Learner repair during task-based peer interactions in an EMI classroom: a case study of a marketing class in a Chinese higher education setting

ABSTRACT

Current research on the in-class discursive realities in English as a Medium of Instruction (EMI) classrooms has been mostly restricted to whole class scenarios, whereas student-student interactive discourse in task-based activities is largely ignored. This study explores peer interactions among university students in an EMI marketing course in China, specifically focusing on repair practices. The data mainly comprised transcribed recordings of 26 peer interactions during two types of tasks, i.e. topic discussions and simulations. Findings reveal that the task type plays an important role in how various trouble sources are addressed. In both task types, there is a dominance of factual repairables, which are addressed through various collaboratively constructed repair trajectories. However, the relatively informal yet concept-related topic discussions elicit more frequent repair in general and linguistic-form related repair in particular than the relatively formal but hands-on natured simulations, though the latter type of task leads to more procedural and processing-related repairs. Implications are discussed regarding the importance of raising teachers’ awareness that different activities hold different language and content learning opportunities in peer interactions in EMI classes.

KEYWORDS:

• English medium instruction
• repair
• task-based interaction
• discourse analysis
• peer interaction

Introduction

Intensified globalisation and the international status of English have propelled many English-as-Foreign-Language contexts to promote English as a Medium of Instruction (EMI), i.e. ‘the use of the English language to teach academic subjects (other than English itself) in countries or jurisdictions where the first language (L1) of the majority of the population is not English’ (Macaro et al. 2018, 37). At the primary and secondary education levels, the same educational setting is usually labelled as Content-and-Language-Integrated-Learning (CLIL) (cf. Dearden and Macaro 2016; Schmidt-Unterberger 2018). In the Chinese tertiary context, EMI for domestic students usually includes bilingual formats where English is used for at least 50% of the lecturing time (Guo et al. 2018). Though EMI courses/programs remain at the peripheral sphere of China's education system, they are gaining popularity due to various economic and academic motives at national, institutional, and individual levels (Zhang 2018). A recent survey of the websites of 116 key institutions included in Project 211 by the Chinese Ministry of Education showed that more than 80% of them reported offering EMI courses (Kong and Wei 2019, 45).

It seems that the application of collaborative tasks that necessitate peer interaction is quite compatible with the learning objectives of EMI. From the perspective of language learning, interactionist researchers maintain that adequate comprehensible input alone is not sufficient for students to acquire satisfactory target language proficiency, and learners should be additionally provided with ample opportunities to receive corrective feedback and produce modified output through teacher-student or peer interaction (García Mayo and Alcón Soler 2013; García Mayo and Lázaro Ibarrola 2015; Ward 2008). Incidentally, content subject literature and higher education reform initiatives, including business education, have also drawn attention to the importance of a more collaborative and participatory paradigm as a way to facilitate knowledge construction and develop problem-solving skills or thinking strategies needed in an increasingly competitive and multicultural workplace (Bratianu, Hadad, and Bejinaru 2020; Dole, Bloom, and Kowalske 2016; Safarini 2019; Tan and Vicente 2019). More specifically in the Chinese tertiary context, some empirical studies evoking student perceptions have suggested that students prefer a more student-centred, interactive pedagogical approach (see Beckett and Li 2012; Huang 2018; Jiang and Zhang 2019; Wang and Curdt-Christiansen 2019), though current EMI classes are reported to feature a prevalence of monologic, non-interactive teaching styles (eg. Chen, Han, and Wright 2020; Li 2012; Yang 2017).

Despite the purported advantages of adopting collaborative tasks in language/content subjects and EMI contexts, surprisingly little research has been done to examine language use, including repair, in peer interaction in task-based activities in EMI settings. Repair, which concerns how interlocutors deal with various aspects of problems or troubles in conversation, has generally been regarded as facilitative to language learning in L2 classrooms as it allows learners to make necessary conversational adjustments to make complex language accessible to each other (Markee 2000). Many interrelating factors, such as interlocutors’ language proficiency, face considerations (Ahvenainen 2021) and grouping method (García Mayo and Imaz Agirre 2019), may influence how problems are addressed in L2 interaction. Among these factors, the potential influence of task types can hardly be neglected. As argued by Llinares (2015, 68), the tasks or activities designed by the teacher will shape the ‘context of the situation’, which, together with the ‘context of culture’ (i.e. subject types and their genre characteristics), constitute two vital context dimensions to understand language use in EMI.

In response to the scarcity of earlier research on peer interaction in EMI task-based contexts and with an understanding that language/content learning opportunities in doing tasks largely depend on what students do with language, we focus on how students engage in trouble-solving or repair practice in two types of tasks in an under-graduate EMI marketing course at a Chinese university. The two task genres specifically refer to 1) topic discussions, where students discuss topics that connect textbook concepts with real marketing phenomena, and 2) simulations meetings, where students act as marketplace stakeholders and engage in communicative activities.

Two research questions will be pursued in this study:

1. What are the main types of troubles that the students attend to during task-based interactions across the two task types mentioned above?

2. To what extent are repair practices influenced by the two task types?

The organisation of repair

In their seminal work on repair in conversation, Schegloff, Jefferson, and Sacks (1977) maintain that trouble sources, which are usually referred to as repairables, encompass more than apparent errors or mistakes; meanwhile, the presence of errors does not necessarily set a repair into action. Such a perspective allows participant-relevant interpretation of what constitute centrally relevant problems in the talk-in-interaction (Smit 2014, 180). Across studies, there is a variation in terms of what are picked up as repairables due to different analytic focuses and pedagogical contexts. Buckwalter’s (2001) research on repair sequences in Spanish L2 dyadic discourse focused on linguistic forms such as pronunciation, lexicon, morphology, and syntax problems. In EMI settings where both language and curriculum content learning goals are present, it is necessary to go beyond looking only at linguistic repairables and also consider aspects related to content knowledge construction or interactional function (e.g. Smit 2014).

