Computational linguistic investigation in schizophrenia and autism spectrum disorders

Computational linguistic analysis has been increasingly used to capture formal thought disorder in schizophrenia. Despite promising outcomes, investigations of the computational linguistic disturbances of schizophrenia in a transdiagnostic context are limited. Particularly, shared characteristics, neurodevelopmental origins, and the role of speech in the diagnosis of schizophrenia and autism indicate a need to explore both the commonalities and distinctions in the computational linguistic profiles of these groups. In this study, we investigated the semantic and structural properties of speech samples of 35 patients with schizophrenia spectrum disorder, 25 patients with autism spectrum disorder, and 25 healthy controls in free speech and picture description tasks. Our findings showed that only 5 of 45 features differed between the clinical groups. All of these were from the structural domain, while semantic features did not differ between these neurodevelopmental disorders. The clinical groups demonstrated elevated local and global semantic similarity, and negative sentiment compared to controls. Moreover, the speech of autism spectrum disorder included lower unique word frequency in picture description, alongside shorter pronouns and adverbs in free speech relative to other groups. Schizophrenia spectrum disorder used shorter adjectives than autism spectrum disorder and controls in free speech. Importantly, adjective frequency in schizophrenia spectrum disorder was lower than in autism spectrum disorder in free speech. Overall, our findings demonstrated an extensive dominance of similar computational linguistic traits between schizophrenia and autism spectrum disorders, indicating shared communication disturbances in these disorders. This outcome highlights the critical role of transdiagnostic and neurodevelopmental perspectives in computational linguistic investigations.