Learner agency and its effect on spoken interaction time in the target language

Universitat Internacional de Catalunya ebarbera@uoc.edu This paper presents the results of how four dyads in an online task-based synchronous computer-mediated (tb-scmc) interaction event use their agency to carry out speaking tasks, and how their choices and actions affect time spent interacting in the target language. A case study approach was employed to analyse the language functions and cognitive and social processing that occurred in audio recordings of spoken interaction between four dyads, alongside other indicators of pre-task behaviour, triangulated with results from learner questionnaires. The study revealed that whilst all cases engaged in overt spoken interaction, some cases also avoided the designed task and engaged in covert pre-task planning. Learners’ ability to reconfigure 1) the time mode of the task design; 2) the ways in which technological tools were used and 3) language choice, all impacted on their time spent interacting in the target language. The findings highlight tensions between learners’ choices across the three dimensions that they had reconfigured, raising questions as to how to support time in synchronous interaction in the target language whilst supporting learners’ agency. The implications are presented and discussed.


introduction
Increasing global access to and widespread use of Information Communication Technologies means that new tools to facilitate synchronous spoken interaction are emerging.This in turn implies potentially new and diverse opportunities for learners to exercise agency whilst practicing their speaking skills.Although the actual impact of practising speaking skills via synchronous computer-mediated communication (henceforth voice-based scmc) has not yet been sufficiently examined, oral scmc still provides excellent opportunities for students to practise their English with flexibility of time and place (Yang & Chang, 2008).Having enough opportunities and time to practice speaking is not only a concern for teachers and designers (Appel, Robbins, Moré & Mullen, 2012;Hampel & Hauck, 2004;O'Dowd, 2000& Wang, 2006) but also learners (Hurd, 2007).Students can be concerned with developing fluency, having enough practice and finding opportunities to talk to others (Hurd, 2007).Therefore, maximising opportunities for speaking practice is not only an important goal from these perspectives but it is also deemed that language is best learned and taught through interaction, which in turn contributes to language gains (de la Colina & Garcia Mayo, 2007;Gass & Mackey, 2006).Specifically, rapid real-time interaction, which is necessary to develop fluency, can be facilitated through spontaneous synchronous tasks.Not only can tasks be designed to allow for maximum speaking time for students (Stickler, Batstone, Duensing & Heins, 2005) but advances in Voice over Internet Protocol (VoiP) and related tools such as Skype boost the possibilities to facilitate synchronous peer-to-peer spoken interaction in the target language (henceforth tl).Although 'sufficient practice' can be understood as the number of opportunities to speak on offer during a language learning course, it can also mean length of time interacting.
Despite the pedagogical and technological factors that contribute to the task design to support spontaneous synchronous tasks, learners may make and act on choices that may not run in accordance with this goal.Learners may exercise agency in ways that do not necessarily optimise opportunities for spoken interaction in the tl, and which appear contrary to their own desire to practice as part of their (assumed) goal of learning a language.The problem therefore is that although the exercising of agency may have inherent value (Schwartz & Okita, 2009) based on the belief that learners are agents who "play a defining role in shaping qualities of their learning" (Dewaele, 2009, p. 638), learners may use agency in ways that may not make optimal use of the opportunities to interact orally with others in the tl.
This study attempts to explore how students exercise agency in two online tasks and how this affects time spent interacting in the tl.It describes a task-based scmc event for spoken interaction and the choices learners make in relation to the 'task-as-workplan' (Breen, 1987) or task (design) expectations for synchronous, spontaneous interaction.Some learners' choices result in a reconfiguration of task dimensions that run contrary to taskas-workplan.These are: 1) language choice of either first language (l1) and/or the tl; 2) choices relating to technological tools and 3) choices relating to time mode (i.e.synchronous and/or asynchronous).Specifically, we are interested in learners' choices and actions in relation to each dimension and how they intersect.Because the interaction time in the tl is important for teachers, designers and learners, we also focus on how these three dimensions affect this task outcome.

