Comparing Single Vs. Co-Occurrence Disfluency Features in Explaining Oral Language Proficiency

In language testing research, fluency is typically operationalized through single, isolated temporal features. However, in natural speech, fluency (or disfluency) is often characterized by the clustering of multiple temporal features, collectively revealing the speaker's automaticity in speech production. In this study, we compare disfluency co-occurrence features with commonly used, single disfluency features in their prediction of language proficiency. We examined 31 (dis)fluency features of IELTS speaking responses by 278 L2 English speakers across three L1 backgrounds (Mandarin Chinese, Punjabi, and Arabic). These features were subjected to principal component analysis and ordinal regression analysis to predict IELTS band scores. The findings indicate that both single and co-occurrence disfluency features can effectively explain oral language proficiency, although disfluency co-occurrence features add nuances to the interpretation and implications for L2 speaking assessment research.