Akademik Veri Yönetim Sistemi
2026 ANNUAL CONGRESS of the Schizophrenia International Research Society, Florence, İtalya, 25 - 29 Mart 2026, ss.764-765, (Özet Bildiri)
Background: Patients with psychosis can exhibit a wide range of cognitive and language impairments. One important domain of cognition is social cognition, which refers to understanding others’ emotions and thoughts. In addition, semantic-level language impairments have been identified as a key feature of speech disturbances in schizophrenia. Recent findings suggest that during the prodromal phase of psychosis, subtle alterations in both social cognition and language can already be observed. Therefore, it may be important to detect these impairments before the onset of full psychotic symptoms.
Previous studies have demonstrated semantic language impairments in individuals at clinical high risk for psychosis (CHR-P) using objective natural language processing (NLP) tools. However, research investigating the relationship between language abnormalities and social cognitive deficits in CHR-P remains limited.
The present study therefore aimed to examine the association between automated semantic language features and social cognition in individuals at clinical high risk for psychosis.
Methods: Help-seeking youth at clinical high risk for psychosis (CHR-P) were recruited based on criteria for one of three prodromal syndromes, assessed with the SIPS/SOPS.
For the NLP analysis, we collected speech samples from 44 individuals at CHR-P and 24 age- and sex-matched healthy controls during descriptions of eight Thematic Apperception Test (TAT) pictures, which were then manually transcribed. Texts were preprocessed using NLTK and the SpaCy Turkish model, and sentence embeddings were obtained with SBERT. Cosine similarity was computed between neighboring (local) and all possible (global) sentence pairs, and mean, variance, 5th, and 95th percentiles were extracted. Image-to-text similarity was also calculated using CLIP, and all metrics were averaged across pictures.
Social cognition was measured with the Reading the Mind in the Eyes Test (RMET), and the Hinting Task, assessing both mental state decoding and reasoning components of theory of mind (ToM).
Results: Individuals at CHR-P showed increased sentence-level semantic similarity at both local and global levels, indicating a narrowed semantic space (mean local similarity: p < .001; variance and 95th percentile local and global similarities: p < .001; mean global similarity: p = .006). The CHR-P group also exhibited reduced image-to-text similarity, suggesting difficulty in describing visual stimuli compared to controls (p = .008).
Regarding social cognition, individuals at CHR-P scored lower on the RMET and Hinting tests (both p < .001), reflecting impairments in mental state decoding and reasoning.
Correlation analyses revealed negative associations between social cognition scales and sentence-level similarity measures (except the 5th percentiles), indicating that a narrowed semantic space in prodromal speech may reflect social cognitive deficits (Pearson’s r = -0.265 to -0.509). Additionally, image-to-text similarity was positively correlated with Hinting scores (Pearson’s r = 0.300), suggesting that reduced ability to describe visual stimuli is linked to poorer mental state reasoning.
Discussion: This study extends previous research examining the relationship between automated semantic language features and social cognition in individuals at CHR-P. Our findings suggest that a narrowed semantic space may serve as an indicator of social cognitive deficits during the prodromal phase of psychosis. These findings contribute to the growing field of digital mental health, emphasizing the value of automated language analysis as a non-invasive tool for early detection of language and social cognition impairments in psychosis risk.