Development of second language speech fluency in foreign language classrooms: A longitudinal study

1 L2 fluency development

Speech fluency is a temporal feature of speech and can be understood in both a broad and a narrow sense (Lennon, 1990). In the broad sense, fluency is considered the equivalent of oral proficiency. In the narrow sense, it is defined as a set of acoustic properties that can determine optimal, smooth, and fluid delivery of L2 speech (Segalowitz, 2010). This narrow understanding of speech fluency is often categorized into three sub-constructs (Skehan, 2003, 2009; Tavakoli & Skehan, 2005): breakdown fluency (how many times and for how long a continuous speech is interrupted), speed fluency (how many words or syllables are delivered), and repair fluency (how frequently learners repeat the same words and repair their utterances) (Lambert & Kormos, 2014). Fluency, therefore, entails rapid delivery of speech with few breakdowns and little hesitation.

In order to understand how oral L2 fluency is achieved, it is important to consider the underlying speech production mechanism. According to the widely-accepted models, that speech production involves three stages, denoted as conceptualization, formulation, and articulation (Kormos, 2006; Levelt, 1989). When generating speech, depending on their communicative needs, speakers first determine what information they wish to convey and the manner of delivery. Based on these factors, they create a preverbal plan by referring to world knowledge stored in their long-term memory. This preverbal plan is then implemented in the formulation stage, whereby it is linguistically encoded using lexical, syntactic, and phonological information. In the final (articulation) stage, these linguistically encoded structures are pronounced using articulatory gestures. Throughout these three phases, the monitoring mechanism checks, both internally and externally, whether the outcomes of each phase match the previously produced plan. As fluent speech performance reflects efficient functioning of the speech production mechanism, seamless execution of this process is thought to be indicative of automatization (Kahng, 2014; Segalowitz, 2010).

In extant research, performance in each of the aforementioned three stages comprising L2 speech production is assessed via the previously defined objective L2 fluency measures (i.e. breakdown, speed, and repair fluency) (e.g. Lambert, Kormos, & Minn, 2017; Saito et al., 2018b). It has been argued that breakdown fluency is related to the conceptualization and formulation stages, and is reflected in pause location (Skehan, Foster, & Shum, 2016). Between-clause pausing can be indicative of content planning and conceptualization, whereas within-clause pausing can signal linguistic retrieval and sentence construction in the formulation (Götz, 2013). Repair fluency is posited to reflect the monitoring system effectiveness (Kormos, 1999). Finally, speed fluency is purported to signal the overall efficacy of the three stages of the speech production mechanism.

Building on this theoretical and methodological framework, extensive research has been conducted to examine how learners develop L2 fluency in terms of these key measures depending on the amount of L2 experience. For example, using pre-/post-research design, Leonard and Shea (2017) investigated fluency development in learners who participated in a three-month study abroad program. They reported significant improvements in breakdown and speed fluency measures over time. Although, as a group, the learners achieved non-significant gains, some individuals showed pronounced changes in repair fluency depending on their initial proficiency. While the number of self-repairs increased among intermediate learners, it declined among the advanced learners upon completion of the study abroad program. Lambert et al. (2017) examined the effect of performing the same communicative task six times on several dimensions of L2 fluency. They showed that the number of final-clause pauses decreased during the first two performances only, while the number of mid-clause pauses gradually decreased up to the fifth performance. Self-repairs started to improve in the fifth and sixth performance, whereas participants’ speech rate continuously improved throughout the fifth performance. More recently, Saito et al. (2018b) classified their participants according to their L2 proficiency (beginner, intermediate, and advanced) and investigated the development of fluency measures in each cohort. Their results revealed that between-clause pauses differentiated only beginner and intermediate learners, while all three groups differed in terms of within-clause pauses and articulation rate, but not repair fluency. These findings suggest that speed, breakdown, and repair fluency do not develop concurrently as the learning experience progresses but rather emerge at different rates across the stages of L2 learning.

