object than with its location which is use-ful for interpreting the grasping action as directed to thegoal of acquiring a specific object Nothing in this experi-ment however demonstrates that infants do not simplydiscriminate between end states It does not show that theyrelate actions to end states in an lsquoin-order-torsquo clause thehand performed the reaching and grasping sequence inorder to acquire a certain object

An action is goal directed if it is performed not for itselfbut to achieve an end in other words if it is an instrumen-tal action Whether an action is instrumental in relationto an end state or not can be intuitively tested by consider-ing if we would expect it to be performed when it is notneeded for goal achievement Pulling out a cork from awine bottle for example can be an instrumental action toaccess the wine but only when the bottle is not emptyEven when the bottle is not empty we would not expectto see this action if all the glasses were filled with wineAs this example suggests interpreting something as aninstrumental action that has been performed to achieveand gets its meaning from a particular goal state requiresconsideration of many thingsmdashnot just the action and itsend state but also the environment in which it occurs Wehave demonstrated that nine-month-old infants arealready capable of doing this trick and depending on theenvironment interpret observed actions as goal directed

These studies (Gergely et al 1995 Csibra et al 1999)presented infants with computer animations Since theseminal work by Heider amp Simmel (1944) several experi-ments provided evidence that high-level sophisticatedsocial interpretation of actions does not require featuralidentification of agents (for a review see Scholl amp Tremoulet2000) Triangles and circles moving on a two-dimensionalsurface are readily interpreted as if they were people


engaged in various kinds of social interactions One expla-nation of this phenomenon is that during development themotion patterns of human behaviour are abstracted awayfrom their usual context and from human features andwhen these patterns are recognized in artificially createdanimations we tend to project further human attributesto those abstract figures An alternative explanation whichI defend here suggests that certain motion patterns thatallow interpretation of the behaviour of the abstract figuresas goal-directed actions will always attract such aninterpretation and this tendency does not depend onextensive experience with human behaviour This expla-nation predicts that young infants will as readily attributegoals to animated figures as they do to human agents

Our experiments repeatedly presented infants with asimple animation (figure 1) in which a ball approachedand contacted another ball by jumping over an obstacleIn this event the jumping action can be interpreted as aninstrumental action because it is necessary to achieve theend state (the spatial position beside the other ball)Whether or not the infants arrived at the same interpret-ation was tested by showing them two modified versionsof the event and measuring their looking time In the testevents we removed the obstacle which changed therelevant aspects of the environment in a way that madethe jumping action unnecessary for goal achievement Oneof these events (old action) displayed the same jumpingapproach which however was no longer necessary andwas therefore inefficient while the other event (new action)displayed a straight-line approach to the same positionwhich was an efficient action in the new situation Ifinfants interpreted the original event as a goal-directedaction they should find the new instrumental action morecompatible with their interpretation and should respondwith longer looking at the old inefficient action

This is exactly what we found in 9- and 12-month-oldinfants (see figure 1) Further control studies were per-formed to verify whether these looking time differences wereindeed attributable to the interpretation of the habituationevent The control studies involved the same test eventsbut they were preceded by a habituation event which dif-fered only slightly from the original habituation event butdid not show a proper instrumental action In this eventthe lsquoobstaclersquo was positioned not in between the two ballsbut behind the moving ball (see figure 2) In this environ-ment the jumping action cannot be considered to be aninstrumental action and the infants were not expected toattribute the end position as the goal of the moving ballrsquosaction Their looking times confirmed this prediction

