Incorporating Co-occurrence Into the Operationalization of Speech Disfluency for Second Language Pronunciation and Oral Proficiency Assessment

Current L2 utterance fluency literature tends to operationalize disfluency as isolated, individual features. However, disfluency features often co-occur at one location or across multiple locations in one utterance. This study explores the co-occurrence of L2 disfluency features in a speech corpus from 71 L1 and L2 speakers of English across proficiency levels on an elicited imitation task and an oral listen-to-summarize task. We segmented each participant's speech into analysis of speech (AS) units (k = 2,704), extracted 2,972 individual disfluency chains based on 15 disfluency variables, and subjected them to principal components analysis and hierarchical-based K-means clustering analysis to identify disfluency co-occurrence patterns and speaker profiles across tasks. Results showed that different disfluency co-occurrences can be interpreted around various repair behaviors, and these repair behaviors also differ across tasks. Further analysis concerning the disfluency-proficiency relationship suggests that whereas some disfluency co-occurrences are meaningfully associated with proficiency, others might not be.