According to Schegloff, Jefferson, and Sacks (1977), the repair proper after a repairable can consist of initiation and outcome, which are performed by the current speaker (SELF) or other participants in the talk (OTHER). While this model seems clear at first sight, Smit (2014) has pointed out two aspects of terminological ambiguity: Firstly, the term repair is used both for the whole exchange as well as the reaction to a repairable. Secondly, there is some empirical difficulty when applying this construct to describe repair sequences that take longer than three moves. She has thus integrated this model and the negotiation of meaning approach as described by Varonis and Gass (1985) into her ‘interactional repair plus’ model. According to this model, a repair exchange consists of one or more repair trajectories. Each trajectory consists of a repairable and the subsequent repair practice, which can be a one-off move carried out by either SELF or OTHER, or consists of initiation and response interactively accomplished by both SELF and OTHER. The response can lead to a reaction. In long repair exchanges, loops of response and reaction moves may occur. Since this model improves on terminological clarity, we have adopted it in the current descriptive analysis (see Extract 1)

This sequential understanding of repair is different from what most Second Language Acquisition (SLA) literature has narrowly used the term for, i.e. teacher correction within the traditional tripartite Initiation-Response-Feedback (IRF) sequence, or learner correction of an error following the teacher's corrective feedback (Lee 2013; Morris 2002; Nassaji 2011). The kind of discourse in which participants orient to linguistic norms during task-based interactions has been captured by notions such as language-related episodes (LREs, i.e. ‘any part of a dialogue where the students talk about the language they are producing, question their language use, or correct themselves or others’, see Swain and Lapkin 2000, 268), corrective feedback, focus on form, etc., which usually anchor in input-based language acquisition theories that imply a focus on interlocutors’ approximation to standardised language use.

Though these SLA-motivated concepts allow insights into how linguistic-oriented troubles are treated, the current study mainly adopts a conversation analysis (CA) approach to repair because 1) it offers a more reciprocal and sequentially-based perspective of how students enact their interactional roles, and 2) it enables discussion of various problems that go beyond a mere linguistic focus. But this study also incorporates an SLA perspective because the researchers intend to understand the language learning opportunities involved in peer interaction. At the same time, the current study does not adopt the CA approach in a pure sense. CA typically interprets the emerging data through localised microanalysis without an attempt to force any ‘order on the apparent chaos of classroom interaction’ (Walsh 2006, 54), but such practice has evoked criticism that it usually results in the impressionistic presentation of analytic categories. To better meet the research needs, it is possible to draw on the data-driven strength of CA analysis, but also flexibly refer to previous literature that has provided some general analytic categorisation to better account for the complex interactional situations.

Repair in L2 task-based oral peer interaction

In spontaneous daily conversation, Schegloff, Jefferson, and Sacks’s (1977) analysis has revealed interactants’ decisive preference for self-repair over other-repair, which may be potentially caused by face considerations: it seems to be more polite to put the blame on SELF than to denote that OTHER has failed to correct the trouble source. However, language use in traditional classrooms might modify this preference structure because of the asymmetric teacher-student power relationships, resulting in a prevalence of the teacher's other-initiation in L1 classrooms (McHoul 1990), or a strong presence of teachers’ other-repair in Foreign Language (FL) classes (Kasper 1985; Van Lier 1988; Varonis and Gass 1985), and also in early semesters of some EMI higher education contexts (Smit 2014). Previous studies comparing teacher repair in CLIL and EFL instructional settings have generally suggested that CLIL teachers are less likely to bring attention to language forms and provide explicit corrective feedback types than language teachers (e.g. De Graaff et al. 2007; Lorenzo, Casal, and Moore 2010; Milla and García Mayo 2014). Different from teacher-led classroom settings which manifest a hierarchically defined role allocation system (Van Lier 2001, 95), peer interaction represents a space where ‘power is arguably decentralised and epistemic hierarchies flattened’ (Komori-Glatz 2018, 288). Students are offered extended opportunities for turn-taking sequences and functional use of language, including repair (Ballinger 2021; Fourlas and Wray 1990; Kumpulainen and Wray 2002).

Among the studies that examine learners’ focus on language during peer interaction in meaning-focused tasks in language learning classrooms, some have suggested that students are most likely to pay spontaneous attention to lexical problems, but are much less likely to focus on grammatical problems, and rarely address phonological troubles (e.g. Buckwalter 2001; Philp, Walter, and Basturkmen 2010; Williams 1999). However, this ranking does not necessarily apply to pedagogic tasks like information gap, text reformulations, dictogloss, etc. (e.g. Calzada and García Mayo 2021; Leeser 2004; Xu, Fan, and Xu 2019). Other interactionist studies have shown students’ focus on form in peer interaction has been influenced by factors such as learner proficiency, paring/grouping methods, or task types. As to the relevance of language proficiency, data from several studies have suggested that students with a lower proficiency display a higher number of language problems but attend to language norms less frequently than their high-proficiency counterparts (e.g. Leeser 2004; Williams 1999). With regard to the influence of pairing/grouping methods, Leeser (2004) reported that lower proficiency learners produced more LREs when being paired with higher proficiency learners than with low proficiency students. Calzada and García Mayo (2021) examined pair interaction among EFL children in dictogloss writing tasks and found high-high proficiency pairs produced more LREs turns than high-low and low-low pairs in almost all categories examined, with the only exception of turns of grammar features (other than 3rd-s), which were most generated by high-low pairs. García Mayo and Imaz Agirre (2019) found that self-selected pairs of students generated fewer LREs than teacher-selected and proficiency-paired students partly because self-selection usually led to pairs who were more likely to engage in informal conversation. Highlighting the influence of task features, Williams (1999) revealed that unstructured tasks like discussions and role-plays generated fewer instances of focus on form than structured text-based activities like cloze and repetition. Philp, Walter, and Basturkmen (2010) suggested that the presence/absence of pre-emptive focus from the teacher and the inherent focus of the task (meaning-oriented or form-oriented) may impact on LREs. When it comes to repair sequences, Buckwalter’s (2001) ethnographic study on dyadic repair sequences during meaning-focused speaking activities found Spanish as foreign language adult learners predominantly used self-repair in addressing language problems.

