) The Poster Carousel in the ESL Classroom: What Happens to Learners' L2 Fluency During Same and Parallel Task Repetition? TESOL Quarterly

This paper reports on the impact of an English as a Second Language (ESL) speaking activity — the poster carousel — on English learners’ second language (L2) ﬂuency. Two versions of the poster carousel were developed to observe the effect of talking about (a) the same poster three times (same-task repetition) or (b) three different posters (parallel-task repetition). 46 ESL learners took part, and their performances were audio-recorded, transcribed, and analyzed using PRAAT to detail their L2 utterance ﬂuency. The ﬁndings suggest that learners were more ﬂuent during repeated performances in the same-task repetition poster carousel group. No changes in ﬂu-ency were observed for the parallel-task repetition poster carousel group. A detailed case study explores these observed ﬂuency increases in the same-task repetition group and generates further hypotheses for empirical exploration. Some observations are made which relate to the selection of ﬂuency measures in L2 ﬂuency research.

The Poster Carousel in the ESL Classroom: What Happens to Learners' L2 Fluency During Same and Parallel-Task Repetition? INTRODUCTION M any second language (L2) learners dream of being able to speak their L2 confidently, and L2 teachers want to help their learners achieve their L2 speaking goals.A substantial body of research suggests that task repetition-giving learners the opportunity to perform a communicative task once and then do it again-can lead to improved L2 speaking performance, especially in relation to oral fluency, and at least in the short term (see Bygate, 2018).
However, despite this, language teachers do not necessarily incorporate task repetition into their teaching (Tavakoli & Hunter, 2018) possibly because they fear it might impact negatively on learners' motivation (Ahmadian, Mansouri, & Ghominejad, 2017;van de Guchte, Braaksma, Rijlaarsdam, & Bimmel, 2016).Teachers may also worry that any benefits of task repetition are lexis-dependent and therefore specific to a particular task (Suzuki & Hanzawa, 2022).A further methodological concern is that much of the existing task repetition research involves carefully controlled research environments and the use of monologic tasks that may not reflect the realities of L2 classrooms (Foster, 2020).
Given the considerable research interest in task repetition, alongside calls to bridge the gap between SLA research and language teaching practice (e.g., Tavakoli, 2020), it is timely that we pause to reconsider the pedagogical value of task repetition.This study sheds light on task implementation factors that influence the effectiveness of task repetition as a pedagogical tool and asks: what happens to learners' fluency as they engage in repetitive task sequences in the language classroom and-ultimately-is task repetition actually beneficial?
Fluency in a Second Language Levelt (1989) posits that when producing speech, a person first conceptualizes the message, then formulates the message by drawing on stored linguistic knowledge, before articulating the message as overt speech.Monitoring also occurs at all stages "to ensure accuracy and appropriateness" (Lambert, Aubrey, & Leeming, 2021, p. 332).Levelt's model has since been adapted to describe the L2 speech production process (de Bot, 1992;Kormos, 2006;Segalowitz, 2010) and these revised models point to vulnerability points in the speech production process, especially for lower proficiency L2 speakers (Lambert et al., 2021), whereby non-parallel processing impacts on the quality of the overt speech, especially their fluency.Segalowitz (2010) explains that L2 fluency can be thought of as three interrelated domains.Cognitive fluency represents the underlying speed and efficiency of the L2 speech production process.Perceived fluency refers to the judgment that a listener makes about a speaker's underlying cognitive fluency, and utterance fluency relates to the observable features of overt speech that signify underlying cognitive fluency.
For L2 research, assessment, and teaching purposes, the domain of utterance fluency provides a framework through which researchers and practitioners can attempt to understand the cognitive fluency of L2 learners (Tavakoli & Hunter, 2018).
Speakers can demonstrate (dys)fluency in different ways during speaking, for example, speaking more slowly or pausing more frequently, and SLA researchers have long tended to group these utterance fluency behaviors under three sub-constructs: speed, breakdown, and repair (Skehan, 1998(Skehan, , 2003)).Speed is typically measured through articulation rate, while breakdown is measured through frequency and/or length of pauses.Repair is often operationalized as frequency of reformulations, repetitions, and self-corrections.Many L2 fluency studies also adopt global measures which straddle two or all three of these sub-constructs (i.e., speech rate).
More recently, L2 fluency research has considered the connection between utterance fluency measures and specific stages of the speech production process.This work attempts to distinguish fluency that relates to L2 knowledge from fluency that relates to natural speaking style and idea shifts (e.g., Peltonen, 2018).For example, in relation to breakdown fluency, many now argue that mid-clause pauses indicate cognitive work at the formulation stage of speech production and is therefore a more L2-specific measure (de Jong, 2016) and pauses which occur around clausal boundaries are considered more suggestive of planning at the conceptualization stage-which happens in both L1 and L2 (Felker, Klockmann, & De Jong, 2019;Lambert et al., 2021;Saito, Ilkan, Magne, Tran, & Suzuki, 2018;Skehan, Foster, & Shum, 2016).
While pausing has been explored in this finer-grained way, less work has been done to pinpoint L2-specific speed and repair measures.That said, some researchers are beginning to investigate the location of sound lengthenings (arguably a measure of speed fluency), sometimes referred to in research as drawls (Peltonen, 2018) or prolongations (Rohr, 2016;Williams, 2023).Rohr (2016) explains that, as with pausing, L2 speakers are more likely to drawl on syllables that fall within clause boundaries than L1 speakers.However, drawls are not commonly included as a measure of L2 fluency (Williams, 2023) and are especially lacking in L2 classroom studies (see Felker et al., 2019).When it comes to repair, there is some awareness that repetitions may perform a similar function to pauses and drawls in that they represent the speaker's need to buy time while formulating speech.Reformulations and self-corrections are different in that they may represent increased monitoring and increased cognitive capacity (Witton-Davies, 2014).

