Understanding emotions from standardized facial expressions in autism and normal development

Understanding emotions from standardized facial expressions in autism and normal development autism 2005 SAGE Publications and The National Autistic Society Vol 9(4) 428 449; 056082 1362-3613(200510)9:4 FULVIA CASTELLI California Institute of Technology, USA ABSTRACT The study investigated the recognition of standardized facial expressions of emotion (anger, fear, disgust, happiness, sadness, surprise) at a perceptual level (experiment 1) and at a semantic level (experiments 2 and 3) in children with autism (N = 20) and normally developing children (N = 20). Results revealed that children with autism were as able as controls to recognize all six emotions with different intensity levels, and that they made the same type of errors. These negative findings are discussed in relation to (1) previous data showing specific impairment in autism in recognizing the belief-based expression of surprise, (2) previous data showing specific impairment in autism in recognizing fear, and (3) the convergence of findings that individuals with autism, like patients with amygdala damage, pass a basic emotions recognition test but fail to recognize more complex stimuli involving the perception of faces or part of faces. ADDRESS Correspondence should be addressed to: FULVIA CASTELLI, PhD, California Institute of Technology, HSS 228 77, Pasadena, CA 91125, USA. e-mail: fulvia@hss.caltech.edu KEYWORDS amygdala; autism; emotion; facial expressions; mentalizing Introduction Since Kanner s (1943) original clinical account of children with autism first described their profound lack of affective contact with other people, psychologists have been evaluating the social and affective impairments in autism. The empirical research on affective impairment of children and adults with autism is wide and varied so that it is not surprising that the findings are extremely mixed. Hypotheses of a general affective deficit (Hobson, 1986a; 1986b; Hobson et al., 1988), and a selective emotion recognition deficit (Baron-Cohen et al., 1999; Howard et al., 2000) have been explored. In addition, the theory of mind (ToM) deficit account of autism allowed investigations of selective emotion processing impairment by contrasting recognition tasks that do and do not necessitate the ability 428 www.sagepublications.com DOI: 10.1177/1362361305056082

CASTELLI: UNDERSTANDING EMOTIONS to represent mental states (Baron-Cohen et al., 1993). The present investigations attempt to replicate and extend these findings with children with autism. A general affective deficit in children with autism has been investigated by Hobson and colleagues using cross-modal matching tasks with vocal and facial expressions of emotion (Hobson, 1986a; 1986b; Hobson et al., 1988), and by Tantam et al. (1989) using naming and discriminating tasks with facial expressions. By contrast, Ozonoff et al. (1990) demonstrated in a literature review that children with autism do not show a general deficit in emotion perception if compared with controls of the same language level. Studies based on semi-naturalistic settings reported that children with autism have difficulties in responding to an adult displaying negative emotions (Bacon et al., 1998; Sigman et al., 1992) and positive emotions (Kasari et al., 1993), but they can distinguish adults displays of anger, or distress, from a neutral expression (Corona et al., 1998; Dissanayake et al., 1996). While all these studies measure a lack of social referencing abilities in children with autism, they also index difficulties in mentalizing (e.g. understanding that other people may have different mental states from their own) and executive functions (e.g. switching attention from a salient toy to distant vocal or visual emotional cues). Recently, Grossman et al. s (2000) study indicated that children and adolescents with autism have difficulties in naming facial expressions of emotions that are incorrectly labelled (e.g. happy face labelled either as angry or as orange ) as opposed to expressions associated with correct labels (e.g. happy face labelled as happy ). However, the study also indicated that the autism group s negative performance was significantly correlated with difficulties in executive function tasks. Thus, the question of whether children with autism have general emotion processing impairments remains open to further investigations. The neuroscience of emotion is beginning to be understood and neuroimaging studies have investigated differential neural responses to emotionally relevant material in adults with high-functioning autism/asperger syndrome (HFA/AS). An fmri study by Baron-Cohen et al. (1999) indicated that adults with HFA/AS showed reduced activation of the inferior frontal gyrus and no amygdala activation while understanding complex mental states from people s eyes compared to judging their gender. The authors suggested that the amygdala is a key neural region that is abnormal in autism (Baron-Cohen et al., 2000). The amygdala hypothesis has acquired further support from structural imaging showing enlarged amygdala volume in adults with HFA/AS (Abell et al., 1999; Howard et al., 2000). Howard et al. s (2000) study also reported that the autism group showed a selective impairment in fear recognition. This behavioural finding is in line with the cognitive profile of patients with amygdala lesions (Adolphs et al., 1999; 429

AUTISM 9(4) Fine and Blair, 2000). However, Adolphs et al. (2001) showed that adults with autism performed better than amygdala patients on recognizing simple emotions, including fear, but showed a severe impairment, like the amygdala patients, in a task involving the judgement of trustworthiness and approachability of a person by watching their faces only. It seems, therefore, that Adolphs et al. s (2001) study indicates a dissociation in adults with autism between intact perceptual processing of simple affective signals and impaired retrieval of social knowledge based on facial cues. Thus, it seems that the amygdala hypothesis of autism does not predict a selective impairment in emotion recognition, but rather predicts more general difficulties with complex social judgements, e.g. judgement of facial expressions trustworthiness and judgement of subtle mental states from eye gaze. This conclusion is compatible with the theory of mind deficit hypothesis of autism, i.e. an impairment in the ability to understand and predict behaviour of others on the basis of their mental states such as beliefs and intentions. Baron-Cohen et al. (1993) adopted a methodological approach that assumes that an affective deficit might be secondary to a theory of mind impairment. They tested children with autism on the ability to recognize expressions of happiness and sadness (i.e. reality-based emotions) versus surprise (i.e. belief-based emotion). Results indicated that the children with autism had more difficulties in understanding surprised expressions than happy and sad faces. In a later study, Baron-Cohen et al. (1997a) showed that adults with autism have more difficulties both in recognizing feelings from the eye region than from the whole face, and in understanding complex mental states (e.g. thoughtfulness, interest) than simple emotions. Unlike children with autism, the adults with autism did not show any difficulties in recognizing surprise. However, the recognition task necessitated making a forced choice between two expressions which are almost never confused, namely, surprise and happiness, and never between the two most confusable expressions, namely, surprise and fear (Young et al., 1997). More recently, Buitelaar et al. (1999) showed that children with autism had no difficulties in recognizing either simple or complex emotional expressions (surprise, shame, contempt and disgust were categorized by the authors as complex emotions, and happiness, sadness, anger, and fear as simple emotions). Taken together, it seems that all these studies on autism based on recognition of basic emotions emotions characterized as rapid, failsafe responses to stimuli correlated with basic survival needs (see Ekman, 1992) have adopted different paradigms with different sets of stimuli and different age groups, yielding some inconsistent data. This area is in need of clarification by further experiments. The aim of the present study is to investigate specific emotion recognition processes in children with autism and normally developing children 430

