Autistic children’s language imitation shows reduced sensitivity to ostracism

Autism is characterised by deficits in social communication and social interaction (DSM-5; American Psychiatric Association, 2013), which can manifest in conversational language (Ying Sng et al., 2018). In typical dialogue, linguistic imitation between conversational partners—or alignment—is a common feature (cf. Pickering & Garrod, 2004), and higher levels of alignment are associated with effective, rewarding exchanges between speakers (Fusaroli et al., 2012; Putman & Street, 1984). Although alignment has been considered primarily as arising from automatic priming and cue-based memory (Horton & Gerrig, 2005; Pickering & Garrod, 2004) and/or audience design mechanisms (Branigan et al., 2011; Brennan & Clark, 1996), speakers also align for social-affective reasons (Giles et al., 1991). Since autism is associated with social-affective difficulties, including reduced social orientation and affiliative behaviour (Chevallier, Grèzes, et al., 2012; Chevallier, Kohls, et al., 2012; Klin, 1991), we might expect that social-affective factors would not influence autistic people’s alignment to a typical extent. Here we examine whether verbal autistic children align atypically in response to an experience of ostracism, in ways that could undermine their conversational ability.

Communication and interaction deficits in autism are commonly viewed as resulting from impaired understanding of others’ thoughts, beliefs, and intentions, or ‘theory of mind’ (Baron-Cohen et al., 1985). This impairment involves cognitive deficits in meta-representational capacity (Baron-Cohen, 1988), as well as problems with social-affective relatedness (Dawson & Lewy, 1989; Mundy & Sigman, 1989). Hobson (1989, 1993) has argued that autistic individuals lack the capacity to recognise and respond to the emotional states of others, which limits their development of interpersonal connections and thus their social understanding. In more recent work, conversational difficulties in autism have been directly linked with the failure to identify with a social partner on an affective level. For example, autistic children who show signs of reduced affiliative behaviour (e.g. smiling)—relative to non-autistic children—are less able to infer their interlocutor’s intended meanings (Hobson et al., 2012), and tend to engage in less co-ordinated conversation (García-Pérez et al., 2007). Furthermore, there is evidence that autistic children’s emotional connectedness with a partner correlates with their use of first person plural pronouns (Hobson et al., 2010), raising the possibility that the affective deficits associated with autism might manifest in language. However, we know of no work that examines how affective factors influence autistic children’s alignment in conversation. Since alignment contributes to effective, rewarding communication, this issue is important for understanding how and under what circumstances conversational difficulties might arise in autism.

We base our investigation on the ‘perception-behaviour’ link identified in studies of non-linguistic imitation, which have revealed a bidirectional relationship between imitation and affiliative behaviour in the typical population (Dijksterhuis & Bargh, 2001). This link is observable in very young typically-developing children’s behaviour: For example, being mimicked by an experimenter promotes a pro-social orientation towards others in eighteen-month-olds (Carpenter et al., 2013). In adulthood, people experience increased liking of partners who mimic their posture and movements (Chartrand & Bargh, 1999), and tend to mimic a partner they like more than one they do not (Stel et al., 2010). These effects extend to conversational alignment: speakers who converge on a partner’s vocabulary range are evaluated more favorably than those who do not (Bradac et al., 1988), and such positive affect may generate tangible benefits for the mimicker (van Baaren et al., 2003).

Further evidence for a link between imitation and affiliation comes from studies which have experimentally manipulated social exclusion (specifically, ostracism) and compared the behaviour of participants who experienced ostracism with controls who experienced inclusion. For example, typical adults who are ostracised are more likely than included controls to mimic a social partner’s physical mannerisms (e.g., Lakin et al., 2008). Similarly, young typically-developing children exposed to ostracism imitate an experimenter’s actions more closely than included controls (Over & Carpenter, 2009), a trend which is amplified when imitating social-conventional rather than instrumental actions (Watson-Jones et al., 2014), and when exclusion is enacted by in-group rather than out-group members (Watson-Jones et al., 2016). A recent study applied this experimental approach to typically-developing children’s alignment of word choice (i.e., lexical alignment), and found that those who experienced ostracism in a virtual ball-tossing game displayed increased lexical alignment with a partner, compared to included controls (Hopkins & Branigan, 2020). The effect of ostracism on imitation has been primarily explained in terms of goal activation theory (cf. Aarts & Dijksterhuis, 2000). According to this theory, the goal to affiliate is directly activated when we feel our sense of belonging is under threat; in order to fulfil these affiliative goals, we consequently display more behaviours that communicate our similarity to a social partner (Over & Carpenter, 2009).

