1. Introduction
Online distance education allows educators and instructional designers to leverage innovative technologies to foster engagement and collaboration among expansive groups of learners (Jin et al., Reference Jin, Karatay, Bordbarjavidi, Yang, Kochem, Muhammad and Hegelheimer2022). These learners form diverse learning communities and work together towards shared goals (Beldarrain, Reference Beldarrain2006; Wenger, Reference Wenger1998). Group work is highly regarded in educational settings, encouraging teachers to create student groups of different sizes, including large classes or subdivided groups within larger classes, tailored to specific teaching objectives and contexts (Qiu, Hewitt & Brett, Reference Qiu, Hewitt and Brett2014). A prominent approach in educational research is computer-mediated collaborative writing (CMCW), which utilises tools such as wikis, blogs, Google Docs, and chat platforms to facilitate interaction and collaboration among students, overcoming barriers of time and space.
While previous studies within the CMCW context have examined different aspects of online interaction and collaboration, including collaboration on discussion forums (Lee, Reference Lee2010; Li & Zhu, Reference Li and Zhu2017; Nami & Marandi, Reference Nami and Marandi2014) and co-editing of writing (Alsahil, Reference Alsahil2024; Aydin & Yildiz, Reference Aydin and Yildiz2014; Kessler & Bikowski, Reference Kessler and Bikowski2010; Kitjaroonchai & Suppasetseree, Reference Kitjaroonchai and Suppasetseree2021), most have focused on small groups of three to five members, exploring writing processes (Li & Kim, Reference Li and Kim2016; Li & Zhu, Reference Li and Zhu2013; Storch, Reference Storch2002), the influence of collaboration on writing products (Abrams, Reference Abrams2019; Li & Zhu, Reference Li and Zhu2017), and student perceptions (Alsahil, Reference Alsahil2024; Ismail, Lustyantie & Emzir, Reference Ismail, Lustyantie and Emzir2020; Zhai, Reference Zhai2021). Limited research has delved into the effects of larger group sizes on student learning outcomes, writing products, and perceptions of collaborative group work (e.g. Kessler & Bikowski, Reference Kessler and Bikowski2010; Kooloos et al., Reference Kooloos, Klaassen, Vereijken, Van Kuppeveld, Bolhuis and Vorstenbosch2011; Zha & Ottendorfer, Reference Zha and Ottendorfer2011).
Online distance education removes logistical and resource constraints for distant learners, leading to the rise of large-enrolment online courses that benefit students, teachers, and institutions (Bikowski, Park & Tytko, Reference Bikowski, Park and Tytko2022). However, these large classes require effective strategies for enhancing student engagement and learning quality, underscoring the need for further research. Given the preference for large-group tasks in larger class sizes (e.g. Abe, Reference Abe2020; Zha & Ottendorfer, Reference Zha and Ottendorfer2011), it is crucial to examine the interaction dynamics within these larger groups comprising approximately 10 members and their impact on the learning process in online distance education. Additionally, while previous studies primarily collected data from course-designated public platforms accessible to all groups of students (Lee, Reference Lee2010; Li & Kim, Reference Li and Kim2016; Li & Zhu, Reference Li and Zhu2013, Reference Li and Zhu2017), the popularity of social communication tools, such as Facebook, WhatsApp, and WeChat, promoted the collection of data from private chats (accessible only to each group) to deepen understanding of student interaction processes and their collaborative learning experience. However, limited research explored the group dynamics collected through such private chats.
This study adopts a sociocultural theory perspective to analyse the language functions in online discussion posts gathered from both the public forum and private chats, aiming to facilitate large-group collaborative writing. The analysis focuses on exploring the interaction processes and students’ perceptions by investigating how large groups negotiate and co-construct the writing task. As our main objective is to comprehend the initial collaborative dynamics and decision-making processes rather than the subsequent editing stages, our study does not involve investigations into co-editing and text revision. The study aims to examine the interaction processes within two large groups of eight to 10 students to answer the following research questions:
-
1. What are the learning outcomes of the large-groups’ collaborative writing?
-
2. What are the large-groups’ intra-group interaction patterns in both the public forum and the private group chat?
-
3. How do the two large groups perceive their collaborative learning experience?
2. Literature review
2.1 Interaction patterns in collaborative L2 writing
In collaborative writing, the term “interaction” refers to communication and reactions between individuals. Damon and Phelps (Reference Damon and Phelps1989) introduced two key concepts – equality (the extent of commitment to collaborative writing and control over its direction) and mutuality (the degree of engagement with each other’s contributions) – to understand learner interaction. Leveraging these concepts, Storch (Reference Storch2002) identified four interaction patterns in face-to-face collaborative writing among 10 pairs: collaborative (equal participation and workload sharing), expert/novice (some experts support others), dominant/dominant (individual work without connection), and dominant/passive (some participants control while others follow passively). Expanding on Storch’s (Reference Storch2002) model, Bradley, Lindström and Rystedt (Reference Bradley, Lindström and Rystedt2010) observed three distinct interaction patterns in the co-construction of text and exchange of peer responses in groups of two to three: lack of interaction, cooperative approach (individual work), and collaborative approach (jointly constructing texts).
