BACKGROUND AND MOTIVATION
A primary issue for research on incidental SLA is how to optimize learners’ ability to use language in conjunction with communicative task performance. Definitions of such tasks have varied, but for the purpose of the present study, the term task will be used, following Ellis (Reference Ellis2009), to refer to second language (L2) learning activities that meet four criteria:
-
1. There is a primary focus on meaning.
-
2. There is a gap that necessitates communication.
-
3. Learners must draw on their own resources to complete them.
-
4. There is a communicative outcome beyond the use of language for its own sake.
When these criteria are met, tasks can be argued to play an important role in L2 pedagogy in that they provide learners with opportunities to employ their linguistic repertoire under relatively natural conditions.
One factor in the implementation of such tasks that has been argued to play a crucial role in optimizing language use and that has received considerable theoretical and empirical attention in the L2 literature is the planning processes that learners engage in in conjunction with the performance of these tasks (see Pang & Skehan, Reference Pang, Skehan and Skehan2014, for a recent overview). Ellis (Reference Ellis and Ellis2005) identifies two points in the pedagogic process at which opportunities for planning might be manipulated. The first is before the task begins. At this stage, learners might either be allowed time to plan the language or content that they will use during the task (either in the first language [L1] or in the L2) or be given the opportunity to rehearse their performance of the task. In the latter case, they complete a sequence of tasks in which they either repeat the exact same task or carry out a parallel version of the task with slightly different content (Bygate, Reference Bygate, Bygate, Skehan and Swain2001). The second point at which learners’ opportunity to plan their performance might be controlled is during the task performance itself. Either learners can be allowed unlimited time to complete the task, or time limits might be imposed on their performances (Maurice, Reference Maurice1983; Nation, Reference Nation1989).
In line with the definition of task outlined previously, however, task repetition might be argued to necessitate a different interlocutor each time the task is repeated. A change in interlocutor preserves task integrity by requiring learners to create original meanings each time the task set is performed. In the present study, for example, participants performed monologue tasks in pairs, alternating as speaker and listener with a different interlocutor each time, as they might if they were repeating the tasks in a classroom or real-world setting. This approach could be argued to add to the external validity of the repetition process in that the exposure that learners receive involves both input-based and output-based versions of the task. The construct of task repetition as it is operationalized in the present study thus represents a combination of task repetition and interlocutor input. It is thus referred to as aural-oral task repetition.
Task repetition of this sort might help learners improve their performances by allowing them to activate, refine, and optimize their linguistic resources for the purpose of successfully completing a given task in line with the specific communicative demands that it entails. Empirical studies on task repetition have asked learners to repeatedly engage in exactly the same task (e.g., Ahmadian & Tavakoli, Reference Ahmadian and Tavakoli2011; Bygate, Reference Bygate, Willis and Willis1996, Reference Bygate1999, Reference Bygate, Bygate, Skehan and Swain2001), in the same type of task with slightly different content (Gass, Mackey, Alvarez-Torres & Fernandez-Garcia, Reference Gass, Mackey, Alvarez-Torres and Fernandez-Garcia1999), or in a task that has the same communicative goal but different content and interlocutors each time (Lynch & Maclean, Reference Lynch and Maclean2000, Reference Lynch, Maclean, Bygate, Skehan and Swain2001). Several empirical studies have also examined the effects of repetition in conjunction with decreasing the time available on consecutive performances (e.g., Arevart & Nation, Reference Arevart and Nation1991, Reference Arevart, Nation and Oller1993; de Jong & Perfetti, Reference De Jong and Perfetti2011; Nation, Reference Nation1989). Furthermore, repetition effects have been investigated immediately (e.g., Lynch & Maclean, Reference Lynch and Maclean2000, Reference Lynch, Maclean, Bygate, Skehan and Swain2001; Wang, Reference Wang and Skehan2014) or after days (Gass et al., Reference Gass, Mackey, Alvarez-Torres and Fernandez-Garcia1999) or weeks (e.g., Ahmadian & Tavakoli, Reference Ahmadian and Tavakoli2011; Bygate, Reference Bygate, Willis and Willis1996, Reference Bygate1999, Reference Bygate, Bygate, Skehan and Swain2001). Finally, speakers have been asked to repeat their performance once (e.g., Bygate, Reference Bygate, Willis and Willis1996, Reference Bygate1999, Reference Bygate, Bygate, Skehan and Swain2001; Wang, Reference Wang and Skehan2014) or several times, with the number of times ranging from 3 or 4 (Gass et al., Reference Gass, Mackey, Alvarez-Torres and Fernandez-Garcia1999) to 11 (Ahmadian, Reference Ahmadian2011).
Of the different forms of planning proposed by Ellis (Reference Ellis and Ellis2005), rehearsal (or task repetition) has been shown to have the most robust effects on L2 fluency (Ahmadian & Tavakoli, Reference Ahmadian and Tavakoli2011; Arevart & Nation, Reference Arevart and Nation1991, Reference Arevart, Nation and Oller1993; Bygate, Reference Bygate, Bygate, Skehan and Swain2001; Lynch & Maclean, Reference Lynch and Maclean2000; Nation, Reference Nation1989; Wang, Reference Wang and Skehan2014). Fluency is generally defined as skilled L2 performance, referring to rapid, smooth, and accurate communication of one’s intentions during online processing (Lennon, Reference Lennon and Riggenbach2000, p. 26). Fluent performance thus entails the efficient functioning of speech production processes under the constraints of real-time oral interaction.
