A Longitudinal, Multi-Method Analysis of Pragmatic Language in ASD and Related Neurodevelopmental Disorders
Objectives: 1) Compare pragmatic language profiles and trajectories of pragmatic development in ASD-O and related disorders across multiple language contexts varying in social demand. 2) Assess the utility of computational linguistic tools for identifying ASD-related language impairments.
Methods: This longitudinal study assessed school age males with ASD-O (n=43), FXS-ASD (n=57), FXS-O (n=13), DS (n=22) and younger TD males (n=24) at up to three time points annually. Participants completed a comprehensive battery assessing cognition, structural language, and theory of mind. Pragmatic assessments included 1) parent questionnaire (Children’s Communication Checklist-2nd Edition), 2) a standardized measure (CASL-Pragmatic Judgment Subscale), 3) hand-coding of a structured narrative task, and 4) ratings of conversational behaviors during the Autism Diagnostic Observation Schedule using the Pragmatic Rating Scale-School Age (Landa, 2011). Computational linguistic analyses were applied to narrative and conversational samples using vector semantics to measure coherence and reciprocity.
Results: Males with ASD-O showed deficits in pragmatic language across all contexts that were stable across time points (ps < .05). Within the conversational context, males with FXS-ASD also demonstrated a similar quantity and quality of pragmatic language difficulties to males with ASD-O. Notably, both ASD groups showed significantly greater impairment in the conversational context relative to the structured narrative (p = .004). Computational measures differentiated clinical groups during conversation (ps < .03), and were related to ratings of pragmatic competence in both ASD groups (rs > .4). Assessment context drew on different skills across groups, with standardized measures and narrative quality more strongly related to mental age and structural language (rs > .35), whereas conversational difficulties were not significantly correlated with mental age or structural language in the ASD groups (rs < .27).
Conclusions: Groups showed unique profiles of strengths and weaknesses across different language contexts, with the ASD groups showing greatest difficulties in less structured conversation. Computational approaches appear sensitive to language differences observed in ASD, and may guide the development of efficient, quantitative, objective measures of social communication. Observed overlap both in severity and quality of pragmatic violations in males with ASD-O and FXS-ASD across development suggests specific shared pragmatic skills that may relate to variation in the FMR1 gene.