Li and Zhu (Reference Li and Zhu2017) analysed intra-group interaction patterns in an L2 research proposal writing task with four small groups of three, identifying four interaction patterns: collective (showing a higher degree of mutuality than Storch’s, Reference Storch2002, collaborative pattern), expert/novice, dominant/defensive, and cooperative in parallel (similar to Bradley et al.’s, Reference Bradley, Lindström and Rystedt2010, cooperative pattern). It is worth noting that the interpretation of “expertise” in the “expert/novice” pattern varied across studies, with some considering language proficiency (Storch, Reference Storch2002) and others focusing on topic familiarity (Li & Zhu, Reference Li and Zhu2017).
In addition to investigating interaction patterns in single writing tasks (Abrams, Reference Abrams2019; Arnold, Ducate & Kost, Reference Arnold, Ducate and Kost2012; Li & Zhu, Reference Li and Zhu2017; Ma, Reference Ma2020) research has explored interaction patterns across multiple tasks (Abe, Reference Abe2020; Hsu, Reference Hsu2020; Kitjaroonchai & Suppasetseree, Reference Kitjaroonchai and Suppasetseree2021; Li & Kim, Reference Li and Kim2016; Wang, Reference Wang2019). For example, Li and Kim (Reference Li and Kim2016) observed dynamic interaction patterns among L2 groups working on identical tasks in the same wiki space. The quality of group writing products was analysed in relation to these interaction patterns (e.g. Li & Zhu, Reference Li and Zhu2017), with collective patterns yielding the highest quality, followed by expert/novice patterns, and dominant/defensive and cooperative patterns resulting in lower quality.
2.2 Language functions in collaborative L2 writing
Sociocultural theory (Vygotsky, Reference Vygotsky1978) suggests that human cognition develops through socially mediated mechanisms, with language communication being essential in L2 learning. Building on this theory, Storch (Reference Storch2002) used Damon and Phelps’s (Reference Damon and Phelps1989) concepts of “equality” and “mutuality” to define group interaction patterns, which have subsequently been applied to examine writing and revision behaviours (Arnold et al., Reference Arnold, Ducate and Kost2012; Bradley et al., Reference Bradley, Lindström and Rystedt2010; Li, Reference Li2013). Mak and Coniam (Reference Mak and Coniam2008) first proposed four writing change functions: adding, expanding, reorganising, and correcting. Similarly, Arnold et al. (Reference Arnold, Ducate and Kost2012) analysed meaning- and language-related revisions in an L2 context, including adding, deleting, reordering, and correcting, to determine whether students revised only their own contributions (cooperation) or the entire text (collaboration). Li (Reference Li2013) further identified five writing change functions in L2 collaborative writing: addition, deletion, rephrasing, reordering, and correction.
While earlier research (Arnold et al., Reference Arnold, Ducate and Kost2012; Bradley et al., Reference Bradley, Lindström and Rystedt2010; Mak & Coniam, Reference Mak and Coniam2008) focused on text construction behaviours, more recent studies (Hsu, Reference Hsu2020; Kitjaroonchai & Suppasetseree, Reference Kitjaroonchai and Suppasetseree2021; Li & Kim, Reference Li and Kim2016; Li & Zhu, Reference Li and Zhu2017; Wang, Reference Wang2019) have delved into task negotiation behaviours in collaborative writing, particularly examining language functions in the wiki “Discussion” sections. These language functions, such as suggesting, eliciting, justifying, and questioning (Li, Reference Li2014), similar to those in peer response activities (Lockhart & Ng, Reference Lockhart and Ng1995; Zhu, Reference Zhu2001), shape interaction patterns and influence the final written products.
Li and Kim (Reference Li and Kim2016) developed a systematic coding framework to analyse language functions in small-group interactions during L2 writing tasks, categorising them into initiating and responding types with 11 subcategories, including acknowledging (recognising or praising others’ ideas), agreeing (expressing agreement with others’ viewpoints), requesting (making direct requirements or requests), eliciting (inviting opinions or comments), etc. This framework has been used to study interaction patterns in pair or small-group collaborative writing tasks regarding “equality” and “mutuality” (Hsu, Reference Hsu2020; Kitjaroonchai & Suppasetseree, Reference Kitjaroonchai and Suppasetseree2021; Wang, Reference Wang2019). However, the relationship between the two primary language function categories (initiating and responding) and their 11 subcategories remains unclear.
2.3 Leadership style and student engagement in collaborative learning experience
While leadership research typically concentrates on adult leadership in various contexts, there is limited exploration of student leadership and engagement (Tortosa Martínez, Pérez-Fuentes & Jurado, Reference Tortosa Martínez, Pérez-Fuentes and Jurado2022). Previous studies investigating the relationship between leadership experience and student engagement reveal a strong connection between the two (Garton & Wawrzynski, Reference Garton and Wawrzynski2021).
