7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58

Feedback seeking abilities of L2 writers using ChatGPT: a mixed method multiple case study

Abstract

Purpose: The study investigated the feedback seeking abilities of learners in L2

writing classrooms using ChatGPT as an automated written corrective feedback

(AWCF) provider. Specifically, the research embarked on the exploration of L2

writers’ feedback seeking abilities in interacting with ChatGPT for feedback and their perceptions thereof in the new learning environment.

Design/methodology/approach: Three EFL learners of distinct language proficiencies and technological competences were recruited to participate in the

mixed method multiple case study. The researcher used observation and in-depth

interview to collect the ChatGPT prompts written by the participants and their reflections of feedback seeking in the project. Kybernetes

Findings: The study revealed that: 1) students with different academic profiles

display varied abilities to utilize the feedback seeking strategies; 2) the significance

of feedback seeking agency was agreed upon; and 3) the promoting factors for the

development of students’ feedback seeking abilities are the proactivity of involvement and the command of metacognitive regulatory skills.

Implications: Additionally, a conceptual model of feedback seeking in an AI-

mediated learning environment was postulated. The research has its conceptual and practical implications for researchers and educators expecting to incorporate

ChatGPT in teaching and learning. The research unveiled the significance and

potential of using state-of-the-art technologies in education. However, since we are

still in an early phase applying such tools in authentic pedagogical environments,

many instructional redevelopment and rearrangement should be considered and

POST-PRINT VERSION implemented.

Originality/value: The work is a pioneering effort to explore learners' feedback

seeking abilities in a ChatGPT-enhanced learning environment. It pointed out new

directions for process-, and student-oriented research in the era of changes.

Keywords: ChatGPT; AI-enhanced learning; feedback seeking abilities; mixed

FOR WEBSITE VIEWER ONLY

method case study; corrective feedback

Introduction

RESPECT PUBLISHER RIGHTS

In educational studies, feedback has been understood as one of the most effective and

easy-to-implement instructional interventions. Researchers have agreed upon the positive

Kybernetes Page 2 of 55

effects and critical position of feedback in enhancing learning achievement (e.g., Hattie

and Timperley, 2007; Wiliam, 2018). Similarly, Hyland and Hyland (2019) point out that

7 8feedback remains a central topic for educators and researchers in the field of second 9 10language (L2) writing with its affirmed role in “encouraging and consolidating learning” 11 12 (p. 1). In light of the technological advancements, a research trend to study the impact of 13 14 15automated written corrective feedback (AWCF) has become increasingly visible in recent 16 17 years (Barrot, 2021). 18 19

Since the advent of ChatGPT, a conversational service based on generative

20 21

artificial intelligence (GAI), researchers have enthusiastically explored its capacity for an

22 23 24 array of disciplines. According to several piloting studies, ChatGPT has the potential toKybernetes 25 26

outperform and replace its predecessors as an error corrector or AWCF provider (Wu,

27 28

Wang, et al., 2023; Yan, 2023). ChatGPT’s power to automate tasks relies on prompts,

29 30 31the interactive instructions created by users inclusive of specific job demands, contextual 32 33backgrounds, objectives for the automation, etc. Consequently, L2 writers are expected 34 35

to proactively seek instead of receive feedback from the system with well-crafted prompts

36 37 38(White et al., 2023). In a conventional L2 setting, the concept of feedback seeking 39 40 behavior (FSB) facilitates the understanding of learners’ inquiry and monitoring of 41 42 POST-PRINT VERSION

received feedback from teachers or peers (Papi et al., 2019). However, positioned in the

43 44

new and interactive learning environment, we are facing the scantness of empirical

45 46 47 evidence in students’ actual abilities to seek feedback generated by AI-based systems. 48 49

Against the above backdrops, the study sets out to explore learners’ feedback

FOR WEBSITE VIEWER ONLY

50 51

seeking abilities in L2 classrooms using ChatGPT. Three participants with different levels

52 53 54of language proficiencies and technological competences were recruited for the multipleRESPECT PUBLISHER RIGHTS 55 56case studies employing the sequential explanatory mixed-method design. After a 5-week 57 58

project, the researcher analyzed ChatGPT prompts created by the participants through

Page 3 of 55

7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58

Kybernetes

coding and quantitative document analysis. Then, transcripts of the two in-depth

interview sessions were analyzed for qualitative findings. Through the data triangulation

and interpretation, the following research questions are answered:

RQ1: What characterizes the feedback seeking abilities of the participants in the

interaction with ChatGPT for feedback?

