Lexicogrammatical profiling of ASD: cognitive-functional mapping and diagnostic implications.

Introduction: Previous corpus-based study first established an annotated dataset of autism spectrum disorder (ASD) discourse, and subsequent modeling of lexicogrammatical patterns distinguished ASD from non-ASD discourse with high performance (accuracy 80%, precision 82%, sensitivity 73%, specificity 87%). That line of research further identified 46 statistically significant discriminators, of which 20 were analyzed in detail. The present study examines 18 additional discriminators and situates them within cognitive-functional domains to clarify their diagnostic relevance. Findings refine the language-cognition interface in ASD and extend the utility of lexicogrammatical profiling for assessment. Methods: The Tag Linear Model was employed to identify lexicogrammatical features that distinguish ASD and non-ASD discourse. Logistic regression with 10,000 bootstrap iterations was applied to establish statistical significance. Although DNN models yielded higher predictive accuracy, the linear model provided transparent identification of discriminators. Results: Of the 135 items analyzed, 46 were confirmed as statistically significant discriminators ( Discussion: These findings demonstrate that choice patterns of lexicogrammar in ASD reflect domain-specific cognitive constraints. Interpreting the 18 discriminators within 12 cognitive-functional domains provides a linguistically grounded perspective on the neurocognitive profile of ASD and offers implications for future diagnostic and intervention research.