Computers and Education: Artificial Intelligence 6 (2024) 100228

Contents lists available at ScienceDirect

Computers and Education: Artificial Intelligence

journal homepage: www.sciencedirect.com/journal/computers-and-education-artificial-intelligence

Synergizing collaborative writing and AI feedback: An investigation into

enhancing L2 writing proficiency in wiki-based environments
Watcharapol Wiboolyasarin a, Kanokpan Wiboolyasarin b, Kanpabhat Suwanwihok c,
Nattawut Jinowat d, *, Renu Muenjanchoey a
a
Research Institute for Languages and Cultures of Asia, Mahidol University, Nakhon Pathom, Thailand
b
Faculty of Management Science, Chandrakasem Rajabhat University, Bangkok, Thailand
c
College of Teacher Education, Phranakhon Rajabhat University, Bangkok, Thailand
d
Demonstration School, Suan Sunandha Rajabhat University, Bangkok, Thailand

A R T I C L E I N F O A B S T R A C T

Keywords: Collaborative writing, paired with AI feedback, represents an innovative and effective pedagogical approach with
L2 writing the potential to revolutionize language learning practices. This study explores the influence of a three-step
Collaborative writing collaborative writing intervention, where AI feedback specially enhances the revision and editing stages of the
AI feedback
L2 writing process for Thai-major undergraduate exchange students. Employing a quasi-experimental design, the
ChatGPT
Wiki environment
study included 39 junior exchange students, with 18 in the experimental group and 21 in the control group. The
approach employed a wiki-based collaborative writing strategy, integrating ChatGPT for written corrective
feedback. The 10-week intervention encompassed three writing tasks, involving collaborative outlining, indi­
vidual writing with AI feedback, and collaborative writing. Statistical analyses, comprising the Wilcoxon Signed-
Rank test, Mann-Whitney U test, and Friedman test, revealed substantial improvements in posttest scores for the
experimental group in comparison to the control group. These findings underscore the positive impact of a
collaborative writing intervention improved by AI feedback on specific aspects of writing. This study contributes
valuable insights into advancing L2 writing proficiency through structured collaborative activities and targeted
AI feedback, providing valuable implications for both research and pedagogical practices.

1. Introduction individualized feedback to multiple student groups significantly in­

creases their workload (Zou et al., 2023). Students, in turn, may struggle
The evolving technological landscape demands innovative ap­ to receive the timely and specific feedback essential for their develop­
proaches in language education. Educators are constantly seeking stra­ ment as writers (Han & Sari, 2022). This highlights the need for efficient
tegies to enhance Second Language (L2) writing proficiency. tools that can promote instructional efficiency and support personalized
Collaborative writing has emerged as a valuable method for improving learning paths.
students’ writing skills (Alwaleedi et al., 2019; Beiki et al., 2020; Chen & To assist learners and educators in the writing process, Automated
Ren, 2022; Manegre & Gutiérrez-Colón, 2023; Zou et al., 2023). Often Writing Evaluation (AWE) systems such as Grammarly and QuillBot
integrated with process writing, collaborative writing fosters student have been introduced. These systems utilize natural language processing
engagement and is considered a crucial mechanism for student devel­ to detect and suggest corrections for errors in writing (Escalante et al.,
opment (Chao & Lo, 2011; Liu et al., 2022). These activities are most 2023). However, the emergence of Artificial Intelligence (AI), particu­
effective within a student-centered learning environment, where stu­ larly the advanced capabilities of Large Language Models (LLMs), offer
dents actively identify and address weaknesses in their writing while new avenues for enhancing educational practices. Such technologies
co-constructing drafts. Instructors then act as facilitators and mentors, have not only revolutionized the potential for designing productive
offering feedback and guidance to guide the collaborative process. learning activities but also for developing better technology-enhanced
However, at the university level, collaborative writing presents learning environments (Chen et al., 2020; Hwang et al., 2020). Educa­
challenges. Instructors grapple with large classes, where providing tors and researchers are therefore increasingly turning AI-generated

* Corresponding author.
E-mail address: [email protected] (N. Jinowat).

https://doi.org/10.1016/j.caeai.2024.100228
Received 13 March 2024; Received in revised form 20 April 2024; Accepted 21 April 2024
Available online 24 April 2024
2666-920X/© 2024 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC license (http://creativecommons.org/licenses/by-
nc/4.0/).
W. Wiboolyasarin et al. Computers and Education: Artificial Intelligence 6 (2024) 100228