RQ2: How do L2 writers perceive their feedback seeking abilities in an ChatGPT-

enhanced learning environment?

The study is significant as an innovative effort to explore the learning behaviors

of students in a brand-new environment. The conceptual and practical contribution

directed from the findings would be valuable for the field of AI-mediated languageKybernetes

education. The next section provides a review of significant literature related to the core

concept and theories related to the topic. Section 3 details the methodologies utilized in

the study. Results and the interpretation are presented in Section 4 and 5 respectively

before the concluding remarks, limitations, and future research directions in Section 6.

Literature Review

In this section, literature on feedback seeking behavior and automated feedback providers

is reviewed before the conceptualization of feedback seeking in a GAI-enhanced learning

POST-PRINT VERSION

environment.

Feedback seeking behavior

FOR WEBSITE VIEWER ONLY

FSB, the processes of learners’ seeking feedback information, is one of the

determining factors of feedback effectiveness (Leenknecht et al., 2019). The dyadic

RESPECT PUBLISHER RIGHTS

construct, composing of feedback inquiry and monitoring, was borrowed from Ashford

et al. (2003) to indicate the purposeful acquisition of feedback information from the

7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58

Kybernetes Page 4 of 55

surrounding environment or agents. Papi et al. (2019) recast the term for language

learning to denote students’ utilization of inquiry and monitoring strategies in seeking

feedback from teachers, peer learners and the learning environment.

The introduction of FSB to language learning reflects: 1) the focus on individual

differences among learners in feedback (Winstone et al., 2021); 2) the shift from viewing

feedback as a teaching to a learning resource (Papi et al., 2019); and 3) the emphasize on

the proactive involvement instead of the reception of learners (Malecka et al., 2022). Two

research strands could be identified from existing literature on FSB in L2 learning. On

the one hand, extant studies focused on the relation between FSB, its motivational

antecedents, and/or subsequent learning outcomes (see Xu and Wang, 2023; Yao andKybernetes

Zhu, 2022). On the other hand, a few recent studies have delved into L2 learners’ FSB in

technology-enhanced settings. For example, Kessler (2023) argued that L2 learners in an

online community actively sought feedback through clarification and confirmation

requests with clear individual strategic tendencies. Similarly, Xu (2021) studied L2

writers’ FSB during Covid-19 and discovered the positive correlation between feedback-

seeking orientations and the use of self-regulatory strategies.

However, we still face a paucity of knowledge in how students seek feedback in

POST-PRINT VERSION

emerging environments practicing new feedback paradigm, where the role of teachers is

minimal and students’ information-generation from and sense-making of technology-

produced feedback are pivotal (Jensen et al., 2023). FSB from existing studies remains

of the old paradigm, in which the students, receivers of feedback, manage to strike a

FOR WEBSITE VIEWER ONLY

balance between the explicit inquiry for feedback from teachers and peers, and the

implicit monitoring of the learning environment (PapiRESPECT PUBLISHER RIGHTSet al., 2019). Conversely, the

changing learning environment as well as the role of learners as feedback generators urge

further exploration into students’ feedback seeking in ChatGPT-enhanced settings.

Page 5 of 55 Kybernetes

Automated feedback providers: potential of ChatGPT

Researchers have contributed a plethora of studies exploring the effects of technology-

7 8

enhanced feedback on language learning, especially L2 writing performance. In the

9 10

beginning, corpus-based automated writing evaluation (AWE) systems such as Pigai,

11 12 13iWrite, and iTEST were widely applied in L2 writing classrooms (Zhang, 2021). Soon 14 15the AWE systems gave place to AI-based corrective feedback providers, e.g., Grammarly 16 17

and QuillBot. The advantage of these AI-based solutions were empirically validated (Ngo

18 19 20et al., 2022) and thus remained the most frequent choices for automated feedback in L2 21 22classrooms. 23 24Kybernetes

Since its debut, researchers have attempted to investigate into ChatGPT’s effects

25 26

on L2 writing. Yan (2023) explored L2 writers’ abilities to utilize ChatGPT’s text

27 28 29 generation functionality in writing and their perceptions towards the potential threats to 30 31

academic integrity. Barrot (2023) summarized advantages and pitfalls of ChatGPT’s

32 33

potential usage in L2 writing classrooms, and offered recommendations for L2 teachers.

