THE PERCEPTION OF ACTION-AND-REACTION SEQUENCES IN INFANTS

Transcription

1 THE PERCEPTION OF ACTION-AND-REACTION SEQUENCES IN INFANTS Anne Schlottmann 1, Luca Surian 2 & Elizabeth Ray 1 June 16, Department of Psychology, University College London Gower Street, London, WC1E 6BT, UK 2 Dipartimento di Psicologia, Universita di Trieste via Sant'Anastasio 12, Trieste, Italy Acknowledgement: This work was supported by ESRC project grants R to AS and R to AS and ER. Thanks to Sarah Hesketh for testing many of the babies in this study.

2 Action and reaction 2 ABSTRACT We studied whether 8- to 10-months-olds are sensitive to causation-at-a-distance in a schematic motion event described by older observers as goal-directed action and reaction. In Experiment 1, infants (N= 132) saw a computer-animated red square moving towards a blue square, either rigidly or in a non-rigid fashion described by older observers as animal-like. Infants saw either a reaction or delay event. In the reaction event blue moved away while red was still approaching, whereas in the delay event blue only started after red stopped; red and blue did not come into contact in either event. Infants were habituated to the reaction or delay event, each involving either rigid or non-rigid motion. On test, all four groups saw the habituation event in reverse. Reversal involved the same change in spatio-temporal parameters for both groups, but the reversed reaction additionally involved a change in causal roles. Infants in the reaction group dishabituated more than infants in the pause group, but motion rigidity had little effect. Experiment 2 showed that infants discriminate the rigid from non-rigid motions infants and that they also dishabituate to reversal of a reaction involving a different, less animate, non-rigid motion. In Experiment 3 infants discriminated reaction from launch events, as used in previous studies on the perception of contact causality. In Experiment 4, finally, infants dishabituated to reversal of a reaction without self-initiated motion of the two shapes. Overall, our results indicate that infant sensitivity to causation-at-a-distance depends only on the event configuration, but not on motion onset or pattern typical of animal motion. Distinct percepts for contact and noncontact causality could help infants carve up the world into what eventually becomes the physical versus the social/psychological domain.

3 Action and reaction 3 EIGHT- AND 10-MONTHS OLD INFANTS PERCEIVE ACTION AND REACTION People see other people as psychological agents: They act to reach their goals, influenced by intentions, beliefs and other mental states. But people are also material bodies acting because physical forces act on them. Physics and psychology provide two different systems for understanding behaviour. Infants in their first year employ basic principles of contact mechanics (Spelke et al., 1992; Baillargeon et al., 1995), but less is known on their grasp of psychology. This study considers whether 8 and 10 months-olds are sensitive to causation-at-a-distance in a schematic event with minimally contingent motions that adults described as goal-directed or intentional action and reaction (Kanizsa & Vicario, 1968). Our study thus relates to the issue of how infants recognise psychological agents. Much recent research shows that 3- to 4-year-olds have a well-developed theory of mind (e.g., Surian & Leslie, 1999), with interest now moving towards the infant precursors of this naïve psychology (e.g., Surian, 1997). One precursor to realising that human action depends on internal, mental states may be an ability to identify agents and their external goals, as this can largely be based on perceptible features of the environment. Traditionally, infants understanding of others as persons is seen to arise out of their social interactions, leading to first hand knowledge later projected on the external world. By the end of the first year infants engage in intentional action and triadic interaction. Some believe this signals initial appreciation of others as intentional beings (e.g., Frye, 1991; Carpenter et al., 1998); others argue for intersubjectivity from birth that lacks widespread behavioural expression until this time (Trevarthen, 1979); yet others hold that 1-year-olds actions can still be explained behaviourally (Moore & Corkum, 1994). In the end, one unfortunately cannot tell from infants social actions alone whether they only react to others behaviour or also to their partners goals, let alone whether they infer that their partners have intentions, i.e., mental states directed towards these goals.

4 Action and reaction 4 RECOGNISING GOAL-DIRECTED ACTION To overcome these ambiguities, in some recent work (including the present study), infants merely observe actions rather than actively participate in them. Woodward (1998), for instance, habituated 5- and 6-months-olds to a hand reaching for a toy, with dishabituation when the hand reached for a new toy, but not when it reached for the old toy at a new location. Thus, infants may already organise human action in terms of its goal structure. But infants are highly familiar with hands and may have dedicated neural mechanisms for the spatial elements of grasping (Gallese et al.,1996), so it cannot be ruled out that this result is specific to reaching/grasping. To escape such familiarity confounds, animated cartoon agents may be used. Gergely et al. (1995; Csibra et al., 1999), for instance, habituated infants to one circle jumping a barrier to reach another. Nine- but not 6-months-olds dishabituated when upon removal of the barrier the circle took the familiar curved path, but not when it took a novel, more direct, straight path. Thus 9-months-olds appeared to interpret the action as goal-directed, appreciating that the environment constrains such actions. Of course, this does not yet imply that infants attribute mental states to the circles. The logic of both approaches rests on the equifinality of goal-oriented action (Heider, 1958): we group actions if they satisfy the same goal even if they vary physically. Evidence for goal understanding lies in a tendency to treat as familiar a novel trajectory that reaches an old goal, while treating as novel the trajectory of the habituation event if the goal has changed. Equifinality does not appear to the same extent in repeated mechanical events. Moreover, variations in these are generally random; inert bodies are not capable of intelligent adaptation to changing circumstances (with exceptions, e.g., smart missiles). An object may approach a magnet on variable paths, but with a barrier halfway, we expect it to get stuck, not to jump it. According to Gergely et al. (1995), the distinctive feature of goal-

