Automated lexical analysis of interviews with individuals with schizophrenia

Abstract

Schizophrenia is a chronic mental disorder that contributes to poor function and quality of life. We are aiming to design objective assessment tools of schizophrenia. In earlier work, we investigated non-verbal quantitative cues for this purpose. In this paper, we explore linguistic cues, extracted from interviews with patients with schizophrenia and healthy control subjects, conducted by trained psychologists. Specifically, we analyzed the interviews of 47 patients and 24 healthy age-matched control subjects. We applied automated speech recognition and linguistic tools to capture the linguistic categories of emotional and psychological states. Based on those linguistic categories, we applied a binary classifier to distinguish patients from matched control subjects, leading to a classification accuracy of about 86% (by leave-one-out cross-validation); this result seems to suggest that patients with schizophrenia tend to talk about different topics and use different words. We provided an in-depth discussion of the most salient lexical features, which may provide some insights into the linguistic alterations in patients.