A study titled “Multi-modal Perceived Conversation Success Prediction for Spontaneous Autistic and Non-Autistic Dyadic Interactions” was published on October 26, 2026, on the arXiv preprint server (ID 2610.31682v1), aiming to predict autistic and non-autistic individuals’ subjective perceptions of their own conversational success in spontaneous dyadic interactions.
The study is based on three spontaneous dyadic interaction datasets and integrates four complementary behavioral modalities—speech articulation, facial expressions, acoustic prosody, and turn-taking—to construct a multimodal analysis framework. The results indicate that successful dialogue perception is inherently multidimensional; no single modality can fully capture all relevant information, whereas joint multimodal modeling significantly enhances predictive performance.
Unlike previous studies, which have largely focused on autism diagnostic classification or comparisons of group differences, this work directly addresses the practical need for assessing communication quality, emphasizing participants’ subjective experiences as the predictive target. The paper is currently in version v1 and has not yet undergone peer review; the dataset composition and specific performance metrics are not disclosed in the abstract.