Student engagement, also referred to as student involvement or learning participation, has been increasingly emphasised in recent years. While there are various definitions of student engagement, many studies examine it through the three core dimensions (behavioural, emotional, and cognitive) identified by Fredricks, Blumenfeld and Paris (Reference Fredricks, Blumenfeld and Paris2004). In the context of online learning, student engagement refers to active involvement and participation when using online learning platforms for learning (Hu, Li, Deng & Guan, Reference Hu, Li, Deng and Guan2016). Within these virtual learning settings, some students may assume leadership roles, whether formally assigned or emerging through group interactions, to facilitate discussions. The group leaders’ engagement with peers through inquiry, decision-making, discussion, or problem-solving positively influences group dynamics (Micari, Pazos, Streitwieser & Light, Reference Micari, Pazos, Streitwieser and Light2010; Zha & Ottendorfer, Reference Zha and Ottendorfer2011).
Research indicates that leadership style significantly influences student engagement in online collaborative learning environments, where students work in groups to construct knowledge and solve problems. Yamaguchi, Bos and Olson (Reference Yamaguchi, Bos and Olson2002) examined small groups using various communication channels and found that emergent leadership, whether dominant (task-focused) or democratic (relationship-focused), enhances group effectiveness. However, more limited computer-mediated channels, such as text chat, hinder relationship-focused leadership. Luo, Han, Chen and Nie (Reference Luo, Han, Chen and Nie2022) found that in online team learning, there is a preference for person-focused and facilitative leadership. They noted that excessive leader involvement and personal contributions can negatively impact individual learning. Understanding the effects of different leadership styles on student engagement is crucial for enhancing collaborative learning experiences.
3. Methodology
3.1 Context and participants
Ninety-seven students pursuing master’s degrees in English language education at a university in the Chinese mainland participated in a computer-assisted language learning (CALL) course in a distance education setting. The students were divided into 10 groups of eight to 10 members to complete an online group assignment. The task involved co-constructing a summary that evaluated two to three English vocabulary learning apps or websites using the evaluation principles taught in the course. The summary was expected to include basic information about the apps/websites, their advantages, limitations, and suggestions for improvement. Each group conducted their discussions on an online forum with “Comment” and “Reply” functions. Students could create new threads to initiate topics or reply to others’ posts. After the discussions, each group wrote a summary based on their interactions. The summary writing process was discussed within the groups on the forum. Additionally, each group created a private chat group on WeChat, a messaging app installed on their mobile phones, for discussing work division and making important decisions. The public online forum captured the group interactions and discussions, while the private WeChat groups contained information accessible only to each specific group.
Since most students who took the course were busy in-service language teachers, only two groups, each consisting of 10 students, agreed to participate in the study. Since these two groups were the only ones that voluntarily participated in the study and were willing to engage with the researchers, these characteristics made them an “information-rich and accessible sample” (Patton, Reference Patton2015: 466).
3.2 Procedures
The research was conducted over 10 weeks (see Table 1). In the first week, students attended a three-hour lecture on evaluating vocabulary learning software/websites and were introduced to the collaborative writing project, specifically the task of summary writing. The instructor posted task instructions on the online platform. In the second week, students formed groups of eight to 10 members and selected their group leaders. From Weeks 3 to 6, students engaged in group discussions on the designated public platform and in self-selected WeChat group chats to collaborate on the summary writing task. Weeks 7 and 8 involved assessing, grading, and providing comments on the group writings. Finally, in Weeks 9 and 10, follow-up group interviews were conducted with three students from each group who gave their consent.
Table 1. Procedures
3.3 Data collection
In this study, a mixed-methods approach was used to collect data. The data consisted of (1) writing products (summary of the evaluation of learning apps/websites) of the two groups (around 1,000 words for each writing product); (2) discussion data from the public forum (totalling 21,085 words) and private WeChat groups (totalling 4,695 words); and (3) group interview data on student perceptions of the collaborative learning experience. Semi-structured group interviews (please refer to supplementary material for the interview guidelines) were conducted, which explored five dimensions: (1) the role of the leader, (2) decision-making processes, (3) division of labour, (4) group interaction, and (5) reflections on the learning experience. Each interview lasted about 90 minutes.
3.4 Data analysis
To answer the first question, we analysed the quality of the two group writing products based on the raters’ scores and comments. Each writing product was evaluated independently by two raters: the course instructor and another rater teaching the same subject. The evaluation criteria consisted of five aspects: (1) the choice of apps/websites, (2) the use of evaluation principles, (3) content, (4) structure, and (5) language use. Each criterion was evaluated using a scale ranging from 1 to 20. Given the five criteria, the combined maximum achievable score for all criteria was 100. An acceptable inter-rater agreement was established at ±2 marks for each criterion to ensure consistency. In case of discrepancies, the raters engaged in discussion and reassessment to reach a consensus. The final score for each writing product was determined by calculating the average of the scores the two raters gave.