CASTELLI: UNDERSTANDING EMOTIONS by using fine-grained visual stimuli and both perceptual and semantic tasks. The paradigm adopted for the three experiments is based on computergenerated stimuli derived from a standard set of pictures of the six basic emotions, i.e. anger, disgust, fear, happiness, sadness and surprise (Calder et al., 1996; Ekman and Friesen, 1976; Young et al., 1997). According to the theory of mind hypothesis, it is predicted that children with autism fail to recognize the belief-based emotion of surprise as opposed to the reality-based emotions of anger, disgust, fear, happiness and sadness. Unlike Baron-Cohen et al. s (1993) paradigm, which was based on the recognition of the expressions of surprise, happiness and sadness, the present study adopted a broader range of stimuli, including expressions that are more difficult to recognize (i.e. fear, anger and disgust) than happiness and sadness. In addition, the wide range of stimuli allows for monitoring children s performance in relation to the amygdala hypothesis of autism. This hypothesis suggests a correlation between amygdala abnormality and socioaffective impairments. Since the studies reported above (Adolphs et al., 2001; Howard et al., 2000) were contradictory in their findings of face recognition in adults with autism, the additional purpose of the present study is to observe the performance of children with autism in relation to the expression of fear. The study comprises three different experiments. Experiment 1 (matching task) investigates the perceptual ability to discriminate basic emotions. The challenge for the child is determined by the difference in intensity level of the emotions displayed in the stimuli expressions. The task requires a more abstract ability to extract the salient invariance of the six emotional expressions across different intensity degrees rather than matching fixed stereotypical features. The higher intensity of the facial expression is expected to facilitate the matching task across all emotions, whereas the expressions combining equal intensity of two emotions (e.g. fear mixed with surprise) are expected to elicit responses regularly distributed between the two possible emotions (e.g. fear and surprise). However, evidence that normal adults tend to interpret combined expressions in other different ways (e.g. surprise combined with happiness seen more as happiness; sadness combined with disgust seen more as sadness; anger combined with disgust seen more as disgust; fear combined with surprise seen more as fear) has been reported in a study of patients with brain lesions (Calder et al., 1996). It is therefore of interest to investigate whether children with autism show a preferential bias towards only one target emotion within each combination pair. Experiment 2 and 3 (naming task) investigate the semantic ability of children with autism to discriminate emotions from a wide range of facial expressions. Both the absence of a target and the heterogeneity of the test stimuli control for the possibility that children s performance relies on 431

AUTISM 9(4) perceptual matching strategies without a clear understanding of the meaning of the expression. The difference between the second and third studies is that the latter combines the difficulty of the fine-grained stimuli of experiment 1 facial expressions of emotions with different levels with the naming task of experiment 2. In all three experiments it is expected that happiness is the easiest emotion to recognize, and that the expressions of surprise and fear are the most confused emotions, together with anger and disgust (Young et al., 1997). Interestingly, Young et al. s (1997) study on adults categorical perception of facial expressions indicated that the expression of surprise was sometimes identified as fear, and disgust as anger, but when this happened, it was usually because one adult subject consistently did this. It is therefore of interest specifically to investigate whether the error pattern relative to surprise and fear of children with autism differs from that of normally developing children. Experiment 1: discriminating facial expressions of emotions with different intensity levels Design The experiment involves a 2 (group) 6 (emotion type) 3 (intensity type) design. The emotion variable consists of morphed facial expressions of six emotions: anger, disgust, fear, happiness, sadness and surprise. The intensity type consists of three intensity levels for each emotion: 90, 70 and 50 per cent. The task consists of matching each emotion stimulus with one of the six displayed emotion targets (at 100 per cent intensity level). 1 Participants These comprised a group of 20 children resident in a special school for children with autism diagnosed formally with either autism or Asperger syndrome prior to this study by independent clinicians, and a group of 20 normally developing children attending mainstream schools. Children with autism were assessed using the VIQ score of the WISC (Wechsler Intelligence Scale for children, third edition UK, 1992), whereas normally developing children were assessed by the BPVS II test (British Picture Vocabulary Scale, 1997) (Table 1). The choice between the two IQ tests was entirely determined by the time available for testing each child. The test was not used as a matching criterion, but rather to check that the control subjects were at a developmentally normal language level. For matching purposes, the group of children with autism was divided into subgroups according to their verbal mental age (VMA) (6 7 years, 7.1 9 years, 432