The modular model of L1 speech production proposed by Levelt (Reference Levelt1989, Reference Levelt, Brown and Hagoort1999) and adapted to L2 speech production by de Bot (Reference De Bot1992) and Kormos (Reference Kormos2006) postulates three primary stages of speech production. The first is a conceptualization stage in which the speaker selects information from world knowledge to include in a message and organizes it into an information structure to create a preverbal plan. The second is a complex set of procedures referred to as a formulation stage, in which the preverbal plan is encoded grammatically and phonetically. Concepts and their relational structure are argued to be projected onto a phrase structure driven by lemmas drawn from the speaker’s mental lexicon that have associated semantic, syntactic, morphological, and phonological properties. This phrase structure is then encoded with phonetic and prosodic information to produce a phonetic plan. The third and final stage of speech production is then articulation, in which the phonetic plan is buffered and parsed as syllables at the motor level.
The essence of this model, as it relates to L2 fluency, is that these three stages are assumed to operate in parallel. Although each module is expected to work on specific input and generate specific output for the next module, they are hypothesized to operate simultaneously, provided that processing in the parallel module is sufficiently automatic (Levelt, Reference Levelt1989, Reference Levelt, Brown and Hagoort1999). For proficient speakers, formulation may be largely automatic and allow for parallel processing with other modules, whereas conceptualization and monitoring may both require attention and thus rely on serial processing. For lower proficiency speakers, however, lexical retrieval and grammatical encoding might also require attentional resources and hence might result in varying degrees of breakdown in parallel processing (de Bot, Reference De Bot1992; Kormos, Reference Kormos2006). Second language learners thus face a number of challenges in the processes leading from conceptualization to articulation, one of the most important being limitations in attention and working memory capacity, which are essential in managing speech production (Kormos, Reference Kormos and Robinson2011). The fluency of L2 performance reflects the efficient functioning of speech production mechanisms, including the automaticity of encoding processes, the conceptual demands of planning the content of the message, and the skillful handling of performance breakdowns. This aspect of fluency is referred to as cognitive fluency by Segalowitz (Reference Segalowitz2010).
Cognitive fluency can be inferred from the analysis of utterance fluency, which designates the temporal variables of speech or the “oral features of utterances that reflect the operation of underlying cognitive processes” (Segalowitz, Reference Segalowitz2010, p. 48). In task-based studies, utterance fluency is often argued to consist of three subconstructs: breakdown fluency, repair fluency, and speed fluency (for recent discussions, see Bosker, Pinget, Quené, Sanders, & De Jong, Reference Bosker, Pinget, Quené, Sanders and De Jong2013; De Jong, Steinel, Florijn, Schoonen, & Hulstijn, Reference De Jong, Steinel, Florijn, Schoonen and Hulstijn2012). Breakdown fluency is generally considered to be most accurately measured by pause frequency, repair fluency is usually operationalized as the frequency of overt repairs or reformulations in L2 learners’ speech (see De Jong et al., Reference De Jong, Steinel, Florijn, Schoonen and Hulstijn2012; Götz, Reference Götz2013), and speed fluency is typically assessed by speech rate or the number of pruned syllables uttered per second (Ellis & Barkhuizen, Reference Ellis and Barkhuizen2005). Although it is difficult to relate these fluency measures to a specific speech production mechanism, it can be argued that filled and unfilled pauses between clauses are indicators of conceptualization and content planning, whereas filled and unfilled pauses within clauses signal breakdowns in lexical and syntactic encoding (Butterworth, Reference Butterworth1975), and the frequency of overt self-repairs might be argued to reflect the availability of attentional resources for monitoring (Kormos, Reference Kormos1999). Finally, speech rate is an overall measure of the speed with which an utterance is produced (see Götz, Reference Götz2013).
Particularly relevant to the present study is Wang’s (Reference Wang and Skehan2014) research, which compared various approaches to providing learners with opportunities to plan and found much larger effect sizes for repeating a task once than for other types of planning (see also Skehan, Xiaoyue, Quian, & Wang, Reference Skehan, Xiaoyue, Quian and Wang2012). In Wang’s (Reference Wang and Skehan2014) study, speech rate increased and the average length of unfilled pauses at clause boundaries decreased, whereas the average length of unfilled pauses within clauses and the frequency of reformulations were unaffected by repetition. The question arises as to why this might have been the case and whether subsequent repetitions of the task might not have improved other aspects of learners’ L2 fluency. Wang’s findings suggest that one task repetition enhances the speed with which students can deliver their message and also reduces the time needed to pause at clause boundaries for generating ideas. Nevertheless, subsequent repetitions may be required before learners can acquire sufficient access to L2 resources (lexis and syntax) to avoid breakdowns in speech processing, as reflected in the frequency of midclause pauses, and to devote attention to monitoring their output more carefully, as reflected in the frequency of self-repairs and reformulations.
Based on Levelt’s (Reference Levelt1989, Reference Levelt, Brown and Hagoort1999) speech production model, Bygate (Reference Bygate, Bygate, Skehan and Swain2001) argues that when L2 learners perform a task for the first time, their speech production system needs to execute all the relevant processing steps under time pressure. Hence, based on the limited attentional model of speech production (Kormos, Reference Kormos2006; Skehan, Reference Skehan2009), it could be argued that on the first performance of a given task, learners have to distribute their attentional resources strategically among conceptualization, formulation, and monitoring. They also have to handle performance breakdowns due to incomplete lexical or syntactic knowledge representation or slow speed of access to these representations (Dörnyei & Kormos, Reference Dörnyei and Kormos1998). Task repetition allows L2 learners to rely on previously conceptualized task content and to activate recently used linguistic constructions to express their message. This might reduce the attentional demands on learners to conceptualize, encode, and monitor their messages simultaneously.
