Profiling French-speaking learners' productive (dis)fluency
(2017) International Conference on Fluency & Disfluency Across Languages and Language Varieties — Location: Louvain-la-Neuve, Belgium (15.February.2017)
Files
No attached file found for this publication.
Details
- Authors
- Abstract
- Over the past few decades, the notion of fluency has generated growing interest in the field of second and foreign language acquisition. In several recent studies (e.g. Bosker et al. 2013; Cucchiarini, van Doremalen & Strik 2010; Gilquin & Granger 2015), researchers have analysed the way learners of different proficiency levels or mother-tongue backgrounds make use of a number of devices such as filled and unfilled pauses, self-corrections or other speech management strategies. Despite the new insights that have emerged, however, there is still a lack of agreement among researchers regarding the scope of fluency: while a stream of research restricts the construct to the temporal aspects of speech (speech and articulation rate, pausal phenomena, etc.) (e.g. Ginther, Dimova & Yang 2010; Little et al. 2013; Zellner 1994), other conceptualisations are more far-reaching and encompass elements such as reformulations, false starts, discourse markers, formulaic sequences or pronunciation (e.g. Beliao & Lacheret 2013; Derwing, Thomson & Munro 2006; House 1996). In an attempt to get better insights into this kaleidoscope of components, a number of typologies have been put forward in the literature. Skehan and Tavakoli (Skehan 2003; Tavakoli & Skehan 2005; Tavakoli 2016), for instance, differentiate three subconstructs of fluency, namely speed, breakdown, and repair fluency. They argue that, while speed fluency relates to the speed of delivery and can be evaluated with speech rate measurements, breakdown fluency is concerned with the extent to which the speech flow is interrupted by pausal phenomena, and repair fluency includes phenomena that have to do with self-correction such as reformulations, replacements, false starts and repetitions. Notwithstanding the absence of consensus on the precise definition of fluency, it is increasingly accepted in the literature that fluency results from the conjunction of a number of quantifiable and qualifiable variables and is consequently better analysed as a bundle of features. To date however, many studies have been restricted to the investigation of one of those features at a time without considering its potential interaction with other fluency features, and even fewer studies have considered performance variations within a (seemingly) homogenous dataset. One notable exception is Götz (2013), who analysed a comprehensive set of features and attempted to delineate the “fluency profiles” of German learners of English. In her study, she showed that speakers do have individual preferences for some features and that it is possible to distinguish speaker groups that have different ways of achieving fluency. Against this backdrop, this paper investigates the separate contributions of a wide range of fluency features and examines how these correlate and interact with one another in the interlanguage of French learners of English. The features under investigation include the following: filled and unfilled pauses, restarts, false starts, repetitions, discourse markers and connectors, truncations, vowel lengthenings and foreign words as well as speech rate and mean length of pauses. The objectives are twofold. First, the analysis of the relationship between fluency features aims to investigate whether empirical corpus findings on fluency features support the taxonomy put forward by Skehan and Tavakoli, and more particularly the distinction between breakdown and repair fluency. In this respect, learner and native speaker data are also compared and contrasted in order to identify potential points of divergence. Second, as in Götz (2013) for German learners, the analysis seeks to reveal whether there are different fluency profiles corresponding to different speaker types among a comparable group of French learners. The study is based on the French component of the Louvain International Database of Spoken English Interlanguage (LINDSEI; Gilquin, De Cock & Granger 2010), which is a collection of 50 interviews of Belgian French-speaking university students of English as a foreign language of high-intermediate proficiency level. The native speaker data comes from LINDSEI’s native speaker counterpart LOCNEC (the Louvain Corpus of Native English Conversation; De Cock 2004) that, likewise, contains interviews with 50 British English undergraduate native speakers. In both corpora, each interview contains three speaking tasks, namely a warming-up activity on a set topic, a free discussion and a picture description task, totalling c. 10 hours of learner and native language. The corpora have been time-aligned (which allows for the precise measurement of temporal phenomena) and subsequently annotated with the EXMARaLDA tool (Schmidt & Wörner 2014) for a wide number of (dis)fluency features such as filled and unfilled pauses, truncations, repairs, false starts or repetitions. Results so far indicate that there are statistically significant, mostly moderate, correlations between a number of fluency features. These tend to support, although not perfectly, the distinction between breakdown and repair fluency. False starts, restarts, truncations and repetitions, for example, appear to be mutually related, while filled and unfilled pauses seem to form a separate group. The relationship between the two types of pauses however differs depending on the speaker group: whereas they are highly significantly correlated (r = .45; p < .005) in the native speaker data, it is not the case in LINDSEIFR. The analysis also reveals that the relationship between the two aspects of fluency, i.e. breakdown and repair, might be more complex than commonly assumed: in the data, filled pauses indeed correlate either negatively with false starts (r = -.3; p < .05; LINDSEI-FR) or positively with repetitions (r = .4; p < .005; LOCNEC) – and this finding also highlights an interesting difference between learner and native speaker fluency behaviour. Preliminary results from a hierarchical cluster analysis further show that not all the learners from the dataset behave similarly and that it is possible to identify different clusters among the learners’ fluency performances.
- Affiliations
Citations
Dumont, A. (2017). Profiling French-speaking learners’ productive (dis)fluency. International Conference on Fluency & Disfluency Across Languages and Language Varieties, Louvain-la-Neuve, Belgium. https://hdl.handle.net/2078.5/175733