Though EMI settings represent an ideal environment for students to engage in authentic language practice, only a small number of studies have investigated repair practice in EMI task-based contexts. Grounded in sociocultural theory, Lialikhova’s (2019) research in CLIL lessons demonstrated that high-proficiency groups and mid- proficiency groups of learners were generally able to provide corrective feedback and scaffold each other's cognitive development, whereas low-proficiency groups seldom performed cognitively demanding tasks proactively and collaboratively. In her examination of CLIL classrooms at Austrian secondary schools, Dalton-Puffer (2007) reported that students seldom engaged in linguistic-form related repair during group work discussions. In a content-based language classroom, Jakonen and Morton (2015) employed CA and multimodal analyses in their research on epistemic search sequences in peer interaction and revealed how knowledge gaps were conveyed and treated by peers while working on pedagogic tasks. Several other studies have suggested that L1 is a valuable interactional resource in peer interaction and peer repair in EMI settings (e.g. Pun and Tai 2021)

Whilst previous studies have shed some light on the language performance of peer repair in various L2 classrooms, to our knowledge, current research does not address sufficiently how various types of epistemic or linguistic repairables are managed by students in peer interactive tasks in EMI classes. In particular, our understanding of how task types may potentially influence peer repair in EMI settings is notably underdeveloped. Thus, this case study involves mixed-method research to explore features of peer repair in a marketing course in a Chinese university by paying attention to the various types of trouble sources and repair organisations in two classroom activities, i.e. topic discussions and simulation meetings.

Methodology

Design, contexts and participants

This case study was built upon a larger PhD project which investigated discourse features of task-based peer interactions in an EMI marketing classroom in a key provincial university in a south-eastern city in mainland China. The study was conducted in a class of 29 second-year Chinese L1 users (all aged around 20; 6 males and 23 females) in an undergraduate EMI International Business Programme. The students on average evaluated their English proficiency as 4.57 on a 7-point Likert scale (1 means very poor, 7 means very good). The teacher is a Chinese L1 speaker with three years of overseas education in the UK and around 20 years of EMI teaching experience. Primary data were naturally occurring discourse episodes collected through audio recordings that followed the same four students in an EMI class for the course entitled ‘Marketing Principles’, a 16-week compulsory course that lasted 150 minutes per big class per week.¹ Four students who seemed to be more active participants in class in the pilot phase (the first 4 weeks of the semester) were invited by the first author to be the recorders and focal students with the consideration that they might contribute rich discourse data for analysis. They carried the audio in each interaction they participated in, so each interaction recording would involve at least one focal student.

The data for analysis consisted of 26 interactions in 9 tasks (6 TDs and 3 SMs) conducted in four big classes. Because a few students in the class were not in the same group with the four focal students in all these 26 interactions², the final transcriptions involved 20 students’ discourse. The background information of the 4 focal students and the other 16 students in the resulting 26 interactions is provided below Table 1.

Table 1. Participants’ background information.

Participatory Roles	Pseudonyms	Gender	Previous EMI experience	Proficiency rating^3
Recorders & focal students	LIYI	M	None	Higher intermediate
	HUAN	M	None	Higher intermediate
	QING	M	None	Intermediate
	JUNA	F	overseas exchange (2 weeks)	Intermediate
Other students	Weir	M	None	Intermediate
	Huie	M	None	Intermediate
	Wana	F	None	Higher intermediate
	Zhie	F	None	Intermediate
	Lais	F	None	Intermediate
	Ying	F	None	Intermediate
	Kair	M	None	Lower intermediate
	Kime	F	None	Lower Intermediate
	Jins	F	None	Intermediate
	Xuan	F	None	Lower intermediate
	Xiea	F	overseas exchange (2 weeks)	Intermediate
	Tina	F	None	Intermediate
	Kexe	F	None	Intermediate
	Yuwi	F	None	Intermediate
	Zhen	F	None	Intermediate
	Chek	F	None	Lower intermediate

For ethical considerations, the focal class and the teacher were invited to sign an informed consent form that permitted the researchers to record their discourse and transcribe the lessons in the pilot phase. Though audio data collection may potentially affect students’ behaviours, the participants commented in the interviews that they tended to get used to the presence of the recording device after the pilot phase. The identity of the observer (first author) as a PhD student rather than an authority doing evaluative work also helps decrease the potential influence of the recording on the students’ behaviours.

Two curricular task types

The class agenda followed the usual lesson planning created by the teacher and was not modified specifically for the present study. Though different kinds of activities took place in the class, the focus of our inquiry was restricted to two types of regularly applied tasks, i.e. the topic discussion (TD) and the simulation meeting (SM).