The effects of task repetition, however, might vary depending on the speech production demands of tasks. Communicative tasks can differ in terms of the load they pose on working memory in various stages of speech production (Kormos, Reference Kormos2006; Skehan, Reference Skehan2009). Opinion tasks, for example, might require increased attention to conceptual planning, either at the expense of attention to linguistic encoding or in addition to it. By contrast, tasks in which learners have to convey familiar information—or information provided visually, such as personal or picture-based narrations or instructions—might pose lower attentional demands during conceptualization and allow learners to reallocate attentional resources to linguistic encoding. In studies of task repetition, it is therefore important to examine the extent to which the hypothesized benefits of task repetition vary across tasks with differing demands.
The effects of task repetition on speech processing might also vary depending on the proficiency level of the speakers who complete the tasks. As the automaticity of linguistic encoding is strongly associated with L2 proficiency (Segalowitz, Reference Segalowitz2010), higher proficiency L2 speakers might have more attentional resources available for conceptualizing task content. For this reason, task repetition might result in fewer gains for more advanced learners as one of its primary functions is to allow learners to activate relevant content and linguistic formulation processes. Furthermore, language proficiency may interact with task demands such that higher proficiency learners, for example, would improve as a result of repeating tasks that are high in conceptualization demands, but not on those in which conceptual planning requires less attention. In addition to task demands, it is thus necessary to consider the possible role of proficiency level in moderating the effects of task repetition. To our knowledge, no previous research has investigated how task characteristics and learner proficiency, independently and in interaction with each other, affect fluency gains in repeated task performance.
THE STUDY
It could thus be argued, on the one hand, that performing tasks multiple times within a relatively short time interval is beneficial to the development of learners’ speech processing ability. Most empirical work, however, has only investigated learners’ performance over a small number of task repetitions (for exceptions, see Ahmadian, Reference Ahmadian2011, Lynch & Maclean, 2000, 2001). To date, little is known about how many repetitions are actually needed for learners to optimize the different aspects of their L2 fluency in the short term and how this might vary with task type and learner proficiency level. On the other hand, task repetition might also result in boredom and fatigue (Bygate, Reference Bygate, Bygate, Skehan and Swain2001). It is thus also important to investigate learners’ reactions to repeating tasks and their perception of the value of such repetition. The present study attempts to address both of these issues by considering language production together with the reactions of 32 Japanese learners of English at three proficiency levels to performing three types of tasks six times each within the time frame of a typical L2 lesson.
The study investigates the following research questions:
-
1. How does repeating the same task as speaker and listener (i.e., aural-oral task repetition) with different interlocutors affect L2 fluency in the short term?
-
2. Do the demands of different tasks moderate the effects of aural-oral task repetition on L2 fluency?
-
3. Do speakers’ proficiency levels moderate the effects of aural-oral task repetition on L2 fluency?
-
4. What are the learners’ perceptions regarding the value of aural-oral task repetition, the optimum number of repetition times, and the extent to which repetition brings about boredom or fatigue?
METHODS
Design
The study employed a three-way repeated-measures design. The first within-subjects factor was task repetition at six levels, and the second was task type at three levels (instruction, narration, and opinion). The third between-subjects factor was proficiency level, which was established based on band levels of the TOEIC test (Educational Testing Service, 2008). The four dependent variables in the study were as follows: (a) speech rate, operationalized as the number of syllables per second based on pruned transcriptions of oral discourse; (b) clause-final pausing, operationalized as the ratio of filled pauses between clause boundaries to the total number of pruned syllables produced; (c) midclause pausing, operationalized as the ratio of filled pauses within clause boundaries to the total number of pruned syllables produced; and (d) overt self-repairs, operationalized as the total number of self-corrections or rephrasing to the total number of pruned syllables produced. Postperformance questionnaires were also used to triangulate the findings for speech production, gain deeper insight into the range of variables that might be involved in task repetition, and better understand the responses of learners at different levels of proficiency to repeating the tasks used in the study.
Participants
Thirty-two Japanese English learners, ranging from 18 to 23 years of age, at a large public university in southern Japan participated in the study. These learners were sampled from the entire population of English learners at the university at three levels of proficiency based on their TOEIC scores (Educational Testing Service, 2008). The high-level group was roughly equivalent to the B2 and C1 levels on the Common European Framework of Reference (CEFR; Council of Europe, 2001); the mid-level group to the B1 level on the CEFR; and the low-level learners to the A2 level on the CEFR. The participants’ TOEIC scores were approximately normally distributed both within and across the three proficiency bands (see Table 1).
Descriptive statistics for TOEIC scores across groups
Materials
Based on Yule’s (Reference Yule1997) typology of referential communication tasks, three tasks differing in discourse genre (instruction, narration, and opinion) were used in the study. The instruction task presented participants with a traditional problem in which a farmer has to get a fox, a chicken, and some wheat across a river, but he has a boat big enough only for himself and one other thing, and he cannot leave the fox and the chicken alone together, nor can he leave the chicken and the wheat alone together. It also provided an eight-picture sequence that illustrated how to solve the problem and asked them to explain the solution six times to different interlocutors. The narration task was based on a four-frame picture story taken from the Pre-1 Level of the EIKEN English Test (Nihon Eigo Kentei Kyokai, 2009), which involved a group of shop owners arriving at a solution to a local graffiti problem and asked participants to narrate the story to these six interlocutors. Finally, the opinion task presented the participants with two photographs from a family trip to a zoo and asked them to give their opinions on what they saw in the pictures, what they thought was happening, who they thought took them, and why they thought they were taken to the six interlocutors. In each case, participants were given a one-sentence prompt with which to begin in order to establish the respective discourse frames.