To address the second question, the discussion data from the public platform and the private WeChat group were coded in terms of Li and Kim’s (Reference Li and Kim2016) taxonomy of language functions. In the first level of coding, all online posts were classified as either initiating (proposing new ideas) or responding (reacting to other members’ ideas) language functions. In the second level of coding, all the initiating and responding language functions were further classified into 12 subcategories. Based on frequency counts of the language functions and content analysis of the discussion data recorded on both platforms, interaction patterns for the two groups were derived in terms of “equality” and “mutuality”. Equality was measured by comparing language functions performed by each member to assess their contributions. According to Li and Kim (Reference Li and Kim2016), balanced commitment indicates high equality, while unequal contribution indicates low equality. Mutuality was measured by calculating the ratio between initiating and responding language functions for each group. A high number of responding language functions indicated a higher level of mutuality (Li & Kim, Reference Li and Kim2016). The data coding was done independently by the first and second authors of the article, and the inter-coder reliability was 0.88. Disputed cases were resolved through discussion and mutual agreement. Table 2 presents the definitions of each language function and examples from the discussion data.
Table 2. The taxonomy of language functions adapted from Li and Kim (Reference Li and Kim2016)
Note. Addressing* is a new language function added to the subcategories because some examples of addressing others’ requests were found in the discussion data in the current study.
While some language functions – (5) elaborating, (6) eliciting, (7) greeting, (10) requesting, (11) stating, and (12) suggesting – could be either initiating and responding language functions depending on the context, other language functions – (1) acknowledging, (2) addressing*, (3) agreeing, (4) disagreeing, (8) justifying, and (9) questioning – were more likely to be responding language functions. For example, students may initiate group discussions by greeting group members, elaborating on their ideas, or offering suggestions about the writing content. Take one post from the public platform as an example:
Hi guys (Greeting), we should select one skill or knowledge area, review 2–3 related websites or software according to the evaluating principles learned in Unit 2, discussing their merits, limitations, and areas for improvement on the public platform. (Suggesting)
In this case, the “greeting” and “suggesting” language functions fall into the category of “initiating”.
Conversely, if students extend others’ ideas, make requests, or state ideas previously discussed by the group, these posts fall into the category of “responding”. Take a post from the public platform as an example:
I do believe that you have done a great job with the analysis. (Acknowledging) I couldn’t agree more. (Agreeing)
In posting this, the student responded to others by recognising or praising their ideas and expressing agreement with their viewpoints. In this case, the “acknowledging” and “agreeing” language functions fall into the “responding” category.
As shown in the two examples, one discussion post may serve more than one language function. Therefore, the number of language functions instead of the discussion posts was counted. Furthermore, the new language function of “addressing*” was added to the subcategories because some examples of addressing others’ requests were found in the discussion data in the current study. For example, one group leader posted, “I have made a form. Could you please fill in the form with the information you collected?”
By responding “I have added new information to the form accordingly”, the participant addressed others’ requests or answered their questions. Accordingly, the language function of this post was coded as “addressing”, which falls into the “responding” category.
To answer the third question, the interview data were transcribed verbatim. The qualitative data analysis followed the six-step approach recommended by Braun and Clarke (Reference Braun and Clarke2006). This involved getting familiar with the data, generating initial codes, identifying themes, reviewing and refining the themes, defining and naming the themes, and finally writing up the analysis. The interview data were independently transcribed and coded by the first and second authors of this paper, with an inter-coder reliability of 0.89, indicating a high level of agreement. Any discrepancies or disagreements were resolved through discussion and mutual agreement. Analysing the interview data provided valuable insights into the relationship between students’ intra-group interaction processes and their learning outcomes.
4. Results
4.1 Learning outcomes of large-groups’ collaborative writing (RQ1)
Group 1 received a score of 90 out of 100, while Group 2 received a score of 71 out of 100. Group 1 demonstrated higher overall writing quality, with a good awareness of content, structure, and language accuracy. In contrast, Group 2 had multiple coherence breakdowns, inadequate structure, and inaccurate language use. Table 3 shows the subscores and total scores for each group.
Table 3. Subscores and total scores for each group
In terms of subscores, Group 1 outperformed Group 2 in four out of the five categories, including the appropriate use of evaluation principles, content, structure, and language use. Group 1 provided detailed explanations of the strengths of the apps based on the evaluation principles learnt, while Group 2 listed only features and strengths without providing in-depth explanations. Both groups covered the required content, but Group 1 provided a well-structured and detailed analysis with fewer grammatical errors and a broader range of vocabulary. Conversely, Group 2 used a bullet-point format and made numerous grammatical mistakes, leading to a comment from the rater about the need for improvement in language use.
