S199. HOW TO COLLECT FREE SPEECH TO SPEECH GRAPH ANALYSIS: STANDARDIZED TIME-LIMITED PROTOCOL

• Published in: Schizophrenia bulletin Vol. 46; no. Supplement_1; p. S114
• Main Authors: Simabucuru, Gabriela, Copelli, Mauro, Ribeiro, Sidarta, Mota, Natália
• Format: Journal Article
• Language: English
• Published: US Oxford University Press 18.05.2020
• Subjects: Data collection; Freedom of speech; Poster Session I; Schizophrenia
• ISSN: 0586-7614; 1745-1701
• DOI: 10.1093/schbul/sbaa031.265

BackgroundSpeechGraphs is a computational tool successfully applied to the differential diagnosis of psychosis based on the non-semantic structural analysis of speech graphs. This approach provides quantitative, fast, and low-cost measurements of clinical interest based on free speech, but its replicability needs to understand how different ways to collect speech data can impact such measurements. Stimuli or interruption on a patient’s speech can have a direct impact, especially on SpeechGraphs analysis. We developed a standard for data collection, controlling the time in order to keep the subject talking for a minimum of 30 seconds, and only stimuli the patient’s speech with general instructions. We aim to investigate specifically the impact of considering interviewers interferences marked as paragraphs in transcriptions, and the impact of a time-limited standardized protocol to avoid this bias.MethodsTwo different speech samples from a previous study were compared: 1) using free speech (N = 60, Mota 2014) or 2) using the proposed time-limited protocol (sub-sample of N = 31, Mota 2017). For both samples, we calculated connectedness attributes (such as LSC) in two in two different transcribing conditions: with and without paragraphs. The paragraphs represented the interviewer’s stimuli, and when paragraphs were deleted, we connected the subject’s speech in a unique line.ResultsInterviewer’s interferences marked by line-breaks or paragraphs had an impact on connectedness results for the freely speaking protocol (LSC Schizophrenia x Non-Schizophrenia: p = 0.0051 and without paragraph, LSC Schizophrenia x Non-Schizophrenia: p = 0.7764). The standardized protocol with a time limit was sufficient to avoid this bias: we found that there are no differences in considering or not paragraphs, with reports of 30 seconds (LSC Schizophrenia x Non-Schizophrenia: p = 0.0017 and without paragraph, LSC Schizophrenia x Non-Schizophrenia: p=0.0003)DiscussionThe standardized data collection protocol seems to be robust in comparison to not controlled methods to collect free speech, allowing the automatization of data gathering and transcriptions, preventing methodological errors in future SpeechGraphs applications.