In addition to the inherent differences in discourse demands among the instruction, narration, and opinion genres (for discussions, see Berman, Reference Berman2008; Yule, Reference Yule1997), the format of the tasks could also be argued to differ in terms of the amount of conceptual and linguistic structure that they provided. The instruction and narration tasks had a clear sequential information structure. In the instruction task, learners were given pictures that supplied the specific content to complete the task, and each successive step allowed them to recycle the same lexis and sentence frames. The narration task was similar in that it provided learners with the specific content to be related, but it allowed far less repetition of language from one frame of the story to the next. Finally, the opinion task required learners to supply the content required to complete the task themselves (i.e., what they saw in the pictures, what they thought was happening, who they thought took them, and why they thought they were taken). In other words, the opinion task did not provide a clear information structure in the way that the instruction and narration tasks did (see Tavakoli & Foster, Reference Tavakoli and Foster2008, on the effects of content structure on L2 speech performance). The open nature of the opinion task might also be expected to result in less sense of a clear outcome.
Finally, two questionnaires were used in the study. The first was a brief background questionnaire that focused on English language experience, and the second was a postperformance questionnaire on learners’ thoughts on the value of repeating each of the tasks. For each of the three tasks, participants were asked (a) whether and in what ways they felt their performance improved as a result of repetition, (b) whether their performance improved through the sixth performance and, if not, how many repetitions were enough, and (c) whether and in what ways it was useful to complete the task with multiple partners.
Procedures
Advertisements were circulated to all English learners at the university, offering payment in cash for participation in a 2-hr English conversation research project. Participants were accepted in the order in which they applied and were allocated to groups based on their proficiency level and availability. They all received information about the aims of the research and signed a consent form.
In addition to the three monologue tasks reported in the present study (instruction, narration, and opinion), a dialogue opinion task was also included in a balanced Latin square design when the data were collected. The order in which these four tasks were performed was counterbalanced into four task sets, which were then performed six times each. In other words, the 32 participants were allocated to one of four groups, and each group of eight performed one task set six times during a separate recording session.
Each group of eight participants then worked in pairs. They performed each task in the task set twice (once as the speaker and once as the listener). Speaker and listener roles were alternated after each task performance so that each task in the set was completed by both participants before they moved on to the next one. When participants had finished all four tasks with one partner, they switched partners and repeated the same process with another partner until they had performed the task set six times with different partners each time. Participant pairing was controlled across the design so that each participant worked with partners from each of the three proficiency levels. To accomplish this, each of the eight participants in a given session was given a letter or a number (A-B-C-D-1–2–3–4). They were then systematically alternated in terms of partner and speaker-listener order across the six performances of their task set. Table 2 indicates the specific pairings of learners and the order of speakers used in the study.
Pairing and ordering of participants for each performance of the task set
It can be seen in Table 2 that proficiency level was generally distributed across the design. Each participant worked with six of the seven other participants in the study, but the order in which they worked with speakers of different levels varied across the design. Furthermore, the order in which the participants performed initially in the role of speaker or listener in each pair was alternated across the design so that each participant initially functioned as the speaker or listener a comparable number of times across the design as a whole. However, interlocutor factors (proficiency and order) could not be counterbalanced perfectly. Even if they had, interlocutor effect would have been distributed across the design rather than controlled. The design used in the study thus combines the effect of task repetition with that of interlocutor input. The advantages and disadvantages of the design are considered in the Discussion and Conclusions section of the article.
Finally, participants were given 120 s to complete each task for the first and second repetitions. Performance time was reduced to 90 s for the remaining four repetitions as piloting in classes with similar learners had shown that this time was adequate for completing the tasks at their own pace. This was confirmed in the performances recorded for the project. Most participants finished slightly before the time had expired. Thus, the reduction in time after the second performance in the sequence cannot be argued to have resulted in a pressured performance, reducing or eliminating during-task planning time, as participants generally did not use all of the time that they had available.
Each participant thus performed the three monologue tasks in the role of the speaker 18 times (6 performances of the 4 task types) and in the role of the listener 18 times within a period of approximately 90 min, or the length of a typical university-level English class.
Analysis
The study is based on 576 task performances constituting approximately 16 hr of oral data. Each of the 32 participants produced approximately 30 min of speech (18 task performances). These performances were transcribed into Analysis of Speech (AS) units following the procedures of Foster, Tonkyn, and Wigglesworth (Reference Foster, Tonkyn and Wigglesworth2000). Speech rate was then calculated as the number of pruned syllables per second from the first syllable after the prompt sentence to the end the task performance (Ellis & Barkhuizen, Reference Ellis and Barkhuizen2005).
As automated analysis of unfilled pause data proved impossible due to background noise in the classroom, the frequency of filled pauses was counted manually. Filled pauses were identified as “non-lexical fillers such as er and uhm and elongations of sounds (drawls)” (Gut, Reference Gut2009, p. 80). The ratio of clause-final and midclause pauses to total pruned syllables was then calculated for each performance.
The number of overt self-repairs during each performance was also counted manually. Overt self-repairs were defined as self-initiated, self-completed corrections when the speaker halts the speech flow and executes a modification (Kormos, Reference Kormos1999, p. 313). Footnote 1 The ratio to total pruned syllables was then calculated (see the Appendix for an example of the data analysis).
Of the 576 task performances, 36 (6.25%) were selected across tasks and speakers and were double coded for pause and self-repair frequencies by the second author and a research assistant. Two-tailed Pearson correlations revealed intercoder reliability to be very high for clause-final pauses (r = .956, n = 36, p < .0001), midclause pauses (r = .952, n = 36, p < .0001), and self-repairs (r = .966, n = 36, p < .0001). The research assistant then coded the remainder of the performances.