4.2 Intra-group interaction patterns (RQ2)
4.2.1 Equality
To examine equality among group members, we counted the language functions performed by each member and compared them to assess their contribution to and control over the direction of the writing task during the discussion. Balanced contributions among group members indicated high equality, while unequal contributions indicated low equality (Li & Kim, Reference Li and Kim2016). As shown in Table 4, for Group 1, all 10 members participated in the public and private chat discussions. Six members (Emily*, Aiden, Brian, Cindy, David, and Julia) actively engaged in the discussion on both platforms, serving as experts in CALL evaluation and leading the discussion. The leader, Emily*, performed the highest number of language functions (39), while the other members made more or less equal contributions (21–27 language functions). Overall, Group 1 demonstrated a good level of equality. In contrast, in Group 2, two members were absent from the discussion on the public platform, and one did not participate in the WeChat group discussion. The leader, Kali*, dominated the discussion with 41 language functions, while other members had unequal participation, performing only 0–16 language functions each. This indicates low equality within Group 2. In summary, Group 1 showed a higher level of participation in and commitment to collaborative writing. They performed four times more language functions (90) on the public platform than Group 2 (22) and overall performed two times more total language functions (202) than Group 2 (99).
Table 4. Overall counts of language functions per person on the public platform and private chats
Note. The names with * are leaders selected by the group members in the group.
4.2.2 Mutuality (first-level coding)
In the first level of coding, the discussion was classified as either initiating or responding language functions. According to Li and Kim (Reference Li and Kim2016), a higher number of responding language functions indicates a higher level of mutuality, while a higher number of initiating language functions suggests a lower level of mutuality. Table 5 shows that Group 1 had a much larger total number of language functions on the public platform (69 initiating and 21 responding) than Group 2 (18 initiating and four responding). This suggests that Group 1 engaged in more socialising and interactive discourse. Mutuality was low for both groups on the public platform, as indicated by the ratio between initiating and responding language functions. However, the ratio between responding and initiating language functions for Group 1 (69:21 = 1:3) was higher than that of Group 2 (18:4 = 1:5), indicating a relatively higher degree of mutual engagement in Group 1.
Table 5. Number of initiating and responding language functions performed by each group
As shown in Table 5, the total number of language functions in the private chat (112 for Group 1, 77 for Group 2) was significantly larger than on the public platform (90 for Group 1, 22 for Group 2). Unlike on the public platform, members were much more active and showed a far stronger bond in the private chat, where they could engage in instant conversations and receive immediate replies. Accordingly, the number of responding language functions on the private chat greatly exceeded the number of initiating language functions for both groups. Group 1 had 98 responding language functions compared to 14 initiating language functions, whereas Group 2 had 65 responding language functions compared to 12 initiating language functions. This indicates that group members actively engaged with each other’s ideas, indicating much higher mutuality for both groups in the private chat than on the public platform. Furthermore, the ratio between responding and initiating language functions in the private chat was higher for Group 1 (98:14 = 7:1) compared to Group 2 (65:12 = 5:1). This result is the reverse of that from the public forum, where the ratio for both groups was much smaller (1:3 for Group 1 and 1:5 for Group 2). The result from the private chat also confirms a higher degree of mutual engagement for Group 1.
4.2.3 Mutuality (second-level coding)
To examine mutuality more closely, the students’ discussions on both platforms were further coded and categorised into 12 subcategories based on language functions (see Table 2). Excerpts from both platforms were included to support the frequency counts of sublanguage functions and to explore how meaning and social relationships were mediated through communication.
4.2.4 Group 1
Table 6 shows a total of 83 initiating language functions from both platforms. The 64 stating language functions indicate that each of the 10 members stated their own ideas and posted writing content on the topic. This may have helped provide writing directions for other group members. Additionally, the greeting (5), suggesting (5), eliciting (4), and questioning (3) language functions successfully elicited 119 responding language functions, including suggesting (33), stating (18), acknowledging (16), agreeing (14), and questioning (13) language functions. This indicates that the 10 members interacted with each other to some extent and made joint contributions to task discussion on both platforms.
Table 6. Frequency counts of language functions performed by each group
Note. Addressing* is a new language function added to the subcategories because some examples of addressing others’ requests were found in the discussion data in the current study.
Figure 1 illustrates how active members, such as Brian, responded and contributed to the group discussion initiated by the leader, Emily*, who started the conversation by posting the assignment’s guidelines on the public platform. Brian then elaborated on the points, offered suggestions, and even posted an example to stimulate meaningful responses. This prompted Fanny and Cindy to acknowledge and follow up, setting the writing direction. Brian’s familiarity with the topic, as defined by Li and Zhu (Reference Li and Zhu2017), positions him as an expert guiding the discussion. The group’s active engagement with each other’s ideas fosters positive social relations and facilitates the co-construction of the writing.
Figure 1. Excerpt 1 from the public platform discussion for Group 1.
The discussion data from the private chat revealed results similar to those of the public platform. As shown in Figure 2, certain members of Group 1, such as Henry, assumed the role of experts, guiding the conversation and offering encouragement during the group discussion. Others, such as Grace, actively responded and participated in the discussion, claiming tasks for themselves, which received approval from the group leader. The group’s willingness to engage in collective idea-sharing and co-contribution of text indicated a high level of mutuality. This trend mirrored the behaviour observed on the public platform.