Task Repetition
A pedagogical task in language-learning terms can be considered "a goal-oriented activity . . .that involves meaningful use of language" (Van den Branden, 2016, p. 240).Task repetition, therefore, consists of learners performing a task and then performing it again, either immediately or after some time has passed (Bygate & Samuda, 2005).A distinction can be made between same-task repetition, where repeated tasks are identical in terms of structure and content, and parallel-task repetition, where the structure of the task is the same, but the content is different (Lambert et al., 2021).
When explaining task repetition effects on utterance fluency, earlier task repetition research tended to foreground the conceptual planning argument: the initial performance of the task acts as a dry-run (Bygate, 1996(Bygate, , 2001(Bygate, , 2018) ) and during a subsequent performance, the conceptual work has been taken care of and the speaker can direct resources to formulation and articulation (Lambert et al., 2021).Recent research has also discussed the impact that lexical priming or reuse of constructions (Suzuki, Eguchi, & de Jong, 2022) might have on fluency during task repetition, especially when the repeated performances of the task come soon after each other (Bui, Ahmadian, & Hunter, 2019), because the act of having previously performed a task may mean that a "blue-print" of the utterance is stored in memory (Fukuta, 2016;Lambert, Kormos, & Minn, 2017;Lambert, Philp, & Nakamura, 2017, p. 5).
Seeking to elucidate the underlying processes affected by same-task repetition, Lambert, Kormos, and Minn (2017) and Lambert, Philp, and Nakamura (2017) used a range of specific fluency measures and suggested that stages of the speech production process may be affected during different iterations of the task: conceptualization (as measured by frequency of between-clause pauses) in early iterations and formulation (as measured by mid-clause pauses) during later iterations.A key question, however, is whether swifter formulation is because of recently activated (i.e., primed) language or regularly activated (i.e., proceduralized) language.From a pedagogical perspective, a concern is that, if lexical priming is the primary reason for increased fluency during task repetition, any gains may be lexis-dependent and less likely to be transferable to other tasks and contexts.

Parallel-Task Repetition and L2 Fluency
Although studies are far fewer in number, parallel-task repetition which is sometimes referred to as task type repetition, has also been found to impact on L2 fluency (De Jong & Perfetti, 2011;Lambert et al., 2021;Pinter, 2005;Sample & Michel, 2014).Parallel-task repetition may offer an attractive option for language teachers because it could be more motivating for learners (Ahmadian et al., 2017), and any fluency gains might be more likely to transfer to novel tasks (Lambert et al., 2021).Increases attested to in Parallel-task repetition studies have typically been attributed to familiarity with the general task procedure (Bygate, 2001) and/or familiarity with abstract rules and structural priming (Suzuki et al., 2022).