Preliminary screening of the data revealed some positively skewed score distributions, particularly in the case of the two pause measures. This was corrected with square root transformation. Following transformation, a comparison of skew and kurtosis values with their standard errors revealed no z scores higher than +/−1.96, and the data were thus assumed to be approximately normally distributed for the purpose of the statistical analyses used in the study. In addition, Levene’s tests for each distribution revealed acceptable homogeneity of variance (p < .05) on each of the variables in the study. A multivariate analysis of variance (MANOVA) was then conducted using SPSS22 for Windows followed by post hoc univariate analyses and pairwise comparisons with Fischer’s LSD tests to test the null hypotheses that there were no statistically significant differences in speech rate, clause-final pausing, midclause pausing, or self-repair among (a) the six levels of repetition, (b) the three task types, and (c) the three proficiency levels, and (d) that there were no significant interactions among these variables individually or in combination. Cohen’s d (Cohen, Reference Cohen1988) was then calculated to determine the effect sizes of the relevant pairwise comparisons. Cohen recommends interpreting effect sizes above d = .2 as small, above d = .5 as medium, and above d = .8 as large, and these benchmarks have provided the standard for interpreting effect sizes in social science research. However, based on a comprehensive survey of research in the field of SLA, Plonsky and Oswald (Reference Plonsky and Oswald2014) argue that Cohen’s scale underestimates the range of effects typically obtained in L2 research, and they recommend field-specific benchmarks of small (d = .40), medium (d = .70), and large (d = 1.00) in interpreting effect sizes in SLA research. We have adopted the latter criteria for interpreting the effect sizes of the pairwise comparisons in the present article.
RESULTS
Using Pillai’s trace, multivariate analysis of variance revealed a significant main effect and a very large effect size for aural-oral task repetition on L2 speech fluency, V = 0.957, F(20, 10) = 11.163, p = .000, partial η2 = .957. Cohen (Reference Cohen1988; see also Tabachnick & Fidell, Reference Tabachnick and Fidell2007), for example, recommends using .01, .09, and .25 in interpreting small, medium, and large effects of partial η2, respectively. Furthermore, this analysis revealed no significant interactions among this form of repetition, task type, and proficiency level individually or in combination with small effect sizes in all of these comparisons: (a) repetition × proficiency, V = 0.280, F(40, 580) = 1.092, p = .326, partial η2 = .070; (b) repetition × task, V = 0.150, F(40, 1,160) = 1.130, p = .268, partial η2 = .038; and (c) repetition × proficiency × task, V = 0.244, F(80, 1,160) = .940, p = .627, partial η2 = .061. This indicates that the overall effects of aural-oral task repetition were generally robust for speakers of different proficiency levels completing tasks with differing discourse demands. Finally, follow-up univariate analyses revealed that this repetition effect was statistically significant across all four aspects of L2 fluency, but with a very large effect size for speech rate, and medium effect sizes for the other pause and self-repair phenomena observed: (a) speech rate, F = 122.092, p = .000, partial η2 = .808; (b) clause-final filled pauses, F = 6.306, p = .000, partial η2 = .179; (c) midclause filled pauses, F = 9.243, p = .000, partial η2 = .242; and (d) self-repair, F = 3.072, p = .019, partial η2 = .096. In the following sections, pairwise comparisons with Fischer’s LSD tests are discussed in order to elucidate the specific nature of the repetition effect on four aspects of L2 fluency.
Speech Rate
Pairwise comparisons revealed that the aural-oral repetition effect on the rate of L2 speech production was due to differences between all performances (1–2: p = .000, d = 1.53; 2–3: p = .000, d = 1.48; 3–4: p = .016, d = 0.48; 4–5: p = .000, d = 0.84) except in the case of Performances 5 and 6, for which the difference was not significant (p = .943, d = 0.05). In other words, the effect of immediate aural-oral task repetition continued to increase significantly through the fifth performance of the tasks in the study. However, the effect size was large according to the criteria posited by Plonsky and Oswald (Reference Plonsky and Oswald2014) across the first three performances only and was small to medium thereafter through the fifth performance (see Figure 1 and Table 3). Not only did gains in speech rate begin to level off after the third repetition, but the pattern of development also began to diverge across the task types, with the opinion task showing a less linear pattern of development than either the instruction or the narration task (see Figure 2 and Table 3). Furthermore, the pattern observed for this form of task repetition on speech rate was very consistent across all three proficiency levels (see Figure 3 and Table 3). The results thus indicate a robust connection between task repetition and speech rate and point to the possibility of a ceiling effect for immediate practice on speech rate at five repetitions.
Repetition and speech rate.
Speech rate means and standard deviations across tasks, performances, and proficiency levels (n = 32)
Repetition, task, and speech rate.
Repetition, proficiency, and speech rate.
Clause-Final Pausing
Pairwise comparisons revealed that the frequency of pauses at clause boundaries decreased significantly from the first to the second performance only (p = .047, d = 0.40) (cf. Wang, Reference Wang and Skehan2014). The differences between subsequent repetitions did not reach statistical significance at the .05 level (2–3: p = .146, d = 0.21; 3–4: p = .071, d = 0.36; 4–5: p = .921, d = 0.03; 5–6: p = .576, d = 0.08). The effect of aural-oral task repetition on clause-final pausing was relatively small according to Plonsky and Oswald (Reference Plonsky and Oswald2014) and limited to the difference between the first two task performances (see Figure 4 and Table 4). Furthermore, there was considerable variation in clause-final pausing across tasks. The largest gains were between the second and third repetition for the opinion tasks and between the first and second repetition of the narration task and were negligible on all repetitions of the instruction task (see Figure 5 and Table 5). However, we see a generally descending pattern until the fourth performance on all task types, after which nonlinear patterns of development begin to emerge (see Figure 5). Finally, comparison of clause-final pauses across the three proficiency levels revealed that the two higher proficiency groups showed a consistent pattern of decreased pausing across their first four performances of the task set (see Figure 6 and Table 4). This pattern is consistent with the pattern in Figure 4. The lowest proficiency speakers, on the other hand, showed fluctuating progress after their second performance of the task set (see Figure 6 and Table 4).