CASTELLI: UNDERSTANDING EMOTIONS Table 1 Subjects verbal and performance ability scores, chronological age (CA), and verbal and performance mental ages (MA) Group Score CA (years) MA (years) Mean (SD) Mean (SD) Mean (SD) Autism WISC: 12.3 (2.3) Verbal = 9.2 (2.6) (N = 20) Verbal IQ = 75.2 (16.9) Performance = 10.1 (3.2) Performance IQ = 82.5 (21.3) Control BPVS = 98.0 (18.3) 9.2 (2.4) Verbal = 9.11 (2.7) (N = 20) 9.1 12 years, 12.1 14 years) and matched with the same number of controls of the same CA in each subgroup, assuming that the chronological age of the non-autistic children was roughly equivalent to their VMA. However, the results were analysed relative to the performance of the whole group. Materials The stimuli consisted of laminated cards (6.5 9 cm) representing computer-manipulated photographic quality images of morphed facial expressions of an adult male model with different levels of emotion intensity, obtained by blending two prototype expressions from the Ekman and Friesen (1976) series, e.g. anger happiness, with different proportions (i.e. 90:10, 70:30, 50:50, 30:70, 10:90) (see Calder et al., 1996 for full details of the morphing technique). The target cards represented expressions at 100 per cent intensity level (original prototypes) of six different adult female models. There were four stimuli cards in each emotion at level 90 per cent and 70 per cent, and two stimuli cards for each pair of emotions at 50 per cent of intensity. For each subject there were two sets of 30 stimuli cards each. The target cards were pasted on plastic boxes sized 10 15 3 cm. Procedure Each child was tested individually in a separate room of the school. In the training phase the experimenter showed each target emotion (female model) to the child, asking her to say how she was feeling and to provide an example of the displayed emotion (e.g. Tell me about a time when you were surprised ). If the child showed uncertainty, giving no example, the experimenter provided a standard example (e.g. I was surprised when I opened my birthday present ). Each target emotion was then fixed on an empty box in front of the child, making sure that she/he had a full view of all targets. The display of the six targets was randomly arranged across 433

AUTISM 9(4) subjects. During the practice session the experimenter showed one at a time the expressions of happiness or sadness at 90 per cent intensity level, asking the child to place the card in the box with the similar expression. After practice, the child started the experiment (two sessions with 30 cards each). The cards were given one at a time, and the experimenter kept asking Where does it go? until the routine was established. At the end of the first session, the display of the targets on the boxes was randomly rearranged in order to control for biases due to a preference of a particular position (e.g. central positions versus lateral). At the end of the two sessions, the cards in each box were counted and coded. Results The analysis of the score with emotions at 90 and 70 per cent intensity level was carried out separately from the score with 50 per cent emotions (i.e. stimuli eliciting two correct scores). The error patterns of each group were also analysed to identify consistent mismatches between the expressions of surprise and fear. Analysis of correct performance with emotion stimuli at 90 and 70 per cent levels of intensity Non-parametric analyses were performed on the groups correct matching of each emotion stimulus with its emotion target, with emotions split into higher (90 per cent) and lower (70 per cent) levels of intensity (Table 2). Group-comparison analyses (Mann Whitney tests) revealed no significant group effect on each emotion beyond intensity level (anger, z = 0.68; disgust, z = 0.14; fear, z = 0.12; happiness, z = 0.11; sadness, z = 0.66; surprise, z = 0), no intensity-level effect beyond groups on each emotion at 90 per cent intensity (z scores for anger, disgust, fear, happiness, sadness and surprise were all 0), and no intensity-level effect beyond groups on each emotion at 70 per cent intensity (again z scores for all emotions were 0). A Friedman test of all emotions correct matching revealed a significant effect (chi-square = 32.6, p < 0.0001). Planned comparisons revealed that, as predicted, the correct score for happiness was the highest (F (1) = 14.5, p < 0.001), and the scores for fear and surprise were the lowest (F (1) = 21, p < 0.0001). The results failed to support the ToM prediction of a specific impairment in children with autism in recognizing surprise. In addition, the data revealed no specific difficulty with fear. Analysis of correct performance with emotion at 50 per cent intensity level Table 3 shows the score for the groups correct matches of the emotions at 50 per cent intensity level, and Figure 1 shows the overall performance. There are two correct responses for each of the six emotion 434

Table 2 Overall correct performance in discriminating emotions regardless of intensity level, and groups correct performance in discriminating emotions at 90% and 70% intensity levels (max. score = 4) 435 Emotion Anger Disgust Fear Happiness Sadness Surprise Total Mean 3.1 3.3 2.7 3.8 3.3 2.5 11 SD 1.1 1.1 1.2 0.6 1.6 1.4 11 Intensity level 90% 70% 90% 70% 90% 70% 90% 70% 90% 70% 90% 70% 90% 70% Autism Mean 3.2 3.1 3.3 3.2 3.0 2.6 3.7 3.8 3.3 2.9 2.8 2.4 3.2 3.0 SD 1.2 1.1 1.1 1.2 1.1 1.3 0.9 0.5 1.3 1.5 1.3 1.7 0.8 0.8 Control Mean 2.9 3.1 3.5 3.3 2.5 2.9 4.0 3.8 3.5 3.5 2.4 2.7 3.1 3.2 SD 1.2 0.9 0.9 1.1 1.2 1.3 0.2 0.4 0.9 0.8 1.4 1.2 1.7 0.6 CASTELLI: UNDERSTANDING EMOTIONS