TD was a type of task that engaged students in impromptu discussion in response to concept-related topics which aim to facilitate students’ understanding of newly introduced marketing concepts with connection to real-life examples. For instance, one TD in week 5 is: Which Values and Lifestyle Survey type matches the product of a particular kind of dishwasher (the dishwasher is briefly introduced to students before the discussion session) and what kind of barriers should be overcome to win the Chinese market. The teacher did not group students for topic discussions but simply asked students to discuss with their neighbours in spontaneously formed groups (the resulting data shows that each group consists of between 3–5 students; c.f. Table 2). After each topic discussion task, the teacher would normally invite one or more students in each group to report back to the class their discussion results. Students’ performance would not be directly included as part of the final grading for the course.

Table 2. An overview of the interaction data.

Task	No. of interactions	No. of participants in each interaction	Turns	Words	Words in task type	L1 words in task type	% of L1 use in task type
TD1	4	3;3;3;4	115	1076	13496	5792	42.9%
TD2	4	3;3;4;5	429	3705
TD3	4	3;3;3;4	464	4842
TD4	3	4;4;5	193	1864
TD5	3	3;4;4	90	673
TD6	3	3;4;5	172	1336
SM1	2	5;6	145	1384	3321	653	19.7%
SM2	2	5;6	78	773
SM3	1	5	131	1164
Sum	26	/	1817	16817	16817	6445	38.3%

SM was a more formal and complex type of event and students’ performance accounted for 30% of students’ final grading. The whole class was divided into six groups at the beginning of the semester (each group consisting of 4 or 5 students based on random allocation through a computer system, but students were allowed to switch with members from another group upon mutual consent prior to deciding the final name list of each group), and each of the six groups needed to perform the role of a host group one time throughout the semester. A simulation meeting required the host group to prepare (outside of the class) and conduct (inside the class) a simulated meeting with multiple stakeholders (role-played by other groups such as customers, the media, the supplier company, etc.) in face of a marketing challenge adapted from real marketing scenarios. The whole section lasted between 20-30 minutes and normally contained phases like plenary presentation, plenary discussion, and more private group discussions based on how the host group designed the task. Only the group discussion phases were analysed considering that plenary sections of simulations usually involved self-scripted or prepared student talk. The content of the group discussions for simulations was generally opinion exchanges either within a particular stakeholder group or between a chair team representative and a stakeholder group. This study reports peer interaction data within groups (excluding the plenary section) that occurred during four selected classes which involve peer interactive tasks upon communicating with the teacher.

Data and data coding

A total of 130 min’ discourse data involving 26 task-based peer interactions over four lessons were transcribed. The interactions were transcribed following the VOICE conventions (see the appendix) and imported to the software of MAXQDA 2020. An overview of the 26 interactions in 9 tasks (6 TDs and 3 SMs) is shown in Table 2.

Table 3 shows the coding system based on four repair trajectories and seven repairables types. They are categorised as such based on the features of the current data while also referring to analytical categories proposed in previous literature (cf. Dalton-Puffer 2007; Smit 2014). In this study, trouble source always occupies the first turn of a trajectory.

Table 3. Coding of repairables and repair trajectories.

Repairables	Explanations	Example(s)
Pronunciation	Problems related to the pronunciation of a word/words	Extract 8
Grammar	Morphological or syntactic problems	Extracts 6 Extract 7
Vocabulary	Problems with choosing, retrieving or understanding a particular word (s)/vocabulary;	Extract 3,4,5,9
Processing	Problems related to the formulation of complete sentences; reconstruction or reformulations of utterances as characteristic of spontaneous spoken language (Dalton-Puffer 2007)	Extracts 2 Extract 11
Mishearing	Problems of intelligibility, normally caused by the acoustic condition	Extract 14
Factual	Problems related to factual and conceptual knowledge	Extract 1,4,5,9,10
Procedure	Problems concerning task requirement, including its sequencing, materials use and group setup	Extract 12, 13
Trajectory	Explanations
self-repair	The speaker addresses the trouble source without being prompted to do so within the same turn. The current research excludes instances of repetition and gap fillers.	Extracts 2 Extracts 6
Self-initiation (Other-response)	The speaker of a trouble source explicates the need for the other to resolve a problem	Extract 3,4,5,9,12
Other-initiation (self-response)	A speaker signals that some aspect of the prior turn is problematic.	Extract 1,3,8,0,14
Other-repair	The trouble source is repaired by the speaker other than the trouble-source speaker.	Extract 1,3,7,9,11,13

Notes: Parentheses are used in the trajectories of self-initiation (other-response) and other-initiation (self-response) to indicate that instances where initiations elicit no response are also calculated.

Extract 1 from our data represents a repair exchange that consists of two repair trajectories:

Extract 1: A factual repair exchange involving two trajectories in TD 6

Table

TST	1 Wana: so it has different kinds of variety for different tastes
ORT + TST	2 Lais: no. just one taste
OIT	3 Wana: one taste?
SRT	4 Lais: yeah, one bottle. one taste

Notes for abbreviations for all extracts: TST: trouble source turn; ORT: other-response turn (or other-response turn); OIT: other-initiation turn; SRT: self-repair turn (or self-response turn); SIT: self-initiation turn

Line 1 and 2 form a trajectory of other-repair to address a factual repairable. Line 2, 3 and 4 show another trajectory of other-initiation self-response also to address a factual repairable. The conjunctive Line 2 acts both as the repair turn of the earlier trajectory and the trouble source turn of the second trajectory. Meanwhile, the speaker of Line 2 assumes both the roles of OTHER and SELF. Sometimes there were fuzzy boundaries in coding the nature of specific instances. For instance, technical or discipline-specific terms (such as Extract 4 and 5), in most cases, would be interpreted both as lexical problems and factual problems related to conceptual knowledge. These hardly discernible cases were viewed as combining two types of repairables and were double-coded. To improve the reliability of the data, half of the data were coded again several months after the initial coding and checked for intra-rater reliability, which revealed an agreement rate of 90%. The differences were re-examined and 3 uncertain cases were discussed with a university professor.