Learner agency, speech and task-based learning
There is a plethora of definitions regarding learner agency.Martin's (2004) definition as "the capability of individual human beings to make choices and act on these choices in a way that makes a difference in their lives" (p.135) is used here over the more common "socioculturally mediated capacity to act" (Ahearn, 2001, p. 112).Although many agree that an individual's capacity to act is socioculturally, contextually and interpersonally mediated (Mercer, 2011), Martin's definition allows for a focus on agency as intentional behaviour during task processes.It goes beyond the conceptualisation of agency as a capacity to being a property of the individual.Agency is a fundamental construct in understanding learning processes and learner identities (Miller, 2012), and many perspectives agree that language plays a central role in current thought on agency.At the level of speech, language can be used as an action or process (Swain, 2006) alongside which, learners can use their motor and sensory systems in order to carry out agentic actions (Bandura, 1999).In online tasks, this may mean using language for problem solving or responding to textual instructions, as well as physically navigating through a task for task completion.
Regarding tasks, Breen (1987) highlighted differences between 'task-as-workplan' (concerned with expectations and intentions of task design) and 'task-in-process' (what learners actually do).Conceptualising pedagogical tasks in these two ways helps to highlight "the notion that learners, as active agents in learning processes, can modify activities according to their own intentions -modifications which may or may not be in direct accordance with the initial intentions of that 'task-as-workplan'" (Dooly, 2011, p. 72).Task design intentions and the expectations arising from them can also be conceptualised as 'lines of desire', an arquitectural term that Lukin and Du Boulay (2003) apply to technology-use to highlight the path or trajectories that people make but which are often shortcuts that ignore the given route.

Language choice and language avoidance as choice
Sociocultural perspectives of language learning have suggested that students learning a new language can use their l1 to serve a variety of functions, which can ultimately support learning processes and outcomes.This can include task management (Macaro, 1997), scaffolding while working on a task (Kötter, 2003), identity negotiation (Myers Scotton, 1983), discussing unknown language words (Knight, 1996) and social interaction (Tarone & Swain, 1995).Language choice and control of language use, including codeswitching between languages, can be understood as a naturally occurring expression of learner agency (Garcia, 2009) forming part of students' multilingual repertoire for different or identical practices.Some research indicates that multilingual practices can contribute to the eventual construction of a final monolingual output (Dooly, 2011) scaffolding cognitive and communicative activities which eventually allow speakers to participate in monolingual activities at the end of the process (Borràs, Canals, Dooly, Moore, & Nussbaum, 2009).Some researchers have found that multilingual language learners working towards monolingual task accomplishment tend to shift between different types (or stages) of l1 and target language use (Borràs et al., 2009;Masats, Nussbaum, & Unamuno, 2007).Their codeswitching allows them to overcome communicative obstacles, facilitating a final stage where the learner can maximise the use of the tl (Dooly, 2011).A learner's plurilingual repertoire can be understood as "resources for practicing agency, that is, the right to make and to enact their own linguistic choices, in goal-orientated and context-embedded situations" (Vitanova, Miller, Gao & Deters, 2014, p. 8).Through language use multilingual subjects can exercise agency (Kramsch, 2009).
Reasons cited for learners choosing to use the l1 as outlined above may fall within what Schwartz and Okita (2009) call "causal-focused and rights-focused applications of agency" (p. 7).Whereas the former is concerned with what conditions foster learning, the latter is concerned with protecting or enabling people's access to a particular form or expression of learning.
Other research that is concerned with the conditions that foster learning includes studies that conceptualises language choice as an avoidance strategy.For example, in Musk's (2014) research on learners' language choices and use of Google translation tool he found that learners made incremental choices that avoided the tl (or favoured the l1) during computer-assisted project work: learners acted upon their language preference, opting to read in their l1.He also noted that learners tended to be product-oriented, drawing on their previous experience to get the job done quickly and efficiently, including their experience of and familiarity with the technological tools.He proposed that different learners rely more on their need or choice to translate and use translation tools because of differences in confidence and reading proficiency in the tl.Musk (2014) suggests that the study of language choice may have been impacted by Tarone's (1978) study, which groups language switching not as an avoidance strategy but rather as a subcategory of conscious language transfer: the conscious judgement that something in the native language -most typically -and something in the tl are similar, if not actually the same (Odlin, 1989).
We now present other studies that involve factors relating to choice and avoidance.This is in order to extend the literature review to include studies that are also concerned with what conditions foster learning.