5 Action and reaction 5 directed action is its rationality : Equifinality is suggestive, but the true mark of goaldirectedness is the sensible fit of the action to its environment. RECOGNISING AGENTS In considering how infants single out the agents or animates 1 capable of such goaldirected action, theorists such as Premack (1990, 1995), Mandler (1992, 1998), Leslie (1994, 1995) and Baron-Cohen (1994) argue that at the most basic level infants distinguish objects passively set in motion from those that start on their own. For Premack (1990) and Baron-Cohen (1994), infants are hardwired to endow self-movers with intention; other event features then allow more detailed interpretation of specific goals (see Premack & Premack, 1997). For Mandler (1992, 1998), in contrast, a first notion of animate can be abstracted from real-world motion patterns and does not entail intentional understanding. Leslie (1994) argues that agents are understood progressively: 4-months-olds understand some mechanical properties, specifically that they have internal energy sources and move without external cause. By 6 to 8 months they appreciate actional properties, i.e., that agents pursue goals, but only by 18 to 24 months do they grasp cognitive properties, that agents have mental states. Despite these differences, however, all theorists agree that self-motion is fundamental. Other cues also help infants identify agents (Mandler, 1992). First, there is the manner of animate motion, such as irregular trajectory (Tremoulet & Feldman, 2000) and non-rigid, bio-mechanical deformation (Berthental, 1993). Then there are morphological cues, such as faceness and body orientation (Johnson, 2000). Finally, there are cues from the object s interactions with the environment; these interactions are typically contingent-at-a-distance upon some other feature. By this, Mandler (1992) refers both to infants experience that other objects react systematically to them (learned through repetition and reinforcement, as 1 Some consider the important distinction to be between agents and inert objects, others between animates and inanimates. It is doubtful that the infant acquires a biological category of animates, but nevertheless the typical agent the infant encounters is animate. We use the terms interchangeably.

6 Action and reaction 6 in conditioning) as well as perceptual contingency that involves such factors as one animate following another,, avoiding barriers and making sudden shifts in acceleration (p. 595). Adults tend to see such motions as intentional or goal-directed. Clear evidence on infants use of these cues is scarce: 3-months-olds discriminate biomechanical motion from scrambled versions (Bertenthal, 1993), but it is unclear whether they see this motion as animate. 5-months-olds distinguish self- from externally caused motion of a box, dishabituating to reversal of path in the latter, not the former (Kaufman, 1999). Thus they know that self-movers spontaneously change direction or lack specific expectations. To 12-months-olds, a self-propelled robot is more disturbing than a selfpropelled stranger (Poulin-Dubois et al., 1996). Thus they link self-motion to familiar animates, but this could merely be associative knowledge. 12-months-olds also followed the gaze of a novel furry toy with body and head if it reacted contingently to them or had facial features or both, but not if it acted non-contingently and lacked a face (Johnson et al., 1998). This result is more convincing in that it demonstrates the interplay of various cues. More data exist on contingencies between infant action and other stimuli. From at least 2 months, infants treat contingently reacting stimuli as social in the sense of smiling and cooing at them even if they are inanimate objects (Watson, 1972). Infants also associate familiar animates with such contingencies: In the still face paradigm, 2-months-olds are distressed when communicative contingencies are halted or the mother responds noncontingently (Muir & Nadel, 1998). From 3-months, infants prefer high, but imperfect contingencies to perfect ones (Bahrick & Watson, 1985), as provided in social interaction, for instance, by mothers mirroring their babies affective expressions. Gergely & Watson (1999) argue that social contingency detection contributes to early self-other differentiation and emotional self-awareness. But while it is clear that infants expect contingent reactions from others, such data still cannot show that infants use contingency to identify agents.

7 Action and reaction 7 CONTINGENCY AND CAUSATION-AT-A-DISTANCE Our study concerns contingency in Mandler s (1992) second sense of a perceived dependency between two actions external to the self. Such contingencies between spatially separated objects often involve animate agents rather than inert bodies. Animate agents can perceive one another and thus can interact from afar. We recognise such remote interaction if the pattern of one action resembles or in an obvious way depends on the pattern of the other action. A contingency may become apparent over time, e.g., one person talks, then the second, etc., or through the spatio-temporal configuration of the event. Thus, when a cat chases a mouse, the second trajectory follows the first. The best-known demonstration of this may be Heider and Simmel s (1944) finding that adults (and preschoolers, Berry & Springer, 1993) interpret triangles and squares moving about in complex ways as actions and reactions of agents with motivations and emotions. A PET study also showed specific brain activation for such animations (Castelli et al., 2000). Rochat et al. (1997) found that 5-to 6-mths olds distinguish correlated movements of two dots (a computer-generated cat and mouse sequence) from uncorrelated motion patterns. This goes beyond the conditioning literature, showing early sensitivity to contingencies between external objects, but not that infants see these objects as agents. When adults interpret such stimuli as goal-directed interaction they take the observed contingency to indicate underlying (mental) causation. Mere contingency can appear between inert objects (e.g., lights blinking in sequence), but with animate agents we often understand the contingency as an action causing a reaction. It might be better to speak of causal contingency or contingent reactivity than contingency per se as a cue to agency. But do we take the contingency as causal only because it involves animate agents? Or do some contingencies appear causal even without other agency cues? This is a central issue addressed in the present study.