Findings

Frequency of repairables

Table 4 displays the relationship between different trouble sources and repair trajectories.

Table 4. Cross-tabulation of repairable type and repair types in two task types in general.

	Repairable type	Self-repair	Self-initiation (other-response)	Other-initiation (self-response)	Other-repair	No. repairables	%	No. repairable (%)	No. repairable (%)
linguistic repairables	Vocabulary	16	32	15	12	75	20.9	110 (30.7)	358 (100.0)
	Grammar	21	1		6	28	7.8
	Pronunciation	2	1	1	3	7	2.0
Other repairables	Facts	2	44	35	48	129	36.0	248 (69.3)
	Processing	48			12	60	16.8
	Procedure	2	20	1	7	30	8.4
	Mishearing			29		29	8.1

As shown above, a total of 358 repairables were identified in the 130-minute data of 26 interactions, indicating that repair is a common practice in our data. It should be noted that the number of repairables is substantially lower than the number of actual problems or errors, many of which were left unaddressed either because they went unnoticed or were treated through the ‘let it pass’ (Firth 1996) strategy intentionally.

The first question concerns the main types of trouble sources that students attend to. We can see from Table 4 that linguistic repairables constitute 30.7% of all the repairables. The table suggests that the students’ repair practices predominantly target facts (36.0%) and vocabulary (20.9%), which are followed by processing (16.8%). The instances of pronunciation repairables are rare (2.0%). Procedure, mishearing, and grammar problems assume the middle position of all the identified cases with similar percentages at around 8%. The strong presence of processing-related repairables is not consistent with Dalton-Puffer's (2007) study that reports a low quantity of processing repairables on teacher-student interaction in Austrian CLIL settings. Following Smit (2014) but different from the practice of Buckwalter (2001), the current analysis excludes instances of repetition and gap fillers as instances of self-repair. If these two phenomena were also included, the number of processing repairs would be considerably higher. In the current data, processing problems constitute the most important reason for students’ self-repair, as reflected in constant message deletion, replacement, and abandonment in the data (e.g. Extracts 2). It, therefore, seems that our data point to a general lack of ability of students to process smooth and complete sentences under online conditions in oral communication.

Extracts 2: self-repair of processing problem in TD and SM

1. Wana: we can(.) erm that means we can(.) erm(.) we can practice our ability in many important and erm(.) some international meetings and (2) @ @ yes? (.) (TD 4)

2. Kime: because because I think most of us must considered our considered considered most of our health. (SM 2)

Treatment of linguistic repairables in the two task types

The second research question concerns to what extent repair practices are influenced by task types. This section takes a closer look at repairs related to linguistic forms, i.e. those caused by problems of either vocabulary, grammar, or pronunciation. Overall, 110 linguistic repairables are identified. But to gain an insight into the differences across the two task types, it is necessary to examine the weighted data, as is shown in Table 5.

Table 5. Linguistic repairables and repairs in two task types in normalised data.

Repairable types	Task types	Self-repair	Self-initiation (other-response)	Other-initiation (self-response)	Other-repair	Total repairables
						(Per 10000words)
vocabulary	TD	13.3	23.7	8.9	7.4	53.3
	SM			3.0	6.0	9.0
grammar	TD	13.3			4.4	17.8
	SM	9.0	3.0			12.0
Pronuncia-tion	TD	1.5	0.7	0.7	1.5	4.4
	SM				3.0	3.0
Subtotal	TD	28.1	24.4	9.6	13.3	75.5
	SM	9.0	3.0	3.0	9.0	24.0

We can see that overall TDs generate more linguistic repairs than SMs when the data are normalised by 10000 words. When viewed separately, the three types of linguistic repairables display somewhat different features in repair practice across the two task types.

The most prominent difference in the two task types concerns lexical repairs, as they make up the largest share of linguistic-form related repairs in TDs but are numerically negligible in SMs. In TDs, lexical problems are particularly addressed through self-initiation (other-response) in L1 in the form of word searching (e.g. … 怎么说? {How do you say … }) or seeking peer explanation for a particular topic-related term (e.g. VALS 是什么意思? {What does VALS mean?}). A repair exchange from TD of vocabulary consisting of 5 trajectories is illustrated in Extract 3.

Extract 3: A repair exchange of vocabulary in TD6

Table

TST + SIT	1 Wana: 糖尿病怎么说？ {How to say Tangniaobing^3?}
	2 JUNA: er:m =
	3 Lais:    it it it's can be good for =
ORT + TST	4 JUNA: dia-dia- <1 > diarrhea </1>
	5 Wana: <1 > diarrhea</1>
ORT + TSR	6 Lais: 那个应该是(.) <@>腹泻吧<2>应该是</2> </@>@@我忘了 {that one should be (.)<@>diarrea <2 > should be</2></@>@@I forgot}
OIT	7 JUNA:<2>是嘛? {is that so} @@ </2>
SIT + TST	8 Wana：<pvc><ipa>'daɪə'biːti? </ipa></pvc>
	9 Lais: 那个好像是 {that word seems to be}
OIT + TST	10 JUNA: 那个不是肥胖的吗？ {doesn't that mean obesity?} =
	11 Wana: =sorry=
ORT	12 Lais: obesity才是肥胖的 {means obesity} =

In this exchange, students were required to reflect on the market positioning problems for a particular type of sugar-free drink. Before this extract, JUNA mentioned one positioning problem was that this drink was pretty normal. In Line 1, Wana intended to mention that the drink may benefit people with diabetes. The English expression seemed to have escaped her so she used self-initiation in Line 1 to seek help from other students. In Line 6, ample use of modifications could be found, as shown in hedges like 应该是 {should be}, 我忘了 {I forgot}, and laughter. In Line 7 and 10, repair is tentatively addressed using questions rather than statements. These suggest that modification seems to be more likely to occur in language-related repairables than factual repairables.

