1
British Journal of Developmental Psychology (2005), in press q 2005 The British Psychological Society
The British Psychological Society www.bpsjournals.co.uk
The Valley task: Understanding intention from goal-directed motion in typical development and autism Fulvia Castelli* California Institute of Technology, Division of Humanities and Social Sciences, Pasadena, CA 91125, USA A novel paradigm investigates the ability to understand an agent’s intended goal in children with autism (N ¼ 25), typically developing children (N ¼ 46), and adults (N ¼ 16 þ 12) by watching a non-human agent’s kinematic properties alone. Computer animations depict a circle at the bottom of a U-shaped valley rolling up and down its slopes and getting closer to a target resting at the top of either side of the valley. The circle’s persistent motion and improving attempts evoke the attribution of the intention to reach the target, regardless of whether the circle fails or attains its goal. Children with autism are as able as controls to infer an agent’s intended-goal, disregarding its failure to reach the target. In addition, the study showed that the perception of motion is a sufficient but not a necessary cue for very young children and children with autism to attribute intention to an agent, whereas adults consider the persistent motion cue as a sufficient and necessary cue to attribute intention to an agent.
In everyday social interactions people explain and predict the behaviour of others with a facility and success that is remarkable. Such a useful strategy to make sense of own and others’ actions is based on the ability to represent mental states. The terms ‘theory of mind’ (ToM), ‘mindreading’, or ‘mentalizing’ have all been coined to describe the effortless propensity to search for beliefs, desires, emotions, and intentions behind agents’ behaviour. A host of studies have investigated the difficulties of individuals with autism to represent mental states at different cognitive developmental stages: from early ostensive communication behaviours, communicative gestures and referential language, to the subsequent ability to understand the concept of deception or false belief (BaronCohen, Tager-Flusberg, & Cohen, 1993, 2000). It is now well documented that typically developing children understand propositional attitudes between the ages of 3 and 4 years old (Wellman, Cross, & Watson, 2001), whereas children with autism show a developmental delay. To date, few studies on autism have investigated the ability to
* Correspondence should be addressed to Fulvia Castelli, California Institute of Technology, Division of Humanities and Social Sciences, Pasadena, CA, USA (e-mail:
[email protected]). DOI:10.1348/026151005X54209
2
Fulvia Castelli
represent mental states that are more ‘transparent’ (i.e. observable in behaviour) than beliefs and desires. In particular, these studies have adopted paradigms based on selfattribution of intention (Phillips, Baron-Cohen, & Rutter, 1998; Russell & Hill, 2001) and on imitation of people’s intended act (Aldridge et al., 2001; Carpenter, Pennington, & Rogers, 2001). The present study investigates the ability to understand an agent’s intended goal with a novel paradigm, based on the perception of a non-human agent’s kinematic properties alone. Phillips et al. (1998) showed that children with autism were able to report their own intention to hit a target in a ‘shooting game’ when the outcome was successful, but not when it was unsuccessful. Considering the complexity of the procedure of the shooting game, the authors suggested that an executive function impairment might account for this difficulty. Russell and Hill (2001) reported no difference between the autism group and controls in both the shooting game task (Phillips et al., 1998) and a novel ‘drawing’ task where children were asked to report their own intended action when the final outcome was unexpected. However, it is possible that subject’s response in the drawing task was facilitated by the direct question of whether they thought or meant to draw what turned out to be unexpected. The ability to attribute intention to others has been investigated by Aldridge, Stone, Sweeney, & Bower (2000) and Carpenter et al. (2001) using Meltzoff’s (1995) re-enactment paradigm. Subjects watched an adult performing intended but failed actions, such as pulling a dumbbell toy apart. In both studies the young children with autism showed no difficulties in performing all the intended actions. However, a learning effect through repeated observation might account for their correct performance in the Aldridge et al. (2000) study, whereas a difficulty effect might account for the similar but poor performance of both the autism group and controls in the Carpenter et al. study. To date, studies on the ability to understand their own and other’s intentions in children with autism have yielded discordant results. In addition, each of the tasks used in previous studies seems to have some limitations, suggesting that a new approach is needed. The present study adopts a novel paradigm for investigating the ability to attribute intention to others, based on the perception of simple non-human agents’ motion pattern. Recent studies have investigated the ability to attribute mental states in individuals with autism by using kinematic information alone (Abell, Happe´, & Frith, 2000; Bowler & Thommen, 2000; Castelli, Frith, Happe´, & Frith, 2002; Klin, 2000). Bowler and Thommen investigated the on-line production of narrative account of events by showing the original Heider and Simmel (1944) animation (two triangles and a circle moving within and around a rectangle in a variety of interacting and non-interacting movements sequences) to a group of children with autism. Results indicate that they were able to use mental state terms such as ‘think’, ‘know’, and ‘want’ to describe the characters’ interactions. Interestingly, they showed more difficulty describing interactions involving no physical contact (e.g. chasing) than interactions with physical contact (e.g. fighting). In a study by Klin, adolescents and adults with autism were shown segments of the Heider and Simmel movie and asked to provide verbal descriptions of what was happening. The autism group used fewer mental state terms than the control group. Abell et al.’s (2000) study adopted novel animations depicting two interacting triangles whose different types of motions selectively elicited descriptions in terms of mental states, goal-directed actions or non-deliberate actions (i.e. random movement). Before each presentation, subjects were cued with character roles, for example, the two triangles are a mother and a child. Children with autism gave
Understanding intention in autism
3