Avoidance, spoken interaction and off-screen behaviours
Learner avoidance of interacting in the tl is commonly studied as a communication strategy during spoken interaction as learners avoid syntactic or lexical items, and topics and concepts that pose language difficulties or pronunciation issues (from Tarone's framework, 1981) including in scmc oral tasks (e.g.Kim, 2014).Within studies of avoidance as a communication strategy, Horwitz, Horwitz, and Cope (1986) suggested a relationship between unwillingness to interact verbally and Foreign Language Anxiety.Research suggests that this tendency is influenced by factors such as personality traits or overall unwillingness to communicate (Levine, 2003).The current study, however, focuses on avoidance of carrying out the 'task-as-workplan' because this provides us with a more complete understanding of avoidance in online tasks.
Avoidance and tool use were noted by Appel, Robbins, Moré and Mullen (2012), who explored how different interface versions affected spoken interaction tasks designed for spontaneous, synchronous interaction.They found that over 60% of the learners had looked at the materials and prepared beforehand, therefore avoiding the synchronous mode of the task.Students reported being more nervous when working with the interface version (Tandem tool), which required "a degree of improvisation and spontaneity which put additional pressure on the students" (Appel, Robbins, Moré & Mullen, 2012, p. 18), over another interface version, which allowed preparation.
No other specific studies of avoidance of oral cmc tasks were identified in the research literature.However, there are a number of studies of off-screen and out-of-class behaviours, which are complementary in the sense that they give insight into learner processes.
Studies of learners' off-screen behaviours with cmc largely fall into two groups (Suzuki, 2013): online contexts where learners are in dyads with other learners and native speakers of the tl (González-Lloret, 2011;Tudini, 2010), or in their physical environment (Kitade, 2008;Leahy, 2004).In the latter group, Suzuki (2013) studied one learner in front of the computer during a teacher-led synchronous Japanese class mediated by audio-based conferencing software, where class participants' behaviour in their physical environments were not observable to others.Results revealed that the learner gained significant affordances from the online/off-screen course format, allowing her to create her own learning opportunities including actively taking private turns in her physical environment without being heard by others.Hafner, Li and Miller (2015) studied university learners' non-online out-of-class behaviours in an English course project work.They found that students' computer-mediated interactions (Facebook, WhatsApp and email) were plurilingual, with students drawing on English, Chinese and mixed code to different extents.Different languages were used strategically: whereas l2 was used more in the construction of the final project product, l1 was used more to promote group cohesion.
Hampel and Stickler ( 2012) also noted off-task conversations between students in a videoconferencing session which they stated usually occurred in the mode other than the one the teacher was using.
Other factors related to language avoidance pertain to learner attitudes and beliefs about l1 use in the classroom.Learners' attitudes can affect the extent and function of own-language use in the classroom and its potential contribution to learning (Hall & Cook, 2012).Beginner university learners were found to prefer l1 use for classroom management and suggested that its use reduces anxiety confirming a positive affective role that l1 use can play (Rolin-Ianziti & Varshney, 2008).

Interaction time, time mode preference and planning time for tasks
Interaction has long been acknowledged as one of the most influential factors contributing to language learning (Boonsue, Jansem & Srinaowaratt, 2015).Although many studies on scmc have focused on the number of turns taken as an indicator of the amount of interaction taking place (Blake, 2005;Jepson, 2005), other researchers have used the amount of time in interaction as a measure of user engagement in the tl (Stickler, Batstone, Duensing & Heins, 2005).Interaction time has been considered by various researchers, (although not using this exact term), in relation to the tension between time spent in the l1 or the tl.
Whilst some argue that there is a case for using the l1 in the classroom as learners change languages and l1 use can be very time-efficient in certain situations (Üstünel & Seedhouse, 2005) others argue that speaking in the tl in the classroom should occur as much as possible (Moeller & Roberts, 2013).Levine (2011) proposes that a plurilingual pedagogy, including the positive use of l1 as a language choice, would increase the total interaction time in the tl.He proposes that while the absolute ratio of tl to l1 communication might decrease, the absolute amount of time spent communicating in the tl would increase because students would talk more.
Regarding time mode, scmc is not flexible (Levy & Stockwell, 2006) since learners engage in 'live' communication with partners, so they must schedule specific times for study.Some researchers consider scmc or asynchronous (acmc) time modes as being learner preferences or learning styles (Benbunan-Fich & Hiltz, 2003;Wang, Wang, Wang & Huang, 2006).These are behaviours related to the psychological, cognitive, and affective domains of interaction within learning environments, which also involve learners' preferred ways to receive, process, and recall information during instruction (Aragon, Johnson & Shaik, 2002).Shahabadi and Uplane (2014) found distinct differences in learning styles between learners in different time modes.
The off-screen or private talk (Suzuki, 2013) that learners engage in during online interaction can also be understood in terms of planning time and learner performance in the tl.
Types of task planning are differentiated based on when planning occurs namely pre-task planning and online planning (Ellis, 2005).The planning pertinent to this current study is pre-task planning which can be 'strategic planning time' (i.e.deliberation of content and code) or 'rehearsal time' (i.e. a practice run through of the task), both of which can occur prior to task performance (Ortega 1999;Ellis 2005).Planning allows learners to attend to language as form and studying planning gives an insight into what learners attend to, and what effect it has on the way they use language (Ellis, 2005).Although research suggests that planning affects ways in which learners perform a task, there is very little research about what learners actually do when they plan (Ellis, 2005).Batstone (2005) takes a sociocognitive view of planning highlighting that learners can approach tasks in two ways: requiring economical and efficient communication or providing opportunities for engaging in learning activities.Furthermore, with respect to agency and planning, Batstone suggests that some language learners are more 'face sensitive' (Batstone, 2005;Aston, 1986) than others and "exploit principles of clarity/economy to the hilt out of a concern with self-protection rather more than with self-expression" (Batstone, 2005, p. 288).Ellis (2005) suggests that the context of tasks also shapes how learners plan for and perform tasks such as testing conditions.With regard to pre-task planning, studies have largely demonstrated a benefit for complexity (the use of more advanced or more diverse tl features), accuracy (the avoidance of error during production) and fluency (real-time rapid language production), with studies of accuracy being less consistent (Sauro & Smith, 2010).In spontaneous synchronous spoken interaction tasks, the affordance lies in offering opportunities to develop fluency, or a learner's capacity to mobilize his or her system to communicate in real time.However, some studies have highlighted learners' perceived deficiencies in modalities resulting in anxiety about speaking with synchronous video or audio tools, which give learners little time to rehearse their statements and which can thus create anxiety (Hampel & Stickler, 2012).

