S19. ANALYZING NEGATIVE SYMPTOMS AND LANGUAGE IN YOUTHS AT RISK FOR PSYCHOSIS USING AUTOMATED LANGUAGE ANALYSIS

Abstract

Background: Schizophrenia is characterized by disturbances in thought and language, often resulting in reduced production and complexity of speech, and abnormal pauses; these language disturbances have been associated with negative symptoms (Cohen et al., 2016). Using clinical ratings, we and others have shown that both language disturbances and negative symptoms are evident prior to psychosis onset in youths at clinical high risk (CHR). Automated natural language processing (NLP) analyses of transcribed speech show that both reductions in semantic coherence and syntactic complexity predict psychosis onset in CHR individuals (Bedi et al., 2015). In the current study, we assessed whether NLP syntactic complexity features and aberrant pauses may be associated with negative symptoms in CHR youths. Method(s): Participants included 33 CHR youths (mean (SD) age 21 (4) years; 11 females; ethnically diverse), of whom 5 developed psychosis (CHR+) within 2 years, whereas 28 did not (CHR-). Speech was elicited using open-ended interview. Audio files were transcribed and de-identified; using the Natural Language Toolkit (www.nltk.org), these transcripts were subjected to preprocessing (e.g. lemmatized) and analyzed using latent semantic analysis and part-of-speech tagging to characterize syntax. PRAAT (www.fon.hum.uva.nl/praat) was used for the analysis of pauses in audio files. Based on the criterion set forth by Goldman-Eisler (1968), pauses were defined as any silence longer than 250ms, as pauses less than 250ms are considered to signify breathing and articulation, while pauses longer than 250ms are assumed to reflect higher level cognitive processes, such as planning, or pathological phenomena, such as thought blocking. Mean pause length was calculated, as well as the percentage of time spent during the encounter spent in silences greater than 250ms, defined as the percentage of pauses. Negative symptoms were assessed using the Structured Interview for Psychosis-Risk Syndromes (SIPS). Spearman correlational analyses of demographics, linguistic variables and negative symptoms were performed. Result(s): CHR patients had a mean (SD) total negative symptom score of 13.42(7.77). For pause features, patients had a mean pause length of 1.1 (0.5) seconds, with 50% (SD 17%) percentage of pauses. Total negative symptom severity was significantly associated with mean pause length (r=.50, p 0.4). Negative symptoms were also associated with the use of determiner pronouns such as "which" and "that", which introduce dependent clauses (r=-.38, p