descriptions of the goal-directed animations (e.g. fighting, chasing) and the random animations (e.g. floating, drifting) that were as accurate as the controls, but they gave less accurate descriptions of the ToM animations (e.g. bluffing, deceiving). The same type of stimuli were adopted by Castelli et al. (2002) in a neuroimaging study with adults with autism, but no character roles were given. Adults with autism used fewer mental states attributions and less accurate descriptions of ToM animations than the controls. However, they described both the goal-directed and random animations as accurately as controls. The type of animations used in all these studies allowed the investigation of the ability of people with autism to decode kinematic stimuli in terms of complex mental states such as beliefs, desires, and intentions. However, in all these studies the ability of mentalizing might either be enhanced or impaired by the use of language in their verbal descriptions of the animations. In addition, although these studies indicate that individuals with autism process visual-kinetic information differently from controls, they do not directly investigate the nature of kinematic stimuli that trigger attribution of intention or goal as opposed to belief and desire. Premack and Premack (1994) suggest that the perceptual triggering inputs for attributing intentionality to an agent are the following: (a) goal-directed motion towards the same single item (goal-directed motion), (b) repeated motion (failing and trying again), and (c) variable motion patterns. Thus, the perception of repeated attempts to overcome failure is a strong cue for the attribution of goals and intentions. However, it is unclear whether, when attributing an intention to an agent, the repeated and variable motion is a stronger cue than the final outcome (end-state) of the agent’s goal-directed motion. The novel paradigm of the present study was designed both to conform to Premack and Premack’s (1994) requirements and to address the question of whether subjects rely more on the outcome of the goal-directed motion or on the repeated motion (failing and trying again). Inspiration for the appropriate type of paradigm to investigate children with autism’s ability to attribute intention to an agent came from Montgomery and Montgomery’s (1999) study with typically developing 3- and 5-year-olds, showing that they were able to attribute to an agent (i.e. a circle) the intention to reach a target (i.e. another circle) beyond an obstacle (i.e. a wall) by perceiving only its repeated jumps against the wall (i.e. goal failed) rather than over the wall (i.e. goal attained.) The present study adopts a new computer animated test – the Valley task – which investigates the perception of intended goal-directed motion with innovative features, such as the type of visual context (i.e. U-shaped valley), the agent’s different motion patterns (i.e. simple and complex), and the responses available to the subjects (i.e. multiple choice). The animated sequences depict a yellow circle that rolls up and down the slopes of a U-shaped valley and, at each attempt, gets closer and closer to one of two circles (blue and red) resting on top of either side of the valley (Fig. 1 shows a still frame from the beginning of the sequence, printed here in black and white). The agent’s persistent motion in one direction and its improving attempts to overcome failure evokes the attribution of the intention to reach the target at the top of the valley, regardless of where the agent comes to rest (the circle stops either at the bottom or midway up the slope of the valley, or reaches a circle resting at the top of the slope). (Examples of each condition can be viewed at: http://www.emotion.caltech.edu/ people/castelli/). The advantage of a U-shaped valley as opposed to a flat baseline with a wall (see paradigms by Csibra, Gergely, Biro, Koos, & Brockbank, 1999; Gergely, Nadasdy, Csibra, & Biro, 1995; Montgomery & Montgomery, 1999) is that the motivation of the
4
Fulvia Castelli
Figure 1. Still frame of the computer animated sequence of the Valley task.
agent’s goal-directed motion is implicit in the naturalistic context (e.g. the agent is stuck at the bottom of the valley and wants to get out by reaching the top of the slope), or at least it does not require an explicit explanation of the motivation for the agent to reach the target. Similarly, the valley context does not require an explicit reference to the fact that the circle trying to overcome gravity is an agent whose movement is internally caused. Another innovative aspect is that the motion trajectory of the circle towards the target is either simple or complex. During the simple trajectory the agent starts off from the bottom and rolls along one slope only. By contrast, during the complex trajectory the agent starts off by jumping upwards at the bottom of the valley, and then rolls up and down along both slopes, as if gaining momentum to get to the top. This condition was introduced because the consistent movement of the agent in one direction only could provide the observer with a salient spatial cue (i.e. pointing towards the target) for a correct attribution of its goal. Another original feature of the task is that subjects can choose the agent’s intended goal from among five locations along the valley, namely, the two circles resting at the tops, the bottom and the two midway locations between top and bottom (see Fig. 1). Finally, the Valley task allows for testing whether subjects place more value on the agent’s repeated goal-directed motion towards a target or its final outcome as necessary cue for attributing intended goal. In fact, the animated sequences depict not only a circle that attains or fails to reach a target after repetitive attempts (i.e. constant direction condition), but also a circle that rolls repeatedly in one direction, and suddenly rolls all the way down and up to the opposite direction, ending next to the circle at the top (i.e. changing direction condition). In the constant direction condition, there is only one correct response: the circle wanted to reach the top regardless of its final outcome. In the changing direction condition, there are two possible correct responses: (a) the circle wanted to reach the top where it was directed before changing direction, or (b) the circle wanted to reach the top of the other slope where eventually it stopped. In the first case, the attribution of intended goal is influenced more by the persistent goal-directed motion (motion-based representation) than by the agent’s final outcome. By contrast, in the second case the attribution of intended goal is influenced more by the agent’s proximity to the target (i.e. outcome-based representation) than by the agent’s motion pattern. In the eventuality that subjects choose any other locations, then it must be inferred that the changing direction condition is too confusing.
