The multidimensionality of second language oral fluency : The interface between cognitive, utterance, and perceived fluency

Suzuki, Shungo and Kormos, Judit (2021) The multidimensionality of second language oral fluency : The interface between cognitive, utterance, and perceived fluency. PhD thesis, Lancaster University.

Text (2021SuzukiPhD)
2021suzukiphd.pdf - Published Version
Available under License Creative Commons Attribution-NonCommercial-NoDerivs.
Download (8MB)

Abstract

In the context of the learning, teaching, and assessment of second language (L2) speaking skills, L2 fluency has been regarded as one of the important constructs. However, L2 fluency research has witnessed a long debate over the definition and measurements of L2 oral fluency; scholars have interchangeably used the term, “fluency”, with different connotations, such as speakers’ ability, speech features, and listeners’ perception. In order to distinguish different conceptualizations of fluency, Segalowitz (2010) proposed three subconstructs of fluency: utterance fluency (i.e., observable temporal features of speech), cognitive fluency (i.e., speaker’s ability to manipulate L2 knowledge efficiently), and perceived fluency (i.e., listener’s subjective judgements of fluency). However, it is still unclear how these three subconstructs of L2 fluency are interrelated with each other. The overarching goal of the thesis is to examine the construct of L2 oral fluency, particularly focusing on the interrelationship between cognitive, utterance, and perceived fluency. To this end, this thesis consists of four separate studies. Study 1 took a meta-analytic approach to synthesizing previous findings on the relationship between perceived and utterance fluency. Study 2 compared utterance fluency performance across speaking tasks which were designed to differ in the quality of speech processing demands, operationalized by task design features (i.e., task effects). Study 3 examined the contribution of cognitive fluency to utterance fluency, taking a structural equation modelling (SEM) approach. The study also analysed the stability of the factor structure of utterance fluency (Tavakoli & Skehan, 2005)—speed, breakdown, and repair fluency—and cognitive fluency across speaking tasks. Finally, Study 4 investigated the extent to which L2 utterance fluency can be predicted from L1 utterance fluency with regard to the moderator effects of L2 proficiency on the L1-L2 utterance fluency link. Study 1 collected 263 effect sizes from 22 studies reporting the correlation coefficients between listener-based judgements of fluency and objective measures of temporal features (N = 335–746). Among the pooled utterance fluency measures, Study 1 selected the common measures from four different categories: speed (articulation rate), breakdown (silent pause frequency, silent pause duration), repair (disfluency rate), and composite fluency (mean length of run, speech rate). Methodological moderator variables were selected with respect to the major phases of research into the utterance-perceived fluency link: Speech stimulus preparation (e.g., task type, target L2), Rater background (e.g., L1 vs. L2 listeners), Perceived fluency rating procedure (Definition of fluency, the number of point scales), and Utterance fluency measure calculation (length of pauses, manual vs. automated annotation). Studies 2–4 were conducted using the same dataset which included a set of cognitive and utterance fluency measures from Japanese-speaking learners of English (N = 128). Using a range of psycholinguistic tests, cognitive fluency was assessed in terms of linguistic resources and processing speed at different linguistic levels: vocabulary (vocabulary size, lexical retrieval speed), grammar (sentence construction speed and accuracy, grammaticality judgement speed and accuracy), and pronunciation (articulatory speed). In order to measure utterance fluency, speech data were elicited via four speaking tasks which differed in the quality of speech processing demands: argumentative task, picture narrative task, and text retelling tasks with/without read-aloud assistance. The speech data were analysed in terms of three subconstructs of utterance fluency (speed, breakdown, and repair fluency). The participants’ L1 fluency was also assessed, using another L1 argumentative speech task. Their proficiency scores were operationalized as two factor scores of cognitive fluency (linguistic resources and processing speed) in Study 3. Study 1 demonstrated that perceived fluency was strongly associated with speed and pause frequency (r = |.59–.62|), moderately with pause duration (r = |.46|), and weakly with repair fluency (r = |.20|), while composite measures showed the strongest effect sizes (r = |.72–.76|). A series of moderator analyses also revealed that the utterance-perceived fluency link may be influenced by methodological variables particularly related to speech stimulus preparation (target L2, task type, length of stimuli) and perceived fluency rating procedure (the definition of fluency presented to raters). Study 2 compared utterance fluency across four speaking tasks, using Generalized Linear Mixed-effect modelling (GLMM) with the tasks as a categorical fixed-effects predictor. The results showed that conceptualizing demands (content generation) increased the frequency of filled pauses, while the demands on formulation (activation of linguistic and phonological representations) had an impact on articulation rate, mid-clause pause ratio, and mid-clause pause duration. In Study 3, prior to an SEM analysis, a set of confirmatory factor analyses (CFA) demonstrated that utterance fluency has a three-factor structure (speed, breakdown, and repair fluency) and that cognitive fluency has a two-factor structure (linguistic resource and processing speed). An SEM analysis, based on these factor structures of cognitive and utterance fluency, showed that speed fluency was primarily associated with processing speed, while both linguistic resource and processing speed equally contributed to breakdown fluency. Repair fluency was significantly linked to linguistic resource, only when the content of speech was predefined (picture narrative and text summary tasks). Meanwhile, repair fluency was found to be independent of processing speed in all the speaking tasks. Study 4 examined the L1-L2 utterance fluency link using a set of GLMMs. The results suggested that all the L2 utterance fluency measures were predicted from their L1 counterparts. In addition, significant moderator effects of L2 proficiency on the L1-L2 fluency link were found only in speed fluency measures. The L1-L2 fluency link was weakened as a function of L2 linguistic resource but was strengthened as a function of L2 processing speed. The results of Study 1–4 confirmed that the relative importance of three subdimensions of utterance fluency—speed, breakdown, and repair fluency—can vary, depending on the perspective of assessment (perceived vs. cognitive fluency). These findings provide several practical implications for language assessment, such as the development of assessment tools and guidance for examiner training, as well as for L2 fluency learning and teaching.