Purpose of the study
The aim of this study is to explore the relationship between agency and avoidance by focusing on three dimensions of learner choices relating to technological tools, time mode and language use.We aim to understand why some learners may have chosen to avoid the synchronous spoken interaction tasks, which we understand was a form of planning and as a way for learners to exercise their agency.We aim to give some insight for future design for synchronous spoken interaction tasks, and into how online learners may be scaffolded better.The research questions are: 1.How do learners exercise their agency in a synchronous online spoken interaction event in relation to choice and control of a) technological task features; b) language used; c) time modality?2. How do the choices intersect?
3. What effect do the choices have on time spent in the tl?

Participants
The participants were students in an English as a Foreign Language (efl) class as part of their degree programme at a fully virtual university.The learners were in a B2.1 level (lower intermediate) group.Eight adult students, three male and five female, aged between 26-55 years old were engaged in a virtual synchronous peer-to-peer speaking task.Students were bilingual (Catalan and Spanish) with English as an additional language.

Materials
Learners were asked to complete two previously unseen tasks.They had the same text-based instructions and photo for partners A and B, as well as the means to navigate.Each dyad had to collaborate to complete two out of four tasks on the topic of 'travelling'.The first was an information gap task (spot-the-difference) and the second was an opinion sharing task.Learners had four differences to find.The second task used the same photo as the first, accompanied by an open question: What kind of activities can people do in a holiday destination like this?General instructions and guidelines were available to students as online documents but some instructions were also offered in text form on the interface screen hosting the task.Tasks were compulsory course assignments and there was no time limitation for tasks.

Design
The study adopted an exploratory case study approach where students form four dyads.Case studies in language learning facilitate an understanding of learners' issues, experiences, developmental pathways, insights, or performance within a particular linguistic, educational or social context (Duff, 2014).The dyads came from a data set from a previous study by Appel, Robbins, Moré, and Mullen (2012) that explored the effect of different interfaces within a technological tool in which the results from a learner questionnaire indicated that some learners had scripted or pre-prepared their spoken interaction beforehand.Therefore, a purposive sampling approach was used to further explore this phenomenon and its effect on spoken interaction.Cases that showed evidence of engaging in spontaneous speech (e.g.interuptions, shorter turns) and those that did not (e.g.longer stretches of speech without interruption or overlap and far fewer turns) were included in the sample.

Procedure
Recordings of peer-to-peer spoken interaction was captured using a plug-in for Skype, a free video and audio conferencing tool, and started from the start of the first task until the end of the second resulting in approximately 23 minutes of data.Recordings were transcribed and converted to a text document.The transcriptions were then coded using the analytical framework of peer group interaction developed by Kumpulainen and Muntanen (1999).This supported a microanalysis of evolving peer interactions, focusing on three analytical dimensions: the functions of verbal interaction, cognitive processing and social processing.The functions of verbal interaction supported analysis of what learners are doing with language using codes such as Informative, Expositional, and Organisational.The analysis of cognitive and social processing focused on interactive dynamics as they occurred across the participants.The cognitive processing focused on ways in which students approached and processed learning tasks, highlighting working strategies, situating positions towards knowledge and learning and towards themselves as problem solvers using exploratory or procedural speech, for example.The social processing characterised the social relationship and types of participation in peer groups, such as collaborative and individualistic.