Understanding intention in autism
5
In conclusion, the study investigates the ability in children with autism to attribute an agent’s intended goal using a novel non-language-based paradigm, the ‘Valley task’. The paradigm is based on the perception of a circle’s goal-directed and persistent motion that evokes the attribution of the intention to reach the target, regardless of whether the circle fails or attains its goal (constant direction condition). An additional condition (changing direction condition) was added to the task in order to address the general question of whether subjects attribute goal-directed intention to an agent on the basis of its final destination or its persistent direction of motion.
Methods Design A 3 (group) £ 2 (motion trajectory) £ 2 (direction) £ 5 (outcome location) design was used (see Figs 2, 3 for a schematic view of the constant direction condition and the changing direction condition. Subjects view all conditions during a single presentation). The experimental task consists in deciding about the final goal of the moving circle by clicking (with the computer mouse) on one of the five marked locations along the sides of the valley. (See Materials for details.) Subjects A group of 26 children with high-functioning autism or Asperger’s syndrome, a group of 50 typically developing children and teenagers attending mainstream schools, and a group of 18 adults. The children with autism had all previously been diagnosed according to published criteria (DSM-IV, APA, 1994) by a psychologist or psychiatrist with expertise in autism and all were attending special schools for individuals with autism. Their mean chronological age was 12.2 years, with a range between 9.11 and 17.5 years, and their mean verbal age was 9.7 years. The VMA was assessed using the BPVS II test (British picture vocabulary scale, 1997) on 14 subjects, and the VIQ score of the WISC Top: intended goal
Midway : intention failed
Bottom : intention failed
Figure 2. Constant direction condition in a schematic way. The circle at the bottom of the valley rolls towards one of the other circles (e.g. blue circle resting on the right-side platform) three times, and then it stops at one of the following three ‘outcome’ locations: bottom, midway, or top. Subjects can choose to click on one of the five grey platforms (or on the blue and red circles) in response to the question: ‘Where did the yellow circle want to go?’ Only one response is correct for this condition: ‘top’.
6
Fulvia Castelli Top-Changing direction (agent eventually stops here)
Top–Constant direction (agent aims repeatedly here)
Figure 3. Changing direction condition in a schematic way. The circle at the bottom of the valley rolls towards one of the other circles (e.g. circle resting on the right-side platform) three times and suddenly changes direction and reaches the opposite circle (e.g. the circle resting on the left-side platform). Subjects can choose to click on one of the five grey platforms in response to the question: ‘Where did the yellow circle want to go?’ Only two responses are equally correct (top-constant direction and topchanging direction) for this condition. The type of response reflects a different type of intended-goal representation (based on the agent’s final outcome or on the agent’s persistent motion towards the target).
(Wechsler intelligence scale for children, third edition UK, 1992) on the remaining 12 children. The choice between the IQ tests was entirely determined by the time available for testing each child. The typically developing children’s mean chronological age was 9.1 (SD ¼ 3:6) years, with a range between 5.6 and 15.9. Their VMA was assessed with BPVS II test (3 teenagers were not available for this test. However, given the rigorous selective admission criteria of their school, their VMA was assumed to be in line with if not above their chronological age; CA). The autism group and the control group of all children were matched on VMA. All children in both groups passed the following ToM tests: the Sally-Ann Test (Baron-Cohen, Leslie, & Frith, 1985) and the Smarties Test (Perner, Frith, Leslie, & Leekam, 1989). The adults’ chronological age was 31.3 years, ranging from 19 to 40 years. Their VMA was not assessed and the ToM tests were not administered. All subjects passed a ‘wanting test’ aimed at ascertaining whether they were able to understand questions relative to ‘wanting something’ and ‘someone wanting’ in the absence of verbal stimuli (see Materials for details, Table 1). Table 1. Subjects chronological (CA) and verbal mental age (VMA). The mean, standard deviation (in parentheses), and ranges are reported in calendar years and months
Group
CA (yrs) Mean (SD) range
VMA (yrs) Mean (SD) range
12.2 (2.0) 9.11–17.5 9.1 (3.6) 5.6–15.9 31.3 (6.3) 19–40
9.7 (3.0) 5.3–17 9.7 (3.10) 6.3–15.4 –
Autism–Children (N ¼ 26) Control–Children* (N ¼ 50) Control–Adults (N ¼ 18) *VMA for the control group is from N ¼ 47.