8 Action and reaction 8 To use complex pursuit movements for this, as in Rochat et al. s (1997) study, would pose a problem of interpretation, however: It would be unclear whether the operative cue is the contingent motion itself, or the fact that the contingency takes a sensible form. The cat does not follow all twists and turns of the mouse s trajectory, but moves on the shortest path to its prey s location. Thus in complex animations there may also be rational variation of the motion in Gergely et al. s (1995) sense and this might lead to an interpretation of goaldirectedness rather than the contingent motion per se. THE REACTION EVENT Here we investigate a cat and mouse scenario without rational adjustments of the trajectory, to study whether infants perceive causation in minimally contingent movements at a distance. In the reaction event (Figure 1), introduced by Kanizsa and Vicario (1968), an animated shape moves in a straight line up to a second which moves in turn before the other can reach it. For a fraction of a second both move simultaneously before the first stops and the second continues. Adults described this as A chasing B and B running away from A. Kanizsa and Vicario argued that observers see an intentional reaction at a distance. However, the phrase may be misleading -- intentional usually refers to an understanding in t 0 A B t 1 t 2 t 3 t 4 Figure 1. Schematic reaction event. (Shape A moves towards B which is stationary. B begins to move before contact, slightly before A has reached its final position.)

9 Action and reaction 9 terms of mental states, but it is not clear that observers make such attributions for reaction events. Accordingly, we will merely refer to perceiving goal-directedness. This is a different aspect of goal-directedness, however, than embodied in Gergely et al. s (1995) barrier and Woodward s (1998) reach event. Both events involve action towards a goal, i.e., approach, whereas B s goal in the reaction event is to avoid contact. Kanizsa and Vicario s studies complement Michotte s (1963) work on launching, for which adults report physical not psychological causality. Reaction and launch events differ only in whether B moves before A reaches it or upon contact. This small difference signals that the events belong to different ontological domains. Studies of launch and reaction events thus offer the opportunity to study domain-specific perception of causality in events of comparable minimal complexity. Perception of contact causality is well established in adults (e.g., Schlottmann & Shanks, 1992; Schlottmann & Anderson, 1993; Schlottmann, 2001) and infants (e.g., Leslie & Keeble, 1987; Oakes, 1994). Causation at-a-distance has received less attention, but recent work confirmed that adults report psychological causality for reaction events (Schlottmann & Ray, 2003), and that children as young as 3 years link reaction events to psychological, but launch events to physical causality (Schlottmann et al., 2002). The present studies follow up on our initial work with infants (Schlottmann & Surian, 1999). Our studies employed the reversal paradigm used by Leslie & Keeble (1987) to establish that 6-months-olds are sensitive to contact causality in launch events: Babies were habituated to launch events, with movement, say, from left to right, or to a noncausal event with a delay at contact. Upon test, infants saw the habituation event played in reverse, i.e., with movement from right to left. Thus spatio-temporal direction reversed for both groups. If infants perceive only this, both groups should dishabituate equally. However, cause and effect reversed as well in the launch, but not the delay event. Indeed, infants in the causal

10 Action and reaction 10 group dishabituated more. This suggests that infants are sensitive to the causal structure of the event, not just its spatiotemporal structure, an interpretation later confirmed with a different habituation technique (Oakes,1994). We used the reversal paradigm to study whether infants were sensitive to reaction-ata-distance, as linked to psychological causality by older observers (Schlottmann & Surian, 1999). Nine-months-olds were habituated to causal reaction events or noncausal, delayed control events. Infants dishabituated to reversal of the causal, but not the noncausal sequence. Thus infants also appear sensitive to the causal structure of the reaction event. THE PRESENT STUDIES: THE ROLE OF NON-RIGID AND SELF-INITIATED MOTION The reaction event in our prior study contained other cues to agency besides contingent motion. First, the shapes engaged in self-initiated motion. Second, they did not move rigidly as in Kanizsa & Vicario s (1968) standard sequence, but non-rigidly, by rhythmic expansion and contraction (Figure 2). Michotte (1963) reported that adults describe this caterpillar motion as self-generated, animate motion; we found the same for adults (Schlottmann & Ray, in prep.) and children as young as 3 (Schlottmann et al., 2002). Michotte also argued that adults tendency to describe his reaction events as a form of t 0 t 1 t 2 t 3 t 4 Figure 2: Schematic caterpillar motion (A square expands from the right edge, then contracts from the left edge. The resulting translation appears self-generated and animal-like.)

