Linguistic markers in schizophrenia patients with cognitive deficit: a machine learning study using Korean speech

Abstract

Although the relationship between schizophrenia (SZ), formal thought disorder, and cognition has been extensively studied, comparatively less is known about which automatically extracted linguistic features are associated with cognitive status in SZ. We analyzed lexical, syntactic, semantic, and sentiment features from speech samples of 88 patients with SZ [37 cognitively impaired (Cog-D) and 51 cognitively preserved (Cog-ND)] and 69 healthy controls (HC). Features were extracted using Korean-adapted Natural Language Processing (NLP) pipelines. Medication effects were controlled for using an Extrapyramidal Symptom Rating Scale. Machine learning (ML) classification was conducted as an exploratory analysis. Patients with Cog-D showed marked reductions in lexical richness, syntactic complexity, semantic coherence and sentiment polarity compared with those with Cog-ND or HC. The performance of the ML models showed a discrepancy between the mean outer-fold performance and the performance on the hold-out test dataset. Several linguistic features were correlated with symptom severity and language performance. Linguistic disturbances in patients with SZ were heterogeneous and more pronounced in Cog-D group compared to Cog-ND group. These findings highlight the potential of NLP-based speech analysis for identifying cognitive impairment and support the cross-linguistic generalizability of linguistic markers.