Understanding intention in autism
7
Materials The Valley task The animations were created with Micromedia Director4e and presented on a portable computer. (Examples of each condition can be viewed at: http://emotion.caltech.edu/ people/castelli/). All the sequences featured a small yellow circle, rolling up and down a green valley and getting closer to one of two circles (blue and red) resting on top of either side of the valley. There were four possible outcomes for the yellow circle: (1) it reaches the top of one slope of the valley, (2) it rolls down and stops at the bottom of the valley, (3) it rolls down and stops at the midway point between the bottom and the top, and (4) it rolls repetitively towards the top of one slope, then it suddenly changes direction and reaches the top of the opposite side. The term ‘constant direction condition’ refers to any of the first three outcomes. The term ‘changing direction condition’ refers to the fourth outcome. When the circle reaches its final location, it remains still for 2 seconds, then disappears and the question ‘Where did the yellow circle want to go?’ appears on the top of the screen. The image disappears as soon as the child clicks with the mouse on one of the five possible locations, to be replaced by an interval message: ‘Get ready! Click to continue’. Every four trials the message was: ‘Very good! Get ready for level 1. Click to continue’. The same message appears for level 1, 2, 3, and 4. The total duration of each sequence was 13.5 seconds for frames in motion plus 2 seconds for the still frame, after which the experimental question appeared on the screen. There were 16 movie sequences in total (four outcomes repeated both for simple and complex and for the two targets resting on the left and on right side of the valley) presented in quasirandom order across subjects.
The Wanting test Subjects were presented with four different photographs: two of them depicted a standing woman in front of a shelf too high for her, on which there was a teapot and a clock. In one picture she is reaching out for the teapot, in the other for the clock. The experimenter asked: ‘What does she want? The teapot or the clock?’ The other two photographs depicted a woman and a man standing in front of a high shelf with a clock on it. In one picture it is the man reaching out for the object, in the other it is the woman. The experimenter asked: ‘Who wants the clock? The man or the woman?’
Procedure Each subject was tested individually in a quiet room. The animations were presented ‘full screen’ on a 1500 monitor of a portable computer (diagonal width of the monitor ¼ , 38.1 cm) positioned on a table at a suitable distance for each subject. In the training phase the experimenter explained to the children that they were going to do a very simple computer game in which they had to decide where a yellow circle rolling up and down a valley wanted to go. Children were then showed a practice sequence displaying the simple motion with bottom outcome. If the child answered incorrectly, the experimenter showed the practice trial again. No feedback on the performance was given. After the example, the experimental session was administered. Children were told that the game had difficulty levels in order to encourage them to pay attention to all sequences until the end of the task. Although the speed of the circle was the same in all
8
Fulvia Castelli
animations, the experimenter explained that the circle rolls up and down the valley faster and faster so that the ‘player’ had to be very careful about keeping track of its movement. The subject was also encouraged to stop anytime he/she wanted between trials. The whole experiment lasted an average of 15 minutes. Adult subjects were given the following written instruction: ‘You are going to see several animations showing a circle rolling up and down a valley. When the circle stops a question will appear on top of the screen. You have to click on the spot where the circle wanted to go. Thank you for your participation’. Each subject was presented with a total of 16 films displayed in quasi-random order. The order of presentation was counterbalanced across subjects.
Results Constant direction condition The analysis of the score for the constant direction conditions (i.e. stimuli eliciting one correct response) was carried out separately from the score for the changing direction condition (i.e. stimulus eliciting two correct scores). Figure 2 shows the constant condition (Bottom, Midway, Top) schematically. The minimum requirement for a valid performance was answering correctly on the constant directions baseline condition, namely, when the agent attained its goal in the simple motion pattern in at least one out of two trials. Consequently, seven subjects were eliminated from the analysis: one from the autism group, four from the control group of children, and two from the adults. These subjects were thought to either not have understood the test or not been willing to comply with the instructions. Table 2 shows the groups’ correct score split into the agent’s outcome conditions. The top outcome constitutes the baseline condition (target-attained), whereas both the bottom outcome and midway outcome constitute the test conditions (target-failed). Table 2. Group’s correct responses split into failed outcomes and successful outcomes. Mean and standard deviation in parentheses Correct responses When the agent fails the goal and stops at:
Autism (N ¼ 25) Control–Children (N ¼ 46) Control–Adults (N ¼ 16) Total (max ¼ 4)
Bottom
Midway
2.9 (1.6) 2.9 (1.4) 3.4 (1.0) 3.0 (1.4)
2.6 2.6 2.9 2.6
(1.7) (1.6) (1.4) (1.6)
When the agent attains the goal and stops at: Top (baseline) 3.8 3.7 3.4 3.7
(0.5) (0.8) (0.9) (0.7)
Total (max ¼ 12) 9.3 (3.2) 9.3 (2.9) 9.7 (3.0)
A non-parametric group comparison test (Kruskall–Wallis) on the total correct score revealed no group effect (H ¼ :59, p ¼ ns). A nonparametric two-way ANOVA (Friedman test) was carried out on the correct score for each outcome (top, bottom, and midway) revealing a significant effect (x2 ¼ 15:4, p , :001). A series of non-parametric paired comparisons (Wilcoxon tests) were carried out in order to investigate whether
Understanding intention in autism
9
some outcome locations elicited more correct responses than others. Results indicated that the top was the easiest condition (top vs. bottom: z ¼ 4:2, p , :002; top vs. midway: z ¼ 5, p , :002; p values corrected for multiple comparisons). Finally, a Wilcoxon test on the two types of motion trajectory revealed a slight but significant difference (z ¼ 2:1, p , :05), with the complex motion (M ¼ 4:5, SD ¼ 1:7) eliciting fewer correct responses than the dimple motion trajectory (M ¼ 4:8, SD ¼ 1:5). These results did not show any impairment in children with autism in the attribution of intention on the basis of perceived goal-directed motion.