feedback as a cost-effective and efficient alternative in writing class­ within educational curricula to bolster learning outcomes. Indeed, reg­
rooms at the higher education level (Meyer et al., 2024; Ruwe & ular engagement in collaborative writing activities could enhance
Mayweg-Paus, 2023; Steiss et al., 2024), with AI tools capable of fluency, accuracy, and complexity in student writing by providing
providing personalized feedback that alleviates the instructional burden consistent practice and feedback opportunities. Moreover, the role of AI
(Kasneci et al., 2023). feedback in these collaborative environments remains an area ripe for
LLMs, such as ChatGPT, Gemini, and LLaMA, represent a significant exploration, specifically regarding the potential for AI to augment the
leap forward, extending the capabilities of automated writing feedback writing process through provision of timely, targeted, and adaptive
(Yang et al., 2023). Rooted in transformer architectures (Devlin et al., feedback. Addressing this gap is crucial to advancing our comprehension
2018) and attention mechanisms (Vaswani et al., 2017), these models of the ways in which AI tools can serve as effective complements to
have been refined with extensive textual data, enabling them to discern collaborative writing practices.
language patterns and emulate human-like writing. Unlike conventional To enhance L2 writing curricula, it is vital to not only consider the
AWE systems, LLMs engage users with a natural language interface, theoretical underpinnings, such as sociocultural theory which empha­
simplifying the feedback generation process (Kasneci et al., 2023). sizes the significance of learning through social interactions and cultural
This study investigates the intersection of collaborative writing and tools (Vygotsky, 1978), but also to practically apply these theories by
AI feedback, aiming to demonstrate how their integration can substan­ integrating AI into educational frameworks. Such an approach could
tially benefit L2 learning. This synergy is expected to significantly significantly revolutionize the facilitation and assessment of collabora­
improve the writing skills of L2 students by offering consistent and tive writing, cementing the role of AI as a transformative force in
personalized support. While AI feedback promises to bolster the revision educational practices.
process with its immediacy and adaptability, empirical validation of its
efficacy, particularly in wiki-based collaborative settings, is lacking. 2.2. AI feedback and revision in L2 collaborative writing
Through a quasi-experimental design, this research assesses the influ­
ence of a structured collaborative writing intervention, assisted by AI While traditional feedback sources such as teachers and peers have
feedback, on the L2 writing proficiency of Thai-major exchange stu­ been extensively studied and recognized for their effectiveness in L2
dents, illuminating the potential of these emerging tools to redefine the writing instruction (Alharbi, 2022; Guasch et al., 2013; Zhao et al.,
landscape of L2 writing instruction. 2023), the integration of AI feedback signifies a substantial paradigm
shift. This shift aligns with Vygotsky’s sociocultural theory, which em­
2. Literature review phasizes the critical role of social interaction and tools in learning. Ac­
cording to this theory, cognitive functions are developed from social
The theoretical and empirical foundations of collaborative writing interactions and mediated by cultural artifacts, including advanced
and AI feedback within L2 learning environments are critically exam­ technological tools (Vygotsky, 1978). AI applications, serving poten­
ined, elucidating the historical context of collaborative writing in edu­ tially as virtual tutors, observe students’ learning processes, analyzes
cation, the evolution of AI feedback mechanisms, and their impact on their learning performance, and provides instant, tailored assistance
writing proficiency. This review pays special attention to the integration based on individual needs (Hwang et al., 2020).
of these elements in various educational settings, emphasizing their AI models like ChatGPT are increasingly utilized for their ability to
application in non-dominant languages such as Thai. deliver immediate and targeted corrective feedback, which is particu­
larly crucial during the revision stages of writing (Zheng et al., 2023).
2.1. Collaborative writing in wiki-based environments This feedback is not merely corrective but also scaffolds the learner’s
development within their zone of proximal development, a key concept
Collaborative writing has emerged as a recognized and valuable in sociocultural theory. This zone describes the range of tasks that a
pedagogical tool, fostering interactive learning experiences and skill learner can perform with the guidance of a more knowledgeable other
development in L2 learners (Abrams, 2019; Dobao, 2012; Hsu, 2019; but cannot yet perform independently.
Pham, 2021; Wigglesworth & Storch, 2012). Within wiki-based envi­ Research highlights that technology-driven feedback plays a crucial
ronments, learners actively participate in the dynamic co-construction role in the revision stage, assisting students in addressing specific areas
of written content. This involves exchanging ideas, providing feed­ for improvement and enhancing the overall quality of their work (Li
back, and collectively contributing to the development of their writing et al., 2015; Meyer et al., 2024). The potential of emerging generative AI
skills. Such environments offer several benefits, not only enhancing the tools, such as ChatGPT, lies in their capacity to offer timely, targeted,
overall quality of writing in subsequent assessments (Alghasab & Han­ adaptive, and directly applicable feedback—qualities that are indis­
dley, 2017; Aydın & Boğaziçi, 2014; Hsu & Lo, 2018) but also encour­ pensable for promoting significant improvements in student writing
aging active participation and facilitating the application of knowledge (Biber et al., 2011; Graham et al., 2015). Given the balance required
across various writing tasks. between feedback quality and timeliness, ChatGPT has emerged as a
The wiki platform serves as a shared space for collaborative writing, particularly promising evaluative tool that can effectively streamline the
allowing learners to annotate essays collaboratively, identify areas for revision process (Steiss et al., 2024).
improvement, and refine their writing together (Hosseini et al., 2020). The significance of AI feedback in enhancing student writing be­
For instance, Wang (2015) conducted a study where students were comes even more apparent with increased frequency, consistently
divided into wiki and non-wiki groups. Both groups, composed of four leading to marked improvements (Meyer et al., 2024). This trans­
students each, collaborated on drafting, peer-editing, and revising two formative potential is further bolstered by research highlighting the
written assignments over 12 weeks. Pre-tests and post-tests assessed the positive impact of frequent AI feedback on individual writing tasks,
impact of wiki collaborative work on students’ individual business suggesting that regular and iterative feedback can substantially improve
writing skills. While both groups improved, the wiki group demon­ writing outcomes and assist students in identifying their strengths and
strated a more significant enhancement. This consistency in findings weaknesses and enhancing their proficiency (Annamalai et al., 2023;
underscores the positive influence of collaborative writing, particularly Giessler, 2023; Han & Sari, 2022; Link et al., 2022; Rad et al., 2023;
within wiki environments, on individual writing proficiency. Zhang & Hyland, 2023). For example, Rad et al. (2023) demonstrated
Despite these insights, the literature has not adequately addressed that the experimental group using an AI-based application significantly
the implications of the frequency with which collaborative writing tasks improved writing outcomes, engagement, and feedback literacy
are assigned on L2 proficiency. This represents a discernible gap in the compared to the control group, facilitated by the AI fostering a dialogic
research, particularly in determining the optimal frequency of such tasks interaction that allowed for personalized learning paces.

2
W. Wiboolyasarin et al. Computers and Education: Artificial Intelligence 6 (2024) 100228

Semi-automatic feedback software in AI environments can provide demographic and experiential variables. Participants who did not meet
learning analytics data, enabling instructors to create contextually the criteria of having at least three years of prior Thai language study or
appropriate feedback that enhances student engagement (Ma et al., those who had not been in Thailand for a minimum of seven months
2021; Winstone & Carless, 2020). were excluded from the study. This criterion was established to ensure
Despite these advancements, the specific contributions of AI feed­ that all participants had a sufficient foundational knowledge of Thai and
back to collaborative writing tasks are not well-documented, repre­ adequate exposure to the language environment, which are critical for
senting a significant research gap. Moreover, while much of the existing assessing the impacts of the intervention on L2 writing proficiency.
research on AI feedback focuses on dominant languages like English, To foster a student-centered writing activity, participants organized
there is a noticeable lack of studies concerning non-dominant languages themselves into small groups of three, with each group designating a
such as Thai, especially in contexts where Thai is learned as a foreign leader and collaboratively establishing a wiki for their project. The
language. Our study aims to bridge this gap by exploring the role of AI rationale behind this grouping strategy stemmed from the belief that
feedback in enhancing Thai language proficiency among L2 learners, smaller group sizes would cultivate a more noticeable and equitable
contributing valuable insights to an under-explored area of language distribution of workload and accountability among members, thus
education research. This initiative not only addresses a significant void enhancing the collaborative writing process. Before the study
in the current literature but also highlights the importance of diversity in commenced, the researcher/instructor, with prior experience in con­
language research and the broad applicability of technological in­ ducting online writing courses, introduced a novel approach, given that
novations across languages for a wider educational impact. the students had not encountered online writing courses previously.
To assess the improvement in Thai language proficiency among the
3. Research questions individual participating L2 learners, both pre- and post-writing tests
were administered. These tests, strategically designed, aimed to eval­
Building on a robust theoretical framework, this study seeks to uate the participants’ competence in producing Thai essays on familiar
explore the synergy between collaborative writing practices and AI topics. The pre-test served as a baseline assessment, capturing the initial
feedback. This integration is predicated on the notion that collaborative state of their writing skills and determining whether their writing
writing with AI- driven feedback can significantly improve the writing competencies differed. The post-test was then implemented to gauge the
proficiency of learners by providing relevant written corrective feedback impact of the collaborative writing and AI feedback intervention on
(WCF) that learners can incorporate during the writing process. The their overall language proficiency. This methodological approach not
pedagogical theory supporting this research posits that such an only aligns with the study’s objectives but also provides a robust
approach not only increases learners’ engagement and motivation but framework for evaluating the specific contributions of collaborative
also provides a scaffold that aids in the development of writing skills in a writing and AI feedback in an L2 learning environment.
supportive, iterative learning environment.
Given this theoretical backdrop, the current study investigates the 4.2. Generative AIs and prompts
following research questions.
ChatGPT, an advanced language model developed by OpenAI, served
RQ1: To what extent does the collaborative writing intervention, as a valuable resource for delivering WCF to students. To ensure effec­
complemented by AI feedback, contribute to enhancements in tive implementation, the training process for ChatGPT involved detailed
learners’ Thai writing proficiency? familiarization with the specific rubric criteria used to evaluate Thai
RQ2: Does the frequency of creating collaborative writing assign­ language writing. This preparation was critical for aligning the AI’s
ments by L2 learners positively correlate with improved writing feedback with academic standards specific to L2 writing assessments.
ability? The core of the training phase was the development of a key Thai
prompt: “ประเมินความเรียงต่อไปนี้ โดยใช้เกณฑ์รูบริกส
4. Methods ประเมิน การเขียนภาษาที่ สองของนัก ศึกษาต่างชาติ” (literally translated as
์
“Evaluate the following essay using rubrics to assess second language
This study adopted a quasi-experimental design with a nonequiva­ writing of international students”). This prompt directed ChatGPT to
lent pretest-posttest control group (Creswell & Creswell, 2022) to apply the established criteria to the students’ essays, facilitating the
investigate the impact of a three-step collaborative writing activity with delivery of precise and contextually appropriate feedback that could
and without AI-powered feedback on the L2 writing proficiency of guide improvements effectively.
Thai-major undergraduate exchange students. The choice to utilize ChatGPT over other AI models was substanti­
ated by a series of comparative tests that demonstrated ChatGPT’s su­
4.1. Participants and context perior capability in handling the linguistic intricacies of Thai. An
alternative AI model was tested using the same prompts but consistently
In this study, 39 junior exchange students, consisting of 32 females assigned the same scores to varying pieces of writing, suggesting a lack
and 7 males aged 18–20, participated as enrolled members of Thai of sensitivity to the qualitative differences in the submissions. This was
language courses at a public university in Thailand. They were divided attributed to the model’s limited understanding of Thai, a language
into two groups: 18 participants in the experimental group and 21 characterized by unique grammatical and syntactic challenges.
participants in the control group. With an average of three years of In contrast, ChatGPT exhibited a nuanced comprehension of the
experience in learning the Thai language in their hometowns and a evaluation criteria, enabling it to offer personalized feedback that was
minimum of seven months studying in Thailand, these students brought both specific and actionable. This capacity to discern and articulate
a unique combination of prior foundational knowledge and immersive detailed areas for improvement confirmed the suitability of ChatGPT for
exposure to the language in the country, distinguishing them from those this educational context, ensuring that the feedback provided was not
with less direct exposure or different educational contexts. To approxi­ only accurate but also pedagogically valuable.
mate equivalence between the experimental and control groups, par­ This rigorous evaluation and selection process underscored our
ticipants were matched based on criteria including prior exposure to commitment to transparency and methodological rigor in implementing
Thai, duration of their study in Thailand, age, and initial language AI technology. By carefully calibrating ChatGPT to meet the specific
proficiency levels. Stratified random sampling was employed to assign needs of Thai L2 writing assessment, the study ensured that the AI
students to either the experimental group (18 participants) or the con­ feedback was effectively integrated into the educational framework,
trol group (21 participants), ensuring a balanced distribution across key thereby enhancing the overall integrity and impact of the intervention.

