Multimodal assessment of communicative-pragmatic features in schizophrenia: a machine learning approach

An impairment in pragmatic communication is a core feature of schizophrenia, involving linguistic, non-verbal/extralinguistic, and paralinguistic modalities. This study identifies the communicative-pragmatic features that best discriminate individuals with schizophrenia from healthy controls. Using a multimodal assessment and a Decision Tree (DT) machine learning model, we evaluated various pragmatic phenomena and expressive modalities. The DT showed good overall performance (Accuracy 82%, Sensitivity 76%, Precision 91%). Linguistic irony was the most informative discriminator, followed by violations of Gricean maxims, then extralinguistic deceitful and sincere communicative acts. Findings are discussed in relation to pragmatic theory and clinical implications for assessment and rehabilitation design.