Changing direction condition The changing direction condition was introduced to address the question of how subjects attribute goal-directed intention in the presence of an outcome which is purposefully designed to be seen as either successful or unsuccessful. Figure 3 depicts in a schematic way the changing direction condition. Correct responses are either topconstant direction, where the circle’s repeated attempts were directed before changing direction, and top-changing direction, where the circle eventually stopped. The minimum requirement for a valid performance was to click on one of the two top locations in the simple motion pattern. Consequently, seven subjects were eliminated from the analysis of correct responses: three from the autism group, three from the control group of typically developing children, and one from the adult group. These subjects were thought to either not have understood the test or not been willing to comply with the instructions. A series of non-parametric paired t tests (Wilcoxon) on the performance of the three groups revealed a significant difference only in the group of children with autism: they chose the top-changing location (M ¼ 2:7, SD ¼ 1:4) significantly more often than topconstant location (M ¼ 1:2, SD ¼ 1:4; z ¼ 2:3, p ¼ :02). Typically developing children’s performance was not significantly different: top-changing (M ¼ 2:0, SD ¼ 1:7) vs. topconstant (M ¼ 1:8, SD ¼ 1:2; z ¼ :60, p ¼ ns.), nor the adults’ performance: topchanging (M ¼ 1:3, SD ¼ 1:3) vs. top-constant (M ¼ 2:5, SD ¼ 1:2; z ¼ 1:6, p ¼ ns). Since these results indicated that children with autism’s performance was biased towards top-changing location (they attributed goal-directed intention on the basis of the agent’s final outcome) and their mean score was similar for the smaller group of adults (but indicating a bias towards top-constant direction), a further analysis was carried out after the number of subjects in the adult group was increased. In addition, since the group of typically developing children included children whose age ranged from 5 years to 15 years, it was divided into three age subgroups in order to monitor possible developmental changes from early years to adulthood (see Table 3). The youngest mean CA is 5.2 years (SD ¼ 0:4), ranging from 5.6 to 6.2, VMA 6.01 (SD ¼ 1:0). The young mean CA is 9.5 years (SD ¼ 1:0), ranging from 8 to11.9, VMA 10.01 (SD ¼ 2:06). The teenagers’ mean CA is 14.2 (SD ¼ 4:5), ranging from 13.4 to 15.9 years, VMA 15.09 (SD ¼ 1:0). Since the changing-direction condition was not originally designed to analyse age groups’ performance and between-group difference, the following results are presented as pilot data for future studies. Table 3 shows the mean score and standard deviation of each group’s correct performance (top-constant direction and top-changing direction). A series of non-parametric paired t tests (Wilcoxon) on each of the five groups’ performance revealed significant within-group differences: the location top-changing was chosen significantly more often than top-constant by the autism group (z ¼ 2:3,
10
Fulvia Castelli
Table 3. Changing direction: subjects’ responses regardless of the motion pattern of the agent’s goaldirected motion. Mean (max score ¼ 4) and standard deviation (in parentheses) Subjects’ response type (regardless of motion pattern) Group Autism–children (N ¼ 22) Youngest–control children (N ¼ 15) Young–control children (N ¼ 18) Teenager–control children (N ¼ 10) Adult–control (N ¼ 27)
Top-constant direction
Top-changing direction
1.2 (1.4) 0.7 (0.8) 1.5 (1.6) 2.3 (1.8) 2.7 (1.5)
2.7 (1.4) 3.2 (0.8) 2.4 (1.6) 1.7 (1.8) 1.2 (1.4)
p ¼ :02) and the youngest group (z ¼ 2:9, p ¼ :003). By contrast, the adults responded more often top-constant than top-changing (z ¼ 2:4, p ¼ :015). The type of responses of the teenagers and of the young did not differ (z ¼ 0:5, p ¼ ns., and z ¼ 1:5, p ¼ ns., respectively). Given this change in attitude among the responses given by the three groups, autism, youngest and adult, it is of interest to investigate how the responses differ betweengroups. Mann–Whitney t tests indicated that the adults clearly stand out choosing the top-constant response more often than the youngest children (z ¼ 3:9, p , :0001), the young children (z ¼ 2:4, p ¼ :017), and the children with autism (z ¼ 3:3, p , :001). The teenagers did not differ from the adults (z ¼ 0:6, p ¼ ns.), the young children (z ¼ 1:3, p ¼ ns.), or the children with autism (z ¼ 1:6, p ¼ ns). However, they responded top-constant more often than the youngest children (z ¼ 1:9, p , :05). The analyses on the top-changing responses indicate the opposite attitude: the adults chose this option less frequently than the youngest (z ¼ 3:9, p , :0001), the young (z ¼ 2:5, p ¼ 0:16), and the children with autism (z ¼ 3:3, p , :001). The responses of the teenagers did not differ from the young (z ¼ 1:3, p ¼ ns.), the adults (z ¼ :13, p ¼ ns.), or the youngest (z ¼ 1:8, p ¼ ns). These results indicate a shift of interpretation from early school age to adulthood from an outcome-based to a persistent goal-directed motion-based representation of intended goal. However, given the small sample in each age group, it is not possible to establish whether there are clear developmental changes across children’s age groups and more importantly, whether older children with autism (VMA 9.7 years) – who showed the outcome-based representation as often as the youngest control group (VMA 6.01 years) – show a developmental delay. Future research is clearly needed to further investigate how individuals with autism perceive and represent goal-directed intention.