The Poster Carousel
Many popular ESL/EFL classroom activities such as mingles, fairs or speed dating activities (Bailey & Savage, 1994) are inherently repetitive (Gatbonton & Segalowitz, 2005) and are particularly conducive to language learning because the "learner remains involved in genuine communication while engaged in repetition" (Segalowitz, 2010, p. 176).The poster carousel (Lynch & Maclean, 1994) is an example of an inherently repetitive classroom activity, which requires learners to design a poster and then to present their poster to multiple visitors.It has a "particular combination of text input, task structure and learner interaction" (Lynch, 2018, p. 203) which may appeal to teachers working within a task-based or task-supported paradigm.The poster carousel activity also incorporates both pre-task and online planning (Foster & Skehan, 1996) because learners have the chance to discuss the topic as they design their poster and the poster then supports the learners conceptually during performances.
Adopting a case study methodology, Lynch andMaclean (2000, 2001) showed how the quality of the talk of five L2 English learners improved over the six performances of the poster carousel with notable POSTER CAROUSEL IN THE ESL CLASSROOM improvements in "syntactic accuracy" (2001, p. 148), "more precise and less awkward" performance (2001, p. 142), "greater precision in choice of words " (2001, p. 144) and "less vagueness" (2001, p. 151).They also note the "correct fluent use of some language forms after initial difficulty" (2001, p. 144).What is more, the researchers found that the learners in their study responded positively to the poster carousel and were able to perceive benefits for their L2 speaking skills.
The poster carousel described in Lynch andMaclean (2000, 2001) was designed for their teaching context-English for Medical Conferences-but can be easily adapted, making it particularly suited for use in research studies of task factors and L2 fluency within a language classroom.

THE CURRENT STUDY
Overall, then, task repetition leads learners to speak with higher fluency during subsequent task performances, although an important question remains: what underlying processes drive these fluency increases?To date, there has not been a detailed empirical exploration of the poster carousel and its impact on utterance fluency so it is difficult to link holistic performance improvements described in Lynch and Maclean's research with specific stages of the L2 speech production process.Furthermore, given that parallel-task repetition may also foster fluency increases (Lambert et al., 2021) and that learners and teachers may respond more positively to this type of repetition, it is important to explore the pedagogical value of a parallel-task version of the activity.Research questions in the current study were therefore: 1. What is the impact of a same-task repetition version of the poster carousel on adult ESL students' L2 fluency? 2. What is the impact of a parallel-task repetition version of the poster carousel on adult ESL students' L2 fluency? 3. What possible explanations are there for any observed fluency changes during the poster carousel?

METHODOLOGY Design
Intact classes of EFL learners at a private English language school in Central London were randomly assigned to take part in either a same-task repetition (STR) poster carousel training session (n = 24) or a parallel-task repetition (PTR) carousel training session (n = 22) during their normal scheduled English class.In both conditions, participants gave three oral performances (a narrative task)-in the STR group, the three performances related to the same task and in the PTR group they were three different tasks of the same type-and this performance data was subsequently analyzed quantitatively (see dependent variables section below).A mixed between-and withinsubjects design was used to look for fluency changes within-subjects across three task performances and between-subjects to compare changes between performances for the two different groups.To answer RQ 3, a full set of annotated performances of a single learner was examined.

Dependent Variables
This study adopts three dependent variables which are traditionally associated with speed, breakdown and repair aspects of utterance fluency, respectively.
Articulation rate.Following De Jong and Perfetti (2011), articulation rate was calculated as the total number of raw syllables produced in the 1-minute sample from each performance that was analyzed, divided by total speaking time (excluding all pauses >250 ms), and multiplied by 60.While drawl measurement would have allowed a more precise indication of speed changes related to formulation, it was not performed on the entire sample because of time constraints but is discussed as part of the case study analysis.
Frequency of mid-clause pauses.Skehan et al. (2016), Huensch and Tracy-Ventura (2017), Lambert, Kormos, and Minn (2017) and Lambert, Philp, and Nakamura (2017) have all suggested that between-clause pauses are related to conceptualization and mid-clause pauses to formulation.Arguably, both filled and silent pauses, and combinations of the two (composite pauses) signal processing delays (Kormos, 2006).Frequency of pause, rather than length seems to be most associated with judgments of fluency (Suzuki, Kormos, & Uchihara, 2021).Accordingly, the breakdown fluency measure selected for this study was frequency of mid-clause pauses which was calculated by dividing the total number of mid-clause pauses (filled, silent, and composite >250 ms) by the total sample time and multiplying by 60.A clause was defined as being either an independent or subordinate clause or sub-clausal unit1 (Foster et al., 2000).
Frequency of repetitions.In L2 speech analysis, a range of repair behaviors (repetitions; reformulations; self-corrections) are typically bundled together and an overall frequency calculated.However, as D€ ornyei and Kormos (1998) explain, while repetitions are likely to represent a need to stall (indicative of greater strain on processing) for time, other repair behavior might be more related to monitoring (indicative of reduced strain on processing).Putting these together might have the effect of canceling one another out.Therefore, repair was calculated as the total number of repetitions only divided by total sample time and multiplied by 60.