Results and Discussion
In order to understand the results and analysis, first we present a descriptive summary of two different general trajectories taken by the four cases outlined in Table 1 below, which highlights how learners used their agency to follow 'task-as-workplan' (Breen, 1987) or form their own workplan, trajectory or 'line of desire' (Lukin & Du Boulay, 2003).
The descriptions of the two main trajectories were based on the results from learner questionnaires where some learners indicated they had pre-prepared or scripted their interaction.The coding process of the cognitive dimension revealed that the code 'exploratory talk' was absent in the transcripts of cases 2, 3 and 4, yet learners would have needed to use exploratory talk to find the differences.The coding process also revealed that in cases 2, 3 and 4, turns for negotiation for organisational purposes were largely absent compared to case 1, who made a number of negotiated turns for organizational purposes.This confirmed that organization had already taken place for cases 2, 3 and 4 before the recordings.Cases 2, 3 and 4 also used the uncommon word "awnings" in their talk, indicating they had looked up the word meaning.In contrast, Case 1 used circumlocution to express their intended meaning when they did not know the lexical item as can be seen in example 2 below.
Other indicators and qualitative differences, noted in the interactions, were also evident in the transcripts, which we will subsequently demonstrate.These indicators were deemed sufficient evidence that covert planning for cases 2, 3 and 4 had taken place and that case 1 had engaged directly in overt synchronous interaction.
Indicators included the complete absence of the l1 in the transcripts of cases 2, 3 and 4, but which is present in the transcript of case 1 as can be seen in the examples below.The presence of the l1 (as in case 1) is to be expected in the interaction of bilinguals (Macaro, 2006) at an intermediate level, with whom learners share the same l1.Example 2 from Case 1: Spot-the-difference task M: …then at the bottom of the building I can see ...mmm... I'm sure it's not called umbrella, but you know that part to make the entrance of stores or something darker?H: Yes.
Example 3 from Case 2: Spot-the-difference task L: I see on my picture that there are some window shops with some green awnings about these windows... N: Er, in my picture the awnings are red so it must be the third difference.
Vocalised forms of private speech were also absent, which can also be expected in conversations intermediate level learners as an aid in the mastery of task-related difficulties (McCafferty, 1994).In case 1 private speech occurred when one learner was mumbling the written instructions to himself as shown in example 4 below, before entering into interaction with his partner.
Example 4 from Case 1: Spot-the-difference task M: OK… (mumbles reading the instructions) you can see the same picture.There are four differences.Then I don't know, er, I think this is a tube station, er, because of the signs, er, on the corner of the building… Expressions of task difficulty were also absent in cases 2, 3 and 4, but present in case 1 as shown in example 5 below.

H: Very tough
Table 1 below outlines the trajectories of the four dyads until the start of task one.Case 1 and their choices (choices 1-3) reveal one trajectory, evident in the transcripts.The covert trajectories regarding choices of cases 2, 3 and 4 are unknown, so they are considered as one trajectory.We recognise, however, that each dyad and learner will approach the task in unique ways.The description of the trajectories of cases 2, 3 and 4 is based on the fact that they had a common need to make certain choices, and acted on them in order to get to the point of recording.It is a trajectory that is not overtly evident in the transcripts, apart from the indicators that planning had occurred and which are described above.Therefore, choices 1-7 of cases 2, 3 and 4 are descriptions of covert choices taken before recording started, but are not necessarily an exact representation of the actual order of choices taken.