2 L2 fluency development in foreign language contexts

According to Muñoz (2008), in most FL teaching settings, (1) instruction is typically limited to two to four 50-minute sessions per week; (2) exposure to the L2 during class is limited in source (mainly the teacher), quantity (not all communication in the classroom is conducted in the L2), and quality (teachers’ oral proficiency varies considerably); (3) peer interaction in L2 is very limited; and (4) the L2 is rarely used outside the classroom. Larson-Hall (2008) referred to this type of learning context as a ‘minimal input’ condition (p. 36; see also, Collentine & Freed, 2004). Therefore, FL classrooms are generally viewed as ‘restricted settings’ (Best & Tyler, 2007, p. 19), unsuited for the development of oral proficiency.

Some studies conducted in FL classroom settings confirm this assertion. For instance, Freed, Segalowitz, and Dewey (2004) evaluated and contrasted the development of fluency measures in three learning contexts (FL, study abroad, and domestic immersion) over a 12-week period. Their analyses revealed that students assigned to the immersion program showed significant improvements in terms of total number of words, length of the longest turn, speech rate, and speech fluidity (as a mixed measure of utterance fluency), whereas those in the study abroad program made significant gains in fluidity. In contrast, no improvements were noted for the learners who received instruction in FL classrooms. Similarly, Mora and Valls-Ferrer (2012) reported robust improvements in the six measures adopted in their study (including speed and breakdown fluency measures) for participants assigned to a three-month study abroad program, whereas students partaking in FL classroom instruction failed to make any gains.

Another important issue addressed by previous studies is the role of the quality and quantity of L2 learning experience in oral proficiency development. Freed, Segalowitz, and Dewey (2004) reported that the number of hours dedicated to L2 writing outside the classroom was significantly associated with L2 fluency gains. The authors tentatively attributed this finding to the fact that, when speaking and writing, learners are hypothesized to use the same underlying cognitive mechanisms (conceptualization, formulation, articulation, and monitoring). In other words, any improvement in the efficacy of cognitive mechanisms attained through L2 writing exercises can be transferred to the speaking ability, resulting in L2 fluency development (Blake, 2008). Similarly, Baker-Smemoe et al. (2012) found that the amount of L2 use outside the classroom was significantly correlated with improvements in reading, listening, and speaking skills. Moreover, the regression analyses showed that deliberate use outside the classroom of what the learners had learned in the classroom (e.g. L2 grammar, vocabulary, and expressions) accounted for 20% of their speaking ability. More recently, Saito, Suzukida, and Sun (2018a) showed that Japanese learners enhanced their global comprehensibility as a function of increased classroom experience in the course of one semester. This finding is particularly important because, compared with the naturalistic context in which L2 is used on a daily basis, in FL contexts, communication in L2 outside the classroom is typically limited (Harada, 2007; Muñoz, 2008). Accordingly, learning experiences inside as well as outside the classroom can be considered to play an important role in FL learners’ ability to enhance their L2 fluency.

Although the authors of the aforementioned studies demonstrated some unique aspects of L2 fluency development and the role of L2 experience inside and outside classroom, it is worth noting that their findings are based on short-term observations (over a few months, except for Saito et al., 2018a). In that respect, Saito and Hanzawa’s (2018) investigation is most relevant to the current study, as the authors scrutinized the potential benefits and limits of FL classroom education from a long-term perspective. They tracked various aspects of L2 oral production (fluency, lexicogrammar, prosody, and segmental aspects) of 40 Japanese students over a 12-month period using a human rating method (having native speakers rate speech samples). Of particular relevance to the current study on fluency development is their finding that, even though the Japanese learners showed a gradual improvement over the course of one academic year, these gains did not correlate to their L2 learning experience inside or outside classroom.

