Computational Phenotyping of Schizophrenia: Measuring Language and Facial Expression

Cheryl Corcoran, MD, described her research applying computational methods to the analysis of language, acoustics, and facial expression in individuals at clinical high risk for psychosis.^1,2

Corcoran shared the origins of this work as a collaboration with computer scientists and physicists, when she applied early natural language processing techniques to transcripts from her qualitative research with at-risk individuals whose eventual clinical outcomes were known. That analysis identified 2 predictive language features—semantic incoherence and reduced linguistic complexity—such that "those who would go off track or especially who spoke simply and had shorter sentences" were at greater risk of converting to psychosis. The collaboration with a physicist colleague formed the foundation of an ongoing and expanding research program.

Corcoran described the current generation of this work, which employs large language models to analyze open-ended interview speech not only for structural features but for semantic content. She reported that individuals at clinical high risk who had suicidal ideation or had made suicide attempts demonstrated an angry focus in open-ended speech, and that analyses could detect experiences of identity diffusion, stigma, and affective states. The program has since expanded to include acoustic analysis and automated video-based facial expression coding, enabling quantification of blunted affect and assessment of congruence between verbal and nonverbal expression.

Framing the broader implications, Corcoran identified the absence of objective clinical measures as psychiatry's defining methodological gap, observing that "we're the only field in medicine where you can't just order [a test]." She expressed hope that computational phenotyping of language and behavior could serve as a platform for longitudinal patient monitoring and as outcome measures in clinical trials, and indicated that her APA presentation would include practical instruction enabling clinicians to collect and analyze such data themselves.

Dr Corcoran is a professor of psychiatry at the Icahn School of Medicine at Mount Sinai.

References

1. Corcoran CM, Carillo F, Fernandez-Slezak, et al. Prediction of psychosis across protocols and risk cohorts using automated language analysis. World Psych. 2018;17(1):67-75.

2. Bilgrami ZR, Castro E, Agurto C, et al. Collecting language, speech acoustics, and facial expression to predict psychosis and other clinical outcomes: strategies from the AMP SCZ initiative. Schizophrenia. 2025;11:125.