Case 1
Cases 2, 3 and 4 1. Choose not to look at the answers/press the button with the technological tool to start recording 2. Choose language to start interacting in (L1) 3. Choose when to change language code to TL (Spanish to English) Learners continue to sustain synchronous spoken interaction in English while attempting to resolve information gap task in TL (English) 1. Choose to look at the answers with the technological tool.The spot-the difference task is made redundant because learners have seen answers.Now the task is how to perform the task as if they had not looked at answers.
2. Choose language for discussion in the planning stage (L1 Spanish and/or Catalan and/or TL) 3. Choose (negotiate) time mode preference to manage the interaction for planning (synchronous/asynchronous or both) 4. Choose whether to script or prepare performance in L1 and/or TL (exact strategies unknown) 5. Choose to look up an uncommon word in English "awnings" (one of the differences in the photo) 6. Choose how to carry out the performance (language choice(s) unknown) 7. Choose to perform the task completely in the TL 1a) How do learners exercise their agency in relation to choice and control of technological task features?Case 1 followed 'task-as-workplan' (Breen, 1987) which involved pressing the button to invite their partner to start the task, and checking answers during task-processes with virtual buttons in the tool.Navigational choices in relation to the tool highlight learners' agentic actions using their "motor systems" (Bandura, 1999, p. 4).Their interaction results in a total of 11 minutes 43 seconds in synchronous spoken interaction, with two seconds interacting in the l1.Cases 2, 3 and 4 did not follow 'task-as-workplan' but instead followed their own workplan.They chose to use the tool features to look at the answers, resulting in much less time interacting synchronously in the tl: between 1 minute 38 seconds and 3 minutes 34 seconds.This choice, however, afforded dyads both 'strategic planning time' (i.e.determination of content and code) to be presented and 'rehearsal time' (i.e. a practice run through of the task) (Ellis, 2005).It is not known to what extent and how much time was dedicated to either type of planning time.
In addition, other foci for planning apart from rehearsal and strategic planning (Ellis, 2005) would have needed to be present for cases 2, 3 and 4 in order for the participants to have completed the task.Although they are unknown, the planning foci would need to have included a number of choices: to choose whether to follow 'task-as-workplan' or to look at the answers; to choose how to manage the task once they had looked at the answers (e.g. to script or prepare, and which time mode and tools would be used to communicate with each other); to choose who was going to say what and in what order depending on the degree of scripting vs. spontaneity they had planned and to decide on the co-ordination of their actions around how and when to record their interaction and to choose how to collaborate together so that it appeared that their recorded interaction was spontaneous.Based on this deduction, the number of choices for cases 2, 3 and 4 in relation to technology and management of their language performance appear to be extended and more complex than for case 1.Although their time interacting synchronously in the tl for the pedagogical task is less than in case 1 (see Table 2), their covert interaction and related behaviour means that they spent some considerable time managing the task in their own way.We cannot confirm, however, whether they were more active verbally offline (as in Suzuki's study in 2013) or if they interacted orally/textually, or synchronously/asynchronously during that time.1b) How do learners exercise their agency in relation to choice and control of language used?
The two trajectories revealed learner language choices as being different, but both trajectories relate to l1 use for organisational purposes or 'task management' as observed by Macaro (1997).Case 1 codeswitches from Spanish to English in order to negotiate an organizational move whilst simultaneously starting the task.The fact that case 1 codeswitched in the recording (knowing the teacher will evaluate it) implies a positive attitude to l1 use.The amount of interaction time in the tl was greater than in the other dyads.This supports Levine's (2011) suggestion that if a plurilingual approach is employed (in this case by learners during the task itself) learners can have more interaction time in the tl, because they are generally talking more.The code switch supported task processes (collaboration and negotiation) and task outcomes (task completion and time interacting in the tl).However, if cases 2, 3 and 4 also employed their l1 in the pre-task planning we could also say that a plurilingual approach was employed by learners leading to a final monolingual output (Dooly, 2011;Hafner, Li & Miller, 2015) Cases 2, 3 and 4 do not use their l1 during the recordings.Their organisation of the task has occurred in the planning stage.We do not know what language(s) they used in pretask planning, but the fact that the l1 is totally absent in the recordings can be explained by beliefs or attitudes towards l1 use (Hall & Cook, 2012) or not needing to use the l1 for language purposes such as lexical problem solving.The second explanation, however, is unlikely, given their language level; some l1 use was to be expected because in learning activities, the language of thought for all but the most advanced user of the language is inevitably in their l1 (Macaro, 2006).
It is unknown whether cases 2, 3 and 4 used their multilingual repertoires to control two (or more) languages in authentic communication (Kramsch, 2009) in the planning stage.While preparing their performance they may have written their preparatory work, individually or together or in a combination of both.This process, however, suggests some involvement of both their l1 and tl.The choice to use different languages for different purposes can be understood as a strategic choice by learners as found by Hafner, Li and Miller (2015) with tl use being used for the final product (Musk, 2014;Hafner, Li & Miller, 2015).
Cases 2, 3 and 4 prepared their contributions to perform in English rather than to completely abandon the task, suggesting that the avoidance strategy allowed them to scaffold their orientation towards task completion while working on the performance.Importantly, then, task avoidance is not a final choice but rather temporary until they start recording and begin the pedagogical task.The time it took for these cases to script or prepare for the recording is unknown.Despite this, a sustained amount of time is implied by the choices they would have needed to negotiate alongside the scripting and planning of the content itself.From this perspective, l1 use in the planning stage can be considered an efficient language choice as found in studies by Musk (2014) and Üstünel and Seedhouse (2005) to get the job done (Musk, 2014) as they manage the task implying the use of language for management, which they have not studied (Macaro, 2006).Case 1 also uses the l1 within a codeswitch also related to task management purposes.
Furthermore, the recordings of the interactions had to be sent to a forum and shared with other peers.Therefore, the choice to avoid a direct recording of their spontaneous interaction being made public (including potential errors and conversation in the l1) may be explained by any one or more of the performance anxieties: a fear of negative evaluation, communication apprehension or test anxiety (Horwitz et. al 1986).1c) How do learners exercise their agency in relation to choice of time modality?
Case 1 engaged in the synchronous task mode while cases 2, 3 and 4 reconfigured the time mode to asynchronous or, alternatively, a hybrid version.Although the asynchronous mode afforded learners planning time, the choice to reconfigure the time mode meant that they also reconfigured the language outcome from shared rapid real-time spoken interaction, to speech resembling individual spoken production.In doing so, they reconfigure a task designed to develop fluency into one in which they could focus on accuracy and complexity in their utterances (although whether they achieved it or not is not measured).Given that the recordings were to be evaluated by teachers, this reconfiguration is perhaps not surprising, and highlights Ellis' (2005) observation that context, such as test conditions, can shape planning.The asynchronous time mode afforded rehearsal time and strategic planning.It also afforded time for non-linguistic aspects of the task such as social reasons (Ellis, 2005) or technologically related reasons such as tool familiarity.Furthermore, the mode gave learners a way of reducing any potential anxiety because they were no longer "on the spot" (Hurd, 2007, p. 13).
2) How do these choices intersect?
The results suggest that learners' overarching choice to carry out the 'task-as-workplan' or to work according to their own workplan affected other subsequent choices.Choices relating to the three dimensions of agency analysed suggest that each dimension was (re) configured in order to carry out the task according to dyads' own workplan.Although the exact trajectories and choices of case 2, 3 and 4 during the planning stage are unknown, what is known is that the relationship between the choices across the three dimensions was mutually supportive.The choices were made in pairs, confirming that avoidance as a choice can be collaboratively co-constructed by participants… as a topic of interaction (Markee, 2011) as well as non-avoidance.The mutual use of both the l1/tl and the ability to interact in a non-recorded asynchronous or synchronous time mode afforded learners time to plan the task in their own way until they were willing and ready to engage in synchronicity in the tl.Although avoidance in cases 2, 3 and 4 is not a communication strategy as in Tarone's (1981) typology, it can be considered a communication strategy of a kind: learners controlled the tool and time-mode in order to secure temporary avoidance, which may be an emerging communication strategy specific to scmc spoken interaction tasks.
3) What effect do these choices have on time spent in the tl?
We do not know if and how much time was spent interacting orally in the tl in the planning stage.Case 1 had the longest time interacting in the tl and cases 2, 3 and 4 resulted in minimal time interacting in the tl, as seen in Table 2 below.

