A Longitudinal Study of Semantic Networks in Schizophrenia and other Psychotic Disorders Using the Word Association Task

Abstract

AbstractThe underpinnings of language deviations in psychotic symptoms (eg, formal thought disorder, delusions) remain unclear. We examined whether the semantic networks underlying word associations are useful predictors of clinical outcomes in psychosis. Fifty-one patients with schizophrenia and other psychotic disorders and 51 matched healthy controls generated words in a Cantonese continued word association task. Patterns of word associations were examined using semantic similarity metrics derived from word embeddings (fastText) and the structure of individual semantic networks. A longitudinal design—baseline and 6 months later—enabled investigation of the relationship of changes in semantic associations in patients who were in an acute psychotic state at baseline compared to clinical stabilization 6 months later. The semantic similarity measure increased over time in patients, while it remained stable in controls. Moreover, the change in semantic similarity over time correlated with the changes in patients’ formal thought disorder symptoms. There were differences in individual semantic networks between the groups at both time points. Patients had less structured networks on average, as evidenced by a smaller network diameter and clustering coefficient, and smaller average shortest path lengths. The identification of several state-like semantic measures that change over time with patients’ mental states allows for nuanced comparison with clinical measures. Semantic measures are complex. Semantic similarity was a state-like measure that changed over time with mental state in psychotic disorders, whereas individual semantic network parameters were trait-like and stable over time.