The findings reported by Saito and Hanzawa (2018) provide empirical support for FL classrooms as a potentially beneficial setting for attaining L2 fluency gains, albeit over a longer period than needed in study abroad and immersion programs. However, as the authors adopted human rating for measuring L2 learners’ oral proficiency, it remains unknown how different L2 learning experiences within FL classroom context affect objective fluency measures (speed, breakdown, and repair fluency). As discussed previously, given that various aspects of L2 fluency develop in different learning stages, tracking FL learners’ speech with objective measures over a long period (more than one academic year) can provide valuable insight into FL fluency development.

1 Participants

The initial study sample comprised of 54 students recruited from various arts and social science programs at a large Japanese university. Since the project took place over the course of one academic year, four students did not participate in the T2 and T3 assessments, reducing the final sample to 50 students (31 females and 19 males). The mean age of the participating students was 18.1 years. In line with previous FL studies examining oral proficiency (Saito, 2019; Saito et al., 2018a), participants were required to meet specific inclusion criteria: (1) they had to be native Japanese speakers (with both of their parents being first language Japanese speakers); (2) they had to be in their first year of university at the beginning of the project; and (3) they had to have started learning English in Grade 7 of secondary school and have no prior living abroad experience lasting more than two months. All 50 students met these criteria, thereby ensuring that they had similar linguistic backgrounds and had received relatively uniform secondary English education in Japan.

The students’ general English proficiency prior to the experiments was estimated by an English proficiency test specifically designed for the Japanese university students. The results showed that their English proficiency level fell between B1 and C1 on the Common European Framework Reference of Languages (CEFR) (M = 710.5, SD = 56.0, Range = 586–811 out of 1,000) (Tannenbaum & Wylie, 2008).

2 Speaking task

In the current study, participants’ speech was elicited via a personal narrative task. This task was selected because it encouraged students to focus on meaning rather than form, as they were required to convey their communicative intentions under time pressure. The participants were asked to describe ‘the toughest or most challenging event they had experienced in the past few months’ in English. They were instructed to begin their narrative with the prompt sentence shown in the instruction sheet: ‘The most challenging event I have experienced in the past few months . . .’ The task instruction sheet also included three guiding questions, ‘When did it happen?’; ‘Where did it happen?’; and ‘Why did you find this experience to be the most challenging?’ to indicate what they were required to talk about in this task. Those guiding questions were intentionally provided to help the students produce a speech of certain length without excessive hesitation or dysfluency. In a sense, these questions played a role similar to that of an interviewer in an interview task (Freed, Segalowitz, & Dewey, 2004).

The same personal narrative task was used in all three tests (T1, T2, and T3) in order to elicit comparable speech samples from the participants and allow their progress over the course of one academic year to be monitored; a similar methodological approach was adopted by Derwing, Munro, & Thomson (2008) and by Simard, French, & Zuniga (2017) in their longitudinal L2 speech study. It is worth noting that, when the same task is performed more than once, there is potential for a ‘practice effect’ to arise, which may mask actual changes in learners’ performance. Nonetheless, in the current study, the gains due to the practice effect were considered to be minimal owing to the nature of the personal narrative task. Contrary to other speech elicitation methods, such as picture description and decision-making tasks commonly used in previous studies (N.H. de Jong, Steinel, Florijn, Schoonen, & Hulstijn, 2012a), when repeating the same personal narrative or interview task, speakers can change the content (N. de Jong & Tillman, 2018). Such flexibility can be argued to reduce the potential risk of ‘practice effect’. Moreover, as the prompt required learners to talk about an experience that had occurred in the past few months, given the time span between consecutive tests, they were not permitted to describe the same event more than once. Indeed, none of the participants recounted the same experience more than once.

3 Procedure

To keep track of the participants’ L2 fluency development over one academic year, their spontaneous speech was recorded as they presented their personal narrative at three test points (T1, T2, and T3). On each occasion, participating students were tested individually in a soundproof room at the university by the author or a research assistant, both of whom are native Japanese speakers. The instructions were given in Japanese (the participants’ first language, L1) to ensure that they fully understood how to complete the task. Participants were asked to plan their narration for one minute, aided by the prompt sentence and three guiding questions. Once the preparation time had elapsed, the participants spoke for two minutes about a personal experience that had taken place in the past few months, aided only by the prompt sentence. Finally, the participants completed the questionnaire inquiring about their L2 learning experience, which is described in detail below. All speech samples were recorded using a Marantz PMD 660 recorder and Shure SM 10A-CN microphone, with a 44.1 kHz sampling rate and 16-bit quantification.