While imitation seems to be sensitive to ostracism in the typical population, converging lines of evidence suggest that such effects might not manifest in autism. Firstly, studies of non-linguistic imitation in autistic children and adults report a deficit when the to-be-copied behaviour involves a social component (Wang & Hamilton, 2012; see also Vivanti & Hamilton, 2014, for a review). For example, autistic children are less likely than typically-developing children to over-imitate an adult’s unnecessary actions (Marsh et al., 2013; though cf. Nielsen et al., 2013), and their action imitation is not modulated by social contextual cues, compared with non-autistic controls (Vivanti & Dissanayake, 2014) and children with Williams Syndrome (Vivanti et al., 2016). Importantly, it has been shown that priming autistic adults with a pro-social attitude does not lead to increased levels of non-linguistic imitation (compare to those primed with a non-social attitude), as it does in typical controls (Cook & Bird, 2012).

Secondly, there is evidence for atypical processing of ostracism in autism: Autistic individuals report typical levels of distress in response to ostracism, but display relative hypoactivity in the neural circuitry that deals with rejection (Bolling et al., 2011; Masten et al., 2011; McPartland et al., 2011; Sebastian et al., 2009). Furthermore, although autistic adults experience a heightened physiological response to ostracism, they do not interpret ostracism as emotionally significant to the same degree as non-clinical controls (Trimmer et al., 2017). Such findings may reflect the influence of alexithymic traits—difficulties with recognising and describing one’s own emotional states—which are prevalent in the autistic population (50% compared with 10% in the typical population; Bernhardt et al., 2014; Hill et al., 2004), and which are associated with language impairment (Hobson et al., 2019; Milosavljevic et al., 2016). Taken together, the findings from these two literatures raise the possibility that autistic individuals’ imitative behaviour might not be modulated by ostracism.

Hence in the current study, we predict that autistic children who experience ostracism will not follow the pattern of typically-developing children of increased lexical alignment relative to included controls (Hopkins & Branigan, 2020). Previous studies have found that priming mechanisms giving rise to alignment are intact for autistic children, such that they align to the same extent as groups of typically-developing children matched by either age or verbal ability (Allen et al., 2011; Branigan, et al., 2016; Hopkins, Yuill, & Branigan, 2017; Hopkins, Yuill, & Keller, 2016). But these results were all found in contexts where social-affective factors were not central to the interaction. They therefore do not elucidate whether social-affective factors might modulate alignment. Here we set out to tap the social-affective mechanisms of autistic children’s alignment, by manipulating their inclusionary status before they interact with a conversational partner. If autistic children show an atypical pattern of alignment in response to ostracism, this might partly explain why their conversational behaviour appears unusual. Moreover, given the bi-directional link between imitation and affiliation, atypical alignment could also help to explain why partners might perceive conversation with autistic children as dissatisfying or unrewarding.

The current study adopted the same protocol reported by Hopkins and Branigan (2020). Autistic children experienced either ostracism or inclusion via the Cyberball paradigm (Williams et al., 2000), before playing a picture-naming game with an experimenter, based on the children’s card game ‘snap!’ (Branigan et al., 2016). In the game, each card depicted a familiar object that had two acceptable names (e.g., rabbit vs. bunny). On experimental trials, the child heard the experimenter name her picture with either a favored name (rabbit; as established by a pre-test) or a disfavored alternative (bunny). Two turns later, the child named the same object. Children’s tendency to imitate the experimenter’s use of disfavored names was recorded, and the extent of lexical alignment was compared between the ostracism and control (i.e., inclusion) groups. If autistic children’s priming-based tendency to align were further strengthened by the concurrent action of social-affective mechanisms, then those who experienced ostracism should lexically align with the experimenter to a greater extent than controls, consistent with typically-developing children (Hopkins & Branigan, 2020). However, a lack of difference in lexical alignment between the ostracised and included groups would be consistent with an impairment of social-affective mechanisms of alignment in autistic children.

