Fluency assessment

Abstract Speech fluency is often assessed using articulation rate and pause frequency. However, not all pauses hinder fluency: when placed strategically, they structure discourse and enhance comprehensibility. To better characterize speaker fluency, it is crucial to consider where pauses occur. Traditional approaches rely on categorical syntactic boundaries (e.g., clauses or phrases), but inadequately capture syntactic complexity. We propose a continuous measure of pause placement based on syntactic distance between adjacent words. Using spontaneous English speech from Japanese learners and native speakers, we show that syntactic distance robustly predicts both pause location and duration across proficiency levels. We compare its contribution to proficiency classification against baseline and categorical models. The syntactic distance model outperforms all others, explaining 87% of variance (versus 65% for baseline and 76% for clause/phrase models), with strongest model fit and lowest prediction error. This measure provides a robust and meaningful predictor of L2 speech fluency.