Infants' understanding of object-directed action

Cognition 98 (2005) 137–155 www.elsevier.com/locate/COGNIT

Infants’ understanding of object-directed action Ann T. Phillipsa,*, Henry M. Wellmanb a

Seaver and New York Center of Excellence for Autism Research, The Mount Sinai Medical Center, Mount Sinai School of Medicine, One Gustave Levy Place, Box 1230, New York, NY 10029-6574, USA b University of Michigan, Ann Arbor, MI, USA Received 10 January 2004; accepted 15 November 2004

Abstract When and in what ways do infants recognize humans as intentional actors? An important aspect of this larger question concerns when infants recognize specific human actions (e.g. a reach) as objectdirected (i.e. as acting toward goal-objects). In two studies using a visual habituation technique, 12-month-old infants were tested to assess their recognition that an adult’s reach is directed toward its target object. Infants in the experimental condition were habituated to a display in which an actor reached over a wall-like barrier with an arcing arm movement, to pick up a ball. After habituation infants saw two test displays, for which the barrier was removed. In the direct test event the actor reached directly for the ball, the arm tracing a visually new path, but the action consistent with attempting to reach for the object as directly as possible. In the indirect test event the actor traced the old path, reaching over in an arc, even though the wall was no longer present. This arm movement was identical to that in habituation but no longer displayed a reach going directly to its object. In a control condition infants saw the same movements but in a situation with no goal-object. In the experimental conditions, with a goal object present, infants looked longer at the indirect test event in comparison to the direct test event. In the control conditions infants looked equally at both indirect and direct test events. We conclude that sensitivity to human object-directed action is established by 12-month-olds and compare these results to recent findings by [Gergely, G., Nadasdy, Z., Csibra, G., & Biro S. (1995). Taking the intentional stance at 12 months of age. Cognition, 56, 165–193] and [Woodward, A. (1998). Infants selectively encode the goal object of an actor’s reach. Cognition, 69, 1–34]. q 2005 Elsevier B.V. All rights reserved. Keywords: Infants; Habituation; Human action

* Corresponding author. E-mail address: [email protected] (A.T. Phillips). 0022-2860/$ - see front matter q 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.cognition.2004.11.005

138

A.T. Phillips, H.M. Wellman / Cognition 98 (2005) 137–155

1. Infants’ understanding of object-directed action How do infants understand action? Adults largely understand action in intentional terms, seeing people as having intentional mental states—e.g. beliefs about the world, desires for things—reflected in intentional actions. Over the last 15 years research has demonstrated that preschoolers and toddlers share this ‘intentional stance’—thus, they employ a variety of intentional mental-state constructs to reason about persons’ actions; they conversationally describe and explain human behavior in terms of what the person ‘wants’, ‘thinks’, and ‘knows’; they distinguish intended voluntary actions from unintended biological or physical movements such as a person shaking with fever or being blown down by the wind (e.g. Bartsch & Wellman, 1995; Inagaki & Hatano, 1993; Perner, 1991; Schult & Wellman, 1997). Recent findings suggest that even one 1/2- and 2-year-olds are able to reason about persons’ intentions (Repacholi & Gopnik, 1997). While children at this age are not able to do sophisticated mental state reasoning, they are able to appreciate the difference between intentional and unintentional behavior (Carpenter, Aktar, & Tomasello, 1998; Meltzoff, 1995) and to appreciate that words refer to objects (Baldwin, 1991). That this kind of intentional understanding is in place by the second year of life means it must be built upon an earlier foundation. Thus, there is now considerable interest in whether and when infants understand actions as intentional. Theoretically, some authors argue that infants are ‘hardwired’ to understand intentionality and this understanding is ‘triggered’ by actions of a certain kind, in particular autonomous motion (e.g. Baron-Cohen, 1995; Premack, 1990). Others argue that it is through learning experiences, such as association and histories of reinforcement, that infants come to understand intentional action (e.g. Moore & Corkum, 1994; Rogers & Griffin, 2002). Theorists of the first sort tend to assume that infants’ analyses of intentional action are, from early on, deep, abstract and general. They are deep in the sense of providing the infant with a construal of the actor’s internal mental states (specifically intentions but other states as well), abstract in the sense that intentional analyses apply to humans, to animals, even to geometric computerized figures, and general in the sense that many different sorts of actions and movements all specify to the infant a goal-directed, intentional action. Adult understandings do seem deep, abstract and general in this sense (e.g. intentional actions can be seen in abstract shapes and are seen in such object-less actions as waving good-bye). Theorists of the second sort tend to assume that the infant’s initial understandings are shallow, concrete and specific. They are shallow in the sense of originally capturing certain action regularities (not deeper mental states), concrete in the sense of only applied for some specific agents or contexts (e.g. only for humans, the most familiar agents in the infant’s world), and specific in the sense of extracted only for some acts, for example, only actions that influence objects clearly and directly. There is, however, agreement across these different perspectives that infants must be able to detect and recognize action patterns in the world around them. Theorists of the second sort assume that infants are detecting and learning from specific instances, while theorists of the first sort assume that innate knowledge is triggered by the experience of witnessing events which fall into certain categories. Thus, an important question relevant