conclusion
The findings highlight that there may be a whole host of factors as to why cases 2, 3 and 4 avoided the 'task-as-workplan', but they remain unknown.The analysis revealed that reasons for task avoidance (or choosing to plan) transcended the cognitive explanations that have been a primary research focus in the field of sla.Avoidance as a learning strategy (Musk, 2014) should also take into account other possible social and contextual factors.These include a possible need to be familiar with a peer (who is effectively a stranger) to establish a shared perspective on the task and/or to establish group identity.l1 use may have been used to establish group cohesion as found by Hafner, Li and Miller (2015), or to induce a positive affective role as found by Rolin-Ianziti and Varshney (2008), particularly if anxiety was present.The lack of visibility of paralinguistic features (such as gaze and gesture), specific to audio-based synchronous communication, may also heighten the need to engage in peer familiarisation as an attempt to compensate for the lack of body language (Hampel & Hauck, 2004) or a way to counter the effects of a communication process that can be depersonalized (LeCourt, 1999).Contextual factors may include the online modality.Given that online learners report that they have difficulties in finding time to invest in activities (Romero & Gentil, 2014), learners may be inclined to approach tasks economically and efficiently (Batstone, 2005) in order to get the job done quickly.Similarly, conceptualising online language learning as a social practice, which is concerned with situating language and learning practices in the broader social practices of everyday life (Burton & Potts, 2013), might highlight that these learners' typical social practice is to collaborate asynchronously in other subjects in their l1, and to prepare and produce a joint outcome in the form of projects and assignments.
Therefore, planning can be understood as an extension of their normal social practice.Familiarisation of both the technological tool and the peer with which they will carry out the task may contribute to an increased sense of technical expertise with the tool, preparing themselves for the experience beforehand and being supported through it (Hampel & Hauck, 2004).
In relation to technology, Kress and van Leeuwen (2001) emphasise that multimodal technology makes demands on learners who have to operate several modes (e.g.speech and writing).Indeed, Satar and Özdener (2008) concluded that the (multi-)modality provided by various cmc tools should be considered in relation to divergent groups of learners, especially when differences such as anxiety and proficiency levels are at stake.The task in this study involves learners managing speech, images (icons) and text (instructions) relating to pedagogical and technological aspects of the task alongside navigational demands requiring physical touch.This expanded multimodal scenario may induce more anxiety or need for familiarisation, for example.Alternatively, students may want to use one mode or space for talk between themselves, which is a different mode or space to the one the teacher uses, a behaviour identified by Hampler and Stickler (2012).Testing conditions (Ellis, 2005) may have also been a factor considering the interaction was to be evaluated.
Affective factors may have included fear of negative evaluation, communicative apprehension, test (evaluation) anxiety, anxiety of the synchronous mode (Hampel & Hauck, 2012) or anxiety caused by tool use.Individual differences may also have been a factor including perceived or actual low proficiency level as found by Musk (2014), asynchronous learning styles as identified by Shahabadi and Uplane (2014), and a need for a private space (Suzuki, 2013) and for practice or being 'face sensitive' (Batsone, 2005;Aston, 1986).Considering that learners were going to be evaluated on their interaction, learners may have focused on accuracy over fluency, seeking to avoid making errors as a face-saving strategy.Learner beliefs regarding l1 use during tasks may have also been at play.
Other factors the authors believe may have been present include learners having a different perceived task outcome to the one intended; they may have perceived that it was important to find all the four differences correctly, rather than engage in the interaction in order to do so.
The covert behaviour is central to this study and has consequences for learning (Suzuki, 2013).Results do not wholly support the beneficial role of covert behaviours in other cmc studies however, as found in Sauro and Smith (2010) and Smith (2008).The difference may exist because these studies focused on synchronous text-based cmc, which afforded learners some think-time during interaction.Similarly, factors in Suzuki's (2013) study, although focused on synchronous oral-based cmc, may also have afforded students thinktime as the interaction occurred in a teacher-led, multi-party class of learners, possibly giving learners more think-time than peer-to-peer interaction in pairs.Although all of these technological-interactional configurations may aim and help to develop learners' fluency in real-time (or as near to real-time interaction as possible), the presence of think-time may modify the on-the-spot pressure felt by learners whilst interacting.Although think-time can be seen as an affordance in these studies, because learners could participate and maintain their ability to interact, it may not be desirable if the aim of the task is to simulate real-time "on the spot" (Hurd, 2007, p.13) rapid interaction.Real-time interaction is what many learners face in one-to-one, face-to-face non-online scenarios, given that this is where most spoken interaction occurs.The learners' choice to avoid rapid-real time interaction in order to potentially gain think-time therefore, may be to their detriment as it eliminates the valuable practice of real-time interaction solely with another person.Suzuki (2013), in her discussion of the pilot study for her main study questions whether regular, active offscreen behaviour in an online course setting provides merit for l2 acquisition.If time spent spontaneously interacting in the tl in the overt stage is a measurement of task success and is significantly less than in the planning stage, then we might conclude that some off-task behaviour (looking at answers and scripting) is not always beneficial.This conclusion is despite the fact that research on pre-planning suggests that pre-planning results in more fluent and more complex language (Foster & Skehan, 1996).
Once the reasons for avoidance are known, ways to scaffold learners could be introduced.Anxiety appears to be the most probable cause, given that the task involved many anxiety-inducing factors.Offering time to rehearse the same task type rather than exactly the same task may support familiarisation with their peers and task type, and tools to help "bridge the gap" (Hurd, 2007, p. 15) between private practice or rehearsal and the realtime task.This can "promote comfort and confidence" (Hurd, 2007, p. 15).Reassurance for learners who are less proficient, less confident or fear negative evaluation for l1 use may be needed.The evaluation of the interaction also need not take place until lots of practice has occurred and the tool use becomes normalised.Signalling a shift towards the importance of task process rather than task completion (finding the answers) may also need to be emphasized in courses.Introducing peer-to-peer evaluation may also help to move the focus away from the task as a performance to the quality of the task process, helping to reduce any performance-related anxiety related to teacher and other peers listening to them.

