Ready to CONNECT: Conversation and Language in Autistic Teens

Conversations are a critical medium for success in daily life, but predictors and measures of conversational success are poorly understood. The overarching goal of this proposal is to identify networks of naturalistic and standardized psycholinguistic features that lead to successful conversations. Standardized language assessments often do not capture important linguistic processes in real-world conversations, such as pronominal reference, back-channeling, turn-taking, or phonological, lexical, or syntactic alignment. The double empathy theory further posits that autistic conversational difficulties reflect failures of mutual understanding, rather than autistic deficits, indicating that autistic and neurotypical conversation partners differentially use and understand these linguistic processes. This proposal centers individuals with autism spectrum disorder who have age- appropriate scores on standardized language measures, many of whom nonetheless struggle with communication. We will use machine learning to model conversational profiles based on interactional measures of linguistic processes drawn from spontaneous conversation, and standardized language assessments, to evaluate conversational success in neurotype-concordant and neurotype-discordant interactions. Leveraging the ubiquity of videoconferencing, we will collect clinical and psycholinguistic data from dyadic conversations in a large sample of 500 12–15-year-old adolescents. We will also collect in-person conversational data from a group of n = 60. After providing a canonical speech sample, participants will have conversations with neurotype-concordant and -discordant partners in two contexts: (1) a get to know you conversation, and (2) a collaborative conversation, in which partners each hold one of a pair of pictures that differs in five ways and verbally collaborate to find the differences. We objectively define conversational success as the number and speed of correct identifications in Task 2. In addition, partners will rate their interactions post- hoc on subjective social metrics (e.g., likeability, warmth, boredom) and conversational success metrics (e.g., turn-taking, mutual appreciation, interest in further interaction). Conversations and speech samples will be recorded and then scored by naïve third-party raters on the same metrics. Recordings will be analyzed for acoustic, psycholinguistic, and conversational measures (e.g., fundamental frequency, prosodic range, pause duration, linguistic alignment, turn-taking). We will contrast the power of standardized scores and naturalistic psycholinguistic measures to predict both subjectively and objectively defined conversational success (Aim 1) and compare success in neurotype-concordant and neurotype-discordant partnerships (Aim 2). Aim 3 will leverage this rich dataset of acoustic, linguistic, perceptual, and standardized data to model computational predictor networks of conversational success. Results will advance the field by establishing metrics of conversational success in real-world social interactions and using computational models to form meaningful conversational profile clusters that go beyond simple diagnostic dichotomies to inform personalized supports.