A.T. Phillips, H.M. Wellman / Cognition 98 (2005) 137–155

139

to all current theorizing concerns, what features of observable behavior might plausibly indicate and manifest intentionality? Infants can distinguish mechanical movements vs. biological self-propelled movements via a number of movement features (e.g. Mandler, 1992). Intentional actions, however, have additional movement features, in particular, the hallmark of intentionality is referentiality; all intentional acts have an ‘object’. More specifically, an important subset of intentional actions are goal-directed behaviors, and the prototype for this set of behaviors is goal-directed actions on material objects, or object-directed actions. Two related features, among others, seem to be characteristic of object-directed action. First, object-directed actions typically go directly to their object, as opposed to moving around in an indirect fashion. Second, object-directed actions go around obstacles to get to their object, not simply stop when encountering an initial obstacle. If these features of objectdirected action provide typical, at times distinctive, markers of intentional action, then they could be features infants use in understanding intentionality and in linking their intentional understanding with observable behaviors. At what point do infants show an appreciation for these markers of goal-directed action? Although there is a long history of inquiry into infant social awareness, most recently researchers have extended preferential looking time methods used to investigate infants’ understanding of the physical world (e.g. Baillargeon, 1993; Spelke, Breinlinger, Macomber, & Jacobson, 1992) to examine their understanding of the social world (e.g. Gergely, Nadasdy, Csibra, & Biro, 1995; Phillips, Wellman, & Spelke, 2002; Woodward, 1998). With regard to infants’ understanding of goal-directedness a series of studies reported by Gergely and colleagues has been particularly influential. As shown in Fig. 1, these authors used computerized two-dimensional displays of circles involved in intentional-like motion first with 12-month-olds (Gergely et al., 1995) and then with 6- and 9-month-olds (Csibra, Gergely, Biro, Koos, & Brockbank, 1999). Infants were habituated to a circle moving up to and then over a wall-like barrier, then back down across the barrier to join up with a second circle. After habituation infants saw two test events for which the barrier was removed. In the direct test event the first circle moved directly in a straight line to link up with the second circle. In the indirect test event the circle moved in the same path as in habituation (although the barrier no longer intervened). Nine- and 12-month-old infants looked longer at the indirect test event. Six-month-olds did not do so. Importantly, in a control condition, infants saw a habituation event much like the one in Fig. 1, except

Fig. 1. Displays for the habituation and test events for infants in the research of Gergely et al. (1995).

140

A.T. Phillips, H.M. Wellman / Cognition 98 (2005) 137–155

there was no barrier. When habituated to this event infants looked equally at the direct and indirect test events, dishabituating to both. These results provide evidence that 9- and 12-month-old infants appreciate the goal-directedness of action; their importance inspired us to explore them further. Specifically, there are two ways in which we see beneficial expansion. First, if infants understand goal-directedness this should be evident as well in their reaction to live human actions. That is, infants should show the same pattern of results for people as for computerized shapes. Gergely and his colleagues certainly assume that their data inform us about an infant understanding that encompasses the structure of human action, however, their research has not included displays of live human actors. An additional issue concerns controlling for alternative interpretations. The control condition, used by Gergely et al. demonstrated that infants no longer showed differential looking to the direct and indirect test events when the barrier was removed. This controls for the possibility that infants have a preference for the indirect over direct trajectories in their experimental condition. However, given a focus on object-directed behavior, it seems crucial to examine infants’ appreciation of actions with and without goal-objects, rather than or in addition to actions with and without barriers. It is possible that infants in Gergely et al.’s research might be demonstrating expectations about movement trajectories in the presence and absence of barriers regardless of whether the action is object-directed. For example, infants may simply have expectations that agents will move around barriers in their path, and otherwise move in straight lines (see also Rogers & Griffin, 2002). If infants appreciate goal-directedness, then presence of the goal object should crucially influence their reactions. To address these issues we assess infants’ understanding of human action rather than computer displays of animated shapes. Our experimental condition and control condition directly contrast action with and without an object (rather than action with and without a barrier).

2. Study 1 In the current research infants in the experimental condition saw a person reach over a barrier and grasp an object, as shown in Fig. 2. They saw this event until habituated, and then were shown two test events where the barrier was removed. One test event showed a direct reach for the object. The other test event showed an indirect reach. Like Gergely et al.’s research, this experimental condition contrasts two alternative possible encodings of the actor’s actions. In one alternative, during habituation the actor’s actions are encoded as reaching as directly as possible for the object. If so, then during test, with the barrier removed, the direct reach event preserves that encoding of the action (albeit varying the trajectory of the reach). An indirect reach, however, no longer reaches as directly as possible for the object. The other alternative is that during habituation the actions are encoded in terms of the specific, physical trajectories involved. If so, then during test the indirect reach preserves the arcing trajectory of movement shown in habituation; the direct reach changes this trajectory. In total, if infants encode the actions in terms of physical trajectories, then they should prefer the direct test event, because it is new in the sense

A.T. Phillips, H.M. Wellman / Cognition 98 (2005) 137–155

141

Fig. 2. Displays for the habituation and test events for infants in the experimental condition.

of no longer showing the same physical movement. But if infants encode the actions in terms of going-as-directly-as-possible for the object, then they should prefer the indirect test event, because it is new in the sense of no longer showing a reach that is as direct as possible. In the control condition, as shown in Fig. 3, infants see events identical to those in the experimental condition except that no goal object is involved. 2.1. Method 2.1.1. Participants Thirty-two healthy 12-month-old infants participated in the experimental condition. One was eliminated due to fussiness, two were eliminated because of interference from a parent, and five were eliminated because of observer error. The mean age of the remaining 24 infants in the experimental condition was 11 months-28 days (range 11-12 to 12-22). There were 12 males and 12 females. Thirty healthy infants participated in the control condition. Three were eliminated due to fussiness, three were eliminated because of observer error. The mean age of the remaining 24 infants was 11 months-29 days (range 11-15 to 12-23). There were 14 males and 10 females. Infants’ names were obtained from birth announcements in the local newspaper. Parents were contacted by letters and follow-up phone calls and those that participated were paid $10 as reimbursement for their travel expenses.

Fig. 3. Displays for the habituation and test events for infants in the control condition.

