Content-free speech activity records: interviews with people with schizophrenia

Abstract

AbstractSchizophrenia is characterised by a variety of symptoms, many of which are expressed verbally. However, privacy concerns limit the possibility of collecting and sharing large corpora of schizophrenic speech. As a result, variability in the communicative behaviour of individuals with schizophrenia is currently poorly understood. In this study we explore how far content-free speech activity records can successfully profile the behaviour of patients with schizophrenia. We used data from one of the very few publicly available corpora of conversations with patients with schizophrenia, featuring interviews between a therapist and three patients with different symptoms. Crucially, in this study we annotated only moments of speech and silence during the interview, and we entirely discarded the verbal content of the interview. In this way we perform a type of analysis that fully preserves the speakers’ privacy, while still allowing for comparisons with the full original recorded material. We developed several types of analyses and corresponding visualisations from the activity records. Exemplifying these analyses, clear patient-specific profiles can be derived, based on the dimensions of total silence duration and the speech duration of patients and therapists. These findings are consistent with an independent phenomenological assessment of the three patients.