Repetition and clause-final pausing.
Clause-final pausing means and standard deviations across tasks, performances, and proficiency levels (n = 32)
Repetition, task, and clause-final pausing.
Midclause pausing means and standard deviations across tasks, performances, and proficiency levels (n = 32)
Repetition, proficiency, and clause-final pausing.
Midclause Pausing
Post hoc pairwise comparisons showed weak effects for aural-oral task repetition on midclause filled pauses. None of the differences between any two consecutive performances were statistically significant at the .05 level (1–2: p = .095, d = 0.52; 2–3: p = .110, d = 0.34; 3–4: p = .272, d = 0.13; 4–5: p = .307, d = 0.14; 5–6: p = .840, d = 0.04). Although the effect size for the first repetition was larger than the others, it was still small according to the benchmarks proposed by Plonsky and Oswald (Reference Plonsky and Oswald2014). On the other hand, the difference between every second performance up to the fifth performance did reach statistical significance (1–3: p = .001, d = 0.71; 2–4: p = .003, d = 0.50; 3–5: p = .049, d = 0.30), whereas the difference between Performances 4 and 6 did not (p = .595, d = 0.12). Thus, these stepwise effects of aural-oral task repetition on midclause pausing did not dissipate until around the fourth performance, when the effect size decreased from medium to small (Plonsky & Oswald, Reference Plonsky and Oswald2014) (see Figure 7 and Table 5). Nevertheless, the effects observed in Figure 7 were generally representative of participants’ performance on all three of the tasks (see Figure 8 and Table 5). With the exception of the second performance of the narration task, we see a clearly descending pattern through the third performance on all task types, after which midclause pausing behavior begins to fluctuate (see Figure 8). Furthermore, comparison of the performance of the three proficiency levels reveals that the initial effect of aural-oral repetition on midclause filled pauses was most pronounced between the lowest-proficiency-level learners’ first and second performances of the task set, after which all groups show a generally descending pattern in midclause pausing until the fourth performance (see Figure 9 and Table 5).
Repetition and midclause pausing.
Repetition, task, and midclause pausing.
Repetition, proficiency, and midclause pausing.
Self-Repair
Finally, pairwise comparisons revealed that the frequency of self-repairs and reformulations followed a pattern that was quite distinct from either pause phenomenon. None of the differences between any combinations of the first four performances reached statistical significance at the .05 level (1–2: p = .922, d = 0.02; 2–3: p = .356, d = 0.14; 3–4: p = .614, d = 0.16; 4–5: p = .216, d = 0.14; 5–6: p = .980, d = 0.08). However, the differences between the fifth repetition and the first two performances (1–5: p = .017, d = 0.45; 2–5: p = .01, d = 0.48) and the differences between the sixth performance and the first three performances (1–6: p = .017, d = 0.50; 2–6: p = .015, d = 0.51; 3–6: p = .020, d = 0.49) were significant, although the effect sizes were small according to Plonsky and Oswald (Reference Plonsky and Oswald2014). Thus, significant gains in L2 fluency due to decreased self-repair (see Figure 10 and Table 6) did not begin until decreases in both clause-final and midclause pausing had dissipated. Likewise, when participants’ performance on each task is compared, production on the instruction and narration tasks followed a similar pattern throughout the first five repetitions, but their performance on the opinion task showed a sudden decrease in fluency due to an increase in self-repair during Performance 4 (see Figure 11).
Repetition and self-repair.
Self-repair means and standard deviations across tasks, performances, and proficiency levels (n = 32)
Repetition, task, and self-repair.
However, there was considerable variability among proficiency levels with respect to self-repair (see Figure 12 and Table 6). The frequency of self-repairs is stable between the first two performances of the task set for all groups when clause-final pausing effects are significant (see Figure 4) and is relatively stable through the third performance when midclause pausing effects are the largest (see Figure 7). However, there is considerable fluctuation in self-repair behavior among groups in the latter performances of the task set when significant differences in self-repair behavior begin to emerge, indicating that self-repair may be a fluency variable that is dependent to some extent on proficiency.
Repetition, proficiency, and self-repair.
Speaker Perceptions
Participants’ impressions were elicited regarding (a) the value of repeating each task type, (b) the optimal number of task repetitions for each task type, and (c) the value of repeating tasks with different partners. Table 7 summarizes the results of this analysis.
Perceptions of task repetition
Overall, participants felt repetition to be helpful. Nearly all of them reported that repeating the task was useful on the opinion and narration tasks, and three quarters of them indicated that it was beneficial on the instruction task. The majority of their comments fell into two categories. Approximately 40% mentioned that repetition improved their fluency on the tasks. Furthermore, between a quarter and half of the learners (24% on the instruction task, 35% on the narration task, and 48% on the opinion task) indicated that repetition assisted them in recalling useful words and expressions or that they were able to learn useful words and expressions from their partners. It is interesting to note that usefulness in terms of incorporation was felt to be the most relevant for the opinion task, in which task content was open, and lowest for the instruction task, in which the task was structured and relatively inflexible, with the narration task falling in between (cf. Tavakoli & Foster, Reference Tavakoli and Foster2008).