142

A.T. Phillips, H.M. Wellman / Cognition 98 (2005) 137–155

2.1.2. Procedure Infants sat in a high chair or their parent’s lap facing a table (120!80 cm) 60 cm away. The table supported two objects: a multi-colored plush ball (13 cm in diameter) placed 80 cm from the left edge of the table and a thin rectangular box, (32!26!7 cm) placed 60 cm from the left edge of the table. A female adult, the presenter, sat sideways to the infant facing the ball, and the box formed a barrier between the person and the ball (see Fig. 2). The entire display was surrounded by light blue curtains. The presenter wore a black, long-sleeved shirt and short white gloves to make her hands and arms salient. As in the figure she sat at the edge of the table with her face oriented to the ball, turned two-thirds of the way from the infant. A video recorder, placed above the display, recorded the infant through a hole cut in the curtain. A second camera recorded the display presentation. Live observations of the infant’s looking times were recorded by a hidden observer, who was blind to the infant’s condition, through another hole to the side of the camera. The observer pressed a button when the infant was looking at the display, then released when the infant looked away. The button fed to a computer program which kept track of the infants’ looks, and indicated when habituation had been accomplished. The habituation program computed the habituation criterion by averaging the total looking times for the first three habituation trials. Habituation trials were terminated (and test trials begun) when the infant had three trials in a row, each with looking times 50% or less than this habituation criterion (or when the infant had received 10 total trials). Thus, the minimum number of habituation trials any infant was shown was six and the maximum was 10. A trial (either habituation or test) began when the infant looked at the display for at least 0.5 s. If the infant did not initially look at the display for that long, the observer made a noise behind the curtain to attract the infant’s attention to it. The trial ended when the infant looked away from the display for 2 s. In this way exposure to the trials was controlled by the infant’s looks. 2.1.3. Habituation event—experimental condition (reaching for objects) Infants were habituated to a display in which the female presenter reached over the barrier (with an arcing motion) to grasp and pick up the ball, bringing it to her torso (see Fig. 2). This motion was at the rate of 60 cm/s. The actor then replaced the ball, and repeated the reach and grasp. This cycle continued until the infant looked away for 2 s, which ended the habituation trial. The infant’s attention was then drawn to the display, and another habituation trial was begun, until the infant reached habituation criterion. In all these trials the presenter was shown from the waist up, from the side, with the arm doing the reaching the one closest to the infant. Because the presenter’s body and face were positioned toward the barrier and object, and thus were turned slightly away from the infant, the infant could see little of the presenter’s face. We wished to display an entire person not just an arm, but to concentrate the infants’ focus of attention on the reach rather than the face. 2.1.4. Test events—experimental condition (reaching for objects) After habituation, the barrier was removed and the infant shown the following two test events. (a) Direct reach event (consistent event): the presenter reached directly

A.T. Phillips, H.M. Wellman / Cognition 98 (2005) 137–155

143

(in a straight-line) for the ball and grasped it. In this event the arm traced a visually new path in contrast to habituation, but the action was consistent with the reach going directly to the goal-object, the ball. (b) Indirect reach event (inconsistent event): the presenter’s arm traced the old path (the same as in habituation), reaching in an arc, even though the barrier was no longer present. This display depicted a path which was visually identical to that in habituation; however, this path was now inconsistent with the reach going directly to the goal-object. These test trials were shown three times each, with each test event in alternating sequence and order of test trials counterbalanced between subjects. 2.1.5. Habituation event—control condition (reaching without objects) In the control condition, the infants were also habituated to a display in which a female actor was seated behind a short wall-like barrier placed on a tabletop. However, as shown in Fig. 3 this time there was no ball and hence no visible goal-object on the other side of the barrier. The actor reached over the barrier with the same arcing motion as in the experimental condition. 2.1.6. Test events—control condition (reaching without objects) The barrier was again removed and the following two test events shown. (a) Direct reach event: the presenter now reached directly away from her body (in a straight-line). This was visually different from the path traced during habituation. (b) Indirect reach event: the presenter traced the old path (the same as in habituation), reaching in an arc. 2.1.7. Presentation checks Because the habituation and test events were presented live to the infant, presentation errors were possible. Therefore, tapes of all presentation events were examined to ensure that (a) the arcing arm movement was of the same height and curvature during test as it was during habituation (with the barrier removed, an arcing reach might not be as high or as curved as with it present), (b) the arm motion in the direct reach test events was low and parallel to the table, and (c) the arm motions were neither too fast nor slow (and thus went at the same rate across direct and indirect events). No presentations were flawed in these fashions so that it was unnecessary to exclude any data because of presentation errors. It still might be possible that the presenter’s arm movements significantly differed when there was an object present (experimental condition) vs. when there was not (control) and that infants were responding to these movement differences. In order to check whether or not the arm movements in the displays presented to the two groups were noticeably different from one another, we occluded the monitor so that only the reach was visible, not the goal. We then had four coders who were blind to the condition view all the taped displays and guess which condition the videotaped arm movement came from. Observers were at chance in their ability to guess which condition the tapes came from; they were correct 58% of the time. 2.1.8. Reliability checks Reliability of the primary observer was assessed for all infants by a second observer who recorded looking times from videotapes of the infants. If the primary observer made