AUTISM 9(4) Table 3 Groups correct performance in discriminating expressions blending together two different emotions (at 50% intensity level) (max. score = 2) Combination of Anger Anger Surprise Surprise Sadness Sadness two emotions Disgust Happiness Happiness Fear Disgust Fear (intensity level 50%) Autism Mean 1.8 1.6 1.7 1.6 2.0 1.7 SD 0.4 0.5 0.5 0.7 0.2 0.7 Control Mean 1.6 1.6 1.9 1.9 2.0 1.7 SD 0.7 0.6 0.4 0.3 0.2 0.6 stimuli which represent the combination of two equally blended emotions. The data analysed consist of two different target emotions for each emotion stimulus at 50 per cent (e.g. the stimulus fear surprise matched with fear, or matched with surprise). A non-parametric analysis (Mann Whitney test) was performed on the groups correct matching and no significant effect was revealed (z = 0.23). In addition, in order to investigate possible group bias towards only one emotion target, a series of Mann Whitney tests were performed on each pair of correct matching (e.g. the autism group matching more frequently surprise fear with surprise rather than with fear). No significant difference in a group preference towards only one of the two correct emotion targets was revealed (anger disgust matched with anger, z = 1.4, or with disgust, z = 0.7; anger happiness 2 anger happiness happiness surprise sadness sadness (max = 2) Mean score 0 disgust anger surprise fear disgust fear 2 PAIRWISE EMOTIONS more frequent choice less frequent choice Figure 1 Overall correct (N = 40) performance with emotions at 50% intensity level: subject s score (max. = 2) for each pair of facial expressions combining two equally intense emotions (50% intensity level) 436

DUARTE ET AL.: STRESS IN MOTHERS matched with anger, z = 1.4, or with happiness, z = 1.6; surprise happiness matched with surprise, z = 0.66, or with happiness, z = 0.16; surprise fear matched with surprise, z = 0.38, or with fear, z = 0.76; sadness disgust matched with sadness, z = 0.74, or with disgust, z = 0.49; sadness fear matched with sadness, z = 0.71, or with fear, z = 0.62). Analysis of groups error pattern in matching emotions regardless of intensity level Since possible confusions between emotions are also likely to occur with morphed expressions matching emotions at 90 and 70 per cent intensity levels, a further analysis was carried out on performance relative to the incorrect matching between emotion stimuli and emotion targets (with the exclusion of expressions morphed at 50 per cent) regardless of intensity level. Non-parametric group-comparison tests (Mann Whitney) were carried out separately on each type of error made with each emotion stimulus. The results indicated no differential error pattern in the two groups. Non-parametric paired comparisons (Wilcoxon tests) indicated that the most frequent errors were due to confusion between fear and surprise (Figure 2). 2 Experiment 2: naming facial expression of emotions with natural intensity Design The experiment involves a 2 (group) 7 (expression type) design. The expression type independent variable consists of the six basic emotions and one neutral expression. The experimental task consists of naming each single facial expression with one emotion label, or the neutral expression with a non-emotion label. The dependent variable is the number of correct responses. 3 Subjects The same groups of participants were tested as in experiment 1. Materials The stimuli consisted of laminated cards (9 11.5 cm) representing photographic quality images of facial expressions derived from Ekman and Friesen (1976). The pictures represented the facial expressions of 10 different models (four adult males and six adult females) each displaying the emotions with natural intensity (i.e. 100 per cent intensity level). The cards were arranged randomly into two decks of 35 cards, each composed of pictures of seven expressions displayed by five models (three females and two males). 437

AUTISM 9(4) mean score (max = 6) 8 6 4 2 0 2 4 6 8 anger disgust fear happiness sadness surprise correct matching = upper portion of the bar incorrect matching = lower portion of then bar anger disgust fear happiness sadness surprise Figure 2 Overall performance in matching each emotion regardless of intensity level (N = 40) Procedure Experiment 2 was always carried out after experiment 1, on the same day. Subjects were asked to name each expression (e.g. How is she feeling? ), and were informed that there was an additional expression (i.e. neutral) that showed no particular emotion. If the child provided a different but correct label for the emotion target (e.g. smiling instead of happy ; crying or upset for sad ; sick instead of disgusted; annoyed or cross instead of angry ) the experimenter accepted them as substitutes, and recorded them for each subject. Definitions pertaining to non-basic emotional states were classified as unclear (e.g. grumpy, disappointed, bored, confused ). The pre-naming session was followed by the session during which the child was presented with one card after the other and asked to say how the person in the picture was feeling until routine was established. Each answer was recorded and coded as correct, incorrect or unclear. Results An initial analysis of the correct naming score was carried out, followed by an additional analysis of the erroneous labelling performance, with the aim of exploring groups consistent errors across emotions. 438

CASTELLI: UNDERSTANDING EMOTIONS Analysis of correct naming of emotions Group-comparison analyses (Mann Whitney tests) were performed on the groups correct naming of each expression stimulus with its emotion label (Table 4). Results revealed no significant group effect (anger, z = 0.45; disgust, z = 1.4; fear, z = 0.2; happiness, z = 0.3; sadness, z = 2.3; surprise, z = 0.84; neutral, z = 0.45). A Friedman test on all emotions correct naming score revealed a significant effect (chi-square = 89.8, p < 0.0001). As predicted, all children found the expression of happiness the easiest emotion to name (planned comparison, F (1) = 72, p > 0.0001). Figure 3 shows the correct score for each emotion in order of difficulty: a series of Wilcoxon tests revealed that children named surprise correctly more often than fear (z = 3, p < 0.003) and named correctly equally often fear and disgust (z = 1.8, n.s.). They also named correctly equally often surprise and anger (z = 0.5, n.s.), neutral Table 4 Overall correct score and groups correct score for naming neutral and emotional expressions (max. score = 10) Target label Neutral Anger Disgust Fear Happiness Sadness Surprise Total Autism Mean 7.8 8.0 5.7 5.9 9.9 7.7 7.6 7.5 SD 2.4 1.5 3.3 2.4 0.4 1.9 2.6 2.6 Control Mean 6.7 7.7 4.4 6.0 9.9 6.2 7.4 6.9 SD 3.6 1.3 2.1 2.8 0.3 2.4 2.0 2.8 Total Mean 7.2 7.8 5.0 5.9 9.9 6.9 7.5 // SD 3.1 1.4 2.8 2.6 0.3 2.3 2.3 // mean score (max = 10) 12 10 8 6 4 2 0 9.9 7.8 7.5 7.2 6.9 5.9 5 happiness anger surprise neutral 1 sadness fear disgust E m o t i o n Figure 3 (N = 40) Overall correct score for naming each emotion in order of difficulty 439