Procedure
Through piloting and discussion with class teachers, two new versions of the poster carousel were developed that required learners to create storyboards and talk about extreme life experiences.
Same-Task Repetition Carousel Procedure.

Teacher divides class into AB pairs and gives each pair an
envelope with one of seven different stimulus texts (Appendix S1). 2. Each pair creates a six-frame storyboard on poster paper which illustrates the text they were assigned.3. Teacher pins posters on the wall around the classroom and each pair stands by their poster (Figure 1). 4. As stay with their posters while the Bs move in a clockwise direction to the next poster.Each A uses their poster to tell story to their B visitor and Bs can ask questions (Figure 2). 5. Steps 5 and 6 are repeated until each A has told the same story 3 times to 3 different visitors (Figure 3).6. Learners go back to original pairs and swap places.The process is repeated but the Bs will now be telling the story and the As will be visiting other posters.
For the current research, a slightly modified version of the STR procedure described above was used so that learners in the STR condition first listened to a story from another person before repeating that story three times rather than presenting their own story three times.This was to because participants would be more familiar with the story for which they had designed the storyboard and may have impacted on between-subjects test (described below).The unmodified version is presented above, as this is the version most likely to be used in the language classroom.1. Teacher divides class into AB pairs and gives each pair an envelope with one of seven different stimulus texts.2. Each pair creates a six-frame storyboard on poster paper which illustrates the text they were assigned.3. Teacher pins posters on the wall around the classroom and each pair stands by their poster (Figure 4). 4. As stay with their posters while the Bs move in a counterclockwise direction to the next poster.Each A uses the poster to tell story to their B visitor (Figure 5). 5. Bs and As now SWAP PLACES (Figure 6).6.As move in clockwise direction to the next poster and Bs present this story to their new visitor (Figure 7).7. AAs and Bs SWAP PLACES again.8. Bs move anti-clockwise to the next poster and As present this story to their new visitor.9. Steps 7-10 are repeated until all As and all Bs have told three different stories.
In the PTR condition, speakers spoke about three different topics, but the procedure followed ensured that the stimulus stories were naturally counterbalanced among the participants to limit the possibility of a task effect.

Materials
The stimulus texts were seven written, first-person descriptions of extreme life experiences (see example text in Appendix S1) from the "Experience" column of the Guardian newspaper (www.theguardian.com/experience): Cow, Ravine, Baby, Hippo, Cat, Toddler, and Railings.Careful piloting of the materials revealed that they generated oral accounts that were similar in length and structure.The six-frame, picture-only, storyboards that were created (see Figure 8) supported the students during their retellings.The six-frame storyboard is a highly popular stimulus with a long tradition of use in L2 studies (e.g., Polio & Gass, 1998;Suzuki et al., 2022) and asking learners to create their own storyboards embeds the research within the conventions of a communicative classroom (Lambert, Kormos, & Minn, 2017;Lambert, Philp, & Nakamura, 2017).