How can we explain infantsrsquo early emerging ability toattribute goals to animated shapes We have proposed(Gergely amp Csibra 1997 Csibra et al 2003) that watchingthese animations on a computer screen infants adopt ateleological stance The teleological stance is akin to theintentional stance of Dennett (1987) in that it representsan interpretational strategy that seeks to construe an eventin terms of goals (see Keil (1995) for a different versionof the teleological stance) It is however different fromthe intentional stance in that it does not attribute mentalstates to the agents (I will return to the relation betweenthese two stances later) Note that the teleological stance(and the intentional stance) is not an explicit inferentialsystem but a bias a tendency to construe events in accord


habituation90

80

70

60

50

30

20

10


p lt 005

p lt 001


seco

nds


habituation90

80

70

40

30

20

10



seco

nds







observed action

behaviour action


(efficiency)

physicalcontext


interpreted action










habituation90

80

70

60

50

30

20

10


p lt 005


seco

nds


habituation90

80

70

60

50

30

20

10


p lt 005


seco

nds







habituation90

80

70

60

50

30

20

10


p lt 001


seco

nds







constraints

action

goal


beliefs

intentions

desires


















about





constraints goal

action


directionality

relevance agent

action

referent

reactivity)








(a)

(c)

(b)

100

mea

n lo

okin

g ti

me

(s)

mea

n nu

mbe

r of

ori

enta

tion

s

50

looking time

orientations


20

10

















REFERENCES



































































habituation90

80

70

60

50

30

20

10


p lt 005

p lt 001


seco

nds


habituation90

80

70

40

30

20

10



seco

nds







observed action

behaviour action


(efficiency)

physicalcontext


interpreted action










habituation90

80

70

60

50

30

20

10


p lt 005


seco

nds


habituation90

80

70

60

50

30

20

10


p lt 005


seco

nds







habituation90

80

70

60

50

30

20

10


p lt 001


seco

nds







constraints

action

goal


beliefs

intentions

desires


















about





constraints goal

action


directionality

relevance agent

action

referent

reactivity)








(a)

(c)

(b)

100

mea

n lo

okin

g ti

me

(s)

mea

n nu

mbe

r of

ori

enta

tion

s

50

looking time

orientations


20

10

















REFERENCES



































































observed action

behaviour action


(efficiency)

physicalcontext


interpreted action










habituation90

80

70

60

50

30

20

10


p lt 005


seco

nds


habituation90

80

70

60

50

30

20

10


p lt 005


seco

nds







habituation90

80

70

60

50

30

20

10


p lt 001


seco

nds







constraints

action

goal


beliefs

intentions

desires


















about





constraints goal

action


directionality

relevance agent

action

referent

reactivity)








(a)

(c)

(b)

100

mea

n lo

okin

g ti

me

(s)

mea

n nu

mbe

r of

ori

enta

tion

s

50

looking time

orientations


20

10

















REFERENCES



































































habituation90

80

70

60

50

30

20

10


p lt 005


seco

nds


habituation90

80

70

60

50

30

20

10


p lt 005


seco

nds







habituation90

80

70

60

50

30

20

10


p lt 001


seco

nds







constraints

action

goal


beliefs

intentions

desires


















about





constraints goal

action


directionality

relevance agent

action

referent

reactivity)








(a)

(c)

(b)

100

mea

n lo

okin

g ti

me

(s)

mea

n nu

mbe

r of

ori

enta

tion

s

50

looking time

orientations


20

10

















REFERENCES



































































habituation90

80

70

60

50

30

20

10


p lt 001


seco

nds







constraints

action

goal


beliefs

intentions

desires


















about





constraints goal

action


directionality

relevance agent

action

referent

reactivity)








(a)

(c)

(b)

100

mea

n lo

okin

g ti

me

(s)

mea

n nu

mbe

r of

ori

enta

tion

s

50

looking time

orientations


20

10

















REFERENCES

















































































about





constraints goal

action


directionality

relevance agent

action

referent

reactivity)








(a)

(c)

(b)

100

mea

n lo

okin

g ti

me

(s)

mea

n nu

mbe

r of

ori

enta

tion

s

50

looking time

orientations


20

10

















REFERENCES



































































(a)

(c)

(b)

100

mea

n lo

okin

g ti

me

(s)

mea

n nu

mbe

r of

ori

enta

tion

s

50

looking time

orientations


20

10

















REFERENCES











































































REFERENCES


































































teleological and referential understanding of action in ...cynthiab/readings/csibra2003.pdfpublished...

Documents