One potential influencing factor for more frequent and diversified lexical repairs in topic discussions is that the two task types represent different genres: in the investigated context, TDs are informal in the sense that students are just asked to discuss with their neighbours and their performance is not directly assessed. The group discussion phase in SMs, on the other hand, is a role-play of a relatively more formal type of event. Though it is also not directly assessed, the whole role-play project performance is evaluated and accounts for 30% of the students’ final grades for the course. If the students play their roles well, they will largely follow the typical interactions between marketplace stakeholders, which are much less likely to foster explicit attention to form. Another important factor is the pedagogical focuses of the two task types. TDs require students to discuss real-life marketing phenomena with newly introduced concepts (e.g. connecting marketing behaviours of a particular product to a particular marketing concept, etc.), whereas SMs engage students in doing marketing and have a hands-on oriented nature. This leads to the fact that there are some cases where the distinction between factual and lexical repairs is blurred in TDs: students are dealing with troubles with vocabulary and academic concepts at the same time while working out what these terms mean. These instances epitomise the co-occurrence of language and content learning opportunities (Extract 4).

Extract 4: Self-initiation of vocabulary/facts in TD 2

Table

TST + SIT	QING: What does trendy achievers mean?
	Huie: tren:dy achievers.
ORT	Weir: means fashion. fashion achievers. fashion

But it may be noted that the difficulties in offering fully explicit L2 explanations by OTHER are also evidenced in TDs. In that case, L1 use is a frequently applied strategy to clarify the meaning of unknown terms (Extract 5)

Extract 5: Self-initiation of vocabulary/facts in TD 1

Table

TST + SIT	QING: so. provincials. I couldn't understand this phrase mean. provincials. what does that word mean?
	Huie: province. provincial. sustainer
ORT	Weir: 赶快来查一下 {quickly check the meaning} provincial (2) 乡下人 {country folk}

Compared with lexical repairs, grammatical repairs are numerically closer in weighed data across the two task types. Grammatical problems are mostly self-repaired in the same turn in our data (Extracts 6).

Extracts 6: Self-repair of grammar in TDs and SMs

1. Weir: @@@yili yili fragrant? @@ fragrance (morphological repairable; in SM 3)

2. HUAN: hard to use is because it needs a lot of procedure to get this device readily for use. ready for use (morphological repairable; in TD 3)

3. LIYI: =trade price is for the retailers(.) and sell(.) seller price for the(.) customers (morphological repairable; in SM 1)

4. HUAN: people don't think they. is worth bothering. being bothered to use that (syntactic repairable; in TD 3)

5. Kime: =yeah. have you ever considered the infection through your machine? You know from the video. all of your cups is- are exposed to the air (syntactical repairable; in SM 2)

The above examples show grammatical repairables seem to be less likely to invoke time-costing collaboratively performed trajectories than lexical problems do. This seems to be a reason why informal/formal genre differences do not particularly influence the occurrence of grammatical repair across the two task types.

Yet the informal nature of TDs does seem to explain why the few instances of other-repair of grammar are exclusively found in TDs (Extract 7).

Extract 7: Other-repair of grammar in TD6

Table

TST	Tina: maybe it is attractive to some people who chase(.) healthy=
ORT	Lais: =health=
	Tina:    health. healthy drink? maybe? but

A closer look at the data suggests that these other-repair instances are exclusively morphological problems rather than syntactic troubles, the latter being exclusively self-repaired across the data (Extracts 6: d-e). What is interesting to also note is that the morphological repairables are almost exclusively related to derivational affixes of adjective and noun formation (e.g. Extracts 6: a-c; Extract 7). Moreover, 5 out of the 25 morphological repair exchanges are resolved incorrectly (e.g. Extracts 6: c).

The pronunciation-related repairs are numerically negligible in both task types though they are present in all four trajectories. Other-initiation or other-repair of pronunciation occurs only when pronunciation problems impede the understanding to such an extent that they can easily be confused with a different word (Extract 8), giving an impression that it is generally face-threatening to correct pronunciation problems.

Extract 8: Other-initiation self-response of pronunciation in SM 1

Table

TST	HUAN:   <ipa > [kɑː'tuːn] </ipa > cooperate with the company.
OIT	Kair: (1) 那个好像不是 {that seems not to be} cartoon 吧 {modal participle}?<ipa > [kɑː'tuːn] </ipa> 是卡通 {is cartoon}. 它那个 {that one} =
SRT	HUAN:    carton =
	Kair: =对{right}carton

Treatment of repairables not related to linguistic-form in the two tasks types

We will now turn to the treatment of repairables not related to linguistic-form (i.e. facts, processing, procedural, and mishearing, see Table 6), with emphasis on the similarities and differences across the two task contexts.

Table 6. Repairables and repairs in two task types in normalised data.

Repairable types	Task types	Self-repair	Self-initiation (other-response)	Other-initiation (self-response)	Other-repair	Total repairables (Per 10000w)
Facts	TD	1.5	27.4	17.8	29.6	76.3
	SM		21.1	30.1	24.1	75.3
Processing	TD	27.4			4.4	31.9
	SM	33.1			18.1	51.2
Procedure	TD	1.5	9.6		2.2	13.3
	SM		21.1	6.0	12.0	39.1
Mishearing	TD			18.5		18.5
	SM			9.0		9.0
All linguistic repairables	TD	28.1	24.4	9.6	13.3	75.5
	SM	9.0	3.0	3.0	9.0	24.0

As can be seen in Table 6, when it comes to factual repairs, what is similar in both TDs and SMs is that they represent the strongest repair type, with all the three collaboratively conducted trajectories frequently found in the data. What seems to be different across the two task settings is that factual problems are particularly taken up through other-repair in TDs (Extract 9) and through other-initiation in SMs (Extract 10).