3
W. Wiboolyasarin et al. Computers and Education: Artificial Intelligence 6 (2024) 100228

4.3. Essays and evaluation criteria platform. A detailed comparison of the timeline for activities undertaken
by the experimental and control groups is elucidated in Table 1.
This study was designed to examine the progression of L2 writing Participants were required to complete three L2 writing tasks within
proficiency through both individual and collaborative writing tasks, their groups, each following a structured three-step process conducted
facilitated by the use of modern educational technologies. Participants weekly: collaborative outlining, individual writing with AI feedback,
were tasked with composing five essays over the course of the study and collaborative writing. The detailed description of each step is pre­
(details in Appendix A). Two of these essays functioned as both indi­ sented below.
vidual writing pretests and posttests, requiring participants to craft a
300-word essay on a predetermined topic. These tasks were designed to Step 1 - Collaborative Outlining: In this initial phase, students
measure the development of writing skills before and after engaging in worked in grouped to outline essays on assigned topics. This
the three-step collaborative writing activity. Each participant was collaborative planning phase was crucial for developing a structured
assigned a unique QR code facilitating access to their designated writing approach to writing, encouraging participants to jointly explore and
space on Google Docs, with each writing session constrained to a 90-min agree on the title, main opinion, supporting reasons, and evidence.
time limit to maintain consistency. This activity was designed to foster critical thinking, independence,
The other three essays were integral to the three-step collaborative and autonomy in learning, as noted by Lin (2019). Collaborative
writing activity, which was conducted in small groups of three. Each outlines were documented on a designated wiki page, enhancing
group was provided with a QR code linking to a dedicated wiki via transparency and collective responsibility.
Google Docs, which detailed each step of the writing process and offered Step 2 - Individual Writing and AI Feedback: Following the outlining
guidance to enhance the effectiveness of their joint efforts. phase, students independently drafted essays based on their group’s
The evaluation of all essays was conducted using rubrics developed plans. The experimental group received personalized WCF from
from an extensive review of literature (Becker, 2016; Butvilofsky & ChatGPT, which assessed their essays against established criteria and
Sparrow, 2012; Cheung, 2023; Rakedzon & Baram-Tsabari, 2017; Sol­ prompts mentioned earlier (Fig. 1). This feedback was intended to
tero-González et al., 2012; Uludag & McDonough, 2022) and designed to deepen their understanding of effective writing practices and inform
assess various aspects of writing quality, including content, structure, their revisions. In contrast, the control group wrote their essays
utilization of examples and evidence to support main points, language without AI feedback and did not engage in post-feedback revisions,
use, word count, and spelling. Each of the six evaluation categories was underscoring the study’s focus on the impact of AI feedback.
scored on a scale of 4 points, allowing for a maximum total of 24 points. Step 3 - Collaborative Writing: In this final phase, team members
To ensure consistency and objectivity in the evaluation process, the collaboratively refined their essays by integrating individual con­
rubrics were applied by a panel of three experienced educators who tributions into a comprehensive final piece. This step emphasized
were familiar with the educational objectives of the study and trained in appropriate content arrangement and organization over grammat­
the application of the rubrics. This panel reviewed the essays indepen­ ical precision. Each participant was responsible for enhancing the
dently to mitigate potential biases and ensure that the grading reflected clarity and effectiveness of the group’s collective essay, focusing on
the learning objectives accurately. Discrepancies in scoring were dis­ vocabulary, grammar, and mechanics.
cussed among the evaluators to reach a consensus, thus aligning the
grading process closely with the intended educational outcomes of the Group dynamics and individual contributions were closely moni­
study. The detailed rubrics and scoring methodology are provided in tored through the recording and reviewing of each group’s interactions
Appendix B, offering transparency and allowing for replication and to ensure active participation and equitable involvement. The instructor
validation of the study’s findings. facilitated the process and maintained adherence to the collaborative
framework without directly influencing the content or direction of
feedback. Guidelines and rubrics were consistently applied across both
4.4. Three-step collaborative writing groups to prevent instructor bias and uphold methodological integrity.
This structured three-step process was repeated across two addi­
This study implemented a structured three-step collaborative writing tional writing tasks, culminating in a posttest designed to assess indi­
activity over a 10-week period, designed to foster holistic writing among vidual L2 writing performance.
students. Drawing inspiration from Liu et al. (2022), this process inte­
grated AI feedback, revision, and collaborative learning through a wiki
4.5. Data analysis