11 Action and reaction 11 psychological causality depended heavily on hints of either a verbal or perceptual nature to take the shapes as animate; our own studies with adults confirmed this (Schlottmann & Ray, 2003). Accordingly, in our previous study we presented reaction events with non-rigid animal motion to infants. But this means we cannot tell whether infants appreciation of causation-at-a-distance depends on the contingency, or on the conjunction of contingency with non-rigid and self-initiated motion. In other words, do we need to identify the agents to interpret their interaction, or does the configuration of the interaction itself prompt the interpretation, without requiring other agent cues? Most theorists implicitly assume the former, but the empirical literature is not conclusive. In Woodward s (1998) study, infants did not recognise the action as goaldirected if a mechanical claw reached for the toy. Similarly, Meltzoff (1995) found that toddlers would imitate the intended goal of a failed action (trying to pull a dumbbell apart, but a hand slipped off) when a person, but not a machine, engaged in the act. These studies suggest that perception of goal-directedness can depend on the presence of animate agents. However, observers also report goal-directed action when looking at animated geometric shapes. Such animations usually contain some agency cues, e.g., self-motion typically remains. In Gergely et al. s (1995) work, the circle jumping the barrier self-initiated motion and prior to jumping repeatedly expanded/contracted, as if in greeting, in response to a larger circle (the goal on the far side) doing the same. Csibra et al. (1999) found that infants goal-directed interpretation remained even without self-start and communicative contingency: Infants dishabituated to the familiar but now irrational path with the small circle merely flying in from the side and with a static large circle (Exp 3). However, instead of the circle jumping identically throughout, barrier size varied and the circle adapted, always jumping efficiently with minimal clearance. Thus more evidence was provided of its rationality, which may have made up for the loss of other agency cues. Infants may need

12 Action and reaction 12 more than one bit of information that an object is an agent, but whether this is cumulative or comprises different types of information may matter less. The present studies address which agency cues contingency-at-a-distance, nonrigid motion and self-initiated motion -- contribute to perception of causation-at-a-distance. In Study 1 infants saw either reaction events with rigid or non-rigid caterpillar motion. Study 2 involved a number of control conditions, including a non-canonical reaction event in which the onset of A and B s motion was hidden. Our previous work (Schlottmann & Surian, 1999) suggested that non-rigid and self-initiated motion are not sufficient for an attribution of causation at a distance (since infants in the control group did not dishabituate); the present studies investigate whether these agency cues are necessary. EXPERIMENT 1 Method Infant Participants: The final sample included 132 babies with no known health problems, recruited by advertisement. There were and months-olds, with younger babies ranging from 8 months 1 day to 9 months 1 day, mean age 8 months 17 days (31 girls and 36 boys), and older babies ranging from 10 months 1 day to 11 months 6 days, mean age 10 months 13 days (29 girls and 36 boys). An additional 77 infants were excluded, 43 due to fussiness, parental interference or technical problems, 33 because they failed to habituate in 12 trials, and 1 for ceiling level looking times on both test trials (more than 4 standard deviations away from the group means). Stimuli: Each movie involved a red and a blue square, about 22 x 22 mm in size, initially stationary, Red on the left, Blue in the middle (left-to-right version). Red moved towards Blue, stopping about 9 mm to its left. Blue began to move right either before Red stopped,

13 Action and reaction 13 with about 420 msec overlapping movement at a distance of 7.1 cm between centres in the 'Reaction' movie, or after about 1220 msec delay in the 'pause' event. To equate sequences with and without pause the stationary periods at beginning and end were adjusted, with each cycle symmetrical about the midpoint of the sequence. Movies were made with the MacroMedia Director animation software. One cycle took about 4.8 seconds and repeated up to 10 times, with a 750msec interval between cycles during which the screen went grey. Right-to-left movies were mirror-reversed: Blue moved first, Red second, from right to left; this was the only spatiotemporal difference between the two versions. Half of the infants in each group saw the left-to-right version on habituation, the other half the right-to-left version. On reversal, all infants saw the alternate movie. In the rigid motion condition, each square moved at a constant rate of about 9.4 cm/sec to cover 113 mm distance in 1.2 seconds. In the nonrigid motion condition, corresponding to Michotte s caterpillar stimulus, the squares moved by expansion-contraction. Each expanded for 200 msec at about 18.8 cm/sec to a rectangle of about 4.1 cm length, with the right edge stationary, then contracted in the same way, with the left edge stationary until the original shape was recovered. These steps repeated twice, separated by 40 msec delay. Thus, average translation speed for rigid and nonrigid motion was the same. The shapes also had the same start and end position for rigid and nonrigid motion, and in both reaction and pause events. Design and Procedure: The overall design was a 2 Age (8 or 10 months) x 2 Event (Reaction or Pause) x 2 Shape (Rigid or Nonrigid) between subjects factorial design. Infants of each age were randomly assigned to the four groups. Infants were tested in a semi-dark experimental room, sitting on their caretaker's lap approximately 90 cm away from the monitor (viewable area 19 inches diagonally). All other equipment was hidden. Caretakers had no knowledge of purpose or design of the study.

