Computational Linguistics for Autism-related Social Skills: Assessment for Characterization and Therapy (CLASS-ACT)

Abstract The goal of this R03 is to leverage computational linguistics to identify sensitive, efficient, and scalable measures of social communication that can be used to inform intervention efforts for transition-aged youth on the autism spectrum. Such measures would yield significant benefits for both (a) quantifying post-intervention clinical outcome and (b) precision tailoring for future interventions. By applying sophisticated computational linguistic approaches specifically designed to assess social communication in autistic youth (developed by MPI Parish- Morris and Co-I Cho) to a rich pre-existing dataset from a completed RCT of the START intervention model (extracted from 421 video-recorded conversations from 35 participants collected across 6 timepoints by MPI Vernon), the proposed project will significantly advance the goal of objective, efficient, and scalable clinical outcome measurement for youth with autism. Our collaborative research teams will use computational linguistics to identify objective, quantitative predictors of clinical intervention outcomes in autistic youth and chart individual trajectories of change. We hypothesize that (a) youth assigned to the START intervention group will demonstrate significant gains in linguistic markers of social phenotype in comparison to waitlist controls, (b) across the entire START cohort, modeling individual trajectories of growth in linguistic markers of social phenotype will identify subgroups with unique baseline factors that moderate the slope of change, and (c) linguistic markers that predict social communication success will vary by speaker (male vs. female) and context (same-sex vs opposite-sex conversations). We will also validate our linguistic features as social communication outcome metrics over time. We also anticipate that identified vocal/linguistic features will have high convergent validity with existing measures of social communication outcome and social impression ratings, and high discriminant validity with measures of restricted/repetitive behaviors and interests. This R03 proposal addresses NIDCD Strategic Plan Theme 3 to promote a precision medicine approach to prevention, diagnosis, and treatment of conditions that impact speech and communicative functioning, such as autism. A key output of this project will be validated computational linguistic features that can be used for clinical outcome measurement in autism. These metrics hold promise for both (a) efficiently and objectively quantifying post-intervention clinical outcome in autism and (b) guiding precision tailoring for future intervention implementations. Thus, the proposed project fills an important knowledge gap that could significantly advance the goal of objective, efficient, and scalable clinical outcome measurement and intervention response monitoring for transition-aged autistic youth. This proposal has high