AUTISM 9(4) and sadness (z = 0.5, n.s.), and neutral and anger (z = 1.2, n.s.). They found it easier to name anger than sadness (z = 1.9, p < 0.05). This experiment did not indicate any particular verbal difficulty of children with autism with labelling the emotional expression of either surprise or fear. Analysis of error patterns in naming emotions Non-parametric group-comparison tests (Mann Whitney) were carried out separately on each type of error made with each expression stimulus. In line with the results of the matching task of experiment 1, there were no differences between the two groups in the naming task. A series of non-parametric paired comparisons (Wilcoxon test) showed that the most frequent mistakes were to say that the expression of fear was a surprised face, and that the expression of disgust was an angry face. 4 This result is in line with Young et al. s (1997) findings with healthy adults who found the pairs fear surprise and disgust anger equally confusing. Experiment 3: naming facial expressions of emotions with different intensity levels The aim of this third part of the study was to compare the ability of children with autism and controls to name emotions using more difficult stimuli than in the previous experiment. Design The experiment involves a 2 (group) 6 (emotion type) 2 (intensity type) design. The emotion type consists of the same six basic emotions as the discriminating task of experiment 1. The intensity type consists in only two emotion levels of intensity (90 per cent and 70 per cent). The experimental task consists in naming each emotional expression with one of the six basic emotions labels. Subjects Participants of this study were the same as in experiments 1 and 2, except that the number of children with autism was reduced to 19. Materials The stimulus cards were the same as that for the matching task (experiment 1), with the exclusion of the facial expressions that combined emotions at 50 per cent intensity. The cards were assembled in two blocks of 24 cards (two stimuli at each level for six emotions). The presentation order of the two blocks was counterbalanced across subjects. 440

CASTELLI: UNDERSTANDING EMOTIONS Procedure Children performed this experiment immediately after the naming task of experiment 2 and therefore there was no need to repeat the pre-labelling procedure. Children were invited to take a pause between the two experiments and then were simply asked to do the same as in the previous naming task. Each answer was recorded and coded as correct, incorrect or unclear. Results As in the two previous experiments, the data were analysed relative to the groups correct and incorrect performance. Analysis of correct naming emotions with different levels of intensity Group-comparison analyses (Mann Whitney tests) were performed on the groups correct naming of each expression stimulus with its emotion label with emotions split into higher (90 per cent) and lower (70 per cent) levels of intensity (Table 5). Results revealed no significant group effect (anger, z = 0.5; disgust, z = 0.5; fear, z = 0; happiness, z = 0.1; sadness, z = 0.6; surprise, z = 0.7). A Friedman test of all emotions correct naming (combining the two levels of intensity) revealed a significant effect (chi-square = 37.8, p < 0.0001). Figure 4 shows the correct label for each emotion expression in order of difficulty, regardless of intensity level. As predicted, all children performed at ceiling in naming the expression of happiness (happiness compared to all other emotions, planned comparison, F (1) = 25.5, p < 0.0001). Again, as in the previous naming task, children with autism did not perform differently to the controls, indicating that neither the expression of surprise nor the expression of fear constitutes a verbal obstacle at any level of intensity for both groups. Analysis of errors pattern in naming emotions with different levels of intensity Non-parametric group-comparison tests (Mann Whitney) were carried out separately on each type of naming errors with each emotion stimulus. The results indicated no differential errors patterns in the two groups. Additional series of non-parametric analyses (paired comparison, Wilcoxon tests) were carried out in order to investigate the type of errors occurring most frequently in the overall sample. The analyses revealed that disgust was commonly mistaken and named either unclearly or as sad or as angry. 5 As in the previous tasks, children also made significantly more errors of saying that fear was a surprised face and surprise was a fearful face. 6 441

AUTISM 9(4) 442 Table 5 Overall sample and groups correct score for naming emotions with different levels of intensity (max. score = 4) Emotion Anger Disgust Fear Happiness Sadness Surprise Total Mean (max. = 8) 6.6 4.7 5.4 7.9 6.6 6.2 11 SD 1.8 3.4 2.7 0.2 2.6 2.4 11 Intensity level 90% 70% 90% 70% 90% 70% 90% 70% 90% 70% 90% 70% 90% 70% Autism Mean (max. = 4) 3.5 3.3 2.3 2.1 2.7 2.7 4.0 4.0 3.2 3.0 3.2 3.0 3.1 3.0 SD 0.9 0.9 1.8 1.8 1.6 1.5 0.0 0.2 1.6 1.5 1.2 1.1 1.4 1.4 Control Mean (max. = 4) 3.4 3.2 2.6 2.5 2.8 2.5 4.0 4.0 3.5 3.6 3.3 2.9 3.3 3.1 SD 0.9 1.0 1.8 1.6 1.5 1.5 0.0 0.2 1.1 1.0 1.5 1.4 1.3 1.3