14 Action and reaction 14 They were told not to interfere with the infant, and on the reversal test they were specifically instructed to close their eyes. A camera above the monitor was centered on the infant's face; the experimenter observed the infant on video. A MacIntosh G3 PowerPC was used to control stimulus display and record looking times. Trials began with sounds and a flashing screen to attract the baby's attention. When the baby looked, the experimenter hit a key to start the movie and record onset of a look. When the baby looked away, the experimenter hit another key. If the baby looked away prior to half-point of the first cycle the trial was abandoned, otherwise it ended if the baby looked away for 2 seconds consecutively, or after 10 complete cycles. Habituation continued until average looking time on three consecutive trials fell below half of the average on the first three trials; the minimum number of trials was 6, the maximum 12. After habituation all infants saw a familiar test (the habituation movie) followed by a 30 second break to maximise attention at this crucial point, followed by a novel test (the reversal movie). A second observer without knowledge of purpose or design of the study checked videos for a randomly selected third of the babies. Interobserver reliability was high: The correlation of looking times measured on- and offline was r =.87 and.98, for the familiar and reversal test, respectively. When all trials, habituation and test, were considered, the mean correlation across the 44 babies was r =.98, range.78 to Results and Discussion Habituation Table 1 shows the data on the first 3 and last 3 habituation trials for babies seeing the reaction or pause events (bottom rows); the table also gives the breakdown for the 8 subgroups of the design. The overall decrease in looking from first to last trial was confirmed by a Trial main effect in a 2 Trial x 2 Age x 2 Event x 2 Agent mixed model ANOVA, F (1,124) = , MSE =.001. There were no other effects in this analysis. To

15 Action and reaction 15 normalize the variances, the ANOVAS throughout were computed on the reciprocally transformed data. This reduces the influence of extremely large looking times (Howell, 1992). For readability, tables and figures in the text show the raw data. Table 1: Mean looking times in seconds (and standard errors) to reaction and pause events and in 8 subgroups during habituation. n hab 8 months reaction, nonrigid (4.03) (4.22) (4.64) (1.96) (1.01) (1.38) pause, nonrigid (3.17) (2.80) (2.76) (1.37) (1.06) (0.74) reaction, rigid (3.56) (3.10) (3.25) (1.48) (1.18) (1.09) pause, rigid (3.21) (3.67) (3.08) (2.34) (1.14) (1.12) 10 months reaction, nonrigid (3.06) (2.71) (2.66) (0.82) (0.95) (0.87) pause, nonrigid (3.02) (2.96) (2.77) (1.52) (1.16) (0.98) reaction, rigid (3.56) (3.14) (2.25) (1.15) (1.00) (0.82) pause, rigid (3.47) (3.89) (3.19) (1.26) (1.60) (0.86) overall reaction (1.75) (1.64) (1.76) (0.72) (0.52) (0.53) pause (1.60) (1.65) (1.45) (0.83) (0.61) (0.47) The reaction and pause groups were also similar in the mean looking time per habituation trial, and s, respectively, and the number of habituation trials, 7.52 and There were neither event, nor age or agent differences in ANOVAs on these, or on looking times on any of the first 3 or last 3 habituation trials. The exceptions were an Age x Agent interaction on the third habituation trial, F(1,124) = 5.40, MSE =.003, p =.022, and an Agent main effect, F(1, 124) = 4.00, MSE =.003, p =.048, on the third-to-last trial. Other than that, the groups were statistically equivalent during habituation.

16 Action and reaction 16 Since infants were run to a criterion, spontaneous recovery was expected between the last habituation and subsequent familiar test trial (showing the habituation movie), due to regression towards the mean. This recovery was indeed significant, F(1,124) = 9.85, MSE =.003, p =.002. The analysis also indicated more spontaneous recovery in the pause than in the reaction group, F(1,124) = 6.07, MSE =.003, p =.015. In the pause group, looking time increased from 8.38 to seconds, but in the reaction group, it increased only from 9.16 to seconds. However, the difference in looking time to reaction and pause event on the familiar test trial was not significant, F(1,124) = 2.57, MSE =.003, p =.112, and the decrease in looking time from first habituation to familiar test trial remained significant, F(1,124) = , MSE =.001, p <.001, with no group differences. Finally, the comparison of last habituation and familiar test trial also found an Age x Event x Agent interaction, F(1,124) = 4.45, MSE =.003, p =.037. This effect reflects that 8-months-olds looked longer at reaction events with nonrigid than rigid agents, but they looked longer at pause events with rigid than nonrigid agents. The 10-months-olds, in contrast, showed exactly the opposite pattern, looking longer at reaction events with rigid agents, but at pause events with nonrigid agents. However, this 3-way interaction did not reach significance when the familiar test and last habituation trial were analysed separately. Reversal Test Figure 3 presents our central result, showing looking times on the familiar and reversal test for infants watching reaction and pause events. Both groups reacted to the change in spatiotemporal direction with renewed interest. Importantly and as predicted, looking time increased more in the reaction (left bars) than the pause group (right bars), in line with the argument that infants in the reaction group additionally perceived a change in causal roles upon reversal. This consolidates our earlier finding (Schlottmann & Surian, 1999).

17 Action and reaction 17 Figure 3: Mean looking times and standard errors to the familiar and reversal test in babies seeing the reaction or pause event. (Infants in the reaction group show more recovery of looking than infants in the pause group.) familiar reversal familiar reversal reaction event pause event The statistical analysis confirmed the visual impression with an Event x Test interaction, F (1, 124) = 6.87, MSE =.002, p =.01, in a 2 Test x 2 Age x 2 Event x 2 Agent mixed model ANOVA. This interaction did not differ by Age or Agent, both F < 1. More recovery to reversal of reaction than pause events appeared regardless of age, and for both nonrigid and rigid agent motion. It appears that infant perception of causation-at-adistance does not depend on the nonrigid motion of potentially animate agents. 2 The recovery upon reversal was significant in all four reaction groups, F(1,15) = 12.39, 7.88, and 12.39, MSE =.002,.001,.001 and.002, p =.003,.013,.000 and.003, for 8-months-olds watching nonrigid and rigid agents, and 10-months-olds watching nonrigid and rigid agents, respectively. In contrast, recovery was not significant in three of four pause groups except for 8-months-olds watching pause events with nonrigid agents, F(1,17) = 2 The same pattern of results and significances is found if the nonhabituators are included in the analysis.