34 35 36Furthermore, efforts have been made to incorporate ChatGPT in L2 writing assessment. 37 38Mizumoto and Eguchi (2023) reported convincing accuracy and reliability from the 39 40

automated scoring on L2 writing products by ChatGPT. Accordingly, the potential of

41 42 POST-PRINT VERSION 43ChatGPT as a powerful AWCF providers is posited on: 1) the ability to provide accurate 44 45grammatical and syntactical corrections, thanks to the tremendous volume of trained 46 47

linguistic data (Fang et al., 2023); 2) the mechanism of prompt engineering with which

48 49 FOR WEBSITE VIEWER ONLYusers could tap prompts to iteratively and progressively elicit feedback and responses of 50 51 52 improved quality (White et al., 2023); and 3) the observed enhancement of learners’ 53 54RESPECT PUBLISHER RIGHTS

engagement and involvement in the interaction with ChatGPT (Yan, 2023).

55 56 57 58

Kybernetes Page 6 of 55

Furthermore, ChatGPT brings changes to L2 learning behaviors in seeking and

utilizing feedback. First, the human-ChatGPT communication is bi-directional, making

7 8ChatGPT a learner-aware, human-like, and adaptive feedback provider (Barrot, 2023). 9 10Second, once properly prompted, ChatGPT could generate feedback on a wide range of 11 12

writing constructs and linguistic features, inclusive but not limited to grammar and

13 14 15syntactical correctness (Mizumoto and Eguchi, 2023). Third, multi-turn human-ChatGPT 16 17 communication is encouraged and corroborated to be conducive to performance 18 19

improvement of ChatGPT-generated feedback (Bang et al., 2023). Given the human-AI

20 21

communication interactivity and corresponding changes in learning behaviors, relevant

22 23 24 studies on learning behavior from the learner perspectives are urgently needed.Kybernetes 25 26 27 28Conceptualizing feedback seeking in GAI-enhanced learning 29 30

The study is grounded on interaction theory. As argued by Long (1996), the four

31 32 33constructs involved in learning included input, interaction, feedback and output. For L2 34 35development, Bitchener and Ferris (2012) emphasized the understanding, internalization 36 37

and processing of feedback on the basis of exiting knowledge. Furthermore, Kaptelinin

38 39

and Nardi (2018) pinpointed that the deepening of the effects of technology could be

40 41 42 attained by putting human-computer interaction within the “purposeful human activities” POST-PRINT VERSION 43 44

(p. 3). Correspondingly, a conceptual model (Figure 1) is proposed by integrating

45 46

feedback seeking from GAI-based systems with the subsequent internalization and

47 48 49processing of the feedback.

FOR WEBSITE VIEWER ONLY

50 51 52 53 54RESPECT PUBLISHER RIGHTS 55 56 57 58

Page 7 of 55 Kybernetes

7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24Kybernetes 25

Figure 1. Conceptual Model of L2 writers’ feedback seeking from ChatGPT

26 27 28 According to Papi et al. (2019), FSB consists of feedback monitoring and inquiry 29 30

through which the learners observe and absorb relevant feedback information from the

31 32

environment, and seek further explanation of the feedback. However, the emergence of

33 34 35 GAI-based tools poses new challenges to the operationalization of the two behaviors. 36 37

First, the approaches of feedback inquiry changes. Compared to the traditional teacher-

38 39

fronted settings, learners no longer resort to the explanation and help from teachers or

40 41 42peer (Jensen POST-PRINT VERSION et al., 2023). Conversely, the affordance of GAI systems could replicate the 43 44role of both teachers and peer learners at the cost of extra prompts written and submitted 45 46

through the dialog interfaces. Second, the feedback monitoring as described by Papi et al.