Extract 9: Other-repair of fact/vocabulary in TD 2

Table

TST + SIT	Weir: experiencer 是什么意思? {is what meaning?} so experiencer means that someone has tried the chocolate?
ORT + TST	Huie: try to be different.
ORT	Weir: try to be different? no. experienced.

Extract 10: Other-initiation self-response of fact in SM 3

Table

TST	Weir: =yoghurt=
	QING: =like yoghurt?
OIT	LIYI: are they selling yoghurt too?
SRT	Weir: =yeah=

As to processing repairables, what is similar in the two task types is that they are mainly treated through self-repair in both task types (Extracts 2). The difference is there is a good chance that OTHER execute the repair in SM when students notice their peers are experiencing difficulties in processing smooth sentences (Extract 11).

Extract 11: Other-repair of processing problem in SM3

Table

TST	LIYI: how can they help us in providing a:
ORT	QING: =a good=
	LIYI: =good good merchandise that's like er::cake deserts.

The greater occurrence of other-repair of processing repairables in SMs seems to suggest that learners are more conscious of other interlocutors’ formulation of complete and structurally smooth sentences in the role-plays, especially among students who represent the same stakeholder group.

With regard to procedural repairables, i.e. the ones relating to performing the task, they are typically self-initiated and sometimes other-repaired in both task types (Extract 12 &13).

Extract 12: Self-initiation other-response of procedure in SM 1

Table

TST + SIT	Kair：我们是包装公司吗? {are we the package-making company? }=
ORT	HUAN: =对啊{yes} be responsible. environmental-friendly

The illustrative example above shows that the choice of language for initiating procedural repairs is almost exclusively L1, supporting earlier findings of a general tendency among students to switch to L1 when repairing problems produced by keeping up with the task agenda (Lehti-Eklund 2013).

What seems to be different across the task types is that procedural repairs occur more often in SMs, which perhaps reflects SMs are a more complex type of task in which students more often encounter difficulties in figuring out how to approach the task. However, a few cases in which students remind group members to return to on-task discussion from casual talk or switch to English to comply with the procedural requirement are exclusively found in TDs (e.g. Extract 13).

Extract 13: Other-repair of procedure in TD 6

Table

TST	LIYI: 湖南佬. 辣椒免疫. 辣椒劝退.{Hunan guy. chilli immunity. you had better not eat Chili}
ORT	Zhie: 快点回答问题. 没时间了. {answer the questions quickly. we are running out of time}

As shown above, Zhie reminds LIYI to keep the discussion on task. In our data, students are much more likely to use their L1 and engage in casual talk (almost exclusively in L1) during TDs, potentially caused by the more informal and less constrained nature of TDs.

Finally, it is apparent that in both task types repair of mishearing relies on OTHER to be triggered (Extract 14).

Extract 14: Other-initiation of mishearing problem in TD 2

Table

TST	Wana:<un>*****</un > can save time.
OIT	HUAN: ah?
SRT	Wana: save time(.) for dishwashers

The above extract shows HUAN uses an open question ‘ah?’ to suggest he did not hear clearly. Similar cases occur more frequently in TDs, potentially caused by the relatively more informal genre, in addition to greater noise interference in this task type because TDs usually involve more parallel discussions than SMs.

Discussion and conclusions

Taken together, the current case study reveals that task-based design in EMI can provide ample learning opportunities for students when they are engaged in peer interactions. The study also illustrates that repair is a complex interactional phenomenon and an important discourse strategy as students engage in co-constructing mutual understanding in face of various trouble sources. Moreover, the construct of repair can be a useful tool to shed light on the language and content learning process in different task types.

Students’ repair practices can be mediated by multiple variables, but the task type is clearly an important influencing factor. Before expanding on the differences, let us first look at the aspects of repair that bear more similarities in the two task types. Aligned with earlier suggestions by EMI/CLIL researchers about the strong focus on content in EMI/CLIL classrooms (Nikula 2015; Smit 2014), there is a dominance of factual repairs in both types of tasks in our data, indicating that communicating about facts is the most relevant factor and essential goal in task-based interactions. Moreover, compared with linguistic repairs that are usually self-repaired, factual repairs are mostly collaboratively addressed, suggesting that OTHER serves as a more important source than SELF to resolve factual problems. In addition, factual repairs are more likely than linguistic repairs to receive the potentially most face-threatening repair trajectory, i.e. other-repair, and more likely to be treated without mitigation, corroborating Dalton-Puffer’s (2007) finding that factual trouble sources are generally considered less face-threatening than language problems in CLIL lessons. The negligible use of phonological repairables in both TDs and SMs seems to suggest it is quite face-threatening to correct phonological problems (see also Dalton-Puffer 2007). Across the data, there is a strong presence of processing-related repairables, which constitute the major reason for self-repair in both task types, pointing to a general lack of fluency among students under online conditions.