Analysis
All three performances for each participant were audio-recorded using Sony ICDBX140 Digital Voice recorders.Participants were given up to 2 minutes to tell their stories, but they did not always use all the allotted time so a 1-minute extract was taken from the beginning of each performance.
Recordings were transcribed both orthographically and using the software PRAAT (Boersma & Weenink, 2013).Automatic fluency analysis (e.g., using PRAAT's textgrid to silences feature or De Jong and Wempe (2009) PRAAT syllable nuclei script) was not a viable option in this study because of the levels of background noise as well as the specificity of the fluency measures.Therefore, each extract was annotated manually by the author using the standard textgrid feature.Runs of speech, pauses greater than 250 ms,3 and repair phenomenon were identified through repeated, detailed listening to the recording accompanied by visual inspection of the spectrogram with the screen view zoomed in to, at most, 200 ms (see Figures 9 and 10).Pause position (between-versus mid-clause) and type (silent, filled, composite) were also recorded on the PRAAT textgrids manually by a process of opening the PRAAT file and corresponding orthographic transcription (with clauses marked) simultaneously.Non-verbal phenomena such as laughter, coughing and throat-clearing was discounted from analysis.
A unique PRAAT script was developed which would generate frequencies and durations for the intervals that had been manually created.This output was then used to calculate the specific fluency measures as outlined above.
A 10% sample of the data was re-examined for syllable count (used in the calculation of articulation rate) and pause boundaries by a trained researcher.The second rater coded 22 samples of speech (roughly 10% of the total data) and Spearman's correlations revealed an exceptionally high reliability score (0.993) between scorers.The time-consuming and specialized nature of manual PRAAT analysis meant that it was not possible to get an inter-rater score for all specific calculations and this needs to be considered when interpreting results.

Statistical Analysis
The data were analyzed using SPSS for Windows.There were no outliers in any cell of the design.In general, dependent variables were found to be approximately normally distributed as assessed by inspection of Q-Q plots and Levene's test.Some very slight positive skewedness was observed for frequency of mid-clause pauses.However, given that the skewedness was slight and MANOVA is considered robust to this sort of violation (e.g., Bachman, 2004), a decision was taken not to transform the data.
There was homogeneity of covariances, as assessed by Box's test of equality of covariance matrices.Mauchly's test of sphericity indicated that the assumption of sphericity was met for the measures of articulation rate and frequency of repetition.The assumption of sphericity was violated, however, for the measure of frequency of mid-clause pause for the STR group: v 2 (2) = 6.428, p = .04.For that reason, the Greenhouse-Geisser adjusted values are reported for this measure.
Effect sizes for the MANOVAs are interpreted following Cohen (1988) who suggests that partial g 2 values of 0.01, 0.09, and 0.25 be considered as small, medium, and large effect sizes, and Cohen's d are interpreted as small (d = 0.2), medium (d = 0.5), and large (d = 0.8).Due to the small sample size, observed power is also reported.
To answer RQs 1 and 2, which involved establishing the impact on fluency across performances for each version of the poster carousel, follow-up one-way repeated measures MANOVAs were run for each group separately and results are discussed below.
RQ 1.What Is the Impact of a Same-Task Repetition (STR) Version of the Poster Carousel on Adult ESL students' L2 Fluency?
Articulation rate.For the same-task repetition group, descriptive statistics reveal that mean articulation rate increased from 214.43 sylls/ min during performance 1 to 228.78 sylls/min during performance 2. Mean articulation rate was 227.65 sylls/min during performance 3 (Table 1).Follow-up paired samples t-tests with alpha level set at 0.05 revealed significant differences in articulation rate between performance 1 and performance 2 and between performance 1 and performance 3 but not between performance 2 and performance 3 (Table 2) and this is illustrated in Figure 11.
Frequency of mid-clause pause.For the same-task repetition group, a downward pattern is seen across the three performances for frequency of mid-clause pauses (performance 1 = 7.94, performance 2 = 7.02, performance 3 = 6.27) (Figure 12).Paired samples t-tests fell short of significance between performances 1 and 2 and between performances 2 and 3.However, the difference between performances 1 and 3 is borderline significant (p = .057)(Table 3).The low observed power may be the reason for the weaker significance observed here.