47 48 49(2019) is irrelevant for the setting with ChatGPT. With the conversational interaction

FOR WEBSITE VIEWER ONLY

50 51 with AI and proper prompt writing techniques, learners could instantly and iteratively 52 53

request GAI systems to amend, improve and regenerate feedback in a rather explicit and

54RESPECT PUBLISHER RIGHTS 55

direct manner. Hence, during a feedback episode, the feedback seeking processes could

56 57 58 be explicitly executed for multiple times. Consequently, the original two component of

Kybernetes Page 8 of 55

FSB could be reframed: 1) feedback elicitation, in which the learners use prompts to

iteratively generate feedback with different contents and focus; 2) feedback refinement,

7 8an on-going process in which the learners repeatedly report their inner judgement to the 9 10GAI systems, after which feedback would be regenerated based on the provided 11 12

evaluation. Correspondingly, feedback seeking ability refers to the capability of students

13 14 15to effectively use feedback elicitation and refinement strategies in the interaction with 16 17 GAI systems. 18 19 20 21Methods and Materials 22 23 24DesignKybernetes 25 26 27The study adopted a mixed method multiple case study approach (Yin, 2013). In the fields 28 29of applied linguistics (Duff, 2010) and instructional feedback (e.g., Zhang and Hyland, 30 31

2023), case study has been widely applied as an established means to collect rich data of 32 33

students’ actual learning experiences. Additionally, Adams (2019) argued that the limited

34 35 36 number of research subjects in case studies had its merits in unfolding learner experiences 37 38

in using the feedback other than the feedback design.

39 40

A case was defined as a student’s ability to effectively use feedback seeking

41 42 POST-PRINT VERSION 43 strategies and their perceptions thereof. Internally, a sequential explanatory mixed 44 45method design was selected so that the quantitively displayed characteristics of students’ 46 47

feedback seeking abilities would be explained by their own reflection and perceptions

48 49

FOR WEBSITE VIEWER ONLY

50(Creswell and Plano Clark, 2018). The study was collective in nature, since it allowed the 51 52researcher to generalize the findings across multiple cases to a broader context (Stake, 53 54RESPECT PUBLISHER RIGHTS

1995).

55 56 57 58

Page 9 of 55 Kybernetes

Participants

The study took place in an EFL program at a Chinese university. Students in the program 7 8

were required to take two writing courses. The researcher purposefully sampled three

9 10

participants with distinct level of language proficiencies and technological competences

11 12 13(see The screening criteria included: 1) mean grades in precedent L2 writing 14 15assessments; 2) experiences and interest in computer-assisted language learning and 16 17

feedback reflected in a pre-project survey; 3) performance in two digital humanities

18 19 20courses; and 4) self-rated technological competences. The criteria ensured that students 21 22with various levels of language and technological competence were recruited, thus 23 24Kybernetes

forming representative cases of L2 writers in the program.

25 26 27

Table I for their characteristics). The screening criteria included: 1) mean grades in

28 29 30 precedent L2 writing assessments¹; 2) experiences and interest in computer-assisted 31 32

language learning and feedback reflected in a pre-project survey; 3) performance in two

33 34 2

digital humanities courses; and 4) self-rated technological competences. The criteria

35 36 37ensured that students with various levels of language and technological competence were 38 39recruited, thus forming representative cases of L2 writers in the program. 40 41 42 POST-PRINT VERSION

Table I. Demographic Information and Characteristics of the Participants

43 44 Pseudonyms Gender Age Mean Mean Self-rated Instructors’ 45 Grades Grades Technological Recommendations 46 1 2 Competences** 47 Toby M 21 81/100 91/100 9/10 “His balance in both 48 language proficiency and 49

FOR WEBSITE VIEWER ONLY technology abilities make

50 51 52 1 Writing tasks were developed as per the requirement of College English Test Band 6, an 53 English proficiency examination commonly administered in Chinese universities. TheRESPECT PUBLISHER RIGHTS 54 writing performances were rated against a 100-mark rubric composing five writing 55 assessment dimensions, i.e., content, organization, grammar, vocabulary, and mechanics. 56 2 Measured by a self-reported 10-point survey including the following dimensions: a) general 57 58 digital literacy; b) information retrieval and evaluation; c) communicative and collaborative skills using technology; and d) language-specific technological skills.