Autism is associated with clinically significant impairments of communication and social interaction (DSM-5; American Psychiatric Association, 2013), which manifest in conversation (Ying Sng et al., 2018). Typical adults and children imitate a conversational partner’s language choices (Brennan & Clark, 1996; Giles et al., 1991; Pickering & Garrod, 2004) and such alignment promotes effective, rewarding exchanges between speakers (Fusaroli et al., 2012; Putman & Street, 1984). The current study investigated whether autistic children’s priming-based tendency to align (Allen et al., 2011; Branigan et al., 2016; Hopkins et al., 2017) can be socially modulated, in a way that might promote affiliation within dialogue. Consistent with previous research, we showed that autistic children display spontaneous lexical alignment with an experimenter (Branigan et al., 2016; Hopkins et al., 2017). Yet we found no evidence that they modify their linguistic behavior to achieve affiliative goals: Unlike their typically-developing peers (Hopkins & Branigan, 2020), autistic children who had experienced ostracism did not display increased lexical alignment, relative to controls. This finding suggests that linguistic imitation is unresponsive to social manipulation in autism, in the same way as is non-linguistic imitation (Cook & Bird, 2012).

However, it is argued elsewhere (Slocombe et al., 2013) that we should not assume that the same mechanisms underpin linguistic alignment in two groups, just because the groups show the same level of alignment. This observation seems relevant to the current study: Although the behavioural results of the ostracism and control groups were the same, our analyses suggested group-specific patterns in our data. The patterns are consistent with Branigan et al.’s (2010) proposal that observed alignment behaviors may arise as the outcome of multiple underlying mechanisms. Specifically, although there was a negative relationship between alignment effects and SCQ scores for all children (i.e., reduced lexical alignment was associated with increased autism symptom severity), this relationship was especially strong for the ostracism group. This suggests that, when social-affective mechanisms could be expected to be relevant to alignment, they were—albeit not to the extent that they yielded observable changes in alignment behavior (i.e., significantly stronger alignment effects). Furthermore, there was a significant interaction between prime name and SCQ scores for children in the control group only. This suggests that, in circumstances where there was no specific pressure to affiliate, automatic psycholinguistic mechanisms (i.e., priming) seem to have been more relevant to autistic children’s lexical alignment. Overall, then, our results imply that our experimental manipulation may have engaged different alignment mechanisms in the ostracism and control groups, but not sufficiently to elicit the reliable differences in alignment behaviors observed in typically-developing children (Hopkins & Branigan, 2020). However, since the inclusion and ostracism groups were small in our study, observations of a larger sample would be important in corroborating the finding of reduced affiliative influence on ostracised autistic children’s tendency to align, and the possibility that different mechanisms might differentially contribute to their alignment behaviours.

Hence the question arises: Given that ostracism influences the lexical alignment of typically-developing children (Hopkins & Branigan, 2020), why did we not observe the same effects in an autistic sample? An obvious explanation would be that our sample size was too small to detect such an effect. However, the results of our Bayesian analysis, which provides a measure of the strength of evidence for one hypothesis over another, casts doubt on this possibility. Despite our relatively small sample size, the Bayes Factor quantifying the strength of evidence in favor of the null hypothesis was substantial, which provided very strong evidence against the hypothesis that autistic children’s inclusionary status influenced the extent of their alignment (i.e., the alternative hypothesis; Raftery, 1995).

Another obvious explanation would be that our ostracism manipulation—Cyberball (Williams et al., 2000)—was ineffective. Again, we suggest that this is unlikely to have been the case, since our manipulation check scores suggested that children experienced Cyberball as intended; that is, children in the ostracism group appropriately reported receiving the ball significantly less than did children in the control group.