Table 1 Recognizing the limited size of our sample, we conducted an

Timeline of activities in three-step collaborative writing process. assessment of the normality of data distribution using IBM SPSS Statis­
Week Activity tics version 27. Table 2 presents the outcomes of normality tests,
Experimental group Control group employing Kolmogorov-Smirnov and Shapiro-Wilk statistics.
In the experimental group, the pretest data deviated significantly
1 Pretest Draft outline of task Pretest Draft outline of
1 task 1 from normality, evident in both the Kolmogorov-Smirnov test (p =
2 Individual writing of AI feedback Individual writing of task 1 0.042) and the Shapiro-Wilk test (p = 0.030). Conversely, the posttest
task 1 intervention data demonstrated no significant departure according to the
3 Collaborative writing of task 1 Collaborative writing of Kolmogorov-Smirnov test (p = 0.200*) and the Shapiro-Wilk test (p =
task 1
4 Draft outline of task 2 Draft outline of task 2
0.629). These findings suggest that the posttest data in the experimental
5 Individual writing of AI feedback Individual writing of task 2 group may approximate a normal distribution.
task 2 intervention In contrast, the control group exhibited departures from normality in
6 Collaborative writing of task 2 Collaborative writing of both pretest and posttest data. The pretest data showed significant de­
task 2
partures, as indicated by the Kolmogorov-Smirnov test (p = 0.010) and
7 Draft outline of task 3 Draft outline of task 3
8 Individual writing of AI feedback Individual writing of task 3 the Shapiro-Wilk test (p = 0.083). For the posttest, while the
task 3 intervention Kolmogorov-Smirnov test suggested a lower significance bound of
9 Collaborative writing of task 3 Collaborative writing of 0.081, the Shapiro-Wilk test did not indicate a significant departure (p =
task 3 0.334).
10 Posttest Posttest
Given these findings, we carefully considered the implications for

4
W. Wiboolyasarin et al. Computers and Education: Artificial Intelligence 6 (2024) 100228

Fig. 1. Illustration of AI feedback for individual writing using ChatGPT (see translation in Appendix C.).

Table 2
Assessment of data normality for pretest and posttest between the experimental and control groups.
Group M S.D. Kolmogorov-Smirnovb Shapiro-Wilk

Statistic df Sig. Statistic df Sig.

Experimental pretest 16.94 2.73 0.206 18 0.042 0.883 18 0.030

posttest 19.89 2.40 0.122 18 0.200a 0.961 18 0.629
Control pretest 16.71 2.17 0.219 21 0.010 0.919 21 0.083
posttest 17.05 1.91 0.178 21 0.081 0.950 21 0.334
a
This is a lower bound of the true significance.
b
Lilliefors Significance Correction.

our choice of statistical tests. The presence of significant non-normality required for parametric testing.
in the control group data, especially in Tasks 2 and 3 as detailed in Table 3 provides further insights into the normality assessment
Table 3, led us to select non-parametric tests for our analyses. Non- across the three tasks within each group, using Kolmogorov-Smirnov
parametric tests are less sensitive to the effects of non-normal distribu­ and Shapiro-Wilk statistics. While the experimental group largely
tions and are appropriate when data do not meet the assumptions maintained normality across all tasks, the control group exhibited

5
W. Wiboolyasarin et al. Computers and Education: Artificial Intelligence 6 (2024) 100228

Table 3
Assessment of data normality for three tasks between the experimental and control groups.
Group M S.D. Kolmogorov-Smirnovb Shapiro-Wilk

Statistic df Sig. Statistic df Sig.

Experimental Task 1 19.42 2.65 0.152 6 0.200a 0.973 6 0.910

Task 2 21.17 2.71 0.185 6 0.200a 0.913 6 0.459
Task 3 21.92 1.69 0.236 6 0.200a 0.930 6 0.583
Control Task 1 18.86 1.77 0.257 7 0.179 0.843 7 0.106
Task 2 19.29 1.70 0.346 7 0.011 0.779 7 0.025
Task 3 19.86 1.77 0.325 7 0.024 0.866 7 0.172
a
This is a lower bound of the true significance.
b
Lilliefors Significance Correction.

notable deviations, particularly in Tasks 2 and 3, which reinforced our ranks between the experimental and control groups across the three
decision to use non-parametric tests. This methodological choice ensures writing tasks. The results, summarized in Table 6, revealed a significant
the robustness and validity of our findings, accommodating the non- difference in median ranks for the experimental group across the tasks
normality observed in several key data sets. (χ2 = 9.238, df = 2, p = 0.010). Conversely, the control group exhibited
By opting for non-parametric tests, we aim to provide a more reliable no statistically significant differences (χ2 = 3.100, df = 2, p = 0.212).
analysis that accurately reflects the underlying patterns in our data, These findings suggest that the experimental group’s performance var­
despite the challenges posed by deviations from normality. This ied significantly across the tasks, while the control group’s performance
approach underscores our commitment to methodological integrity and remained more consistent.
the careful interpretation of statistical results.
6. Discussion
5. Results
The present study, to the best of our knowledge, represents a pio­
To assess the difference in writing performance between the pretest neering effort to offer empirical evidence supporting the enhancement
and posttest, a Wilcoxon Signed-Rank test was employed. The results, of Thai writing performance in L2 undergraduate students through the
detailed in Table 4, revealed a significant difference in writing scores for synergistic utilization of collaborative writing and AI feedback. Building
L2 learners in the experimental group (z = 2.903, p = 0.004*) and found upon prior literature that underscores the general effectiveness of WCF
no significant difference in the control group (z = 1.078, p = 0.281). The documented in prior literature (Benson & DeKeyser, 2018; Cho et al.,
sum of the difference scores and the mean rank indicates that this dif­ 2022; Jabulani, 2015; Kim & Emeliyanova, 2019) and its positive impact
ference favors the posttest results. The calculated eta squared value of on AWE outcomes (Barrot, 2023; Li et al., 2015; Link et al., 2022;
0.468 for the experimental group indicates a large effect size, while the Ranalli, 2018; Shang, 2022; Waer, 2023), our study specifically ad­
value of 0.055 for the control group suggests that only a small propor­ dresses a notable gap in empirical evidence regarding the effectiveness
tion of the variance in the posttest scores can be attributed to factors of AI-generated feedback in wiki-based collaborative writing environ­
other than the intervention. Nevertheless, the experimental group’s ments. This research distinctively contributes by exploring AI’s potential
larger eta squared value implies a more substantial impact of the to enhance Thai language proficiency—a less commonly studied,
collaborative writing activity and AI feedback on the posttest scores non-dominant language—and providing a tailored mechanism for
compared to the control group. generating personalized feedback for L2 students.
Our analysis using the Mann-Whitney U test (Table 5) underscores Our findings indicate that AI-generated WCF effectively aids students
the statistical significance of the mean rank differences between the in the revision process within wiki-based collaborative writing envi­
experimental and control groups in both pretest and posttest scores. The ronments, stimulating beneficial affective-motivational processes
findings reveal that, prior to any intervention, the groups exhibited crucial for sustaining motivation (Cen & Zheng, 2024; Meyer et al.,
similar levels of performance, as evidenced by the non-significant dif­ 2024; Ruwe & Mayweg-Paus, 2023). Integrating AI feedback into
ference in mean ranks for the pretest scores (Z = 0.128, p = 0.898). collaborative writing practices presents a novel approach to addressing
However, following the collaborative writing intervention, a note­ the unique challenges associated with L2 writing proficiency. Notably,
worthy difference emerged in posttest scores, with the experimental both the experimental and control groups exhibited improvements in
group outperforming the control group (Z = 2.987, p = 0.003*). This writing scores post-intervention; however, students in the experimental
significant effect indicates that the collaborative writing activity, group benefited significantly more from the AI-generated feedback. This
potentially supplemented by AI feedback, had a positive impact on the outcome aligns with research suggesting that technology, when
writing proficiency of the experimental group compared to the control providing personalized feedback tailored to specific needs, can sub­
group. stantially enhance students’ writing during revision processes (Li et al.,
The Friedman test was employed to analyze differences in median 2015; Meyer et al., 2024; Ruwe & Mayweg-Paus, 2023; Steiss et al.,

Table 4
Wilcoxon test results for the differences between pretest and posttest scores.
Group Ranks N Mean rank Sum of ranks z Sig. η2
Experimental posttest - pretest Negative Ranks 4 4.75 19.00 2.903 0.004* 0.468
Positive Ranks 14 10.86 152.00
Equal 0
Total 18
Control posttest - pretest Negative Ranks 4 12.00 48.00 1.078 0.281 0.055
Positive Ranks 12 7.33 88.00
Equal 5
Total 21

*Significant effect of group with p < 0.05.