144

A.T. Phillips, H.M. Wellman / Cognition 98 (2005) 137–155

an error, such that the infant was allowed to see a test display again after that infant had already looked away for 2 s, or if a trial was cut short while the infant was still looking, that infant was replaced. The secondary observers were blind to the presentation order of the test events, and whether the infant was in the experimental or control condition. The secondary observers’ judgments were compared with the looking times recorded by the live observer in several ways. For the infants included in the study, the percentage of test trials for which the observers’ judgments agreed within 2 s or less was 100%. The percentage of test trials for which the observers’ judgments agreed within 1 s or less was 79%. 2.2. Results Our analysis strategy (here and in the subsequent study) is to focus on the comparison between infants’ looking to direct and indirect test events. If, during habituation, infants appreciate and are thus bored by the actor reaching directly as possible for the object, then during test they should look longer at the indirect reach than the direct reach. Our analyses for this comparison rest on summing the infant’s looking time across the three test trials of any one type. Collapsing looking times across test trials, to achieve a more stable sample of infant response, is a common approach for research of this type (see e.g. Baillargeon, 1986; Spelke et al., 1992). Preferential looking studies consistently produce looking times with highly irregular distributions of the sort requiring data transformation or nonparametric analyses. A standard nonparametric approach for analyzing data of this sort is to compare withinsubjects conditions via Wilcoxon Signed Ranks tests, and to compare between subjects data via the Mann–Whitney U-test (see e.g. Carpenter et al., 1998; Spelke et al., 1992). 2.2.1. Habituation Infants in the experimental condition reached habituation criterion on an average of 6.8 trials. Three infants did not reach habituation criterion, and were shown the maximum of 10 habituation trials.1 Infants in the control condition reached habituation criterion on an average of 7.3 trials. One infant did not reach habituation and was shown the maximum number of 10 habituation trials. On the last habituation trial, experimental infants looked for an average of 6.5 s, control infants looked for an average of 7.5 s. Last trial habituation looking times did not significantly differ in the two conditions—Mann–Whitney test, zZ0.58, PO0.55. 2.2.2. Test As shown in Fig. 4, the 24 infants in the experimental condition looked longer at the indirect reach test event than at the direct reach test event (MsZ7.9 vs. 6.6 s per test trial). 1 Any habituation criterion is arbitrary and some studies of this sort even use a fixed number of familiarization trials. What is most important for this sort of design is that infants receive sufficient exposure to the habituation displays and begin to loose interest. In our data the looking times for the three 12-month-olds in the experimental condition who did not reach criterion nonetheless declined from an average of 20.9 s looking to their first habituation trial to 10.3 s looking to their tenth trial.

A.T. Phillips, H.M. Wellman / Cognition 98 (2005) 137–155

145

Fig. 4. Mean looking times for test events for 12-month-olds in Study 1. Each bar represents the average of three test trials.

To analyze these data we subtracted the sum of the three looking times for one test event from the sum for the other test event in order to compare looking times to the two test events with the Wilcoxon test. Experimental infants looked significantly longer at the indirect reach—Wilcoxon, zZ1.97, P!0.05. Individually, summing across their three trials of each test type an infant could either look longer at the direct or at the indirect test events. Eighteen of the 24 experimental 12-month-olds (75%) looked longer at the indirect reach test event—Sign test (P!0.02). The 24 control infants did not look significantly longer at the indirect reach event than the direct reach event (MsZ7.0 vs. 6.7 s)—Wilcoxon, zZ0.31, PO0.75. Individually, summing across their three trials of each test type, 12 control infants looked longer at the indirect reach while 12 infants looked longer at the direct (N.S.-Sign test). Dishabituation patterns provide a less direct test of our hypotheses. Both test events introduce changes from habituation displays (e.g. removal of the barrier), so in some senses both are novel. However, habituation and both test events remain very similar (e.g. all show object-directed reaching). Accordingly, infants might dishabituate somewhat to both test events or not significantly dishabituate to either test event (but nonetheless look significantly longer at the indirect reach, as hypothesized). Thus, it is the contrast between looks to the two test events that is the most important and sensitive measure in this design. With this caveat, experimental infants did look significantly longer to their first indirect test trial than they did to their last habituation trial (MsZ8.7 vs. 6.5 s, Wilcoxon, zZ1.66, P!0.05 one-tailed), and they did not look longer to their first direct test trial than to their last habituation trial (MsZ6.2 vs. 6.5 s, zZ0.34, PO0.35). Control infants did not dishabituate to either test event (PsO0.20). 2.3. Discussion Twelve-month-old infants exhibited the predicted pattern consistent with encoding the habituation actions in the experimental condition as the actor reaching directly for

146

A.T. Phillips, H.M. Wellman / Cognition 98 (2005) 137–155

the object. Thus, they looked longer at the indirect test event, where the actor is no longer reaching directly for the object. Note that this indirect test event showed the exact same arcing arm motion as the infants had seen repeatedly in habituation. However, with the barrier now removed, this test event no longer showed the person going as directly as possible for the object. Thus, these findings provide evidence that infants were encoding the actor’s habituation actions in terms of the action’s goal-directedness rather than, or more than, the simple bodily movements involved. Performance in the control condition rules out the possibility that infants just prefer (and thus look longer at) curvilinear arm motions. More importantly, the experimental and control conditions provide a conceptually informative contrast between reaching with and without an object. Conceivably, infants’ responses to the experimental displays could be based on appreciation solely of movements and obstacles—something like the principle that in the absence of obstacles movement continues regularly, irregular movements come from the presence of obstacles. However, our experimental and control conditions both display those features. They differ not in the presence and absence of obstacles, but in the presence and absence of goal-objects. Thus, our data confirm that infants are sensitive to the object-directed nature of human actions. Our interpretation is that circumventing obstacles and taking the most direct path are especially relevant for infants when there is a goal-object present. If the action is not object-directed, then reaching in a straight line or reaching in an arc are simply different patterns of movement—neither more interesting nor interpretable than the other. Subsequent to this initial study, Sodian, Schoeppner, and Metz (2004), referring to our results, conducted additional research investigating goal-directed action. They conducted two studies, one using video displays of human actors and another using live puppets. In both cases the person or puppet approached another person or puppet, surmounting a vertical obstacle in a parallel fashion to Gergely et al.’s (1995) design. The control conditions also replicated the control condition used by Gergley et al. With these displays, Sodian et al. also found that 12-month-olds appreciated goal-directedness. But, neither of these studies present infants with live human actors. All together, results now provide support for goal-directed understanding extending across computer animations (Gergley et al., 1995), puppet and video presentations (Sodian et al., 2004), and live human action (Study 1). However, only our research incorporates a no-object control condition, crucial for assessing understanding of goal-directed action by contrasting action with and without goal objects.