Figure 2. Excerpt 2 from the private chat discussion for Group 1.
4.2.5 Group 2
As indicated in Table 6, Group 2’s public platform data show a low degree of reciprocal interaction among members. The most common initiating language functions were stating (15 times) and suggesting (three times). The responding language functions, including two instances of requesting and two of questioning, received no further responses, indicating a communication breakdown and lack of continuity. Group members primarily posted their comments without responding to each other. The group leader, Kali*, largely controlled the conversation, addressing individual members’ posts with questions or requests. This behaviour indicates low reciprocity and mutuality within the group.
Figure 3 provides an example of group interaction on the public platform for Group 2. Lydia and Pan initiated a discussion by posting information about two apps, Momo and Xuele Cloud, respectively. However, their leader, Kali*, did not acknowledge their contributions but directly questioned their comments and requested further explanation from them. As a result, she did not receive any feedback, and the discussion continued as monologues with no reciprocal interaction among group members, resulting in low mutuality.
Figure 3. Excerpt 3 from the public platform discussion for Group 2.
Table 6 and Figure 4 show that in the private chat for Group 2, the leader, Kali*, had an authoritative style, contributing the most and allocating tasks while other members passively followed. There were eight instances of requesting language functions (three initiating and five responding) and two instances of disagreement, a pattern not found in Group 1. An example of this tension was visible when Pan questioned Olivia’s points, leading to an awkward exchange and an apology from Pan. This incident might have created a strained group atmosphere, potentially hindering positive social relationships, idea exchange, and connective knowledge building.
Figure 4. Excerpt 4 from the private chat discussion for Group 2.
Unlike Group 1, where active members elaborated on and suggested ideas for the writing task, Group 2 members passively responded to the leader’s requests and completed assigned tasks. The absence of in-depth discussions and consensus before text construction led to misunderstandings among group members. This approach resulted in loose connections among members, insufficient promotion of social relationships and familiarity, and ultimately low participation and engagement within the group.
4.2.6 Interaction patterns
Based on the quantitative and qualitative analysis of discussion data, Group 1 demonstrated an expert/participant pattern overall, a mixture of expert/novice and collaborative patterns. This was because the “novices” in this group also provided new perspectives and engaged in collective scaffolding. On the other hand, Group 2 demonstrated a dominant/passive pattern, in which the leader took more control of the task while others followed passively. Further details about these patterns are presented in Table 7.
Table 7. Different group interaction patterns for the two groups
4.3 Large-group students’ perception of their collaborative learning experience (RQ3)
To address the third research question, we examined post-task interview data to understand student perceptions of the collaborative learning experience in large groups, based on which we may understand to some extent how their collaborative learning processes influence their writing products. Through qualitative analysis, three themes emerged regarding their learning experience: (1) leadership styles: democratic versus authoritative; (2) student engagement: active versus passive; and (3) learning experience: from positive to negative.
-
(1) Leadership style: Democratic (Group 1)
The team leader plays the role of a mobiliser, the final decision-maker, and an encourager. She initiated the group discussion, and then group members followed up. (Interview: Aiden, Group 1)
When there is a disagreement, and I have to make a decision, I would first inform each member personally, either by text or voice messages, explaining the reasons behind it and welcoming any suggestions/advice before announcing the decision in the WeChat group. Overall, everyone was happy with my decision, and the cooperation went smoothly. It was a hard process though. (Interview: Emily* [leader], Group 1)
-
(2) Student engagement: Active (Group 1)
There was no specific division of labour, but group members claimed tasks for themselves and cooperated well with each other harmoniously. (Interview: Aiden, Group 1)
We formed smaller units of two to three, and each subgroup was responsible for a specific task at each phase, such as collecting information, posting initiating comments, drafting/revising and proofreading. (Interview: Brian, Group 1)
In addition to the discussion in the WeChat group, we also communicated privately with subgroup members through text or voice messages. (Interview: Emily*, Group 1)
Group 1, characterised by a democratic leadership style, demonstrated a more open exchange of ideas and active participation among group members. The leader encouraged members to freely express their opinions in both the chat group and private discussions, ensuring every member had a voice in the decision-making process. In case of disagreement, the leader invested considerable effort in communicating with group members to maintain a harmonious and cooperative atmosphere. She welcomed members’ participation, opinions, and advice, fostering a sense of familiarity and positive social relationships. Group members formed subgroups, each responsible for a specific task. This organisational pattern promoted student engagement within each subgroup as well as within the entire group. Members actively engaged in constructive interactions, providing guidance and support to one another. They utilised various language functions, such as suggesting, acknowledging, questioning, and addressing, to facilitate supportive discourse. These collective efforts led to a truly collaborative writing process and significantly improved the quality of their written work. Consequently, Group 1, demonstrating an expert/participant pattern with balanced equality and high mutuality, produced a high-quality piece of writing, reflected in their rating score of 90.