4 L2 learning experience survey

Relevant information on the participants’ L2 learning experience was obtained via the questionnaire comprising a language contact profile (Freed, Dewey, Segalowitz, & Halter, 2004) and other items previously adopted in precursor research (Saito & Hanzawa, 2018). In the current study, L2 learning experience was operationalized according to the two dimensions that have been related to successful L2 learning in FL classroom contexts: (1) learning experience inside the classroom, and (2) learning experience outside the classroom. The questionnaire was administered at the end of the first (T2) and the second (T3) semester, and students were instructed to report data for the preceding semester. Table 1 provides descriptive statistics of the participants’ L2 learning experiences inside the classroom (number of hours designated for in-class lessons) and outside the classroom (number of hours devoted to extracurricular learning activities) over the course of one academic year.

OPEN IN VIEWER

Table 1.

Summary of the participants’ EFL L2 learning experiences during the project (first semester, second semester, and over the course of one academic year).

	Total hours of form-focused classes			Total hours of content-based classes			Total hours of receptive L2 learning activities			Total hours of productive L2 learning activities
	M	SD	Range	M	SD	Range	M	SD	Range	M	SD	Range
Semester 1	47.55	20.9	19.5–90	18.57	16.19	0–45	53.5	54.8	0–225	53.8	51.7	0–217.5
Semester 2	65.94	23.04	18–118.8	25.29	22.35	0–67.5	50.2	53.3	0–261.2	15.8	21.7	0–150
Total	113.49	41.62	66–205.5	43.86	35.83	0–90	107.3	92.7	8.7–395	66	67.2	0–328.7

For measuring experience inside the classroom, participants were required to report the number of form-focused and content-based English classes they took per week. In the former classes, input and output in L2 are manipulated mainly by teachers to lead learners’ attention to linguistic forms with the goal of improving their linguistic expertise (Spada, 1997). On the other hand, in the latter lessons, L2 is primarily used as a tool for exchanging information, and learners are expected to acquire content knowledge and skills rather than linguistic knowledge of L2 (Lyster & Ballinger, 2011). The distinction between form-focused and content-based classes is crucial in FL classroom contexts because taking content-based classes is considered an active strategy for increasing the amount of L2 learning (Saito & Hanzawa, 2018). The numbers of lessons of each type the participants reported in the questionnaires were separately multiplied by class duration (90 minutes) to obtain the total amount of classroom-based L2 learning experience (form-focused and content-based classes). ¹

For learning experience outside the classroom, students were required to state (1) how many days per week and (2) how many hours per day they engaged in each of the five language skills – reading, listening, speaking, and writing, along with vocabulary learning activities (e.g. memorizing items in a vocabulary list or English book). Although the original language contact profile did not include vocabulary as an extracurricular activity, given that the importance of vocabulary learning is highly emphasized in Japanese FL classroom contexts (Mizumoto & Takeuchi, 2009; Takeuchi, 2003), it was included in the questionnaire. As shown in the Table 1, every week, the participants dedicated a few hours to each learning activity outside the classroom, which is typical for Japanese learners of English in Japan (Saito & Hanzawa, 2016). In keeping with the approach adopted by Freed, Segalowitz, and Dewey (2004), the five L2 learning activities (reading, listening, writing, speaking, and vocabulary learning) were combined to create two new variables for subsequent analyses: receptive L2 learning activities (comprising reading, listening, and vocabulary skills) and productive L2 learning activities (speaking and writing skills).