6
W. Wiboolyasarin et al. Computers and Education: Artificial Intelligence 6 (2024) 100228

Table 5
Mann-Whitney U test results for the differences of pretest and posttest scores between groups.
Test Group Mean rank Sum of ranks Mann-Whitney U Wilcoxon W Z Sig.

Pretest Experimental 20.25 364.50 184.500 415.500 0.128 0.898

Control 19.79 415.50
Posttest Experimental 25.83 465.00 84.000 315.000 2.987 0.003a
Control 15.00 315.00
a
Significant effect of group with p < .05.

likely contributed to the cumulative improvement in writing proficiency

Table 6
over time, affirming the theory that repeated and reflective engagement
Friedman test results for three writing tasks between experimental and control
with content deepens learning (Vygotsky, 1978). Moreover, AI feedback
groups.
has the potential to revolutionize writing assessment, providing imme­
Rank Mean rank diate scoring and augmented feedback on writing quality aligned with
Experimental Control specific criteria. This method can effectively reduce the workload of
Task 1 1.08 1.57 teachers in correcting student writing, mitigate evaluation bias, and
Task 2 2.25 2.07 allow teachers to concentrate on other crucial aspects of writing in­
Task 3 2.67 2.36 struction such as overall structure, coherence, content, and writing
Test Statistics
strategies (Mizumoto & Eguchi, 2023).
N 6 7
Chi-Square 9.238 3.100
Additionally, the integration of AI-generated feedback and collabo­
df 2 2 rative group annotations enhanced the depth and precision of the revi­
Asymp. Sig. 0.010 0.212 sion process. As highlighted by Li (2024), when working in small groups,
a. Friedman Test students could collaboratively annotate the essay on a shared document
such as Google Docs. This involved identifying instances highlighted by
2024). The enhanced performance of the experimental group could be ChatGPT in which the essay could be more specific, characterized by
attributed to higher engagement levels and their prior exposure to comprising multiple components. This collaborative annotation process
technology in learning, which may have facilitated a more receptive facilitated a deeper exploration of nuanced language use and improved
attitude towards AI feedback. Such factors are essential in understand­ the specificity of their essays, aligning with the principles of collabora­
ing why technology-mediated interventions might yield different results tive knowledge building (Li et al., 2021). The integration of
across student groups. This insight is supported by Vygotsky’s socio­ AI-generated feedback and collaborative group annotations created a
cultural theory, which emphasizes the role of tools in mediating learning symbiotic relationship, enhancing the depth and precision of the revi­
and suggests that social interactions within groups can significantly sion process.
impact cognitive development. The improvements in the writing scores post-intervention was sta­
By offering empirical evidence on the effectiveness of collaborative tistically significant compared to the pre-intervention performance in
writing and AI-generated feedback, we contribute distinctively to the the experimental group. This outcome is attributed to the collaborative
literature, often centered on human feedback or limited to English and writing activities on the wiki platform, where students engaged inter­
dominant languages (Hafner & Yu, 2020; Meyer et al., 2024; Mizumoto actively in co-construction of text by exchanging ideas and providing
& Eguchi, 2023; Peng et al., 2023; Ruwe & Mayweg-Paus, 2023; Shang, comments on the content, organization, and language use of their
2022; Tajabadi et al., 2023; Yu, 2021). Significantly, our focus extends writing tasks, thereby contributing to the development of L2 writing
to the potential of AIs in enhancing Thai language proficiency, a skills (Elola & Oskoz, 2010; Hsu, 2019). The knowledge generated
non-dominant language, through a tailored and efficient personalized within this collaborative environment was evidently applied to various
feedback mechanism for L2 students. This synthesis of established writing tasks, leading to enhanced content quality in the subsequent
collaborative writing practices with emerging AI capabilities embodies writing assessments (Hsu & Lo, 2018), especially when combining AI
an innovative approach to L2 writing challenges. feedback (Zheng et al., 2023). This positive impact is underscored by the
While analyzing students writing performance, we observed that fact that the AI-enabled feedback provided instant feedback that proved
both groups exhibited higher posttest scores compared to the pretest. useful and effective for each group. This, in turn, promoted a deep,
However, the posttest scores between the groups were significantly collaborative construction of writing knowledge among the groups.
different. Particularly, L2 students in the experimental group gained Nevertheless, this study has limitations that warrant caution in
additional benefits from receiving AI-generated feedback in wiki-based generalizing the results. The relatively small sample size, though
collaborative writing environments. This aligns with the suggestion by adequate for initial explorations, may limit the robustness and broader
scholars (Li et al., 2015; Meyer et al., 2024; Ruwe & Mayweg-Paus, applicability of our findings, especially given the study’s specific focus
2023; Steiss et al., 2024) that technology, when delivering personal­ on Thai-major undergraduate students. Additionally, the AI-generated
ized feedback with specific improvement suggestions tailored to the feedback and the assessment rubrics themselves could have inherent
work’s qualities, can significantly enhance students’ writing during the biases, reflecting the training data and the subjective interpretations of
revision processes in Step 2. As Meyer et al. (2024) argue, this what constitutes effective writing. Future research should aim to include
enhancement is particularly pronounced when feedback is delivered larger and more diverse participant samples and explore the impacts of
more frequently. AI feedback within different language learning environments to enhance
Throughout the study, we facilitated three rounds of a structured the robustness and reliability of the findings. One of the key benefits of a
three-step collaborative writing process, offering multiple opportunities longitudinal approach would be the ability to observe the development
for L2 students to receive feedback and revise their tasks based on the of writing skills beyond the initial learning phase. It would allow re­
provided feedback. We can therefore assume that the beneficial effects searchers to see if improvements are merely short-lived boosts or if they
of feedback might unfold during subsequent learning opportunities translate into sustained enhancements in writing proficiency.
(Fleckenstein et al., 2023). This approach aligns with the idea that the
cumulative impact of iterative feedback processes can lead to substantial
improvements in writing proficiency over time. This iterative process

