The use of artificial intelligence-based text analysis in the diagnosis of schizophrenia and the examination of the relationship between identified language impairments and cognitive functions

Aim: Thought and language disorders are core features of schizophrenia and are closely related to cognitive impairments. Automated language analysis methods have been employed to quantitatively investigate these thought and language disorders. This study aimed to analyze the linguistic features of patients with schizophrenia and healthy controls using automated methods, examine group differences, and explore the relationship of the obtained data with thought disorders, cognitive impairments, and symptom severity. Material and Method: This cross-sectional, case-control study included 30 patients with schizophrenia and 30 healthy controls. Participants were assessed using the Montreal Cognitive Assessment (MoCA-TR), the Positive and Negative Syndrome Scale (PANSS), and the Thought and Language Index (TLI). During the application of the TLI, participants' speech was recorded and later transcribed. Transcripts were analyzed using Turkish Spacy, grammatical tagging, and Word2Vec for natural language processing. A Random Forest Classifier model was developed to predict patients based on linguistic data. Statistical methods were applied to evaluate the results. Results: Patients with schizophrenia had lower MoCA-TR scores and higher PANSS and TLI scores. Language analysis revealed lower Type-Token Ratio (TTR), reduced use of conjunctions, and increased use of ambiguous pronouns in patients. Average cosine similarity was significantly lower in schizophrenia patients compared to healthy participants. The machine learning model predicted schizophrenia patients with 83% accuracy. Linguistic features showed significant associations with cognitive and thought disorders as well as symptom severity. Conclusion: Automated language analysis reflects the thought and cognitive impairments observed in schizophrenia. The analysis of linguistic data using a machine learning model proved to be a successful tool for clinical differentiation. This method demonstrates the potential to serve as a diagnostic biomarker for schizophrenia.