Discussion The present study investigated the understanding of mental states that are observable in behaviour (transparent mental states, i.e. intention and goal, as opposed to opaque mental states, i.e. beliefs and desires). A new paradigm was created using a computer animated sequence depicting a small circle rolling up and down the slopes of a valley trying to reach one of two circles resting on top of either sides of the valley. Subjects were shown salient visual cues: a constant direct motion of the agent towards a target (failing and trying again), and the agent’s accidental outcome.
Understanding intention in autism
11
The main question of the study was whether children with autism have difficulties in attributing an agent’s intended goal in the presence of its constant goal-directed motion and its unsuccessful outcome. Results showed no impairment in children with autism. Furthermore, they did not show any difficulty when the motion pattern of the agent towards the target consisted of a complex trajectory along both sides of the valley rather than a simple linear trajectory along one side only. The negative finding of the present study is in line with Aldridge et al. (2000), Carpenter et al. (2001), and with Russell and Hill (2001), but it is in contrast with the study by Phillips et al. (1998), which indicated a specific impairment in autism in understanding own intention in the presence of an unsuccessful outcome. Some differences between Phillips et al. and the present study have to be taken into account. First of all, the Phillips et al.’s study investigated the understanding of one’s own intention rather than another’s intention. In addition, the group of children with autism had a lower verbal mental age (CA mean: 13.4 yrs, SD ¼ 3 yrs, VMA: 6.2 yrs, SD ¼ 2.3 yrs) than those in the present study (CA mean: 12.2 yrs, SD ¼ 2 yrs, VMA: 9.7 yrs, SD ¼ 3 yrs). It could be that younger children with autism are less familiar with the possibility that their own intentions, while playing a game, might result in failure. However, Montgomery and Montgomery (1999) have shown that typically developing children as young as 3 years old are able to attribute intention to others by watching an agent’s accidental, failed outcome. An alternative explanation for the contrasting results of the Phillips et al.’s and the present one is that the former used a more demanding task, which was based on distinguishing two discordant representations (intention with expected outcome vs. intention with unexpected outcome), and with no support of salient visual cues indicating the intended goal. By contrast, the present study investigated the ability to represent an agent as acting to bring about a future state of affairs by using simple visual stimuli in motion and a no language-based response. The premise for creating this test was that one of the simplest forms of visual information for judging an agent’s intention comes from observing the agent’s motion trajectory (Heider & Simmel, 1944; Michotte, 1963). Several studies have investigated individuals with autism while watching multiple agents’ complex movement patterns (Abell et al., 2000; Bowler & Thommen, 2000; Castelli et al., 2002; Klin, 2000), showing mentalizing difficulties (indexed with language-based measures). However, these studies did not distinguish between motion cues that evoked the attribution of goal-directed intentions (i.e. transparent mental states) compared with beliefs and desires (i.e. opaque mental states). More specifically, previous studies have adopted paradigms that did not distinguish between repetitive, goal-directed movement and more complex, variable pattern displays. Finally, the discussion turns to the theoretical distinction between different levels in the ability to represent agency suggested by Leslie (1994). According to the tripartite theory of agency, there are two distinct levels for representing an agent beyond its mechanical properties, namely, one for its ‘actional’ properties, and the other for its mental properties. At a ‘lower’ cognitive level, there might be a system concerned with the pursuit of a future state of affairs (goal-directed action) and, at a ‘higher’ cognitive level, there might be a system concerned with states that are ‘about something’, namely, propositional states (e.g. ‘believing that’, ‘hoping that’). Thus, the Valley task taps the lower ability to represent possible or future states of affairs in order to understand the agent’s goal-directed behaviour without necessarily involving the higher process of representing the agent’s mental state.