3. Study 2 The findings from Study 1 deserve replication and extension. In particular, confirmation of the contrast between understanding of action with and without goal objects seems important. Study 2 provides a replication where infants see an actor on videotape. There are several reasons for using video. One concerns the inevitable variability incurred by using live actors. Although we carefully reviewed video tapes of all the presentations, live actors repeating the same actions again and again inevitably display some variations in their movements, postures, and timings. Video displays can

A.T. Phillips, H.M. Wellman / Cognition 98 (2005) 137–155

147

be precisely equivalent across infants, from habituation to test, and across experimental vs. control conditions. Further, infants’ comprehension of video tapes of human actors is a topic of substantive interest in its own right. Of course, it is possible that infants do not react to videotaped action as they do to live actors. But there is preliminary data suggesting that they might, as provided by Sodian et al. (2004). However, Sodian and colleagues do not provide a direct comparison of infant responses to video displays and to equivalent displays of parallel live actions. Study 2 uses videotaped displays of the exact same reaching actions as in Study 1; this both adds to the small body of research using videotaped action to test infants understanding of persons (Sodian et al., 2004; Mumme & Fernald, 2003; Baldwin, Baird, Saylor, & Clark, 2001) and helps validate the use of videotaped actions in research with infants.

3.1. Method 3.1.1. Participants Thirty-five healthy 12-month-old infants participated in the experimental condition. Six were eliminated due to fussiness, one was eliminated because of interference from a parent, and four were eliminated because of observer error. The mean age of the remaining 24 infants in the experimental condition was 12 months-2 days (range 11-3 to 12-22). There were 13 males and 11 females. Thirty-four healthy infants participated in the control condition. Six were eliminated due to fussiness and four because of observer error. The mean age of the remaining 24 infants was 11 months-29 days (range 11-12 to 12-19); there were 11 males and 13 females. Infants were recruited as in Study 1.

3.1.2. Procedure Infants sat in a high chair in front of a 23!34 cm video monitor. The monitor was directly behind a rectangular hole in a dark blue large wooden partition so that the infant saw the monitor screen alone, flush with the hole. The room lights were dim and indirect to make the actions displayed on the screen more visually salient. The mother sat behind the infant. A video camera, placed directly below the video monitor, recorded the infant through a hole cut in the wooden partition that framed the display. Live observations of the infant’s looking times were recorded by a hidden observer through another hole to the side of the video screen.2 The observer recorded infants’ looking onto a computer program which, just as in Study 1, kept track of the infants’ looking. The computer both collected the looking time data and presented the displays based on the infants’ looking toward 2 Unlike the live experiment, this primary observer was not blind to the infant’s condition. One potential advantage of the video method is that a single adult can test each infant. In doing so, however, that experimenterobserver is privy to the infant’s condition. In Study 2, however, as in Study 1, the secondary observer who coded from video tapes of the infant’s looking was always blind. As explained later our primary data analyses utilized the observations from this second, blind observer.

148

A.T. Phillips, H.M. Wellman / Cognition 98 (2005) 137–155

and away from the displays. In exactly the same way as Study 1, the habituation program computed the habituation criterion, terminated habituation and began the test trials. Just as in Study 1 the minimum number of habituation trials any infant was shown was six and the maximum was 10. 3.1.3. Events The habituation and test events were the same as in Study 1 except that every infant saw the same video of a single male actor engaging in the reaching actions. To begin a trial, the video monitor showed a flashing red and white checker board to draw the infant’s attention to the display. When the observer saw the infant attend to the display he or she pushed a button to begin the presentation. Then, in the experimental condition, infants saw the same actions as shown in Study 1 and depicted in Fig. 2. In the control condition, infants saw a video of the same male actor executing identical actions but in this case with no ball, as depicted in Fig. 3. 3.1.4. Presentation checks The video tape presentations were filmed and edited to ensure that (a) the arcing arm movement was of the same height and curvature during test as it was during habituation, (b) the arm motion in the direct reach test events was low and parallel to the table, (c) the arm motions were neither too fast nor slow (and thus went at the same rate across direct and indirect events) and (d) the arm motions, postures, etc. were identical for the experimental and the control tapes. 3.1.5. Reliability checks Reliability between the live primary observer and the secondary observer who recorded looking times from videotapes was assessed for all infants. The secondary observers were completely blind to the video displays that any infant was seeing and thus were blind to the infants’ condition. If the primary observer made an error, such that the infant was allowed to see a test display again after that infant had already looked away for 2 s, or if a trial was cut short while the infant was still looking, that infant was replaced. As noted earlier, in this study eight infants were replaced for these reasons. For the infants included in the study, the percentage of test trials for which the observers’ judgments agreed within 2 s or less was 96%. The percentage of test trials for which the observers’ judgments agreed within 1 s or less was 86%. 3.2. Results Because primary observers were not blind to an infant’s condition, all analyses were conducted on the data as observed by the blind, secondary observers. However, the pattern of results is identical if the data from the primary observer is used instead. 3.2.1. Habituation Infants in the experimental condition reached habituation criterion on an average of 9.2 trials. Seventeen infants did not reach habituation criterion, and were shown the maximum of 10 habituation trials. Control infants reached habituation criterion on an average of 9.3

A.T. Phillips, H.M. Wellman / Cognition 98 (2005) 137–155

149

Fig. 5. Mean looking times for the three test trials of each type for the 12-month-olds in Study 2.