RQ3. What Possible Explanations Are There for any Observed Fluency Changes during the Poster Carousel?
A nested case study (Kimmy) was selected on the basis of typical sampling (D€ ornyei, 2007) because Kimmy's utterance fluency profile was largely representative of the STR group overall.A deductive approach was taken to the case study analysis in the sense that the transcriptions (orthographic and PRAAT) were investigated to shine light on the utterance fluency findings (see Appendix S1 for full transcriptions).
Locating Speed Increases.To explore articulation rate changes in this study, Kimmy's performances were analyzed and syllables longer than 250 ms4 (drawls) were identified through inspection of the PRAAT spectrogram and highlighted on the orthographic transcriptions.All drawls in Kimmy's performances were coded according to their location (mid-clause versus between-clause) and were also measured for length.Details of this analysis can be found in Table 5 alongside the same information relating to Kimmy's articulation rate.
The number of between-clause drawls remain relatively stable across Kimmy's three performances.The number of mid-clause drawls, on the other hand, drops considerably, from 12 during performance 1 to 3 in performance 2. In Kimmy's case, then, some of the increase in articulation rate between performance 1 and 2 can likely be explained by this substantial decrease in frequency of mid-clause drawls.
Lexical Priming and Verbatim Repetition.The transcriptions of Kimmy's performances reveal that Kimmy recycles some lexical content from previous performances (Table 6).During narrative detail 1, performance 1, Kimmy lengthens "fo~r" accompanied by a mid-clause pause before the subsequent low-frequency lexical item "charity."During the second performance, there is no hesitation before "charity," suggesting that the word is now primed for use.During narrative detail 5, we see a mid-clause drawl of 0.65 s during the article "a~" which precedes the noun "stick" and another mid-clause drawl during the  verb "wa~s" which precedes the prepositional phrase "in her chest".During the second performance, there are no drawls and the phrases "a big stick of wood" and "was in the chest of the pregnant woman" are uttered without significant hesitation.This suggests that lexical priming may be involved in fluency increases in Kimmy's case.
In detail 3, we see how a complete phrase is re-used in each of the subsequent performances.However, it is also noticeable that Kimmy does not engage in very large stretches of verbatim repetition and she seems equally likely to play with the order of elements in her narrative (detail 1, performance 3), abandon some elements (detail 2, performance 3) and introduce others (detail 4, performance 3).As she does this, we see hesitation around clause boundaries, which is in line with other studies that have considered the impact of (re)conceptualisation on L2 fluency (e.g., Felker et al., 2019) but a reduction in mid-clause hesitation.This suggests that whether language is re-used or not, the speaker is still finding it easier to formulate her utterances in subsequent performances.
Hesitation Clusters and Problematic Language.The analysis of Kimmy's performances highlights particular moments of high dysfluency where drawls co-occur with other hesitation phenomena like filled and silent pauses in hesitation clusters (Figure 14).
The lengthiest clusters tend to occur around lower frequency or otherwise problematic language.Table 7 highlights how Kimmy seems to struggle with her explanation of the car leaving the road.This is evidenced by the mid-clause hesitation cluster around verb + preposition "went away" (performance 1) and "get away" (performance 2) FIGURE 14. Hesitation cluster during Kimmy's first performance.

POSTER CAROUSEL IN THE ESL CLASSROOM
suggesting that she is unsure of the target construction (presumably, "went off" or "veered off").What is interesting is that, during the third performance, this detail is abandoned altogether.Other constructions which cause mid-clause hesitation are abandoned after the first performance, such as the detail regarding the steering wheel (Table 7).
On the other hand, Kimmy's third performance includes a hesitation cluster which precedes the use of the genitive case, the first time that Kimmy has attempted this structure across the performances (Table 6, detail 5, performance 3).It is possible that the mid-clause drawling seen here relates to the processing and production of that more complex structure, suggesting that not all problematic language is abandoned in repeated performances.
Summary of Case Study.Speed increases in Kimmy's case may be related to a reduction in drawls in the mid-clause position, suggestive of swifter formulation during subsequent iterations.Some of this swifter formulation may be related to recently activated (primed) language (e.g., the words "charity," "heavily," "stick").However, the case study also revealed Kimmy might be consciously selecting language which is more familiar to her (proceduralized language).Kimmy also seems to experiment with more complex language during the final performance.This study found that articulation rate increased between same-task repetition performance 1 and performance 2. Articulation rate is a crude measure, potentially linked to both conceptualisation and formulation.The articulation rate gains for Kimmy corresponded with a considerable reduction in drawls in the mid-clause position between performances 1 and 2, suggesting an association with the formulation stage of the speech production process.Interestingly, the speed of articulation did not increase further between performances 2 and 3.There is some evidence to suggest that this could be associated with increased experimentation with language during performance 3 leading to dysfluency.
The only other study which has looked at same-task repetition and its effect on articulation rate is De Jong and Perfetti (2011).They found that articulation rate increased significantly during same-task repetition, but the largest gains were between performances 2 and 3 which is in contrast with the findings of the current study.An explanation could be that it was only the final 2-minute performance which constituted time pressure for the learners in their study.An alternative explanation is that the tasks placed different demands on participants; the fact that the task in the current study was a narrative task as opposed to an opinion task, and perhaps the specific combination of repetition and planning inherent in the poster carousel made earlier increases in articulation rate more likely.
Previous studies have argued that speed increases during task repetition could be linked primarily to speakers reusing constructions (Suzuki et al., 2022;Thai & Boers, 2016) and there is evidence to suggest this is also the case here.However, the case study suggested that the speaker was also reconceptualising the message, abandoning certain elements, playing with order and even experimenting with more complex language.In the poster carousel, participants may be benefitting from opportunities to reuse some constructions but may also reinvest increases in cognitive fluency to monitor their speech and use the information to make strategic changes during subsequent performances.
Frequency of mid-clause pauses reduced gradually over the three performances resulting in a borderline significant decrease between performances 1 and 3.This result mirrors that of Lambert, Kormos, and Minn (2017) and Lambert, Philp, and Nakamura (2017), who similarly found a step-wise reduction in frequency of mid-clause pause.
The current study adds to our understanding of pausing during task repetition with the indication that mid-clause pauses tend to appear around problematic language such as in the middle of phrasal verbs.
In the current study, there was a reduction in the frequency of repetitions between the first and second performances, following a similar trajectory to articulation rate.This is in contrast with other studies which have found no effect for same-task repetition on repair (e.g.Wang, 2014).Lambert, Kormos, and Minn (2017) and Lambert, Philp, and Nakamura (2017) also found no effect for task repetition on repair (frequency of overt self-corrections) over the first three performances of the tasks, although they did find a reduction in the need to repair on subsequent task performances.A possible explanation for the differences in findings here is that this study used a repair measure that targeted stalling as opposed to monitoring behavior.