18 Action and reaction , MSE =.001, p =.002. However, from Table 2 it is clear that infants in the corresponding reaction group showed substantially more recovery of looking. Table 2: Mean looking times in seconds during the familiar and reversal test (standard errors in brackets), number of babies, percentage of babies looking longer at reversal, and recovery score (the difference between the mean reciprocal looking times on familiar and reversal trials, as used in the ANOVA); * indicates significant recovery, as listed in the text. familiar reversal n % showing recovery recovery score 8-months-olds reaction, nonrigid %.0576 * (2.41) (3.64) pause, nonrigid %.0380 * (1.20) (1.47) reaction, rigid %.0352 * (1.19) (2.18) pause, rigid %.0074 ns (1.61) (3.14) 10-months-olds reaction, nonrigid %.0505 * (1.02) (2.01) pause, nonrigid %.0275 ns (2.23) (2.05) reaction, rigid %.0534 * (1.23) (2.53) pause, rigid %.0210 ns (1.10) (1.94) These results were confirmed nonparametrically. Overall, babies watching reaction events showed more recovery of looking upon reversal than babies watching pause events (Mann-Whitney U = , p =.036, 1-tailed). In all four reaction groups, the number of babies looking longer at the reversed than the familiar event exceeded chance (p =.011,.038,.002 and.011, sign test, 1-tailed, groups listed in the order of Table 2). In the pause groups, in contrast, the number exceeded chance only for 8-months-olds watching pause events involving nonrigid agents (p =.004, 1-tailed), but not for the other three groups (p >.10). Thus the results of the parametric and nonparametric analyses correspond. The present study shows that infants are sensitive to causation at a distance in the

19 Action and reaction 19 reaction event. This confirms our previous finding (Schlottmann & Surian, 1999) and complements previous work showing that infants are sensitive to contact causality in the launch event (Leslie & Keeble, 1987). Moreover, the present study shows that perception of causation-at-a-distance does not seem to depend on non-rigid motion, taken as animate by adults and talking-age children. Our results raise a number of further issues addressed by 3 control experiments below. EXPERIMENT 2 First, there are different interpretations of the finding that infant perception of causality does not depend on the non-rigid versus rigid motion manipulation: This might mean that, unlike adults, they are unaffected by the presence of animate agents, or more simply that they do not perceive our non-rigid motion as animate. In the most extreme case, they may not discriminate non-rigid and rigid motion in these animations. To test this, Experiment 2a investigated whether infants discriminate our non-rigid and rigid motions. Experiment 2b, in contrast, considered whether infants would also react to role reversal when the agents moved with a different type of non-rigid motion, not considered as animate by older observers (Schlottmann & Ray, in prep.). Method Infant Participants: Experiment 2a involved 17 infants, 9 girls and 7 boys, mean age 9 months 18 days, range 8 months 1 day to 10 months 29 days. Experiment 2b involved 17 infants, 11 girls, 6 boys, mean age 9 months 0 days, range 8 months 4 days to 10 months 22 days. Twenty further infants were excluded for parental interference, experimenter error, fussiness or failure to habituate in 12 trials. Stimuli and Procedure: Stimuli and procedure were as in Experiment 1, with the following exceptions: In

20 Action and reaction 20 Experiment 2a, 9 infants were habituated to shapes exhibiting rigid motion, then tested for discrimination on non-rigid motion. A further 8 infants were habituated on non-rigid motion and tested on rigid motion. All infants saw reaction events on habituation and test trials. No break was given between the familiar and novel test stimuli, to avoid that renewal of attention was simply due to this break rather than intrinsic features of the test stimulus. In Experiment 2b, infants were tested for a reaction to role reversal in reaction events as in Experiment 1. However, the events involved a non-rigid motion considered less animate by adults (Schlottmann & Ray, in prep.). This stimulus was created by having a square expand and contract perpendicular to its direction of translation, rather than in this direction (Figure 4). In this way, the non-rigid deformation was independent of the translation across the screen, in contrast to the standard in which the translation occurred through deformation. The amount of deformation, however, was as in the standard. t 0 t 1 t 2 t 3 t 4 Figure 4: Schematic of object motion in Experiment 2b (The square expands and contracts bilateraterally in the vertical direction while moving across the screen horizontally.) Results and Discussion Experiment 2a: Rigid versus non-rigid motion discrimination. Mean looks decreased