5 Analyses

Over the course of one academic year, 150 speech samples were generated (50 students × three test points). Prior to the analyses, the prompt sentence given on the instruction sheet was removed from all recordings. The speech samples were annotated using a free sound analysis software PRAAT (Boersma & Weenink, 2016). Two trained research assistants identified filled and unfilled pauses with an aid of the script developed by N.H. de Jong and Wempe (2009), which automatically detected gaps (silences) in speech and their duration. In this script, the lower threshold for unfilled silent pauses was set to 250 ms, excluding micropauses (e.g. bursts before plosive sounds) (Bosker et al., 2012). At the same time, the research assistants transcribed all speech samples based on the Analysis of Speech Unit (ASU) (Foster, Tonkyn, & Wigglesworth, 2000), allowing them to code self-repairs and repetitions. Finally, the author manually checked the location and duration of pauses, transcriptions, and self-repairs and repetitions of all speech samples (for a general description of the speech samples, see Appendix A in supplemental material).

As discussed in Section II, students’ speech samples were investigated in terms of three dimensions (breakdown, speed, and repair fluency) which are hypothesized to correspond to three distinct stages of the L2 speech production model (conceptualization, formation, and monitoring). For the current study, these three dimensions were further divided into six measures as follows:

First, as a measure of breakdown fluency, pause location is argued to be indicative of different speech processing stages (i.e. between-clause pauses correspond to conceptualization and within-clause pauses to formation) (Skehan et al., 2016). Therefore, between-clause and within-clause pauses were separately coded based on AS units. Second, in the current study, the pause phenomena were assessed in terms of frequency and length (for the means, SDs, and 95% CIs for the frequency and duration of filled and unfilled pauses, see Appendix B in supplemental material). This approach is in line with that adopted by other authors who considered both pause frequency and length as a measure of breakdown fluency (Baker-Smemoe et al., 2014; Bosker et al., 2012; N.H. de Jong et al., 2012b, 2013; Kahng, 2014; Kormos & Préfontaine, 2011; Y. Suzuki, 2020; S. Suzuki & Kormos, 2020), while others focused solely on pause frequency (Lambert, et al., 2017; Leonard & Shea, 2017; Saito et al., 2018b). Findings yielded by these studies revealed that pause frequency and length may have different influence on perceived fluency (S. Suzuki & Kormos, 2020) and fluency development (Y. Suzuki, 2020). Second, in the current study, articulation rate (the number of syllables spoken during the total speech time, excluding pause length) was adopted as a measure of speed fluency. Note that articulation rate differs from speech rate, which is based on the total speech time that includes pause length. This decision was made to avoid any overlaps between breakdown and speed fluency, by clearly categorizing pause length as a part of breakdown fluency. Third, although repair fluency may not be a strong indicator of perceived fluency or cognitive fluency (Bosker et al., 2012; Saito et al., 2018b), it was included in the current study because the number of repair attempts was shown to change after attending a study abroad program (Leonard & Shea, 2017) or completing repetition training (Lambert et al., 2017).

To ensure measurement reliability, 10% of the data set was re-measured by the author a few months after the first measurements with >90% agreement for all fluency measures. In addition, Pearson’s correlational analyses were conducted, showing some significant correlations between the six fluency measures (see Appendix C in supplemental material). ³ This issue is discussed further in Section VII.

1 Group improvement

The results of the Levene’s test showed acceptable homogeneity of variance (p > .05) for the six fluency measures at each testing point (T1, T2, and T3). Therefore, a repeated measures ANOVA was conducted to examine whether there were significant changes over time for each of the six measures, with Time as a within-participant factor. When a main effect was significant in the repeated measures ANOVA, pairwise comparisons of the findings related to the three testing points were performed for those fluency measures that were significant. In the comparisons, Cohen’s d was calculated to determine the effect sizes. Following Cohen’s (1988) benchmarks for interpreting effect sizes in social science research, effect sizes above d = .20 were interpreted as small, those above d = .50 as medium, and those above d = .80 as large.

