Detecting semantic memory abnormalities in psychosis : quantification using semantic space modelling and network approaches

Abstract

Background: Language disorganization is central to the conceptualization of psychosis. Prior literature has consistently reported the presence of various semantic memory impairments in psychosis across a range of neuropsychological tests. However, the underlying factors of these semantic abnormalities remain inconclusive. The quantification of semantic abnormalities have improved with the recent advancements in semantic space modelling and network analyses. The current study applied such computational methods to a continued word association task, using immediate response to cue-words as a means to explore semantic associations. This study aims to quantify and characterize semantic abnormalities in patients with psychotic disorders, in comparison to healthy controls. Furthermore, a prospective 6-month longitudinal study design was employed to investigate the semantic abnormalities during an active psychotic episode compared with the same patients after remission 6-months later, in order to clarify the state-trait status of the semantic abnormalities, and their relationships with clinical symptoms and cognitive functioning.

Methodology: Fifty-one patients with psychotic disorders and 51 healthy controls matched in terms of age, gender and years of education have been included in the study. The study assessed a battery of clinical, cognitive, functioning and semantic associations in patients at two time points. A battery of cognitive assessments and semantic associations were also collected from healthy controls at two time points. A continued word association task was used to elicit three association responses per cue from a set of 200 Cantonese cue-words. The association responses were used to quantify and characterize semantic memory processing through three distinct levels of analyses, namely semantic similarity, individual-level and grouped-level semantic networks.

Results: First, the current study showed that the semantic similarity changed over time with the change in patient’s symptom severity, while it remained stable across time in healthy controls. In particular, semantic similarity change over time significantly correlated with the change in formal thought disorder (FTD) symptoms. Second, there were significant structural differences between patients and controls on individual-level and grouped-level semantic networks at two time points. Third, there were significant changes across time in the control’s individual-level network structure, while the grouped-level network structural anomalies found in patients remained stable over time. Lastly, the individual-level semantic network properties were found significantly contribute to FTD symptoms and affective symptom severity at baseline.

Discussion: This is an exploratory attempt in applying a novel approach through semantic space modelling and network analyses to investigate the semantic abnormalities in Cantonese-speaking patients with psychotic disorders. Generally, differences in semantic memory processing between patients and controls have been demonstrated across all three levels of analyses. Semantic similarity has been observed to be a state-dependent measure that changes along with patient’s mental states. Whereas, the semantic network structures remained stable over time despite the changes in mental states, suggesting a trait-like measure. This study is one of the first studies to identify potential differences in semantic deficits in psychotic disorders due to language differences.