21 Action and reaction 21 from s to 6.94 s from the first to the last habituation trial. Infants looked for 7.98 s on the familiar test trial involving the habituated agent motion, for s on the novel test presenting the alternate motion. This recovery was the only significant effect in an ANOVA, F(1,15) = 7.76, MSE =.002, p =.007 (1-tailed), with a recovery score of This was also confirmed non-parametrically, Z = , p =.02 (Wilcoxon, 1-tailed). Thus infants indeed distinguish between a square shape moving rigidly or non-rigidly, in a manner considered animal-like by older observers. This adds to Bertenthal s (1993) data that infants distinguish non-rigid dot pattern motion derived from realistic bio-mechanical models from randomised or up-side down versions of the same patterns. Infants may be sensitive not only to realistic, but also to schematic biological motion, as introduced by Michotte. Experiment 2b: Non-rigid, vertical deformation orthogonal to object translation. Mean looks decreased from s to s from the first to the last habituation trial. Infants looked for s on the familiar test, for s on reversal. Although the recovery score was quite low,.0270, it was significant, F(1,16) = 4.53, MSE =.002, p =.025 (1-tailed). This was also confirmed nonparametrically, Z= , p =.011 (Wilcoxon, 1-tailed). Accordingly, infants may perceive causation-at-a-distance with this stimulus as well, consistent with the view that babies reaction to role reversal does not depend on the manner of motion. EXPERIMENT 3 Another issue is how the perception of contact and non-contact causality relate. Several proposals reviewed earlier hold that infants from early on, if not innately, divide the world into broad domains based on salient differences in object movement and interaction (Leslie, 1994; Mandler,1992; Premack, 1990, Baron-Cohen, 1994). All argue that an initial distinction between contact and self-motion may provide a blueprint for what eventually becomes the physical versus the social and psychological domain. Distinct percepts for

22 Action and reaction 22 contact and noncontact causality could help infants carve up the world in this way. All the proposals cited above assume that contact causality is linked to the physical domain. But it is also possible that perceptual causality is a domain general, undifferentiated ability to see relationships between events ( A does something to B ) that becomes domain specific only with further learning and maturation. Empirically, it is not clear yet whether the distinction between contact and noncontact causality is a cause or consequence of infants early, possibly innate, ontological knowledge (Carey & Spelke, 1994). From Leslie s and our results we know that infants are sensitive to both forms of causality, but not whether they distinguish the two. Leslie (1982, 1984) showed that infants distinguish launch from contiguous events without contact. These are not reaction events, however, as there is no simultaneous motion of the two shapes and in adults, they elicit fewer causal reports (Schlottmann & Ray, 2003). To test formally whether infants distinguish launch and reaction events, in Experiment 2c we habituated infants on either launch or reaction events, then tested them on the other event. Method Infant Participants: Experiment 3 involved 16 infants, 7 girls and 9 boys (mean age 9 months 15 days, range 8 months 2 days to 10 months 23 days). Five further infants were excluded for fussiness or failure to habituate in 12 trials. Stimuli and Procedure: Half of the infants were habituated to standard reaction events with either rigid or nonrigid agents, then tested for discrimination on launch events with the same agents. The other half was habituated to launch events (in which A stopped and B began to move upon contact) then tested on reaction events. Again, there was no break between familiar and novel test trial.

23 Action and reaction 23 Results and Discussion Mean looks decreased from s to 6.27 s from the first to the last habituation trial. Infants looked for 7.34 seconds on the familiar test and for seconds on the novel test, with a recovery score of The corresponding effect of trial reached, F(1,15) = 6.33, MSE =.002, p =.018 (1-tailed); there were no other significant effects in the ANOVA. The result was also confirmed nonparametrically, Z= , p =.025 (Wilcoxon, 1-tailed). Thus, infants do distinguish launch and reaction events. This is a first step towards showing that infants distinguish contact causality from causation-at-a-distance, but it is still not conclusive: This is because infants may discriminate the events based on spatio-temporal features, and yet might consider launch and reaction events equivalent in causality. EXPERIMENT 4 Finally, there is the issue of how self-initiated motion contributes to the perception of non-contact causality. Integral to a canonical reaction event is that the reacting shape B selfinitiates movement when A comes close. As outlined above, self-initiated movement is generally considered the most basic agency cue. Is this feature necessary for the perception of a causal reaction? Unfortunately, this is impossible to test with a standard reaction, because B s self-motion cannot be eliminated. However, the standard event is easy to modify while still maintaining an impression of chase and escape and self-start can be eliminated from some such modifications. The canonical event studied so far maximises similarity to launch events and parallels between contact and non-contact causality. In both cases, A moves towards B, then B moves away, so that A causes B to move. However, in the reaction event we can change the order of the motions: If B moves away first, then A chases after it, we still see a causal reaction, but now B acts and A reacts. In complex cat and mouse sequences with multiple adjustments of trajectory, as in Rochat et al. s work (1997), each might in turn act and react.

24 Action and reaction 24 If A reacts to B rather than B to A, we can eliminate self-initiated motion, by having first B, than A move in from the side, with the motion onset off-screen, similar to the way Csibra et al. (1999) tested the role of self-initiated motion in their barrier event. However, if we keep individual trajectories as before, then this non-standard reaction involves simultaneous motion across the whole screen. This is substantially more than in the standard event, which might provide additional agency information and make up for lack of the selfstart cue. To avoid this, we showed infants only the central part of the event, containing the same overlapping motion as the standard. This was done through use of an occluder with an opening in the middle, giving the impression of seeing the events through a window. In sum, Experiment 4 considered the role of self-initiated motion in the perception of a causal reaction. To this purpose, we habituated infants to non-standard, abbreviated reaction events, with the same minimally contingent motion as the standard, but without self-motion, then tested them on reversal. Method Infant Participants: Experiment 4 involved 24 infants, 14 girls and 10 boys (mean age 8 months 27 days, range 7 months 28 days to 10 months 19 days). Sixteen further infants were excluded for fussiness or failure to habituate in 12 trials. Stimuli and Procedure: Infants were habituated to a modified reaction event (Figure 5), with the shapes moving at the same speed and distance from another as in the standard. Initially the screen was blank, with the shapes moving in from the side rather than starting from rest onscreen. Blue appeared first (left-to-right version) from behind an occluder with a window-like aperture, sized so that just as Blue reached its usual starting position in the centre, Red also became visible. The shapes then moved in the standard overlapping motion. Upon reaching