7
W. Wiboolyasarin et al. Computers and Education: Artificial Intelligence 6 (2024) 100228

7. Implications collaborative writing activity serves as a blueprint for educators

seeking to implement similar interventions in diverse educational set­
This study’s examination of AI feedback in L2 writing instruction tings. This model promotes active engagement and continuous
offers significant practical implications for educators, demonstrating improvement among students, characteristics that are essential for the
how AI-driven feedback mechanisms can be integrated into language development of proficient language skills.
courses to enhance student writing capabilities. Assignments specifically Furthermore, the integration of AI not only supports the individual
designed to utilize AI feedback, such as online essay writing or responses writing process but also enhances collaborative learning by providing
to prompts, can significantly augment the revision process (Escalante immediate, personalized, and actionable feedback. This approach allows
et al., 2023; Pham, 2021). Here, AI tools provide immediate, actionable students to critically engage with their work and iteratively refine their
guidance that helps students refine their drafts before final submission. skills, which is a cornerstone of successful language learning. The
AI tools are intended to supplement rather than replace traditional deployment of AI as a feedback mechanism also offers the potential to
teaching methods. For example, AI feedback can serve as an initial alleviate some of the instructional burdens on educators, enabling them
screening tool, allowing students to correct basic errors independently to devote more time to addressing higher-order writing complexities
before submitting their work for teacher review (e.g., Rad et al., 2023). such as argument structure, coherence, and content depth.
This strategy effectively frees up instructor time, enabling a focus on
advanced aspects of writing such as structure, argumentation, and style. CRediT authorship contribution statement
Additionally, AI feedback offers personalized support (Seo et al., 2021),
adjusting to individual student needs, which is particularly beneficial in Watcharapol Wiboolyasarin: Writing – original draft, Validation,
diverse classrooms with students at different proficiency levels. Methodology, Formal analysis, Conceptualization. Kanokpan Wiboo­
One of the key advantages of AI feedback is its immediacy (Chien lyasarin: Writing – review & editing, Conceptualization. Kanpabhat
et al., 2024; Wang et al., 2024). Tools like ChatGPT provide instant Suwanwihok: Investigation. Nattawut Jinowat: Writing – review &
feedback, essential in educational settings where timely responses can editing, Validation. Renu Muenjanchoey: Writing – review & editing,
significantly impact learning outcomes. Immediate feedback facilitates Validation, Investigation.
real-time learning, enabling students to understand and correct their
errors promptly, which is critical for effective learning retention and Declaration of competing interest
adjustment of behaviors.
AI’s ability to provide feedback tailored to the individual needs of The authors declare that they have no known competing financial
students enhances personalization within the learning process (Bhu­ interests or personal relationships that could have appeared to influence
toria, 2022). By analyzing specific student inputs, AI systems cater to the work reported in this paper.
unique linguistic and stylistic needs, ensuring that feedback is not only
general but also focused on personal areas of weakness and strength. Appendix A. Supplementary data
This targeted approach promotes a balanced and individualized learning
experience, encouraging students to refine specific skills while rein­ Supplementary data to this article can be found online at https://doi.
forcing their overall writing proficiency. org/10.1016/j.caeai.2024.100228.
However, there are notable risks associated with the reliance on AI in
educational settings. Over-dependence on AI tools might lead to a References
diminished appreciation for human judgment and the nuanced insights
that experienced educators offer (Fortino, 2023). Such reliance could Abrams, Z. I. (2019). Collaborative writing and text quality in Google Docs. Language,
Learning and Technology, 23(2), 22–42, 10125/44681.
undermine the personalized teaching and human interaction crucial to
Alghasab, M., & Handley, Z. (2017). Capturing (non-) collaboration in wiki-mediated
education, potentially leading to superficial engagement with the collaborative writing activities: The need to examine discussion posts and editing
learning material. acts in tandem. Computer Assisted Language Learning, 30(7), 664–691. https://doi.
org/10.1080/09588221.2017.1341928
Another concern involves users’ perceptions of AI feedback. Some
Alharbi, M. A. (2022). Exploring the impact of teacher feedback modes and features on
may view feedback from a non-human source as less credible or moti­ students’ text revisions in writing. Assessing Writing, 52, 1–12. https://doi.org/
vating than that from human instructors (Nargesi, 2021; Ruwe & 10.1016/j.asw.2022.100610
Mayweg-Paus, 2023), which could negatively impact their engagement Alwaleedi, M. A., Gillies, R. M., & Hamid, M. O. (2019). Collaborative writing in Arabic
as a second language (ASL) classrooms: A mixed-method study. Language Culture and
and the effectiveness of the feedback. Moreover, AI-generated feedback, Curriculum, 32(2), 157–172. https://doi.org/10.1080/07908318.2018.1521422
while clear and direct, might lack the emotional support and encour­ Annamalai, N., Eltahir, M. E., Zyoud, S. H., Doundrarajan, D., Zakarneh, B., & Al
agement that human feedback often provides, which are vital for stu­ Salhi, N. R. (2023). Exploring English language learning via chabot: A case study
from a self determination theory perspective. Computers & Education: Artificial
dents facing the challenges of mastering a new language. These factors Intelligence, 5, 1–8. https://doi.org/10.1016/j.caeai.2023.100148
highlight the need for careful integration of AI tools into the educational Aydın, Z., & Boğaziçi, S. (2014). Use of wikis to promote collaborative EFL writing.
process, ensuring they complement rather than replace the critical Language, Learning and Technology, 18(1), 160–180, 10125/44359.
Barrot, J. S. (2023). Using automated written corrective feedback in the writing
human elements of teaching and learning. classrooms: Effects on L2 writing accuracy. Computer Assisted Language Learning, 36
(4), 584–607. https://doi.org/10.1080/09588221.2021.1936071
8. Conclusion Becker, A. (2016). Student-generated scoring rubrics: Examining their formative value
for improving ESL students’ writing performance. Assessing Writing, 29, 15–24.
https://doi.org/10.1016/j.asw.2016.05.002
Our study contributes to the existing literature by providing empir­ Beiki, M., Gharagozloo, N., & Raissi, R. (2020). The effect of structured versus
ical evidence on the potential of AI as a WCF provider in wiki-based unstructured collaborative pre-writing task on writing skills of the Iranian EFL
students. Asian-Pacific Journal of Second and Foreign Language Education, 5(18), 1–29.
collaborative writing environments. The significant improvements in https://doi.org/10.1186/s40862-020-00092-0
writing scores observed in the experimental group underscore the Benson, S., & DeKeyser, R. (2018). Effects of written corrective feedback and language
promising potential of integrating advanced large language models with aptitude on verb tense accuracy. Language Teaching Research, 23(6), 702–726.
https://doi.org/10.1177/1362168818770921
collaborative writing practices. This research lays the groundwork for
Bhutoria, A. (2022). Personalized education and artificial intelligence in the United
future initiatives focused on refining and expanding such interventions, States, China, and India: A systematic review using a human-in-the-loop model.
ultimately contributing to the evolution of L2 learning practices. These Computers & Education: Artificial Intelligence, 3, 1–18. https://doi.org/10.1016/j.
findings can be leveraged by educators and researchers to inform the caeai.2022.100068
Biber, D., Nekrasova, T., & Horn, B. (2011). The effectiveness of feedback for L1-English
design of language courses that capitalize on the power of collaborative and L2-writing development: A meta-analysis. ETS Research Report Series, 2011(1),
writing, revision, and AI technologies. The structured three-step i–99. https://doi.org/10.1002/j.2333-8504.2011.tb02241.x