-
(1) Leadership style: Authoritative (Group 2)
Our leader’s role involved organising, making decisions, and coordinating team members. She assigned tasks, communicated with us, and took the lead during the assignment. (Interview: Nina, Group 2)
As the due date was approaching, our leader designed an evaluation form and encouraged group members to focus on two apps to increase efficiency. (Interview: Olivia, Group 2)
I downloaded and tested the apps proposed by group members on both the public platform and the WeChat group and decided to focus on two apps that have been well evaluated by two group members respectively. I made the decision because time was limited, and I think it was the evaluating principles, not the choice of apps, that matters. (Interview: Kali* [leader], Group 2)
-
(2) Student engagement: Passive (Group 2)
I asked the two members who proposed the apps to fill in the evaluation form first, with the other members providing supplements. (Interview: Kali*, Group 2)
As I was assigned the task to evaluate Cake English based on the evaluating principles chosen by our leader, I downloaded and used the app and posted my opinions on the public platform. Then I filled in the form provided by our leader and shared it in the WeChat group. (Interview: Nina, Group 2)
We responded to our leader on both platforms, and she then merged and integrated our opinions into the final document. (Interview: Olivia, Group 2)
Conversely, Group 2, characterised by an authoritative leadership style, demonstrated a generally more restricted exchange of ideas between the leader and active members. The leader took charge of the writing task and made most contributions. Without discussing with group members, she decided on the focus of the topic and evaluating principles and instructed group members to complete a form she designed. Group members worked in isolation with less engagement with each other’s contributions. As a result, their written product represented a mere amalgamation of individual efforts rather than a truly collaborative endeavour. The lack of mutuality in task negotiation and text construction resulted in a writing output that lacked structure, coherence, and language accuracy. Thus, Group 2, exhibiting a dominant/passive interaction pattern, produced a piece of writing of relatively poor quality, as indicated by their rating score of 71.
-
(3) Learning experience:
Positive learning experience (Groups 1 and 2):
I gained a deeper understanding of evaluation principles and can now choose from various apps to meet my needs. Additionally, proofreading others’ writing improved my organisation of ideas and word choice. (Interview: Aiden, Group 1)
Everyone has his/her strengths and weaknesses. We looked at others’ ideas critically and learned the good points from them. (Interview: Emily*, Group 1)
I would think critically whether the evaluation criteria we chose were appropriate for assessing a specific app. (Interview: Nina, Group 2)
I learned how to express my thoughts in an organised way through this writing. (Interview: Olivia, Group 2)
Negative learning experience (Groups 1 and 2):
We hesitated to post everything on the online forum due to its public nature, allowing members from other groups to access and read our comments and even post their own comments. (Interview: Brian, Group 1)
The forum is not so convenient for discussion. It takes time for us to go to the website and log in. I prefer more immediate communication tools like WeChat. (Interview: Nina, Group 2)
In conclusion, the study revealed that the two groups exhibited different leadership styles and student engagement, which may have influenced their different writing outcomes and learning experiences. While both groups generally had positive learning experiences, it is essential to note that one student from each group expressed concerns about the limitations and inconveniences of the public forum used for collaboration. These concerns may have impacted their participation level and the writing outcomes to varying degrees.
5. Discussion and implications
This study examined the interaction patterns within two large groups by analysing group discussions on a public platform and in private chats and their possible influence on student writing outcomes. The findings of this research shed light on intra-group interactions within large groups, providing valuable implications for language teaching and group task design.
One key finding of this study is that mutual engagement is crucial in fostering interaction and collaboration, leading to improved learning outcomes in large groups. This supports previous research by Storch (Reference Storch2002) and Paulus and Roberts (Reference Paulus and Roberts2006), highlighting the importance of mutuality in small-group settings. The study suggests that mutual engagement is equally essential to large groups and contributes to successful collaborative L2 writing tasks.
Interestingly, the study also revealed that high mutuality might be more crucial than high equality for achieving better writing outcomes. The higher-performing group, which displayed mid-level equality and high-level mutuality, outperformed the lower-performing group. This supports Li and Zhu’s (Reference Li and Zhu2017) claim that equality alone does not guarantee high-quality writing without mutual engagement. To enhance the quality of group writing, teachers should design tasks that promote peer interaction and foster high levels of mutuality. Furthermore, one student from each group raised concerns about the limitations and inconveniences of the public forum used for collaboration. One possible solution is to introduce private group discussions, allowing members to have more confidential and focused conversations confined to their own groups. Additionally, improving the accessibility of the forum should be considered to ensure prompt log-ins and enhance overall productivity.
Students were more engaged and established stronger relationships in private WeChat groups compared to public platforms, as evidenced by both quantitative and qualitative data. The private chat conversations displayed increased reciprocity and stronger bonds among members, evident through various language functions like acknowledging, agreeing, questioning, and suggesting. This aligns with Thorne’s (Reference Thorne2003) findings that language learners may prefer certain culture-specific communication tools (e.g. WeChat for Chinese learners) and that individual and collective experience greatly influences student engagement in technology-mediated communication, subsequently affecting language development processes and outcomes. Therefore, educators should promote private chats through social communication tools alongside public platforms to foster strong connections and positive social relationships within large groups.