This leaves two alternative explanations of our findings, which may not be mutually exclusive. One possibility is that children’s manipulation check scores reflected their experience of inclusion/ostracism but failed to capture how this experience affected them emotionally. We intentionally avoided having a manipulation check that (i.) required introspection (an area of difficulty for autistic children; Robinson et al., 2017) and (ii.) explicitly drew attention to the purpose of the manipulation—unlike other studies that have used Cyberball with an autistic population (e.g., Sebastian et al., 2009, who asked participants to rate statements such as “I was excluded”). However, our choices meant that we were unable to determine that children’s assessment of how often they received the ball corresponded with their feelings about being included/ostracised by the confederate avatars. It is plausible that children did not feel included/ostracised in the way (and/or to the extent) that we had anticipated, despite their responses on the manipulation check. Such an explanation would be consistent with evidence of discrepancies between how autistic people process versus how they report experiences of ostracism (Bolling et al., 2011; Masten et al., 2011; McPartland et al., 2011; Sebastian et al., 2009; Trimmer et al., 2017).

A closely related possibility is that although the experience of ostracism might have been emotionally significant for autistic children, this might not have increased their motivation to affiliate with others; such a pattern would be consistent with other findings on social motivation in autism (Chevallier, Grèzes, et al., 2012; Chevallier, Kohls, et al., 2012). In turn, under a goal-activation account of affiliation (Aarts & Dijksterhuis, 2000), any reduced motivation to fulfil affiliative goals in autistic children would have meant that ostracised autistic children would not have been more likely to engage in affiliative behaviours, like their ostracised typically-developing peers (Hopkins & Branigan, 2020; Over & Carpenter, 2009; Song et al., 2015). This explanation would be coherent with evidence for atypical social modulation of non-linguistic imitation in autism (Wang & Hamilton, 2012; Vivanti & Hamilton, 2014; Marsh et al., 2013; Vivanti & Dissanayake, 2014; Vivanti et al., 2016; Cook & Bird, 2012), and with studies that show that autistic children can recognise but not always act to rectify their social isolation (Bauminger & Kasari, 2000).

Hence we have proposed different but complementary accounts of how autistic children’s experience of ostracism might preclude knock-on effects on lexical alignment. We assume that, if autistic children did not feel the negative effects of ostracism, and/or were not motivated to deploy behavioral strategies (e.g., imitation) to compensate for these negative effects, then they would not have displayed alignment above the level generated by automatic priming mechanisms (Branigan et al., 2016; Hopkins & Branigan, 2020; Hopkins et al., 2017). Future research in this area should establish what—for autistic children—disrupts the relationship between ostracism and increased affiliative behaviour that is seen in the typical population (Carpenter et al., 2013; Dijksterhuis & Bargh, 2001). For example, one way of addressing the potential gap between autistic children’s experience versus their reporting of ostracism would be to take additional measures during Cyberball—both physiological and psychological—as per Trimmer et al. (2017), which could help to determine whether the absence of heightened alignment effects reflects atypical emotional processing or something else.

The lack of difference in alignment behaviors between the groups in our study carries practical implications for autistic children and their social partners. Our findings suggest that a potentially valuable form of behavioral adaptation is unavailable to autistic children, which could offer a new perspective on why their conversational behavior appears unusual (Ying Sng et al., 2018). Furthermore, given the bi-directional link between imitation and affiliation in typical speakers (Dijksterhuis & Bargh, 2001), the reduced sensitivity of autistic children’s language imitation to ostracism—and potentially social affiliative considerations more generally—could result in their conversational partners experiencing interactions with them as unrewarding, and hence result in a reduced motivation for these partners to engage in further interaction with them. This could further entrench the social difficulties faced by autistic children, which include poorer quality friendships and increased feelings of loneliness (Bauminger & Kasari, 2000); social isolation at school (Kasari et al., 2011) and even strained familial relationships (Kaminsky & Dewey, 2001).

We conclude by suggesting that the social-affective mechanisms contributing to alignment are impaired for autistic children. In the present study, autistic children imitated the lexical choices of a partner, but those who experienced ostracism were as likely to imitate lexical choices as those who experienced inclusion; this is a different pattern to that found in typically-developing children, who show a stronger tendency to lexically align when they have been ostracised (Hopkins & Branigan, 2020). Intriguingly, however, our study offers evidence that autistic children’s language alignment might be underpinned by different mechanisms in different contexts; we showed that linguistic ability predicted children’s alignment in the control group, whereas social-communicative skills were predictive in the ostracism group. Such findings imply that, for autistic children, alignment mechanisms may be engaged in a way that is selective and contextually appropriate, but beneath a threshold to induce noticeable modulation of alignment behaviour.