The first aim of the statistical analyses was to ascertain whether and to what extent the participants as a group improved their L2 fluency (as indicated by the six measures) over the first (T1→T2) and the second (T2→T3) semester, as well as in the course of one academic year (T1→T3). Table 2 provides the descriptive statistics of the six measures at each time point. As can be seen from the results, the temporal changes in the six fluency measures vary, and are in some cases unobservable. The main effect of Time is significant in two measures: the length of within-clause pauses, F(2, 98) = 15.01, p < .001, partial η 2 = .23, and the length of between-clause pauses, F(2, 98) = 4.50, p = .01, partial η 2 = .08. Although changes in the articulation rate did not reach a statistically significant level, improvements were still noted, F(2, 98) = 2.61, p = .078, partial η 2 = .05. Subsequent pairwise comparisons with Bonferroni correction further revealed that the length of within-clause and between-clause pauses differed in the degree and timing of changes. Participants significantly reduced the length of within-clause pauses between T1 and T2, and T2 and T3, with a medium effect size (d = 0.53 and d = 0.68, respectively), resulting in a significant difference between T1 and T3, with a larger effect size (d = 0.82). On the other hand, a significant reduction in between-clause duration was detected only between T1 and T3 with a medium effect size (d = 0.52). These findings indicate that, while learners’ within-clause pausing duration consistently declined over the course of one academic year, improvements in between-clause pausing occurred more gradually. Nonetheless, no significant differences in articulation rate were found across the three tests.

OPEN IN VIEWER

Table 2.

Descriptive statistics of six fluency measures at T1, T2, and T3.

	T1			T2			T3
	M	SD	95% CI	M	SD	95% CI	M	SD	95% CI
Articulation rate	2.99	0.36	2.89–3.09	3.11	0.39	3.00–3.22	3.09	0.40	2.98–3.21
Between-clause pause frequency	0.11	0.05	0.95–0.12	0.12	0.06	0.10–0.13	0.10	0.04	0.09–0.11
Within-clause pause frequency	0.23	0.08	0.21–0.26	0.23	0.07	0.21–0.25	0.22	0.06	0.20–0.24
Between-clause pause length	1.46	0.72	1.25–1.66	1.34	0.76	1.13–1.56	1.12	0.59	0.96–1.29
Within-clause pause length	1.08	0.37	0.98–1.92	0.90	0.31	0.81–0.99	0.82	0.25	0.75–0.89
Repair frequency	0.09	0.04	0.07–0.10	0.09	0.05	0.08–0.11	0.10	0.05	0.08–0.11

The lack of statistically significant changes in other fluency measures is due, in part, to a considerable variance in curricular as well as extracurricular L2 study patterns among participating students (see Table 1). For example, certain students dedicated much more time to L2 practice (reading, listening, writing, speaking, and vocabulary) outside the classroom than others. For this reason, further correlation analyses were performed to examine whether these differences in L2 learning experience are related to the observed variations in L2 fluency development.

2 Influence of L2 learning experience on fluency development

To address the second research question, statistical analyses were performed to examine whether and to what extent the observed variations in the six fluency measures could be attributed to differences in L2 learning experience inside and outside the classroom among the study participants. In line with previous FL studies (Saito & Hanzawa, 2016; Saito et al., 2018a), participating students’ initial proficiency varied considerably, probably due to differences in their English learning experience prior to entering university (i.e., the knowledge and skills gained during six-year FL instruction prior to joining the current study). In order to isolate the changes that occurred over the course of one academic year from those due to dissimilar initial proficiency, a set of partial correlation analyses was performed. In this series of analyses, differences in participants’ scores at the end of the first semester (obtained by subtracting the scores achieved at T1 from those at T2) and the second semester (obtained by subtracting the scores achieved at T2 from those at T3) were set as dependent variables. In addition, two independent variables were created to capture the total amount of L2 learning experience (both inside and outside the classroom) over time while controlling for participants’ scores achieved at T1 and T2, respectively.

