Given the essential role of oral fluency in second language (L2) pedagogy and assessment, scholars have examined L2 oral fluency as one of the primary constructs of oral proficiency (Tavakoli & Hunter, 2018; Tavakoli and Wright, 2020). Previous studies have attempted to identify temporal characteristics of speech that play an essential role in listeners' perception and proficiency assessment, by investigating oral fluency from three different perspectives (Segalowitz, 2010, 2016)-listener-based perceptions (i.e., perceived fluency), objective temporal features (i.e., utterance fluency) and underlying linguistic knowledge (i.e., cognitive fluency). Although these lines of fluency research have offered insights into the construct definition and valid operationalization of L2 oral fluency, L2 fluency research has suffered from the lack of research looking into fluency in dialogic speaking tasks. Fluency in dialogic speaking tasks can be a theoretically different construct from the one in monologic tasks (Peltonen, 2021; Suzuki et al., 2021; Tavakoli, 2016). In addition, to offer insights from previous studies into setting realistic curricular objectives for speaking skills, it is essential to understand which aspects of oral fluency tend to be developmentally ready according to different proficiency levels (cf. Baker-Smemoe et al., 2014; Tavakoli et al., 2020). Taken together, the current study aims to fill this gap in L2 fluency research, examining what temporal and dialogic features can differentiate between the CEFR levels of fluency, using interactional speech data.
A total of 80 Japanese learners of English were recruited, and they completed an oral proficiency interview consisting of seven topics with varying levels of difficulty, such as social media and globalization, determined by the CEFR manual (Council of Europe, 2018). All the interview sessions were conducted via the video conferencing tool, Zoom. Their interview data were annotated for a range of disfluency features, including silent pauses, filler, and self-repair. We calculated a comprehensive set of utterance fluency measures, following previous studies (De Jong et al., 2013; Suzuki et al., 2021; Tavakoli et al., 2020). Three raters received a training session about using the CEFR descriptor of oral fluency and then independently assigned fluency scores (A1 to C2) to each speaker. We employed a Rasch analysis to determine students' CEFR level while controlling for the scoring variability across raters. As a result, our participants' CEFR levels ranged between A2 to C1 levels.
A series of Bayesian ANOVAs showed the main effects of proficiency levels on most temporal and dialogic features, whereas end-clause pause duration and between-turn pause duration may not differ across four CEFR levels. The results of post-hoc tests indicated that articulation rate may distinguish higher proficiency levels (B1 vs. B2, B2 vs. C1), while mid-clause pause ratio can differ between all the adjacent levels from A2 to C1 level. Meanwhile, between-turn pause ratio and mean length of turn only differed between lower levels of proficiency (A2 vs. B1). We will discuss these findings in relation to a potential developmental pattern of dialogic fluency performance in light of L2 speech production mechanisms.