Task distribution within large groups promotes interaction and mutual engagement. This is supported by Kooloos et al.’s (Reference Kooloos, Klaassen, Vereijken, Van Kuppeveld, Bolhuis and Vorstenbosch2011) study, which found that splitting tasks in large groups enhances participation and learning outcomes. In the high-performance group of this study, members formed subgroups for task division, promoting collaboration on smaller tasks. Each subgroup shared their work with the entire group, taking responsibility for their portion. In contrast, the low-performance group worked individually and reported to the leader, with only the leader responsible for integrating individual contributions. To promote balanced group dynamics, teachers could facilitate task division or designate specific roles to members to maximise the learning of students with different roles within online groups. In addition, teachers can monitor students’ discussions by joining each group or experimenting with various roles in the group discussion. This allows students to identify potential flaws in task design, improve instructions for peer-led learning, and promote peer interaction and mutuality, creating a more equitable and supportive learning environment. For example, in the case of the group that demonstrated a dominant/passive interaction pattern, if the teacher had intervened and provided timely support, the group may have changed their interaction pattern.
The study highlights the influence of leadership style and student engagement in collaborative learning experiences. This corresponds with the research conducted by Luo et al. (Reference Luo, Han, Chen and Nie2022), which revealed that a person-focused and facilitative leadership style positively affects the learning of individual members. In the higher-performing group, the leader encouraged member participation and valued their input, fostering social relationships and collegiality. In contrast, the lower-performing group had limited opportunities for initiatives due to an authoritative leadership style, hindering the development of personal ties. This divergence in outcomes, despite both leaders playing pivotal roles, aligns with Kukulska-Hulme’s (Reference Kukulska-Hulme2004) findings that leaders who value individual contributions and prioritise relationships lead more effective groups. This underscores the significance of a dedicated leader who focuses on task completion and member satisfaction to motivate group members toward their collective goals.
6. Conclusion
This study examined intra-group interaction in large-group collaborative writing within a distance learning context. The findings revealed interesting patterns of interaction related to “equality” and “mutuality”, which may be of interest to researchers and educators seeking to cultivate effective collaborative writing in large groups. Additionally, this study provides some suggestions on task design that could potentially foster collaborative knowledge construction, enhance writing performance, and promote L2 learning in the context of large groups, despite the challenges involved in designing collaborative writing tasks for large groups.
This study has several limitations. First, the small sample size of only two groups may make it difficult to generalize the results to other similar settings. To enhance the generalizability of the findings, future research should include a larger and more diverse sample size, encompassing a wider range of cases and contexts. Second, the fact that students formed their groups could have affected the results. Certain characteristics of the individuals comprising the high-performing group may have also contributed to their group dynamics and hence shaped their interaction pattern. Random group assignments may have mitigated such possible influence and produced more reliable findings. Future research should investigate how to manage group formation and leadership selection to reduce the potential effects of pre-existing group dynamics. Third, while this study looked at discussion data from public forums and private chats, it did not explore the actual text co-construction process. Future studies should investigate how intra-group communication affects the quality of large-group writing by examining both discussion and editing modes. These efforts will lead to a more thorough understanding of how intra-group interaction impacts the quality of large-group collaborative writing.
Supplementary material
To view supplementary material referred to in this article, please visit https://doi.org/10.1017/S0958344024000211
Ethical statement and competing interests
Informed consent was obtained from all participants, and they were provided with clear information about the purpose of the study, the procedures involved, and any potential risks or benefits. Pseudonyms were used for each participant. The authors declare no competing interests in this research. The authors declare no use of generative AI.
About the authors
Fang Mei is a research associate and an EdD candidate at the Education University of Hong Kong. She has been a TOEFL teacher, a teacher trainer, and now a researcher. Her research interests include CALL, corpus-based language teaching and learning, and teacher training.
Qing Ma is an associate professor at the Department of Linguistics and Modern Language Studies, the Education University of Hong Kong. Her main research interests include second language vocabulary acquisition, corpus linguistics, CALL, and mobile-assisted language learning. Her current research focuses on how to theorise and validate empirically a corpus-based language pedagogy.
Jinlan Tang is the dean and professor at the Institute of Online Education and a researcher at the Artificial Intelligence and Human Languages Lab, Beijing Foreign Studies University, China. Her research covers the areas of language assessment, teacher training, and EFL teaching and learning in the e-learning environment. She also serves as the secretary-general of the China Computer-Assisted Language Learning Association.
Bin Zou is an associate professor at the Department of Applied Linguistics, Xi’an Jiaotong-Liverpool University, China. He received his PhD from the University of Bristol, UK. His research interests include artificial intelligence, ELT, EAP, and educational technology. He has published many papers in the CALL field.
Open access