8
W. Wiboolyasarin et al. Computers and Education: Artificial Intelligence 6 (2024) 100228

Butvilofsky, S. A., & Sparrow, W. L. (2012). Training teachers to evaluate emerging Language Teaching Research, 25(2), 234–255. https://doi.org/10.1177/
bilingual students’ biliterate writing. Language and Education, 26(5), 383–403. 1362168819831406
https://doi.org/10.1080/09500782.2011.651143 Li, R. (2024). A “Dance of storytelling”: Dissonances between substance and style in
Cen, Y., & Zheng, Y. (2024). The motivational aspect of feedback: A meta-analysis on the collaborative storytelling with AI. Computers and Composition, 71, 1–10. https://doi.
effect of different feedback practices on L2 learners’ writing motivation. Assessing org/10.1016/j.compcom.2024.102825
Writing, 59, 1–17. https://doi.org/10.1016/j.asw.2023.100802 Li, Y., Li, X., Zhang, Y., & Li, X. (2021). The effects of a group awareness tool on
Chao, Y.-C. J., & Lo, H.-C. (2011). Students’ perceptions of Wiki-based collaborative knowledge construction in computer-supported collaborative learning. British
writing for learners of English as a foreign language. Interactive Learning Journal of Educational Technology, 52(3), 1178–1196. https://doi.org/10.1111/
Environments, 19(4), 395–411. https://doi.org/10.1080/10494820903298662 bjet.13066
Chen, W., & Ren, W. (2022). Educating L2 learners about collaborative writing: Li, J., Link, S., & Hegelheimer, V. (2015). Rethinking the role of automated writing
Exploring the relationship between knowledge of collaborative writing and writing evaluation (AWE) feedback in ESL writing instruction. Journal of Second Language
products. Language Awareness, 31(3), 371–391. https://doi.org/10.1080/ Writing, 27, 1–18. https://doi.org/10.1016/j.jslw.2014.10.004
09658416.2021.1969403 Lin, C.-J. (2019). An online peer assessment approach to supporting mind-mapping
Chen, X., Xie, H., Zou, D., & Hwang, G.-J. (2020). Application and theory gaps during the flipped learning activities for college English writing courses. Journal of Computers in
rise of artificial intelligence in education. Computers & Education: Artificial Education, 6(3), 385–415. https://doi.org/10.1007/s40692-019-00144-6
Intelligence, 1, 1–20. https://doi.org/10.1016/j.caeai.2020.100002 Link, S., Mehrzad, M., & Rahimi, M. (2022). Impact of automated writing evaluation on
Cheung, A. (2023). Developing and evaluating a set of process and product-oriented teacher feedback, student revision, and writing improvement. Computer Assisted
classroom assessment rubrics for assessing digital multimodal collaborative writing Language Learning, 35(4), 605–634. https://doi.org/10.1080/
in L2 classes. Assessing Writing, 56, 1–16. https://doi.org/10.1016/j. 09588221.2020.1743323
asw.2023.100723 Liu, P.-L., Ginting, A. M. G., Chen, C.-J., & Yeh, H.-C. (2022). Students’ performance and
Chien, C.-C., Chen, H.-Y., & Hou, H.-T. (2024). Learning by playing with generative AI: perceptions of wiki-based collaborative writing for learners of English as a foreign
Design and evaluation of a role-playing educational game with generative AI as language. Sage Open. https://doi.org/10.1177/21582440221144953
scaffolding for instant feedback interaction. Journal of Research on Technology in Ma, J., Wang, C., & Teng, M. F. (2021). Using learning-oriented online assessment to
Education. https://doi.org/10.1080/15391523.2024.2338085 foster students’ feedback literacy in L2 writing during COVID-19 pandemic: A case of
Cho, H., Kim, Y., & Park, S. (2022). Comparing students’ responses to synchronous misalignment between micro-and macro-contexts. The Asia-Pacific Education
written corrective feedback during individual and collaborative writing tasks. Researcher, 30, 597–609. https://doi.org/10.1007/s40299-021-00600-x
Language Awareness, 31(1), 1–20. https://doi.org/10.1080/ Manegre, M., & Gutiérrez-Colón, M. (2023). Foreign language learning through
09658416.2021.1937194 collaborative writing in knowledge building forums. Interactive Learning
Creswell, J. W., & Creswell, D. (2022). Research design: Qualitative, quantitative, and mixed Environments, 31(3), 1364–1376. https://doi.org/10.1080/10494820.2020.1836499
methods approaches (6th ed.). SAGE. Meyer, J., Jansen, T., Schiller, R., Liebenow, L. W., Steinbach, M., Horbach, A., &
Devlin, J., Chang, M. W., Lee, K., & Toutanova, K. (2018). Bert: Pre-training of deep Fleckenstein, J. (2024). Using LLMs to bring evidence-based feedback into the
bidirectional transformers for language understanding. arXiv preprint. 1810.04805v2. classroom: AI-generated feedback increases secondary students’ text revision,
https://doi.org/10.48550/arXiv.1810.04805 motivation, and positive emotions. Computers & Education: Artificial Intelligence, 6,
Dobao, A. F. (2012). Collaborative writing tasks in the L2 classroom: Comparing group, 1–10. https://doi.org/10.1016/j.caeai.2023.100199
pair, and individual work. Journal of Second Language Writing, 21, 40–58. https://doi. Mizumoto, A., & Eguchi, M. (2023). Exploring the potential of using an AI language
org/10.1016/j.jslw.2011.12.002 model for automated essay scoring. Research Methods in Applied Linguistics, 2, 1–13.
Elola, I., & Oskoz, A. (2010). Collaborative writing: Fostering foreign language and https://doi.org/10.1016/j.rmal.2023.100050
writing conventions development. Language, Learning and Technology, 14(3), 51–71, Nargesi, M. N. (2021). Comparing AI and human feedback: The role of source in feedback
10125/44226. perceptions [Master’s thesis, San Francisco State University]. ScholarWorks. https://
Escalante, J., Pack, A., & Barrett, A. (2023). AI-generated feedback on writing: Insights scholarworks.calstate.edu/downloads/r494vr478.
into efficacy and ENL student preference. International Journal of Educational Peng, C. X., Storch, N., & Knoch, U. (2023). Greater coverage vs. deeper processing?
Technology in Higher Education, 20(57), 1–20. https://doi.org/10.1186/s41239-023- Comparing individual and collaborative processing of teacher feedback. Language
00425-2 Teaching Research. https://doi.org/10.1177/13621688231214910
Fleckenstein, J., Liebenow, L. W., & Meyer, J. (2023). Automated feedback and writing: Pham, V. P. H. (2021). The effects of collaborative writing on students’ writing fluency:
A multi-level meta-analysis of effects on students’ performance. Frontiers in Artificial An efficient framework for collaborative writing. Sage Open, 11(2), 1–11. https://
Intelligence, 6, 1–11. https://doi.org/10.3389/frai.2023.1162454 doi.org/10.1177/2158244021998363
Fortino, A. (2023, September 5). Thinking with AI - pros and cons — language, logic, and Rad, H. S., Alipour, R., & Jafarpour, A. (2023). Using artificial intelligence to foster
loops. NYU School of Professional Studies. https://www.sps.nyu.edu/homepage/me students’ writing feedback literacy, engagement, and outcome: A case of wordtune
taverse/metaverse-blog/Thinking-with-AI-Pros-and-Cons-Language-Logic-and-L application. Interactive Learning Environments. https://doi.org/10.1080/
oops.html. 10494820.2023.2208170
Giessler, R. (2023). EFL writers’ cognitive engagement with AWE feedback. Language Rakedzon, T., & Baram-Tsabari, A. (2017). To make a long story short: A rubric for
Awareness. https://doi.org/10.1080/09658416.2023.2269088 assessing graduate students’ academic and popular science writing skills. Assessing
Graham, S., Hebert, M., & Harris, K. R. (2015). Formative assessment and writing: A Writing, 32, 28–42. https://doi.org/10.1016/j.asw.2016.12.004
meta-analysis. The Elementary School Journal, 115(4), 523–547. https://doi.org/ Ranalli, J. (2018). Automated written corrective feedback: How well can students make
10.1086/681947 use of it? Computer Assisted Language Learning, 31(7), 653–674. https://doi.org/
Guasch, T., Espasa, A., Alvarez, I. M., & Kirschner, P. A. (2013). Effects of feedback on 10.1080/09588221.2018.1428994
collaborative writing in an online learning environment. Distance Education, 34(3), Ruwe, T., & Mayweg-Paus, E. (2023). “Your argumentation is good”, says the AI vs
324–338. https://doi.org/10.1080/01587919.2013.835772 humans – the role of feedback providers and personalised language for feedback
Hafner, C. A., & Yu, C. (2020). Language socialization in digitally mediated collaborative effectiveness. Computers & Education: Artificial Intelligence, 5, 1–14. https://doi.org/
writing: Evidence from disciplinary peer and teacher feedback. RELC Journal, 51(1), 10.1016/j.caeai.2023.100189
14–32. https://doi.org/10.1177/0033688220901347 Seo, K., Tang, J., Roll, I., Fels, S., & Yoon, D. (2021). The impact of artificial intelligence
Han, T., & Sari, E. (2022). An investigation on the use of automated feedback in Turkish on learner–instructor interaction in online learning. International Journal of
EFL students’ writing classes. Computer Assisted Language Learning. https://doi.org/ Educational Technology in Higher Education, 18(54), 1–23. https://doi.org/10.1186/
10.1080/09588221.2022.2067179 s41239-021-00292-9
Hosseini, M. S., Bavali, M., & Rezvani, R. (2020). Wiki-based collaborative writing in EFL Shang, H.-F. (2022). Exploring online peer feedback and automated corrective feedback
classrooms: Fluency and learners’ attitudes in focus. Cogent Arts & Humanities, 7(1), on EFL writing performance. Interactive Learning Environments, 30(1), 4–16. https://
1–28. https://doi.org/10.1080/23311983.2020.1826649 doi.org/10.1080/10494820.2019.1629601
Hsu, H.-C. (2019). Wiki-mediated collaboration and its association with L2 writing Soltero-González, L., Escamilla, K., & Hopewell, S. (2012). Changing teachers’
development: An exploratory study. Computer Assisted Language Learning, 32(8), perceptions about the writing abilities of emerging bilingual students: Towards a
945–967. https://doi.org/10.1080/09588221.2018.1542407 holistic bilingual perspective on writing assessment. International Journal of Bilingual
Hsu, H.-C., & Lo, Y.-F. (2018). Using wiki-mediated collaboration to foster L2 writing Education and Bilingualism, 15(1), 71–94. https://doi.org/10.1080/
performance. Language, Learning and Technology, 22(3), 103–123, 10125/44659. 13670050.2011.604712
Hwang, G.-J., Xie, H., Wah, B. W., & Gašević, D. (2020). Vision, challenges, roles and Steiss, J., Tate, T., Graham, S., Cruz, J., Hebert, M., Wang, J., Moon, Y., Tseng, W.,
research issues of Artificial Intelligence in Education. Computers & Education: Warschauer, M., & Olson, C. B. (2024). Comparing the quality of human and
Artificial Intelligence, 1, 1–5. https://doi.org/10.1016/j.caeai.2020.100001 ChatGPT feedback of students’ writing. Learning and Instruction, 91, 1–15. https://
Jabulani, S. (2015). Efficacy of written corrective feedback on university students’ doi.org/10.1016/j.learninstruc.2024.101894
writing. Journal of Social Sciences, 45(2), 84–95. https://doi.org/10.1080/ Tajabadi, A., Ahmadian, M., Dowlatabadi, H., & Yazdani, H. (2023). EFL learners’ peer
09718923.2015.11893490 negotiated feedback, revision outcomes, and short-term writing development: The
Kasneci, E., Sessler, K., Küchemann, S., Bannert, M., Dementieva, D., Fischer, F., effect of patterns of interaction. Language Teaching Research, 27(3), 689–717.
Gasser, U., Groh, G., Günnemann, S., Hüllermeier, E., Krusche, S., Kutyniok, G., https://doi.org/10.1177/1362168820951207
Michaeli, T., Nerdel, C., Pfeffer, J., Poquet, O., Sailer, M., Schmidt, A., Seidel, T., … Uludag, P., & McDonough, K. (2022). Validating a rubric for assessing integrated writing
Kasneci, G. (2023). ChatGPT for good? On opportunities and challenges of large in an EAP context. Assessing Writing, 52, 1–12. https://doi.org/10.1016/j.
language models for education. Learning and Individual Differences, 103, 1–9. https:// asw.2022.100609
doi.org/10.1016/j.lindif.2023.102274 Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., &
Kim, Y., & Emeliyanova, L. (2019). The effects of written corrective feedback on the Polosukhin, I. (2017). Attention is all you need [Paper Presentation]. December 4–9.
accuracy of L2 writing: Comparing collaborative and individual revision behavior.