implications
With such a range of opportunities for agency across the three dimensions, learners can almost completely reconfigure the terms and conditions in which they prefer to and are able to interact.Not all learners' choices relating to technology, language or time-mode lead them to time interacting synchronously and spontaneously in the tl, which is necessary for developing fluency.Despite the fact that these online learners work primarily within an asynchronous mode, it is still important to provide opportunities for fluency because they offer the conditions that resemble (as closely as possible) the real time face-to-face interactions in the tl that many learners face in their non-virtual lives.The task time mode, however, may not match the learning style of many online learners who have chosen or prefer asynchronicity.If this is the case, this creates a particular challenge for designers of synchronous tasks.Indeed, this study brings into question whether one can effectively design fluency activities that are going to be evaluated by the teacher in the context of an online course in efl.Further research is needed to identify factors contributing to the temporary avoidance of synchronous speaking tasks.Contextual, affective, social or individual factors need consideration as well as cognitive factors; these may be technology or tool-related, and specific to online learners.

Example 1 from
Case 1: Spot-the-difference task M: ¿Pues empezamos en inglés?[translation: Shall we start in English?] H: OK.I'm ready to start in English M: OK.Me too.H: Um er... Who a start?You or me?M: You can start if you want.

Example 5 from
Case 1: Spot-the-difference task M: Oh yeah yeah yeah yes mine is a little bit white and yours is blue... H: Yes.Wuff... M: Wuff...

Table 2 .
Total interaction time in the target and first language