The two task types also show similar frequency in terms of grammar-related repair. These instances, in particular syntactic problems, are exclusively self-repaired in the same turn, but morphological problems are sometimes dealt with by others (these few instances exclusively occur in TDs). It seems that these Chinese students’ spontaneous attention to morphology-related repairables is more salient than CLIL peer interaction realities in Austria reported by Dalton-Puffer (2007). Even though morphological features are generally not essential for meaning-making or comprehension (Williams 1999) and are more face-risking than lexical items (Dalton-Puffer 2007), in our study, students do sometimes monitor, correct and negotiate morphology-related problems, and around 20% of the exchanges are resolved incorrectly. The struggle to find the correct morphology as shown in our data seems to point to the students’ challenge in acquiring the English derivational rules for noun and adjective suffixes, which is perhaps caused by the fact that derivational suffixes in English are more productive than those in Chinese.

A prominent difference between the two task types is that almost all lexical repairs occur in the topic discussions. In other words, TDs seem to be more conductive to shared treatment of lexical problems in this study. This result may be explained by the fact that TDs represent a more informal context where students find it less face-threatening to attend to linguistic norms than in formal contexts. Another reason is that TDs explicate the need to connect concepts with subject-relevant real-life phenomena. As students have to understand these terms first to perform the task smoothly, they usually try to resolve these problems first through self-initiation before entering into the core discussion. When examining the typical repair trajectories to lexical repairs, we find a general preference for self-repair. However, such a preference seems to be less strong compared with previous studies which report a marked tendency to use self-repair in peer interaction (e.g. Buckwalter 2001; Markee 2000). In our data, students are also found to frequently apply other-initiation and other-repair in TDs to address language problems, suggesting that students in the local context do sometimes assume a teacher-like role and collaborate on language difficulties. This might be partially explained by the coexistence of imbalanced language competence and close relationship among group members. The former factor holds greater potential to necessitate other-initiation and other-repair and the latter factor reduces the face-threatening element involved in pointing out and resolving linguistic-norm related problems.

Another kind of repair more frequently found in TDs than SMs concerns mishearing problems, potentially caused by the more informal genre of this type of task and the fact that more parallel activities are held at the same time in TDs than SMs.

Unlike the general trend of more repairs in TDs, processing difficulties are more likely to be recognised and picked up by other students in SMs than in TDs, especially among students who role-play the same stakeholder group. This finding potentially mirrors the more cost-efficiency and high-stakes nature of the SMs than TDs. Students seem to be more attentive to co-constructing the message because SMs place them in a more interdependent setting if they want to perform the task well.

This study has some pedagogical implications for EMI practitioners. The findings imply that task-based interactions can create a space for both content and language learning with opportunities deriving from both SELF and OTHER. Moreover, different trouble sources may evoke the attention of SELF and OTHER differently. The initial results are suggestive of the importance of raising teachers’ awareness that different task types hold different learning opportunities in peer interaction. The more informal and concept-related tasks like TDs may allow students to attend to form more frequently than more formal and hands-on natured tasks like SMs. In particular, TDs that require students to use concepts to explain real-life phenomena prove particularly helpful in engaging students’ attention in subject-specific words that epitomise language and content integration. But it is important to note that the difficulty in formulating full explanations to this specialised terminology is also found in the data. Moreover, students are more likely to use their L1 and casual talk in TDs, which sometimes evoke other-repair of procedures in TDs. In this connection, optimising the content/language integrative learning effect of TDs requires more delicate scaffolding from the teacher by methods like designing more interesting concept-related topics, gauging the difficulties of conceptual understanding before assigning topics for discussion, and perhaps encouraging students to speak English more often. Although relatively more formal tasks like SMs seldom result in lexical repairs, they nevertheless also provide opportunities for speakers to construct disciplinary knowledge through various trajectories of factual repairs. Other aspects in which students seem to benefit from doing simulation is that students are more likely to use English and more ready to assist each other in formulating complete L2 sentences in this type of task.

Due to practical constraints, this paper does not include analyses of non-verbal strategies like head movement or use of gaze. In addition, it is a small-scale investigation that involves only 26 interactions in one specific EMI course. Despite these limitations, it nevertheless reminds us of the importance of understanding intricate relationships among discursive particularities, activity types, disciplinary features, and the accompanying content and language affordances. More investigations of various aspects of in-class discursive practices like question use, informal speech, and other relevant micro-analytical details are needed to shed light on whether and/or how task-based activities afford learning opportunities for academic and/or non-academic success.

Acknowledgements

We would like to thank the anonymous reviewers for the insightful comments on the article. We are also greatly indebted to Lecturer Liu Hui (PhD) from Nanjing University of Aeronautics and Astronautics for her very constructive suggestions on the structure and content of the first versions of this paper.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by the International Office of University of Vienna under KWA Grant number LVR 008.

Notes

1 A big class consisted of 2 sections with 15 minutes’ break in between.

2 It seemed that some students intentionally avoided being recorded because even if they sat near one of the focal students they would rather stay alone and remain silent.

3 Tangniaobing means diabetes in English

4 Participants’ proficiency was co-evaluated by the researcher and the EMI teacher

Appendix: transcription conventions

Transcriptions follow the VOICE transcription conventions (https://www.univie.ac.at/voice/page/transcription_general_information).

Table

<1 > word</1>	overlap (simultaneous utterances get the same number in the brackets)
(.)	brief pause
(2)	longer pause with the number of seconds in parentheses
=	other-continuation
<un>***</un>	unintelligible speech
<soft > word</soft>	utterance in a soft voice
<loud > word</loud>	utterance in a loud voice
word fragment-	word fragment
word.	falling intonation
word?	rising intonation
Chinese word {English translation}	Chinese utterance and corresponding translation into English
@	laughter
<pvc > word</pvc>	pronunciation variations & coinages
<spel > words or abbreviations</spel>	spelled out words or abbreviations
<ipa > phonetic representation</ipa>	phonetic representation
SS	utterances assigned to more than one speaker