9
W. Wiboolyasarin et al. Computers and Education: Artificial Intelligence 6 (2024) 100228

In The 31st conference on neural information processing systems (NIPS 2017). Long Yang, S., Nachum, O., Du, Y., Wei, J., Abbeel, P., & Schuurmans, D. (2023). Foundation
Beach, CA, USA. models for decision making: Problems, methods, and opportunities. arXiv preprint
Vygotsky, L. S. (1978). Mind in society: The development of higher psychological processes. arXiv:2303.04129. https://doi.org/10.48550/arXiv.2303.04129
Harvard University Press. Yu, S. (2021). Feedback-giving practice for L2 writing teachers: Friend or foe? Journal of
Waer, H. (2023). The effect of integrating automated writing evaluation on EFL writing Second Language Writing, 52, 1–14. https://doi.org/10.1016/j.jslw.2021.100798
apprehension and grammatical knowledge. Innovation in Language Learning and Zhang, Z., & Hyland, K. (2023). The role of digital literacy in student engagement with
Teaching, 17(1), 47–71. https://doi.org/10.1080/17501229.2021.1914062 automated writing evaluation (AWE) feedback on second language writing.
Wang, L., Chen, X., Wang, C., Xu, L., Shadiev, R., & Li, Y. (2024). ChatGPT’s capabilities Computer Assisted Language Learning. https://doi.org/10.1080/
in providing feedback on undergraduate students’ argumentation: A case study. 09588221.2023.2256815
Thinking Skills and Creativity, 51, 1–14. https://doi.org/10.1016/j.tsc.2023.101440 Zhao, Y., Zhang, F., Schunn, C. D., He, P., Li, D., & Zhao, Y. (2023). Feedback, feedback-
Wang, Y.-C. (2015). Promoting collaborative writing through wikis: A new approach for on-feedback and re-feedback: Effects of written dialogic peer feedback on English as
advancing innovative and active learning in an ESP context. Computer Assisted a foreign language writing. Assessment & Evaluation in Higher Education. https://doi.
Language Learning, 28(6), 499–512. https://doi.org/10.1080/ org/10.1080/02602938.2023.2278017
09588221.2014.881386 Zheng, L., Fan, Y., Chen, B., Huang, Z., & LeiGao & Long, M. (2023). An AI-enabled
Wigglesworth, G., & Storch, N. (2012). What role for collaboration in writing and writing feedback-feedforward approach to promoting online collaborative learning.
feedback. Journal of Second Language Writing, 21, 364–374. https://doi.org/10.1016/ Education and Information Technologies. https://doi.org/10.1007/s10639-023-12292-
j.jslw.2012.09.005 5
Winstone, N., & Carless, D. (2020). Designing effective feedback processes in higher Zou, D., Xie, H., & Wang, F. L. (2023). Effects of technology enhanced peer, teacher and
education: A learning-focused approach. Routledge. self-feedback on students’ collaborative writing, critical thinking tendency and
engagement in learning. Journal of Computing in Higher Education, 35, 166–185.
https://doi.org/10.1007/s12528-022-09337-y

10