CASTELLI: UNDERSTANDING EMOTIONS mean score (max = 8) 8 7 6 5 4 3 2 1 0 7.9 6.6 6.6 3.1 5.4 4.7 Happiness Anger Sadness Surprise Fear Disgust Figure 4 Overall correct score for naming each emotion in order of difficulty regardless of intensity level (N = 40) Note: Nonparametric paired comparisons (Wilcoxon tests, p < 0.0025 corrected for multiple comparisons) revealed that the scores for surprise and fear did not differ significantly (z = 1.5, p = not sig.) as well as fear and disgust (z = 1, p = not sig.). The equal scores for anger and sadness were both higher than fear (z = 2.3, p < 0.05; z = 2.7, p < 0.01, respectively) and disgust (z = 3, p < 0.01; z = 2.4, p < 0.05; respectively). Finally, the comparison between the scores of surprise and disgust revealed the latter as the most difficult expression to name (z = 2.1, p < 0.05). Discussion The present study sought to investigate both the perceptual and semantic abilities of children with autism and normally developing children in recognizing basic emotional states of others through their facial expressions. The results revealed that children with autism were as able as controls to recognize all six basic emotions from facial expressions. This was shown not only when they were required to match pictures of emotional expressions with different intensity levels (i.e. morphed expressions with 90, 70 and 50 per cent intensity), but also when they were asked to provide a label for emotional expressions with natural intensity (prototype expressions, with 100 per cent intensity level). The findings of the present study contrast with at least two previous studies indicating specific emotion recognition impairment in autism. Baron-Cohen et al. (1993) found a specific impairment in autism in recognizing the belief-based expression of surprise as opposed to the realitybased expressions of happiness and sadness. Howard et al. s (2000) study showed a specific impairment in recognizing the expression of fear. Some differences between the Baron-Cohen et al. research and the present study have to be taken into account. First, in the Baron-Cohen study children with autism had a considerably lower verbal mental age (CA mean 12.6 years,vma 5.3 years) than those in the present study (CA mean 12.3 years, VMA 9.2 years), and thus may have been less familiar with surprise than happiness and sadness. In fact, typically developing 5-year-olds find 443

AUTISM 9(4) surprise the most difficult emotion to recognize, tending to confuse it with happiness while at the same time associating it with the semantic category of feeling bad (Kestenbaum, 1992). Second, the stimuli used by Baron- Cohen et al. were also different: they used drawings and pictures of individuals showing facial expressions, whereas the present study adopted Ekman and Friesen s (1976) standardized series of facial affect. It is possible that the standardized expression of surprise is easier to recognize than drawings and pictures. Third, the Baron-Cohen study focused only on a limited range of emotions, i.e. happiness, sadness and surprise, rather than all six basic emotions including fear, which has been found highly confusable with surprise. It might be that an effect of featural complexity accounts for the Baron-Cohen result. In fact, the expression of surprise involves coding both eyes and mouth whereas happy and sad expressions can be coded simply by reference to the mouth alone. The wide range of expressions investigated in the present study might have contributed to eliminate this effect. Another point of discussion concerns the theoretical distinction between belief-based emotions and reality-based emotions upon which the Baron-Cohen et al. study and the present study based their prediction. The expression of surprise can be primarily characterized as a basic emotion, in the sense of a rapid reaction response to certain events. It can also be considered an approach emotion (Davidson, 1992), which is associated with a call for further information. In the latter case, it would be of interest to explore the ability of individuals with autism to associate the expression of surprise with an appropriate response behaviour. A final issue regards the clinical samples used in both studies, which were taken from children based in specialized schools. Given the heterogeneity of individuals diagnosed with high-functioning autism/asperger syndrome it is an open question how far it is possible to generalize from these results to individuals with autism more generally. Future research on emotion recognition abilities in autism needs to employ much larger samples of individuals, with diagnosis ascertained using standardized instruments. One purpose of the present study was to assess the performance of children with autism with regard to the hypothesis that amygdala abnormality could contribute to social and emotional processing deficit in autism (Baron-Cohen et al., 1999; 2000). In particular, Howard et al. s (2000) study showed concomitant evidence of amygdala abnormality and selective fear recognition impairment in high-functioning individuals with autism. However, children in the present study showed no impairment in fear recognition. This finding is unlikely to be attributable to the type of test materials used since these have been shown to be sensitive enough to reveal 444

CASTELLI: UNDERSTANDING EMOTIONS selective emotion impairments in different clinical populations other than individuals with autism (Adolphs et al., 1999; Blair and Coles, 2000; Calder et al., 1996; Stevens et al., 2001). Again it is possible that diagnostic heterogeneity could explain the differing results in this and the Howard et al. study. A replication of these findings, along with a description of each subject s pattern of performance, is clearly needed. A convergence of findings also show that individuals with autism, as well as patients with amygdala damage, can pass tests of basic emotions recognition but fail to recognize more complex stimuli, i.e. judging trustworthiness and approachability of people from their faces (Adolphs et al., 2001), judging mental states from people s eye gaze (Baron-Cohen et al., 1997b) or identifying emotions by watching people s face associated with inappropriate labels (Grossman et al., 2000). Interestingly, both the trustworthiness and approachability task and the eye task rely on the ability to understand the mental states of others, and to predict their behaviour on the basis of their appearance. It is likely, therefore, that a mentalizing deficit accounts for these results. In conclusion, both the role of the amygdala within the distributed neural system involving mentalizing (Castelli et al. 2000; 2002; Fletcher et al., 1995; Gallagher et al., 2000) and the mentalizing ability of patients with amygdala damage need to be further investigated. Although the theory of mind hypothesis was not supported by the present findings on basic emotion recognition, other studies have found impairments on tasks that tap mentalizing processing. It is plausible that individuals with autism used compensatory strategies to bypass their basic deficit in emotion recognition. Everyday exposure to such basic stimuli is a constant source of information for learning the association of facial expressions and feelings or needs. It is also common for adults to provide reinforcement cues to children in order to decode emotional signals (e.g. a parent pretending to be crying because she wants the child to understand he did something upsetting). On the other hand, it is possible that individuals with autism have no impairment in recognizing those emotion expressions that have evolved with adaptive functions, but have difficulties in linking the perceptual level of emotion recognition with the higher level of understanding the social meaning of different expressions. In this sense, it is important that research on basic emotion recognition should be oriented towards investigating very young children, and possibly with stimuli that bypass compensatory strategies by controlling for both executive function deficit and mentalizing deficit, which are theoretically distinct from the process to be investigated. In conclusion, a final note relative to children s performance in both groups. The results showed that, consistent with both cross-cultural data 445