12
Fulvia Castelli
The present study indicates that the ‘lower’ level of the intentional representation system is unimpaired in individuals with autism, whereas more than a decade of research has shown severe impairment at a ‘higher’ level. If the development of the ability to meta-represent in autism reaches only the lower level without extending to the higher level, then it would be of interest to further explore its consequences in terms of compensatory strategies. It is plausible that a spontaneous strategy for people with autism who are impaired in mindreading is to ‘lower the reading’ at the level of their actions. The shift in the ability to meta-represent from impaired higher level and unimpaired lower level is likely to be graded across individuals. It would be of interest to explore compensatory strategies adopted by individuals with autism within the metarepresentational system in order to identify accurate processing components. This observation is closely associated with the additional condition of the Valley task, the changing direction condition, aiming at exploring the nature of the representation of an agent’s intended goal. In particular, this condition allowed the investigation of the impact of the agent’s motion cue and the agent’s final outcome on the process of intention attribution. As part of a pilot study, the group of children was divided into three age groups and the number of subjects in the adult group was increased. Results indicated a shift from early school age to adulthood in the representation of an agent’s intended goal, from an outcome-based representation to a persistent goal-directed motion-based representation. For children with a verbal mental age of 6 years, the perception of the agent’s persistent motion is not a necessarily relevant cue for goal attribution. The representation of the agent’s persistent motion towards a circle was overridden by the perception of both the outcome and the proximity of the agent to another circle. The autism group showed the same result. The situation was reversed in the case of the adults, who considered the agent’s outcome accidental rather than intentional. The small samples in each group might account for the lack of significant difference between the intermediate age groups (young children and teenagers) and the autism group. Future studies need to investigate the perception of motion and outcome cues in larger samples, with particular attention in matching participants’ IQ. An interesting result from the pilot study is that the group of children with autism with a verbal mental age of 9.7 years responded in the same fashion as the 5- and 6-year-old controls. Future studies will need to assess whether children with autism’s performance reflects a developmental delay. In the meantime, it is possible to speculate on the difference between the performance of both the youngest control children and the children with autism compared with the adults’ performance. The change of direction of the agent’s goal-directed motion seems to be the crucial variable in determining the different interpretation between the children and the adults. It is plausible that ambiguous situations trigger executive function in adults – more specifically, the ability to take into account discordant but equally valid situations before making a decision. By contrast, this type of ability may not be fully developed in very young children and in children with autism, so that the perception of discordant events would trigger a less demanding process of matching the agent’s end-state with the agent’s intentional state. In conclusion, the paradigm based on the perception of an agent’s goal-directed motion was sensitive enough to capture an intact ability to represent an agent’s goaldirected intention, indicating no impairment at a ‘minimal’ level of mentalizing. In addition, the study showed that the perception of motion is a sufficient but not a necessary cue for very young children and children with autism to attribute intention to
Understanding intention in autism
13
an agent, whereas adults consider the persistent motion cue as a sufficient and necessary cue to attribute intention to an agent.
Acknowledgements This research was conducted in partial fulfilment of the requirements for the degree of PhD at University College London. The author was supported by the European TMR Marie Curie Research Training grant No.ERBFMBICT972667. I am indebted to Uta Frith, James Blair, and Francesca Happe´ for their invaluable advice and comments throughout this project.
