2026 ECNP Congress

PS03-3234

Automated speech analysis in psychosis: identifying linguistic and acoustic features across the psychosis spectrum

Peter Uhlhaas — V.R. Poulton, P. Baek, C.U. Correll, A. König, F. Menne, M.E. Macheleidt · Charite, Department of Child and Adolescent Psychiatry- Psychosomatic Medicine and Psychotherapy, Berllin, Germany

2026 ECNP Congress

Introduction

Language anomalies are an essential aspect of psychotic disorders, such as schizophrenia, and have been considered a core feature of the disorder since the early formulation of the clinical concept. Formal thought disorder (FTD) [1,2] refers to abnormalities in the content and form of speech and comprises positive, negative, and disorganized dimensions and has been assessed traditionally through clinical assessments. More recently, natural language processing (NLP) approaches have recently complemented analysis of FTD through automated examination of speech based on machine learning algorithms. These studies have shown that schizophrenia patients are characterized by reduced discourse coherence, syntactic complexity and language connectedness. NLP-derived parameters classify individuals with schizophrenia with high diagnostic accuracy and predict clinical outcomes in individuals at clinical high-risk for psychosis (CHR-P) [3], suggesting a potential biomarker for early detection and diagnosis.

Aims

This study investigated candidate speech features using NLP in individuals at CHR-P, clinical high-risk for bipolar disorder (CHR-BP), first episode of psychosis (FEP) patients and healthy controls (HC).

Methods

A group of CHR-Ps/CHR-BPs (n=22), FEP patients (n=12) and HC (n=28) were recruited (age range: 11-29) at the Dept. of Child and Adolescent Psychiatry, Charité Universitätsmedizin Berlin, and speech samples were elicited through 10 images of the Thematic Apperception Test (TAT) [4]. Group differences were examined using mass univariate testing on 48 Linguistic and 45 Acoustic automatically extracted speech features (Sigma library), and a principle component analysis (PCA) [5] was applied with varimax rotation across features.

Results

We identified 11 core NLP features with group differences included adjectives (CHR<<FEP: adj. p<0.05), energy of speech (HC>>CHR: adj. p<0.01), syntactic complexity (HC>>FEP: adj. p<0.01; HC>>CHR: adj. p<0.05), reduced speech intelligibility (HC<<FEP: adj. p<0.01; HC<<CHR: adj. p<0.05), articulatory control (HC>>FEP: adj. p<0.05), disorganized speech (HC<<FEP: adj. p<0.05), emotional tone (CHR<<FEP: adj. p<0.01), negative words (HC<<CHR: adj. p<0.05; CHR>>FEP: adj. p<0.01), emotional speech (CHR>>FEP: adj. p<0.05), emotional constancy (HC>>FEP: adj. p<0.05), emotional coherence (HC>>FEP: adj. p<0.05) and loudness (HC>>CHR-P: adj. p<0.05). Group differences were observed for two PCA-factors, „Sentiment“ (19.4%; HC>>FEP: adj. p<0.05; CHR>>FEP: adj. p< 0.05) and “Linguistic Grammar“ (21.1%; CHR>>FEP: adj. p<0.05), which accounted for 40.5% of variance explained. The factor “Sentiment” correlated positively with emotional stress (r=0.784), negative statements (r=0.773), emotional constancy (r=0.748), emotional response (r=0.702), stability of emotional tone (r=0.674), emotional engagement (r=0.616), verbal complexity (r=0.573), and emotional intensity (r=0.568). The factor „Linguistic Grammar“ correlated positively with nouns (r=0.826), determiners (r=0.811), adjectives (r=0.750), adpositions (r=0.725), unique words (r=0.608), lexical density (r=0.570) and understanding ambiguity (r=0.516).

Conclusions

11 core NLP features and two PCA-factors („Sentiment“ and „Linguistic Grammar“) differentiated CHR- and FEP-groups from HC. These results suggest that automated speech analysis provides objective, quantifiable markers of language anomalies. Integrating NLP into clinical workflows may refine risk stratification and enhance the early detection of psychotic disorders by capturing subtle linguistic and acoustic shifts that precede overt clinical symptoms.

References

References:

  1. Andreasen, N. C., & Grove, W. M. (1986). Thought, Language, and Communication in Schizophrenia: Diagnosis and Prognosis. Schizophrenia Bulletin, 12(3), 348–359.
  2. Stolar, N. (2017). Formal thought disorder. In The sage encyclopedia of abnormal and clinical psychology (Vol. 7, pp. 1495-1496). SAGE Publications, Inc.
  3. Corcoran, C.M., Carrillo, F., Fernández-Slezak, D., Bedi, G., Klim, C., Javitt, D.C. Bearden, C.E. and Cecchi, G.A. (2018), Prediction of psychosis across protocols and risk cohorts using automated language analysis. World Psychiatry, 17: 67-75.
  4. Murray, H. A. (1943). Thematic apperception test. Harvard University Press.
  5. Hotelling, H. (1933). Analysis of a complex of statistical variables into principal components. Journal of Educational Psychology. 24(6), 417–441.