AUTISM 9(4) (Ekman, 1992) and adult data (Young et al., 1997), all children found both surprise and fear the two most difficult expressions to discriminate at a perceptual level. The close relationship between these two expressions was noted early by Darwin (1872/1998) who considered these emotions as part of the same continuum, pointing out that fear is often preceded by and sometimes mixed with surprise. Data from children s performance with the expressions combining 50 per cent level of surprise and fear showed no particular preferences towards one expression or the other, indicating that the two expressions display ambiguous/unambiguous features to the same extent. Confusion cannot be attributed to the ambiguity of the 50 per cent stimuli in general, since when children had to decide to match the 50 per cent expression of surprise happiness, they chose more often the expression of happiness, as expected on the basis of the categorical perception study by Etcoff and Magee (1992) and on the results from Calder et al. s (1996) study. Overall, the close resemblance between surprise and fear confused the children with autism no more than the controls. Notes 1 The dependent variable is the number of correct matches of emotion stimuli with the emotion targets placed on top of the boxes. The types of errors are defined henceforth as name of emotion stimuli in name of emotion target (e.g. fear in surprise means that the card representing the expression of fear has been incorrectly matched with surprise). Each emotion stimulus at 90 and 70 per cent intensity level has one correct emotion target to be matched with, whereas each emotion stimulus with 50 per cent intensity level, which represents the combination of two different emotions, has two equally correct emotion targets. 2 Fear in surprise (mean = 1.5, SD = 1.6) versus fear in sadness, z = 2.7, p = 0.006; versus fear in happiness, z = 3.9, p < 0.0025; versus fear in disgust, z = 3.6, p < 0.0025; versus fear in anger, z = 3.5, p < 0.0025. Surprise in fear (mean = 2.2, SD = 2.4) versus surprise in anger, z = 3.9, p < 0.0025; versus surprise in disgust, z = 4.0, p < 0.0025; versus surprise in happiness, z = 4.2, p < 0.0025; versus surprise in sadness, z = 3.7, p < 0.0025 (all p values corrected for multiple comparisons). 3 Similarly to experiment 1, the stimulus cards that subjects have to name are defined as expression stimuli, whereas the labels they provide are defined as expression label. The types of error are defined henceforth as stimulus labelled as target (e.g. fear as surprised indicates that the expression of fear is labelled incorrectly as surprised). 4 Fear as surprised (mean = 2.5, SD = 1.9) versus fear as neutral, z = 5, p < 0.0001; versus fear as angry, z = 3.8, p = 0.0002; versus fear as disgusted, z = 5, p < 0.0001; versus fear as happy, z = 4.3, p < 0.0001; versus fear as sad, z = 4.7, p < 0.0001; versus fear as unclear, z = 5, p < 0.0001. In addition, the other most frequent error was to say that the expression of disgust was an angry face, i.e. disgust as angry (mean = 3.4; SD = 2) versus disgust as neutral, z = 5.2, p < 0.0001; versus disgust as fearful, z = 5.2, p < 0.0001; versus disgust as happy, z = 5.2, p < 0.0001; versus disgust as sad, z = 5.2, p < 0.0001; versus disgust as 446

CASTELLI: UNDERSTANDING EMOTIONS surprised, z = 5, p < 0.0001; versus disgust as unclear, z = 5.2, p < 0.0001. (Results reported with p value corrected for multiple comparisons, p < 0.002.) 5 Disgust as angry = disgust as unclear, z = 0.06, p = n.s.; disgust as sad = disgust as angry, z = 1.6, p = n.s.; disgust as sad = disgust as unclear, z = 1.5, p = n.s. (Results reported corrected for multiple comparisons, p < 0.0025.) 6 Fear as surprised versus fear as angry, z = 3.3, p < 0.0025; fear as disgusted, z = 3.4, p < 0.0025; fear as happy, z = 3.3, p < 0.0025; fear as sad, z = 2.8, p < 0.0025; fear as unclear, z = 3.6, p < 0.0025. Surprise named as a fearful face versus surprise as angry, z = 3.2, p < 0.0025; surprise as disgusted, z = 3.1, p < 0.0025; surprise as happy, z = 2.2, p < 0.0025; surprise as sad, z = 3.7, p < 0.0025; surprise as unclear, z = 2.7, p < 0.0025. (Results reported corrected for multiple comparisons, p < 0.0025.) References ABELL, F., KRAMS, M., ASHBURNER, J., PASSINGHAM, R., FRISTON, K., FRACKOWIAK, R., HAPPÉ, F., FRITH, C. & FRITH, U. (1999) The Neuroanatomy of Autism: A Voxel-Based Whole Brain Analysis of Structural Scans, Neuroreport 10: 1647 51. ADOLPHS, R., TRANEL, D., HAMMANN, S., YOUNG, A. W., CALDER, A. J., PHELPS, E. A., ANDERSON, A., LEE, G. P. & DAMASIO, A. R. (1999) Recognition of Facial Emotion in Nine Individuals with Bilateral Amygdala Damage, Neuropsychologia 37: 1111 17. ADOLPHS, R., SEARS, L. & PIVEN, J. (2001) Abnormal Processing of Social Information from Faces in Autism, Journal of Cognitive Neuroscience 13: 232 40. BACON, A.L., FEIN, D., MORRIS, R., WATERHOUSE, L. & ALLEN, D. (1998) The Responses of Autistic Children to the Distress of Others, Journal of Autism and Developmental Disorders 28: 129 42. BARON- COHEN, S., SPITZ, A. & CROSS, P. (1993) Do Children with Autism Recognize Surprise? A Research Note, Cognition and Emotion 7: 507 16. BARON- COHEN, S., JOLLIFFE, T., MORTIMORE, C. & ROBERTSON, M. (1997a) 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 22. BARON- COHEN, S., WHEELWRIGHT, S. & JOLLIFFE, T. (1997b) Is there a Language of the Eyes? Evidence from Normal Adults and Adults with Autism or Asperger Syndrome, Visual Cognition 4: 311 31. BARON- COHEN, S., RING, H.A., WHEELWRIGHT, S., BULLMORE, E.T., BRAMMER, M.J., SIMMONS, A. & WILLIAMS, S.C. (1999) Social Intelligence in the Normal and Autistic Brain: An fmri Study, European Journal of Neuroscience 11: 1891 8. BARON- COHEN, S., RING, H.A., BULLMORE, E.T., WHEELWRIGHT, S., ASHWIN, C. & WILLIAMS, S.C. R. (2000) The Amygdala Theory of Autism, Neuroscience and Behavioual Reviews 24: 355 64. BLAIR, R.J. & COLES, M. (2000) Expression Recognition and Behavioural Problems in Early Adolescence, Cognitive Development 15: 421 34. BUITELAAR, J. K., VAN DER WEES, M., SWABB- BARNEVELD, H. & VAN DER GAAG, R.J. (1999) Theory of Mind and Emotion-Recognition Functioning in Autistic Spectrum Disorders and in Psychiatric Control and Normal Children, Development and Psychopathology 11: 39 58. 447

