S108. AUTOMATED DETECTION OF INCREASED METAPHOR PRODUCTION IN OPEN-ENDED SPEECH IN SCHIZOPHRENIA AND ITS RISK STATES

BACKGROUND: Clinicians have long observed peculiarities in the use of figurative language by individuals with schizophrenia, as part of larger expressive deficits. Natural language processing has been used to predict psychosis onset (Bedi et al., 2015; Corcoran et al., 2018), identifying classifiers such as semantic coherence. Here we examine use of metaphor across the psychosis spectrum. METHODS: Participant groups across the psychosis spectrum (healthy controls, clinical high risk for psychosis and individuals with schizophrenia) participated in open-ended interviews lasting approximately one hour and encouraged to express themselves narratively. Each interview was transcribed, then deidentified and run through the metaphor analysis algorithm. The results were then used to determine an overall metaphor frequency rate for each participant/transcript. RESULTS: The algorithm detected a significantly higher proportion of the words in transcripts of patients with schizophrenia as metaphorical (6.5%) than in healthy controls’ transcripts (5.7%) (p < 0.01, t57 = 2.68). Patients in the CHR group also produced more metaphorical words tagged by the algorithm (6.6%; p < 0.01, t95 = 3.69). DISCUSSION: These results demonstrate that metaphor usage, as measured using automated algorithms, is a robust and powerful indicator of symptoms, whether defined via current schizophrenia diagnosis or by clinical high-risk status. The ability to estimate metaphoric content in speech and written samples can be a significant contribution to the systemization of psychosis symptoms and and may be useful as a screen for the general population to identify individuals who may be at risk for psychosis.