Social Impairment in Conversation: Disfluency and Compensatory Mechanisms

Background:  Social impairment is a defining clinical feature of ASD. However, little is known about how it concretely unfolds during social exchanges: how interlocutors pick up and react to disfluency, and how patterns of interaction are affected. A better understanding of the dynamics of interactions with adults with ASD will help us understand how social impairment affects the life of people with ASD and which compensatory mechanisms can be used to minimize its effects.

Objectives: We want to develop automated quantitative methods to assess dysfluency and compensatory dynamics in conversation. Using simple measures of conversational turn-taking, we ask the following questions: i) How does autistic social impairment manifest itself in conversations? ii) How does the interlocutor react? iii) Are these dynamics related to specific clinical features?

Methods:  17 ASD and 17 matched Typically Developing (TD) adults were interviewed about the details they could recall of a standardised event they had participated in. Time-coded transcripts were used to calculate turn duration, inter-turn latency, number of turns for each interlocutor and percentage of spoken time in the conversation produced by the interviewee. Mixed effects models were employed to assess the relationship between these measures and diagnosis in interviewer and participants. Finally, we assessed whether participant behavior would predict the interviewer’s behavior.

Results: Participant’s Behavior: There was no main effect of ASD in turn duration or inter-turn latency (p>0.7), but significant interaction with time: TD participants increased their turn duration over time, while participants with ASD did not (Beta= -0.13, SE=0.04, p=0.003), an effect modulated by the ADOS Communication scores as well (Beta=-0.03, SE=0.015, p=0.036). Inter-turn latency increased less over time for ASD than for TD participants (Beta=-0.01, SE=0.005, p=0.03). Severity of clinical features did not affect inter-turn latency. Interviewer’s Behavior: The interviewer did not show any effect of diagnosis in contributions’ length and interturn latency, but provided more turns per unit of time in conversations with participants with ASD (Beta:0.26, SE=0.05, p<0.001), with a median time between turns of 5.4 seconds, against 9.75 seconds with TD participants. This effect was predicted by the participant’s inter-turn latency (Beta=1.4, SE=0.4, p=0.0007) and turn duration (Beta=-2.4, SE=0.64, p=0.0002), but did not interact with diagnosis (p>.3). This might also explain why the amount of information provided by the participants was not affected by diagnosis (Beta=-0.06, SE=0.03, p=0.07) or severity of clinical features (p>0.5).

Conclusions: Using simple turn-taking measures, we observe clear compensatory dynamics at work in the interviews. The more disfluency is displayed in the participant, the more the interviewer provides scaffolding. Future work will investigate whether these effects are modulated by practice and context, and how they affect the success of the interactions and the experience of the participants.