AUTISM 9(4) CALDER, A.J., YOUNG, A.W., ROWLAND, D. & PERRETT, D. I. (1996) Facial Emotion Recognition after Bilateral Amygdala Damage: Differentially Severe Impairment of Fear, Cognitive Neuropsychology 13: 699 745. CASTELLI, F., HAPPÉ, F., FRITH, U. & FRITH, C. (2000) Movement and Mind: A Functional Imaging Study of Perception and Interpretation of Complex Intentional Movement Patterns, Neuroimage 12: 314 25. CASTELLI, F., FRITH, C., HAPPÉ, F. & FRITH, U. (2002) Autism, Asperger Syndrome and Brain Mechanisms for the Attribution of Mental States to Animated Shapes, Brain 125 (8): 1839 49. CORONA, R., DISSANAYAKE, C., ARBELLE, S., WELLINGTON, P. & SIGMAN, M. (1998) Is Affect Aversive to Young Children with Autism? Behavioral and Cardiac Responses to Experimenter Distress, Child Development 69: 1494 502. DARWIN, C. (1872/1998) The Expression of the Emotions of Man and Animals. London: Harper Collins. DAVIDSON, R.J. (1992) Prolegomenon to the Structure of Emotion, Cognition and Emotion 6: 245 68. DISSANAYAKE, C., SIGMAN, M. & KASARI, C. (1996) Long-Term Stability of Individual Differences in the Emotional Responsiveness of Children with Autism, Journal of Child Psychology and Psychiatry 37: 461 7. EKMAN, P. (1992) Are There Basic Emotions?, Psychological Review 99: 550 3. EKMAN, P. & FRIESEN, W. V. (1976) Pictures of Facial Affect. Palo Alto, CA: Consulting Psychologists Press. ETCOFF, N. L. & MAGEE, J. J. (1992) Categorical Perception of Facial Expressions, Cognition 44: 227 40. FINE, C. & BLAIR, R.J. R. (2000) Mini Review: The Cognitive and Emotional Effects of Amygdala Damage, Neurocase 6: 435 50. FLETCHER, P. C., HAPPÉ, F., FRITH, U., BAKER, S.C., DOLAN, R.J., FRACKOWIAK, R.S. & FRITH, C. D. (1995) Other Minds in the Brain: A Functional Imaging Study of Theory of Mind in Story Comprehension, Cognition 57: 109 28. GALLAGHER, H.L., HAPPÉ, F., BRUNSWICK, N., FLETCHER, P. C., FRITH, U. & FRITH, C. D. (2000) Reading the Mind in Cartoons and Stories: An fmri Study of Theory of Mind in Verbal and Nonverbal Tasks, Neuropsychologia 38: 11 21. GROSSMAN, J. B., KLIN, A., CARTER, A.S. & VOLKMAR, V. F. (2000) Verbal Bias in Recognition of Facial Emotions in Children with Asperger Syndrome, Journal of Child Psychology and Psychiatry 41: 369 79. HOBSON, R. P. (1986a) The Autistic Child s Appraisal of Expressions of Emotion, Journal of Child Psychology and Psychiatry 27: 321 42. HOBSON, R. P. (1986b) The Autistic Child s Appraisal of Expressions of Emotion, Journal of Child Psychology and Psychiatry 27: 671 80. HOBSON, R.P., OUSTON, J. & LEE, A. (1988) What s in a Face? The Case of Autism, British Journal of Psychology 79: 441 53. HOWARD, M.A., COWELL, P. E., BOUCHER, J., BROKS, P., MAYES, A., FARRANT, A. & ROBERTS, N. (2000) Convergent Neuroanatomical and Behavioural Evidence of an Amygdala Hypothesis of Autism, Neuroreport 11: 2931 5. K A N N E R, L. (1943) Autistic Disturbances of Affective Contact, Nervous Child 2: 217 50. KASARI, C., SIGMAN, M.D., BAUMGARTNER, P. & STIPEK, D. J. (1993) Pride and Mastery in Children with Autism, Journal of Child Psychology and Psychiatry 34: 353 62. 448