Parallel-Task Carousel and Fluency
In the current study, and in contrast to a number of other studies (De Jong & Perfetti, 2011;Lambert et al., 2021;Pinter, 2005;Sample & Michel, 2014), there was no significant change in participants' fluency during the parallel-task version of the poster carousel.This might suggest that it is primarily the content repetition, or lexical priming, during same-task repetition that drives the increases in fluency that were observed in this study.Fluency gains for parallel-task repetition in De Jong and Perfetti (2011) study may have been because the participants in their study were performing their tasks under increasing time pressure and participants had less time to fill with speech in later performances. 5In Sample and Michel (2014) investigation of paralleltask repetition and oral performance, gains in fluency could be attributed to content repetition because the repeated versions of the task were actually identical in terms of the lexis they required of students.Lambert et al. (2021) found that parallel-task repetition was associated with almost identical speed fluency gains as the same-task repetition condition.However, they used a global measure of fluency-speech rate-while the current study opted for a pure speed measure (articulation rate).It is possible, then, that speech rate captured changes (e.g.between-clause pauses) not detected by the measures used in the current study and which may be associated with task familiarity/structural priming.
An alternative explanation is that, because the parallel-task carousel involved passing stories from one participant to the next, details of the narrative may have been omitted leading to conceptual difficulties during the retelling.In addition, the interlocutors in the parallel-task group tended to ask more questions.This is likely because they knew they would have to tell the same story and wanted to be sure they understood the details.It meant that speakers in the parallel-task group were less likely to launch into a prepared narrative and had to (re)conceptualize the message, potentially creating further dysfluency (Felker et al., 2019).

Implications
The study has shown that the poster carousel can be a useful tool to explore task repetition effects in the classroom context and echoes findings of other research into the impact of same-task repetition on L2 fluency, thereby providing support for the use of this inherently repetitive task sequence in the language classroom.Hopefully teachers are inspired to experiment with the carousel technique and explore task factors, learner outcomes, and attitudes further.
The current study has demonstrated that articulation rate alone can be rather blunt, but when taken alongside an analysis of drawl length and location, provides a more nuanced picture of underlying cognitive fluency.Similarly, while many studies combine a range of repair behaviors for fluency analysis purposes, the findings of the current study add support to the hypothesis that repetitions may perform a very different function to other repair measures.This may provide an explanation as to why a growing number of studies have found that repair measures do not correlate with subjective ratings of fluency (Bosker, Pinget, Quen e, Sanders, & De Jong, 2013, Tavakoli et al., 2020;Suzuki et al., 2021) and that studies of task conditions seem to have contradictory findings when it comes to effects on repair.Indeed, there is support here for an approach to L2 utterance fluency analysis which groups dependent variables in terms of the conceptualisation, formulation and monitoring stages of speech production (Lambert et al., 2021) as opposed to speed, breakdown and repair.
Finally, numerous studies have sought to capture longer term or transfer fluency benefits of task repetition.The case study presented here has shown that increased fluency during task repetition may not be expected to translate into longer term increased fluency but may POSTER CAROUSEL IN THE ESL CLASSROOM have been invested in other ways, giving rise to increased complexity or accuracy in the longer term, for example.

