Analyzing Discourse Patterns of Young Adults with ASD

Background: Most previous work in analyzing speech transcripts from individuals with Autism Spectrum Disorder (ASD) has focused primarily on quantitative syntactic and lexical metrics and qualitative measures of pragmatics, which are usually applied in the restricted context of a narrative retelling or picture description task. Less research has been devoted to systematically examining higher-level discourse structures in less constrained conversational interactions, particularly among young adults on the spectrum.

Objectives: The objectives of this research are: (1) to confirm previous findings about low-level lexical and syntactic characteristics of language in high-functioning young adults with ASD; and (2) to detect differences in the high-level discourse patterns of college-aged males with and without ASD.

Methods: All participants in the study were male university students. Typically-developing (TD) participants were recruited to match the ASD participants according to age, college major, and GPA. Transcripts were produced for 6 male subjects (ASD n=3, TD n=3) from recordings of three tasks designed to elicit conversation with a TD confederate in (1) a usability task for a web application, (2) a collaborative game, and (3) an interactive math and computing tutoring task. Standard measures of syntactic complexity, vocabulary diversity, and disfluency frequency were automatically extracted from the transcripts. In addition, an existing discourse-based annotation scheme was used to manually code the discourse units (“dialog acts”) in the transcripts in order to characterize the way that individuals with ASD acknowledge their interlocutors, seek actions, and ask for information.

Results: As previously reported in the literature, few between-group differences were found in the lower-level syntactic, pragmatic, and lexical features. Several differences in discourse features, however, were observed. In the cooperative game task, individuals with ASD were less likely to acknowledge (oh alright) their interlocutor’s comments and were far more likely to suggest (I’d say start out by moving past and jumping over...) -- and considerably less likely to request (ok push the block on to the ramp) -- that their interlocutors perform an action than their TD counterparts. When compared to their TD counterparts across all tasks, individuals with ASD were considerably more likely to (1) answer questions with “no”, (2) apologize for something they had said or done, (3) request clarification, and (4) thank their interlocutor.

Conclusions: Many of the lower-level features typically used to characterize language in children with ASD seem to lose their discriminative power when applied to young adults, underscoring the need for the development of more complex and nuanced methods of language analysis. The study suggests that discourse analysis, particularly when applied to conversations produced in a collaborative and relatively unconstrained context, can provided insight into the linguistic differences that persist into adulthood. As results for lower-level features align with previous literature, future work will focus primarily on discourse-based comparisons between groups.