trials; 17 infants received the maximum of 10 habituation trials.3 On the last habituation trial, experimental infants looked for an average of 9.0 s, control infants looked for an average of 6.9 s, not significantly different—Mann–Whitney, zZ0.84, PO0.40. 3.2.2. Test Fig. 5 shows the data from Study 2. As shown in that figure, the 24 infants in the experimental condition looked significantly longer at the indirect reach test event than at the direct reach test event (MsZ9.3 vs. 6.8 s per test trial)—Wilcoxon, zZ3.51, P!0.005. Individually, summing across their three trials of each test type an infant could either look longer at the direct or at the indirect reach events. Twenty of the 24 experimental infants (83%) looked longer at the indirect reach events—Sign test (P!0.01). Fig. 5 also shows the data from the control infants. In this condition infants did not look significantly longer at the indirect path test event (6.8 vs. 5.7 s)—Wilcoxon, zZ1.17, PO0.20. Individually, 15 of 24 infants looked longer at the indirect test event—N.S. Sign test (PO0.20). For the same reasons outlined in Study 1, dishabituation patterns are less important and less revealing. In Study 2, experimental infants looked only somewhat longer at their first trial indirect test event than they did at their last habituation trial (MsZ10.4 vs. 9.0), Wilcoxon, zZ1.03, PZ0.10 one-tailed. In contrast, they looked minimally less at first trial direct test events (MsZ8.7 vs. 9.0), Wilcoxon, zZ0.54, PO0.29. Control infants did not dishabituate to either test event (PsO0.25). 3

Infants found it more difficult to habituate to the video displays than live displays. Descriptively, we observed that infants would become bored and frustrated with the video displays, yet nonetheless continue looking. This is one reason we set our maximum habituation trials at 10; in pilot studies more trials meant a very high number of infants fussed-out in the video procedures. Crucially, however, we used a 10 trial maximum in both Studies 1 and 2, for comparability. Also, as mentioned in Footnote 1, what is most important for this sort of design is that infants receive sufficient exposure to the habituation displays and begin to lose interest. In our data the looking times for the infants who did not reach criteria nonetheless declined from an average of 10.3 s looking on their first habituation trial to 8.0 s looking on their tenth trial.

150

A.T. Phillips, H.M. Wellman / Cognition 98 (2005) 137–155

3.2.3. Comparing studies 1 and 2 A Mann–Whitney comparison between experimental infants in Study 1 vs. Study 2 showed no significant difference—zZ0.45, PO0.60. In both studies experimental infants similarly looked longer at the indirect rather than direct test events (as clear comparing Figs. 4 and 5). A parallel comparison between control infants in both studies yielded no significant difference—zZ1.07, PO0.25. In both studies control infants’ similarly looked equivalently at indirect and direct test events. These equivalences allowed us to compare looking across the two studies. A key prediction is an interaction showing preference for the indirect event for experimental but not control infants. Following Spelke et al. (1992), we tested for this interaction by subtracting the sum of each infant’s three looking times at the direct test event from the sum of their three looking times at the indirect event. This difference score indexes the infants’ preference for the indirect event. Experimental infants’ preference (MZ5.6 s) significantly exceeded control infants’ (MZ1.2 s)—Mann–Whitney, zZ2.83, P!0.005.4 Considering individuals, 35 of the 48 experimental infants looked longer at the indirect reach test events—Sign test, zZ3.61, P!0.002. Twenty-six of 48 control infants did so— N.S. Sign test. 3.3. Discussion The data in Study 2 replicate those in Study 1 quite closely: infants respond to videos of human action similarly to parallel live actions, at least for simple displays of single actions of the sort used here. If anything, infants’ data were somewhat less variable and so somewhat more significant with the videoed displays, presumably because the use of video eliminated the inevitable variability of live actors. However, infants also responded significantly in Study 1 to live actors. If infants are understanding, or learning from, real-life behavior of ordinary people in everyday interactions, then, of course, they should perform significantly with live actors as well as videos. In this regard, the analyses comparing infant looking times to the live vs. the video displays revealed no differences; the only effect was that in both studies infants consistently looked longer at the indirect rather than direct reaches during test trials for the experimental condition (but not the control condition).

4. Final discussion These studies examine infant appreciation for object-directed reaching and show that by 12 months infants have come to recognize human reaching toward objects as goal-directed. In Study 1 when presented with live actors, 12-month-olds looked longer at indirect test events over direct test events. In Study 2 when presented with videos of 4

A similar comparison was not significant in Study 1 alone but significant at PZ0.046 in Study 2. Combining across studies appropriately increases the power of the test and shows a consistently significant result.