Other comments did not fit into these two categories but provided information on participants’ perceptions of the tasks and other situational variables. One participant, for example, said that speaking with a high-proficiency interlocutor helped her improve her English speaking skills. Only 3 of the 32 participants (9.4%) expressed boredom after the fourth performance. On the opinion and instruction tasks, all three were in the high-level group. On the narration task, one participant from each group expressed boredom. In addition, a few participants recalled that during the later repetitions, they stopped incorporating new linguistic constructions into their performance.
Table 8 summarizes participants’ responses with regard to how they felt about the number of times they repeated the tasks. Between 28% and 53% of the participants indicated that six repetitions of the tasks were useful in improving their performance. The participants who reported that their performance did not continue to improve until the end were asked to specify the number of repetitions that they thought was sufficient in each case. As reflected in Table 8, this subset of respondents (47% on the narration task, 59% on the opinion task, and 72% on the instruction task) indicated that three or four performances were sufficient. This thus triangulates the results of the fluency analysis, which showed improvement most clearly across the first four performances of each task. Furthermore, it indicates that participants generally did not become bored or fatigued but appreciated the chance to complete the same tasks with different partners across the time frame of a typical lesson.
Perceptions of number of productive repetitions
DISCUSSION AND CONCLUSIONS
In answer to the first research question, concerning the effect of aural-oral task repetition with different interlocutors on short-term L2 fluency, our findings indicate that (a) participants’ speech rate improved markedly over the first three performances and then gradually until the fifth performance; (b) clause-final pausing decreased only between the first two performances; (c) midclause pausing decreased in a stepwise fashion between the first and third, second and fourth, and third and fifth performances; and (4) significant changes in overt self-repair behavior were only detected between the fifth and first two performances and the sixth and first three performances respectively. Immediate massed repetition of input-based and output-based versions of tasks in a sequence may thus have differential effects on L2 speech production. These findings have implications for past and future research on task repetition as it relates to theories of L2 speech production mechanisms and attention during performance.
The findings suggest that up to five performances of a task may be required to prime, activate, and optimize students’ linguistic encoding processes on a task so that they can avoid breakdown and monitor their performance efficiently. In terms of overall speech rate, our results provide additional evidence for the beneficial effects of task repetition on fluency. They are in line with previous research that has found significant gains in speech rate when tasks were repeated once (Ahmadian & Tavakoli, Reference Ahmadian and Tavakoli2011; Arevart & Nation, Reference Arevart and Nation1991, Reference Arevart, Nation and Oller1993; Bygate, Reference Bygate, Bygate, Skehan and Swain2001; Lynch & Maclean, Reference Lynch and Maclean2000; Wang, Reference Wang and Skehan2014) and several times (Ahmadian, Reference Ahmadian2011). However, speech rate is likely to reflect efficiency in carrying out all aspects of L2 speech production: content planning, linguistic encoding, articulation, and monitoring (Götz, Reference Götz2013; Kormos, Reference Kormos2006). Performance of the task for the first time builds a macro-structure conceptual plan for the content to be conveyed (Skehan, Reference Skehan and Skehan2014). It also assists the speakers in the linguistic formulation of their message (Levelt, Reference Levelt1989, Reference Levelt, Brown and Hagoort1999) in terms of priming lemmas in the mental lexicon as well as the syntactic, morphological, and phonological information associated with them, and activating syntactic building procedures (Skehan, Reference Skehan and Skehan2014). Wang (Reference Wang and Skehan2014), for example, has shown that even the articulation phase of speech production benefits from previous performance. These positive effects of previous performance on all three major processes of speech production are reflected in the large gains in speech rate in the present study. However, the other measures of fluency observed provide more nuanced insights into the benefits of repetition.
Our results for clause-final pausing complement Wang’s (Reference Wang and Skehan2014) findings in that the frequency of pauses at clause boundaries decreased after one repetition of our task. Previous research has shown that clause-final pausing is used by speakers for content planning, activating background knowledge, and task-specific schemas, as well as for ordering the information to be conveyed (Butterworth, Reference Butterworth1975; Götz, Reference Götz2013). Hence, it is closely connected with the conceptualization stage of L2 speech production (Kormos, Reference Kormos2006; Levelt, Reference Levelt1989, Reference Levelt, Brown and Hagoort1999). The results of the present study indicate a reduction in clause-final pausing for all learners on the first task repetition, suggesting that one of the benefits of the first task repetition is the enhancement of the conceptualization process.
The fact that no significant difference was detected in midclause pause frequency between the first and second performance in the present study is also parallel to the findings of Wang (Reference Wang and Skehan2014), who found no reduction in the length of midclause pausing after only one repetition. Between the first and third performances, however, the participants in our study reduced midclause pausing (see Figures 7-9). Midclause pausing has been found to signal breakdowns in the linguistic encoding process and tends to occur because of difficulties in lexical access or syntactic encoding (Götz, Reference Götz2013; Kormos, Reference Kormos2006). In particular, midclause pausing reflects difficulties in retrieving relevant lemmas and accessing the morphosyntactic information associated with them (Götz, Reference Götz2013; Skehan et al., Reference Skehan, Xiaoyue, Quian and Wang2012). Decreased midclause pausing is thus an indicator of more efficient linguistic encoding mechanisms. These encoding mechanisms can be facilitated by priming effects from the students’ own previous performances as well by the lexical items and syntactic constructions in their partners’ speech (McDonough & Trofimovich, Reference McDonough and Trofimovich2009; Pickering & Ferreira, Reference Pickering and Ferreira2008). The finding that midclause pausing diminished significantly only between the first and third performances and that this reduction only partially overlapped with changes in clause-final pausing suggests that dividing attentional resources efficiently between conceptualization and linguistic encoding processes is challenging for L2 learners.