Limitations and Directions for Research
Firstly, and similarly to Suzuki et al. (2022), the sample size used in this study (n = 46; n = 22 for one-way analyses) was smaller than recommended in the a priori power analysis performed using G-Power.Care has been taken to present observed power and effect sizes, but caution is advisable when interpreting inferential statistics.Manual fluency analysis of classroom data is time-consuming, but replications could be carried out which make use of existing PRAAT functions and specially created scripts to measure fluency automatically (De Jong & Wempe, 2009) on a larger scale, perhaps in conjunction with multiple nested case studies which provide more targeted fluency data.
Crucially, further empirical research is needed to test hypotheses generated here in relation to location of speed increases during task repetition, how formulation relates to both lexical priming and strategic message changes as well as whether the specific task demands of the poster carousel are responsible for some of the findings.Finally, it would be valuable to explore learners' perspectives on their fluency increases during the poster carousel activity.This could be achieved through stimulated recall and post-intervention interviews, for example (Kahng, 2014).
Finally, carrying out research in an authentic classroom and with genuinely communicative tasks means that the conditions cannot be as controlled as in a laboratory environment.The tasks used in this study were planned very carefully for internal validity however task factors and group dynamics may have played a role in how these interventions impacted on fluency.Future research might seek to disentangle some of the threads discussed here.

CONCLUSION
This paper has reported on a classroom-based study of the poster carousel and L2 fluency.The study suggested that the immediate same-task repetition inherent in the poster carousel allows learners to perform tasks with increased oral fluency.For this particular classroom activity, it seems that the fluency changes are more likely related to reductions in the time needed to formulate content and this may be due in part to lexical priming and also to the learners selecting language which is more familiar.The parallel-task repetition version of the poster carousel did not seem to lead to higher L2 fluency in learners, although the specific nature of the parallel-task repetition carousel could have been an intervening factor.It is hoped this study underlines the value of classroom-based enquiry, and the use of authentic tasks in L2 fluency research.

FIGURE 1 .
FIGURE 1. Classroom set-up for STR poster carousel with pairs of learners standing in front of their posters.One pair is highlighted for illustrative purposes.

FIGURE 2 .
FIGURE 2. Bs moving to next poster.Asterix shows the speaker.

FIGURE 3 .
FIGURE 3. Bs moving to the next poster.Asterix indicates the speaker.

FIGURE 4 .
FIGURE 4. Showing classroom set-up at the beginning of the PTR poster carousel.One pair is highlighted for illustrative purposes.

FIGURE 5 .
FIGURE 5. Showing Bs moving to the next poster.Asterix indicates the speaker.

FIGURE 6 .
FIGURE 6. Showing As and Bs swapping places.

Note:
Pauses are in parenthesis -m = mid-clause; b = between-clause; s = silent; f = filled; c = composite.Bold indicates a drawl.Between-clause pauses and drawls are underlined.

TABLE 1 Descriptive
Statistics for the Three Dependent Variables of Articulation Rate, Frequency of Mid-Clause Pause and Frequency of Repetitions RQ2.What Is the Impact of a Parallel-Task Repetition (PTR) Version of the Poster Carousel on Adult ESL students' L2 Fluency?

TABLE 2
Pairwise Comparisons STR Group-Articulation Rate (AR) FIGURE 11.Means across performances for articulation rate.POSTER CAROUSEL IN THE ESL CLASSROOM comparisons were therefore not carried out for the PTR group but the group means can be observed alongside the STR means inTable 1 and Figures 11-13.

TABLE 5
Analysis of Kimmy's Three Performances During Same-Task Repetition Poster Carousel

TABLE 7
Example from Same Point in Narrative Across the Three Performances