A.T. Phillips, H.M. Wellman / Cognition 98 (2005) 137–155

151

human action paralleling the live displays of Study 1, 12-month-olds showed the same pattern of results. Evidence from the control conditions, which presented the same reaching movements in the absence of an object, rules out several alternative explanations for longer looking at the indirect test events and underscores the importance of the presence of a target-object for infants to appreciate reaching as goal-directed. Although using very different methods, our findings complement the earlier studies by Gergely and his colleagues (Csibra et al., 1999; Gergely et al., 1995). Both sets of studies involve object-directed action that takes a direct or indirect path to the goal-object, and show that 12-month-olds appreciate object-directed action. Importantly, we tested infants with human actors engaged in human action and provide a key control condition, one with no goal-object. In the absence of such a control, it was possible that infant responses to the computerized displays used by Gergely and his colleagues, or to the displays used by Sodian et al. (2004), were simply showing an appreciation of typical, object-less movement trajectories. Our results are not open to this interpretation, however. Taken together our studies along with those of Gergely and colleagues as well as Sodian and colleagues provide broad-based support for an important form of goal-directed understanding in 12-month-olds. One other series of studies has examined object-directed human reaching. Woodward (1998, 1999) has demonstrated that infants as young as 6-months encode a connection between the grasp of a human hand and the objects grasped. In her research, infants were habituated to a hand reaching to and grasping one of two toys which were adjacent to each other about 18 in. apart. After habituation to this repeated reaching event, infants saw two test events in which the locations of the objects were switched. In the new object/old path event the hand reached to the old location and thus now grasped a different object. In the old object/new path event the hand grasped the same object as in habituation, but it was now of course in the other location. If infants perceived or encoded the habituation event simply in terms of the spatial movement of the hand, then the new object/old path event would seem familiar (because the arm and hand execute the same trajectory as habituation), and infants should look longer at the old object/new path event because it shows a novel arm movement. However, if the infants perceived or encoded the habituation movement in terms of the hand’s association or connection with the target object, then they should look longer at the new object/old path event because there the hand is connected with a new object. In a series of studies, 6- and 9-month infants looked longer at the new object/old path event. Woodward’s results, like ours, were found with live human actors, but Woodward also contrasted human action with the movements of physical objects, and found that in her paradigm objects link only to human grasps. Specifically, infants did not make a similar connection between a rod with a claw and an object, when they saw identical displays where the arm-hand was replaced by a rod-claw of similar size, shape and color to the arm (Woodward, 1998). Note that our studies and Gergely’s are designed to address different aspect of the connection between action and goal objects than Woodward’s series of studies. Woodward addresses whether infants encode actions as connected ‘with’ objects in special ways; our studies address infants understanding that actions may go ‘for’ objects. Specifically, our studies and Gergely’s address whether infants appreciate the two characteristic features of

152

A.T. Phillips, H.M. Wellman / Cognition 98 (2005) 137–155

object-directed action that we outlined in our introduction: that object-directed actions go directly to their object, and that object-directed actions go around obstacles to get to their object. How do these results clarify the nature of the origins of intentional understanding? This requires consideration of what 12-month-olds have achieved, what precedes that understanding, and what 12-month-olds have yet to achieve. Assume that one prototypic intentional action involves a person acting on and affecting a specific object—a woman putting up a towel (Baldwin et al., 2001), a hand grasping a teddy bear (Woodward, 1998), a person picking up a ball. Gergely and colleagues, in clarifying their position, argue that it is possible for infants to appreciate such actions as goal-directed without necessarily appreciating intentions (Gergely, 2002; Gergely & Cisbra, 2003). We favor this general sort of proposal. That is, imagine an early stage of intentional understanding where infants construe action in terms of its directedness to goal-objects—its action states—but a construal that does not penetrate to the actor’s intentions—his/her internal states. That is, potentially, an infant could identify the goalstate and goal-object that an actor is moving toward, without identifying the mental states that propel the actor (the actor wants the object); the infant could focus on objectified action not subjective states. In their proposals, Gergely and Cisbra (2003) contrast what they call the teleological stance with the intentional stance. A teological stance interprets goals as manifest in actions. In contrast, an intentional stance focuses on internal states of intention, even in the absence of goal-directed action. Moreover, in their proposal, the teleological stance allows infants to make judgments about the physical efficiency of the behavior. And, Gergely and Cisbra (2003) argue that a teleological stance is abstract with regard to the sorts of entities it applies to (e.g. people, abstract shapes) although at the same time minimalist in the sense just outlined. They argue that this initial, minimalist conception can be expanded later in development to include internal mental states thereby transitioning into the intentional stance. What sort of action features might manifest or cue such a goal-directed action construal by the infant (or on other accounts might possibly cue a full-blown intentional understanding by the infant)? Recent research points to several candidates: Object presence. A goal-object is integral to such an analysis, so (at first) a visible object might be key. Note that, indeed, our data show that 12-month-old infants construe differently action in the absence of an object as opposed to the exact same action in the presence of a goal-object. Action manner. An action that wanders haphazardly and then contacts an object, is perceivably different from one that goes as directly as possible to contact an object. In Gergely et al.’s control condition, for example, being habituated to an indirect action without a barrier but toward a goal-object, apparently does not specify goal-directedness to 9- and 12-month-old infants. Action ‘success’. Meltzoff (1995) has shown that older infants (18-month-olds), in an imitation paradigm, will produce a completed, successful action when presented an incomplete, unsuccessful one. Specifically, if a model tries to hang a ring on a hook, the infant imitates not the failed act, but the unobserved ‘successful’ one that was probably intended. While failed actions may be understood by older infants in terms of

A.T. Phillips, H.M. Wellman / Cognition 98 (2005) 137–155

153

an unobserved goal state, younger infants (at first) might require an observed obtainment of the goal-object to specify goal directedness. Whether or not infants begin with a general, deep and abstract understanding, or instead an understanding more specific to certain action features and/or types of actors, our findings taken together with those of Gergely et al. and Sodian et al. demonstrate that infants’ understanding of goal-directed action is notably general and abstract by the time they reach 12 months of age. Not only do 12-month-olds demonstrate expectations about a novel person, in a novel situation, but they apply the same understanding to computerized shapes with which they most likely have no experience whatsoever, as well as puppets and videos. Further, Johnson (2000) shows that, under several conditions, 12-month-olds treat a nondescript furry device as goaldirected. But these results cannot definitively answer the question of the origins of the appreciation of goal-directed action nor tell us how this understanding has developed in the first year of life. In particular, current data do not tell us whether an understanding of goal-directedness is present innately, or primarily driven by experience. Relatedly, the data do not tell us whether this understanding is first specific to human actions or is more abstract and thus (even at first) maps onto abstract entities. This remains an open question because no data exist for younger infants, before 6- to 12-months. Woodward (1998) has the earliest findings, and, as noted earlier, her data suggest that early achievements in intentional analyses (at 5- and 6-months) might be specific to humans. In addition, the imitation data of Meltzoff (1995) suggests that more complex analyses (e.g. of ‘failed’ actions) might first be specific to humans. But this is an issue in need of further research, especially with younger infants. What about developments beyond 12-months? Under the minimalist interpretation of 12-month-old achievements outlined above, infants have much yet to achieve. Here our data provide one piece of important evidence, in this case from our control condition. In fact, our no-object control condition provides more than a control, in that it examines infants’ understanding of a specific sort of objectless action. Our data demonstrate that at 12 months infants fail to appreciate goal-directedness in the case where action affects no visible object. Thus, infants at this age are limited in their ability to perceive goal-directed behavior in the absence of a goal object. Yet, an adult understanding encompasses a wide range of goal-directed action, beyond reaching for visible objects, and does not require the physical presence of a goal object. Note that the point at which infants appreciate that even object-less behaviors may be goal-directed could represent an intriguing developmental step. Not only would it show an increasingly sophisticated appreciation of goal-directed action, it might demonstrate a step toward understanding goals, subjective states of the agent, beyond objectified action. An appreciation of behavior as goal-directed in the absence of a goal object may only be possible once the transition from a teleological to an intentional level of understanding has taken place. In any event, appreciation of goal-directedness is intricately related to intentional understanding. Our studies help demonstrate that by 12 months infants can recognize observable features of intentional action, namely crucial features manifested by objectdirected actions.