25 Action and reaction 25 its usual stopping point Blue disappeared behind the occluder, with Red continuing until it also reached the edge. Piloting showed that babies quickly lost interest in a stimulus of the same duration as the standard, because for much of the time all movement was hidden behind the occluder. To ensure sufficient interest, we shortened these periods to an overall sequence length of about 3.4 seconds. Half of the babies were habituated on the left-to-right event, half on its reversal, then all were tested with the alternate movie. Results and Discussion Mean looks decreased from s to 7.53 s from the first to the last habituation trial. Infants looked for 9.45 s seconds on the familiar test and for seconds on the reversal. This recovery is impressive, especially given the shorter (and potentially duller) nature of this stimulus compared to the standard reaction. The recovery score was.0572, F(1,22) = 20.80, MSE =.002, p <.001; whether the shapes moved rigidly or non-rigidly made no difference, F < 1. The recovery was also confirmed non-parametrically, Z = , p =.0005 (Wilcoxon, 1-tailed). Thus infants perceive causation-at-a-distance with this nont 0 A B A t 1 t 2 t 3 t 4 Figure 5. The non-canonical reaction used in Experiment 2d. (Shape B moves in from the side towards its standard start position. Once it reaches this, A moves out from behind the occluder and both move in parallel, as in the standard event. At its standard stopping position, B disappears from view and A continues on its own.)

26 Action and reaction 26 canonical stimulus as well. The perception of a causal reaction does not seem to depend on the moving shapes engaging in self-initiated motion. GENERAL DISCUSSION These studies confirm that infants as young as 8 months are sensitive to causation-ata-distance of a type that adults and young children link to the psychological domain. Infants detected causal structure not only when agents self-initiated motion of a non-rigid, animallike type, but also when the agents moved rigidly or with a non-rigid motion considered less animate; this was not due to lack of discrimination between rigid and non-rigid motion. Moreover, infants detected causal structure even without self-initiated motion. Our results complement earlier data on infant perception of contact causality in launch events (Leslie & Keeble, 1987; Oakes, 1994). Taken together, the data raise the question whether infants distinguish contact from non-contact causality. As a first step towards an answer, we showed that infants distinguish launch from reaction events. These findings do not just indicate that babies react more to reversal of continuous than discontinuous, delayed motion. Leslie (1984) already showed that infants show little dishabituation to reversal of a continuous, but noncausal motion of a single object and that infants distinguish causal motion from the motion of a single object changing colour halfway. Our own finding that infants distinguish launch from reaction events also shows discrimination of different continuous motions. It would be difficult to argue against infant perception of causality on grounds that babies do not actually analyse the internal structure of this type of animation events. PERCEPTION OF CAUSATION-AT-A-DISTANCE Kanizsa and Vicario (1968) made two claims: First, that there is not only perception of contact causality, but also a parallel perception of noncontact causality. Second, that causation-at-a-distance in the reaction event appears as an impression of goal-directed action

27 Action and reaction 27 and reaction. The latter was recently confirmed with adults (Schlottmann & Ray, 2003) and children (Schlottmann et al., 2002). That we found sensitivity to the causal structure of reaction events in young infants as well is even more suggestive. Michotte (1963) took issue with this view, arguing that the impression of action and reaction was not due to a perceptual process as in the perception of contact causality, but that it involved interpretation. He argued this because he found causal impressions for reaction events only with experimenter hints, for instance, suggestions of animacy through the way in which the agents moved. Non-rigid agent motion indeed strongly enhances adults impression (Schlottmann & Ray, 2003), but it affects children less (Schlottmann et al., 2002) and infants in this study detected causality with rigid and non-rigid agent motion alike. Whether or not they actually perceived the non-rigid motion as animate, this shows that infant perception of causation-at-a-distance does not depend on suitably suggestive agent motion contrary to Michotte s and in agreement with Kanizsa and Vicario s view. Infants sensitivity to causation-at-a-distance does not imply that they like older observers -- recognise goal-directed action-and reaction in the reaction event, in particular since we do not know for certain whether they distinguish between contact and non-contact causality. Finding that infants distinguish launch from reaction events is a first step in this endeavour, but could still be a purely spatio-temporal discrimination. In contrast, had we found that infants detect causation-at-a-distance only with animal-like, non-rigidly moving agents, this would have implied some distinction from contact causality -- the latter does occur with rigid agent motion as typically found with inanimate objects. Future work will need to search for other (agent) effects differentiating contact and non-contact causality. As it stands, we know that infants are sensitive to both contact and noncontact causality, but cannot be sure yet that they differentiate between the two. In sum, two options remain open: Infants may have an unspecific impression of A