Syntactic complexity, specifically the number of nominal subjects per clause in spoken language, declines in the six months following a first episode of psychosis among individuals who later receive a schizophrenia diagnosis. In a cohort of 26 first-episode psychosis patients and 12 healthy controls, automated analysis of speech samples from the Thought and Language Index interview showed that a 50% decrease in mean nominal subjects per clause after six months was explained by the presence of first-episode psychosis with 95.4% probability. Among those with psychosis, a 30% decrease predicted a schizophrenia diagnosis with 95% probability. This longitudinal decline distinguishes schizophrenia from other psychotic disorders.
Pause dynamics in speech—silent intervals between utterances—add unique information beyond semantic coherence for assessing formal thought disorder (FTD) in schizophrenia spectrum disorders. Across three datasets (naturalistic diaries, picture descriptions, and dream narratives; total 255 participants), models combining pause features with semantic coherence predicted clinician-rated FTD severity more accurately than coherence alone. Late fusion of both feature types raised average Spearman correlation from 0.413 to 0.455. The most informative pause patterns varied by context, indicating that pause dynamics and semantic coherence capture complementary aspects of thought disorganization. This multimodal framework offers a scalable, automated approach to assessing disorganized speech.