154

A.T. Phillips, H.M. Wellman / Cognition 98 (2005) 137–155

Acknowledgements This research was supported by NICHD grant 34004 to Wellman. Our thanks to the parents and children who participated, to Jenny Sootsman for her help in all stages of the research Study 1, and to Nicole LaLonde for her help in all stages of Study 2.

References Baillargeon, R. (1986). Representing the existence and the location of hidden objects: Object permanence in 6- and 8-month-old infants. Cognition, 23, 21–41. Baillargeon, R. (1993). The object concept revisited: New directions in the investigation of infants’ physical knowledge. In C. E. Granrud, Visual perception and cognition in infancy. Carnegie-Mellon symposia on cognition (Vol. 23). Hillsdale, NJ: Erlbaum. Baldwin, D. A. (1991). Infants’ contribution to the achievement of joint reference. Child Development, 63, 875–890. Baldwin, D. A., Baird, J. A., Saylor, M. M., & Clark, M. A. (2001). Infants parse dynamic action. Child Development, 72, 708–717. Baron-Cohen, S. Mindblindness. Cambridge, MA: MIT Press. Bartsch, K., & Wellman, H. M. (1995). Children talk about the mind. New York, NY: Oxford University Press. Carpenter, M., Aktar, N., & Tomasello, M. (1998). 14- to 18-month old infants differentially imitate intentional and accidental actions. Infant Behavior and Development, 21, 315–330. Cisbra, G., Gergely, G., Biro, S., Koos, O., & Brockbank, M. (1999). Goal attribution without agency cues: The perception of ‘pure reason’ in infancy. Cognition, 72, 237–267. Gergely, G. (2002). The development of understanding of self and agency. In U. Goswami (Ed.), Blackwell’s handbook of childhood cognitive development (pp. 26 –46). Oxford, NY: Blackwell. Gergely, G., & Csibra, G. (2003). Teleological reasoning in infancy: The naı¨ve theory of rationale action. Trends in Cognitive Science, 7, 287–292. Gergely, G., Nadasdy, Z., Csibra, G., & Biro, S. (1995). Taking the intentional stance at 12 months of age. Cognition, 56, 165–193. Inagaki, K., & Hatano, G. (1993). Young children’s understanding of the mind–body distinction. Child Development, 64, 1534–1549. Johnson, S. C. (2000). The recognition of mentalistic agents in infancy. Trends in Cognitive Sciences, 4, 22–28. Mandler, J. M. (1992). How to build a baby: II, Conceptual primitives. Psychological Review, 99, 587–604. Meltzoff, A. N. (1995). Understanding the intentions of others: Re-enactment of intended acts by 18-month-old children. Developmental Psychology, 31, 838–850. Moore, C., & Corkum, V. (1994). Social understanding at the end of the first year of life. Developmental Review, 14, 349–372. Mumme, D. L., & Fernald, A. (2003). The infant as onlooker: learning from emotional reactions observed in a television scenario. Child Development, 74, 221–237. Perner, J., & Davies, G. (1991). Understanding the mind as an active information processor: do young children have a “copy theory of mind”? Cognition, 39, 51–69. Phillips, A. T., Wellman, H. M., & Spelke, E. S. (2002). Infants’ ability to connect gaze and emotional expression to intentional action. Cognition, 85, 53–78. Premack, D. (1990). The infant’s theory of self-propelled objects. Cognition, 36, 1–16. Repacholi, B. M., & Gopnik, A. (1997). Early reasoning about desires: Evidence from 14- and 18-month olds. Developmental Psychology, 33, 12–21. Rogers, T. T., & Griffin, R. (2002). Goal attribution without goal representation: A PDP approach to infants’ early understanding of intentional actions. Presented at international conference of infant studies, Toronto, Canada.

A.T. Phillips, H.M. Wellman / Cognition 98 (2005) 137–155

155

Schult, C. A., & Wellman, H. M. (1997). Explaining human movements and actions: Children’s understanding of the limits of psychological explanation. Cognition, 62, 291–324. Sodian, B., Schoeppner, B., & Metz, U. (2004). Do infants apply the principle of rational action to human agents? Infant Behavior and Development, 27, 31–41. Spelke, E. S., Breinlinger, K., Macomber, J., & Jacobson, K. (1992). Origins of knowledge. Psychological Review, 99, 605–632. Woodward, A. (1998). Infants selectively encode the goal object of an actor’s reach. Cognition, 69, 1–34. Woodward, A. L. (1999). Infant’s ability to distinguish between purposeful and non-purposeful behaviors. Infant Behavior and Development, 22, 145–160.