With regard to the frequency of overt self-repairs, decreases were only observed in the fifth and sixth performances. It can be argued that the frequency of self-repairs is indicative of how much attention speakers devote to monitoring processes of L2 speech production (Kormos, Reference Kormos1999). As participants optimize their conceptualization and encoding processes, it could be argued that they need to pay less attention to monitoring. The results of the present study provide some support for this hypothesis. The reduced frequency of self-repairs in the fifth and sixth performances suggests that the participants’ accuracy and efficiency in linguistic encoding had improved by this point such that fewer reformulations and overt self-corrections were necessary.
Our second and third research questions then asked whether any observed effects of aural-oral task repetition would vary across tasks of differing discourse demands and speakers of differing proficiency levels respectively. A MANOVA indicated no significant interaction between task and repetition overall, suggesting that this form of task repetition affected fluency independently of the differing discourse and structural demands of the tasks used in the study. The MANOVA also confirmed that aural-oral task repetition with different interlocutors had positive effects on speed, breakdown, and repair fluency for participants at all three levels of proficiency, which demonstrates the robustness of task repetition on L2 performance. Even language learners at relatively high levels of proficiency (C1 and B2 levels) can be expected to improve their fluency and optimize the use of their linguistic resources and speech encoding processes as a result of repeating the same task as speakers and listeners with different interlocutors within a given instructional sequence. It is possible, however, that because the linguistic encoding processes of higher level learners are more automatized than those of lower level learners, the allocation of attentional resources and the benefits of task repetition for conceptual and linguistic processing vary at different levels of proficiency. Therefore, future research that investigates the linguistic production of students together with their pausing profile is needed to uncover potential proficiency-related benefits of task repetition pertaining to the division of attentional resources at different stages of speech production.
Our final research question then investigated participants’ perceptions concerning the benefits of aural-oral task repetition and how many repetitions of each task they found useful. Very few participants provided overt indications of fatigue in repeating these tasks six times with different partners. Benefits with regard to improvements in fluency were mentioned by a large number of participants, and they also reported instances of learning from their peers and of the priming of existing knowledge representations. For the proportion of participants who did not feel that six performances of a task were necessary, the optimal number of repetitions they perceived as being useful for the narration task was four, slightly less for the opinion task, and closer to three for the instruction task. The nature of the tasks themselves might partly explain these results. The narration task has a clear structure together with variation in content from frame to frame. Success is perceived when the story is told to conclusion, and the variation in the content allows learners to refine their resources in each stage of the story. In other words, the task is structured enough for learners to focus their efforts, but at the same time challenging enough to incorporate variation. The opinion task, on the other hand, is open ended. As participants hear entirely different opinions about the photos each time they repeat the task, they may perceive a plateau in their improvement. Repetition may thus have been perceived to be less useful (see Table 8). Finally, the instruction task is much easier than the other two tasks due to the fact that the content is very repetitive across the picture frames, and the plateau effect might have been felt at an earlier stage. Participants’ responses thus generally triangulate the results of the L2 production data.
Two primary limitations need to be kept in mind in interpreting the findings of the study. First, the decision to collect the data in a classroom context with students working in pairs with each other rather than being recorded performing the monologue tasks alone in a laboratory environment resulted in a level of background noise that made it impossible to analyze the location and duration of unfilled pauses reliably with automated software. The results of the study are thus limited in that only filled clause-final and midclause pauses were considered. Furthermore, controlled studies are now needed to reveal how additional features of fluency such as the mean length of runs and the length of unfilled pauses are affected by massed task repetition. Second, the decision to have learners work with each other and alternate roles of speaker and listener across the task sequence resulted in the observed effects for task repetition being combined with the effects of interlocutor input and any number of other interlocutor variables such as gender, personality, conversation style, and so on. Although efforts were made to distribute these effects across the design through partial counterbalancing (see the Procedures section), the study combines the effect of task repetition with that of interlocutor input. The decision to have students work in pairs and alternate speaker and listener roles might also have contributed to the lack of boredom and fatigue expressed in participants’ responses to the tasks.
As we argued earlier in this article, however, the benefit of the approach used in the study is that it preserves the integrity of the tasks as communicative tools. It also adds external or ecological validity to the results in that they are more reflective of task repetition as it is likely to occur inside and outside the classroom. Nevertheless, it is necessary that the results be interpreted in terms of optimizing speech production in the short term when input-based and output-based tasks are repeated with different interlocutors in a classroom situation. The order in which learners received input from speakers of different proficiency levels varied across the six repetitions of the task set, and this variation could have affected uptake at different points in the sequence. Future research on the effects of repetition independent of interlocutor input and other interlocutor effects is now needed to provide a more fine-grained picture of the effects of repetition as “practice” rather than repetition as “communication” on the efficiency of L2 speech production.
APPENDIX
SAMPLE OF PAUSING AND OVERT SELF-REPAIR ANALYSES
. . . because they have a trouble with er (FP mid) a graffiti problem. And it’s just it is not just (RPR) at this time. They had this problem before. So uhm (FPfinal) then, they tried to solve this problem. So they have a meeting with er (FP mid) shop owners. But they think about the solutions. But the security cameras and street patrol is good way to solve it, but it’s it needs (RPR) too much cost. So they can’t get nice idea at this meeting. But that night, a woman go went (RPR) back her home. And when she find found (RPR) she’s her (RPR) son is was (RPR) writing some picture, suddenly she come up with nice idea which er (FP mid) she think graffiti is problem, but if it’s art, it doesn’t it’s not (RPR) problem. er (FPfinal) so they gather children a month later er (FPfinal) and make children write some picture on her shop’s wall before er (FP mid) graffiti problem comes again.
Note. FP mid = filled pause in midclause position; FPfinal = filled pause at clause boundary; RPR = overt self-repair.