As shown in Table 3, the total amount of L2 learning experience was significantly correlated with changes in between-clause pause frequency and self-repair frequency. First, between-clause pause frequency at T2 was significantly correlated with the total amount of L2 learning during the second semester. As expected, the increase in between-clause pause frequency was negatively correlated with the total learning time, indicating that students who spent more time learning English reduced the number of pauses at the end of clauses. Second, the change in self-repair frequency was significantly correlated with the total amount of L2 learning in both semesters. Interestingly, these correlations were positive, suggesting that students who dedicated more time to learning English were more likely to make a greater number of self-repairs over time. This may be counterintuitive, given that it is generally assumed that the need for self-repairs should decline as learners become more fluent (Lambert et al., 2017). No significant correlations were found between the amount of L2 learning experience and the other four fluency measures.

OPEN IN VIEWER

Table 3.

Partial correlations between the six fluency measures and the total amount of L2 learning experience over time.

Fluency variable	First semester		Second semester
	r	p	r	p
Articulation rate	.06	.64	.16	.25
Between-clause pause frequency	−.17	.22	−.30*	.03
Within-clause pause frequency	−.07	.60	−.19	.18
Between-clause pause length	−.14	.33	−.06	.64
Within-clause pause length	−.06	.66	−.23	.10
Repair frequency	.30*	.03	.36*	< .01

Notes. * α < .05. Participants’ proficiency scores at T1 and T2 were partialled out for the T1→T2 and T2→T3 results, respectively.

3 L2 learning experience types and fluency development

The objective of the final analyses was to examine whether and how the four types of L2 learning experience (form-focused classes, content-based classes, outside receptive activity, and outside productive activity) examined in this study differentially related to changes in students’ scores over the first (T1→T2) and the second (T2→T3) semester. Similar to the approach adopted in the preceding analyses, partial correlation analyses were conducted to examine correlations between the four types of L2 learning experience and changes in participants’ scores during the two semesters while factoring out their scores at T1 and T2, respectively, for each fluency measure. Because learning experience during the two semesters was represented by four independent variables in the analyses, the alpha values were set to .012.

As summarized in Table 4, a significant correlation was found only between repair frequency and L2 productive activities outside the classroom during the second semester. Again, the correlation was positive, suggesting that the students who spent more time engaged in activities aimed at improving their productive skills outside the classroom were more likely to increase the number of self-repairs. No significant correlations were found between the four types of L2 learning experience and the other fluency measures.

OPEN IN VIEWER

Table 4.

Partial correlations between the six fluency measures and the four types of L2 learning experience.

Fluency variable	First semester								Second semester
	Form-focused classes		Content-based classes		Outside receptive activity		Outside productive activity		Form-focused classes		Content-based classes		Outside receptive activity		Outside productive activity
	r	p	r	p	r	p	r	p	r	p	r	p	r	p	r	p
Articulation rate	.00	.95	.31	.03	−.06	.65	.11	.45	−.18	.20	−.09	.54	.23	.11	.24	.09
Between-clause pause frequency	−.11	.43	−.11	.44	−.13	.34	−.15	.29	−.21	.15	−.13	.37	−.25	.72	−.20	.15
Within-clause pause frequency	−.08	.55	−.30	.03	−.12	.37	−.03	.82	−.08	.60	−.04	.77	−.24	.08	−.03	.79
Between-clause pause length	−.11	.43	.02	.89	−.16	.25	−.06	.54	−.13	.39	.12	.40	−.76	.60	−.53	.72
Within-clause pause length	−.07	.61	.01	.97	−.10	.49	.00	.98	−.17	.26	−.05	.74	−.21	.13	−.21	.13
Repair frequency	.15	.29	.31	.03	.16	.26	.31	.03	.07	.62	.13	.38	.27	.06	.41*	< .01

Notes. * α < .012. Participants’ proficiency scores at T1 and T2 were partialled out for the T1→T2 and T2→T3 results, respectively.