References Abell, F., Happe´, F., & Frith, U. (2000). Do triangles play tricks? Attribution of mental states to animated shapes in normal and abnormal development. Journal of Cognitive Development, 15, 1–20. Aldridge, M. A., Stone, K. R., Sweeney, M. H., & Bower, T. G. R. (2000). Preverbal children with autism understand the intentions of others. Developmental Science, 3, 294–301. Baron-Cohen, S. (1994). How to build a baby that can read minds: Cognitive mechanisms in mindreading. Cahiers de Psychologie Cognitive/Current Psychology of Cognition, 13, 513–552. Baron-Cohen, S. (1995). Mindblindness: An essay on autism and theory of mind. Cambridge, MA: MIT Press. Baron-Cohen, S., Jolliffe, T., Mortimore, C., & Robertson, M. (1997). Another advanced test of theory of mind: Evidence from very high functioning adults with autism or Asperger syndrome. Journal of Child Psychology and Psychiatry, 38, 813–822. Baron-Cohen, S., Leslie, A. M., & Frith, U. (1985). Does the autistic have a Theory of Mind? Cognition, (21), 37–46. Baron-Cohen, S., Leslie, A. M., & Frith, U. (1986). Mechanical, behavioural and intentional understanding of picture stories in autistic children. British Journal of Developmental Psychology, 4, 113–125. Baron-Cohen, S., Tager-Flusberg, H., & Cohen, D. (1993). Understanding other minds: Perspectives from autism. Oxford: Oxford University Press. Baron-Cohen, S., Tager-Flusberg, H., & Cohen, D. (2000). Understanding other minds: Perspectives from developmental neuroscience. Oxford: Oxford University Press. Bellagamba, F., & Tomasello, M. (1999). Re-enacting intended acts: Comparing 12- and 18-montholds. Infant Behavior and Development, 22(2). Bowler, D. M., & Thommen, E. (2000). Attribution of mechanical and social causality to animated displays by children with autism. Autism, 4, 147–171. Carpenter, M., Pennington, B. F., & Rogers, S. J. (2001). Understanding of others’ intentions in children with autism. Journal of Autism and Developmental Disorders, 31(6), 589–599. Castelli, F., Frith, C., Happe´, F., & Frith, U. (2002). Autism, Asperger syndrome and brain mechanisms for the attribution of mental states to animated shapes. Brain, (125), 1839–1849. Csibra, 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. Frith, U. (2001). Mind blindness and the brain in autism. Neuron, 32, 969–979. Gergely, G., Nadasdy, Z., Csibra, G., & Biro, S. (1995). Taking the intentional stance at twelve months of age. Cognition, 56, 165–193. Happe´, F. (1995). The role of age and verbal ability in the theory of mind task performance of subjects with autism. Child Developement, 66, 843–855. Heider, F., & Simmel, M. (1944). An experimental study of apparent behavior. American Journal of Psychology, 57, 1377–1259.
14
Fulvia Castelli
Hobson, R. P. (1993). Autism and development of mind. Hove, UK: Erlbaum. Klin, A. (2000). Attributing social meaning to ambiguous visual stimuli in higher functioning autism and Asperger syndrome: The Social Attribution Task. Journal Child Psychology and Psychiatry, 41, 831–846. Leslie, A. M. (1987). Pretense and representation: The origins of Theory of Mind. Psychological Review, 94, 412–426. Leslie, A. M. (1994). ToMM, ToBy, and Agency: Core architecture and domain specificity. In L. A. Hirschfeld & S. A. Gelman (Eds.), Mapping the mind: Domain specificity in cognition and culture (pp. 119–148). New York: Cambridge University Press. Leslie, A. M., & Happe´, F. (1989). Autism and ostensive communication: The relevance of metarepresentation. Development and Psychopatology, 1, 205–212. Leslie, A. M., & Thaiss, L. (1992). Domain specificity in conceptual development: Neuropsychological evidence from autism. Cognition, 43, 225–251. Leslie, A., & Roth, D. (1993). What representation teaches us about metarepresentation. In S. Baron-Cohen, H. Tager-Flusberg & D. J. Cohen (Eds.), Understanding other minds: Perspectives from autism (pp. 83–111). Oxford: Oxford University Press. Meltzoff, A. N. (1995). Understanding the intentions of others: Reenactment of intended acts by 18-month-old children. Developmental Psychology, 31(5), 1–16. Michotte, A. (1963). The perception of causality. London: Metheun. Montgomery, D. E., & Montgomery, D. A. (1999). The influence of movement and outcome on young children’s attributions of intention. British Journal of Developmental Psychology, (17), 245–261. Perner, J., Frith, U., Leslie, A. M., & Leekam, S. R. (1989). Exploration of the autistic child’s theory of mind: Knowledge, belief and communication. Child Development, 60(3), 688–700. Phillips, W., Baron-Cohen, S., & Rutter, M. (1998). Understanding intention in normal development and in autism. British Journal of Developmental Psychology, 16, 337–348. Premack, D. (1990). The infant’s theory of self-propelled objects. Cognition, 36, 1–16. Premack, D., & Premack, A. J. (1994). Moral belief: Form versus content. In L. A. Hirschfeld & S. A. Gelman (Eds.), Mapping the mind: Domain specificity in cognition and culture. New York: Cambridge University Press. Rochat, P., Morgan, R., & Carpenter, M. (1997). Young infants’ sensitivity to movement information specifying social causality. Cognitive Development, 12, 441–465. Roth, D., & Leslie, A. M. (1998). Solving belief problems: Toward a task analysis. Cognition, 66, 1–31. Russell, J., & Hill, E. (2001). Understanding intention in autism. Journal of Child Psychology and Psychiatry, 42, 317–328. Schlottmann, A., & Surian, L. (1999). Do 9-month-olds perceive causation at-a-distance? Perception, 28, 1105–1113. Sigman, M., & Capps, L. (1997). Children with autism: A developmental perspective. Cambridge, MA: Harvard University Press. Wellman, H. M., Cross, D., & Watson, J. (2001). Meta-analysis of Theory of Mind development: The truth about false belief. Child Development, 72, 655–684. Wimmer, H., & Perner, J. (1983). Beliefs about beliefs: Representation and the constraining function of wrong beliefs in young children’s understanding of deception. Cognition, 13, 103–128. Received 22 October 2004; revised version received 26 May 2005
