Learning to Interact: Developmental Trajectories of Linguistic Alignment in ASD

Background:  Interactive alignment (IA) has been argued to underlie not only successful social interactions (Pickering & Garrod, 2004), but also language development. By re-using each other’s words, the child and the caregiver engage each other, and the child acquires new linguistic items and receives feedback on their use (Nguyen & Delvaux, 2015; Messum & Howard 2015). IA has been shown to be unimpaired in high-functioning children and adults with ASD during highly structured task-oriented conversations (Hopkins et al, 2014; Slocombe et al 2014; Allen et al 2011). However, little is known about how the ability to align develops, and whether it is impaired in everyday unstructured interactions.

Objectives: We ask two questions: i) does IA evolve over time as children becomes a more competent speakers? and ii) is IA impaired in children with ASD?

Methods:  We analyzed spontaneous speech in parent-child dyads from a longitudinal corpus (6 visits over 2 years), consisting of 30 minutes of controlled playful activities between parents and 66 children (33 ASD; MA = 33 months) and 33 initially-language-matched (Mullen EL, RL) typically developing (TD; MA = 20 months) (cf. Goodwin et al. 2012). Lexical and syntactic alignment was calculated between neighboring utterances as the probability of words (lexical) or parts of speech (syntactic) to be repeated, normalized by the length of the utterances involved. Syntactic alignment was calculated excluding repeated lexical items to maximally distinguish it from lexical alignment. To assess the development of alignment we used mixed-effects growth curve models. The models included gender, Mullen scores (Mullen 1995), and ADOS (Gotham 2009) scores as fixed factors.

Results: Across groups, we observed a general increase of alignment over time, with two primary effects: (1) Development had significant linear and quadratic components (lexical linear: β=0.76, SE=0.2, t-stat=3.88, p=0.0001; lexical quadratic: β=-0.56, SE=0.18, t-stat=-3.07, p=0.002; syntactic linear: β=1.12, SE=0.44, t-stat=2.53, p=0.011; syntactic quadratic: β=-0.95, SE= 0.45, t-stat=-2.12, p=0.034). (2) Children with ASD displayed a lower degree of alignment(lexical: β=-0.01, SE=0.01, t-stat=-1.96, p=0.05; syntactic: ASD: β=-0.07, SE=0.04, t-stat=-2, p=0.045). There were main effects of ADOS and Mullen scores: higher Mullen is related to higher lexical (β=0.01, SE=0.01, t-stat=1.94, p=0.05) and lower syntactic alignment (β=-1.16, SE=0.33, t-stat=-3.48, p<0.0001); higher ADOS is related to lower lexical (β=-0.13, SE=0.05, t-stat=-2.6, p= 0.01) and syntactic alignment (β=-0.09, SE=  0.04, t-stat=-2.47 p=0.013). Gender had no impact. None of these factors interacted significantly with time.

Conclusions: Our data support the hypothesis that children’s propensity to lexical and syntactic alignment develops over time. Contrary to previous findings on older populations, in our analysis of unstructured play between children and caregivers, we found that interactive alignment was impaired in the ASD group, as a function of symptom severity and intellectual functioning. This has important implications for understanding linguistic and social development in ASD. More work is needed to investigate how alignment develops at later ages, how alignment in unstructured contexts relates to alignment in structured task-oriented conversations, and what effect reduced alignment has on language development in ASD.