Kybernetes Page 10 of 55

7 8 9 10 11 12 13 *Mean Grades 1: mean grades in precedent L2 writing assessments. 14 Mean Grades 2: mean grades in precedent assessments in digital humanities courses. 15 16 Based on students’ performance in pre-project language assessments and 17 18teachers’ recommendations, Toby represented those with high language proficiency and 19 20

technological competence; Lisa stood for those with medium language proficiency and

21 22 23high technological competence; and Isabella exemplified those with high language 24Kybernetes 25proficiency and medium technological competence. 26 27 28 29 Data Collection 30 31

The project was a five-week teacher-scaffolded informal L2 writing practicum with a

32 33 34 total number of 68 learners (including the three participants). During the project, all 35 36

learners were required to complete and revise three argumentative essay each no less than

37 38

250 words. Additionally, two sessions of live demonstration and peer discussion were

39 40 41arranged for each week (see Appendix S1 for practicum details). To ensure students’ 42 POST-PRINT VERSION 43engagement with the practicum, they were allowed to choose one of the essays completed 44 45

in the practicum as assignment (accounting for 15% of the final grade) of an ongoing

46 47 48writing course. 49

FOR WEBSITE VIEWER ONLY

50 During the project, students were asked to keep track of the prompts they have 51 52

written to interact with ChatGPT in learning logs (see Appendix S2). To overcome the

53 54RESPECT PUBLISHER RIGHTS

potential bias in the focus and scope of ChatGPT-generated feedback due to different

55 56 57 feedback seeking abilities and ChatGPT prompting strategies, all the participants were 58

Page 11 of 55 Kybernetes

trained and required to use the same ChatGPT prompting template (see Appendix S1).

The official GPT-3.5 version of ChatGPT from OpenAI was used in the study. Students

7 8were encouraged to explore ChatGPT’s functionality at their own paces, hence no time 9 10restraint was posed on their interaction with ChatGPT. The learning log could be filled 11 12

with multimedia files as supplementary documents, e.g., screenshot, videorecording, etc.

13 14 15Furthermore, to record all the learning behaviors, the project took place at the 16 17 computerized language laboratories equipped with real-time screen recording, 18 19

keylogging, and video-based classroom monitoring devices. Upon the completion of each

20 21

session in the project, students are asked to review and member-check the contents of

22 23 24 learning log.Kybernetes 25 26

To obtain students’ perceptions of their feedback seeking abilities in the

27 28

ChatGPT-enhanced learning environment, two sessions of face-to-face in-depth

29 30 31interview were arranged for each participant at the third and fifth week respectively. Each 32 33interview session lasted for approximately 30 minutes. The interview processes strictly 34 35

followed the interview protocol (see Appendix S3). The participants used their first

36 37 38language (Chinese) so that they could better express their understandings and perceptions. 39 40 All contents and procedures of the interview were audio recorded, translated into English, 41 42 POST-PRINT VERSION

and transcribed verbatim. To enhance the trustworthiness of the collected data, all

43 44

translated texts were checked by a co-researcher who was a certified and accredited

45 46 47 English-Chinese translator before member-checked by the participants (Birt et al., 2016). 48 49

FOR WEBSITE VIEWER ONLY

50 51Coding and Analysis 52 53

Data obtained from observation and interviews were analyzed and organized in

54RESPECT PUBLISHER RIGHTS 55 56accordance with the within-case and cross-case analysis techniques recommended by 57 58seminal works of case study (Stake, 2005; Yin, 2013).

Kybernetes Page 12 of 55

First, quantitative document analysis was applied to the ChatGPT prompts. In

practice, a coding schemes of ChatGPT prompt patterns developed based on the

7 8experiences from the pilot study and the prompt engineering mechanism (White et al., 9 102023) was applied in the study (see Appendix S4). Four coders were recruited from the 11 12

teaching faculty to code the patterns of students’ prompts. The inter-rater reliability

13 14 15measured by the Fleiss’ Kappa statistics (Kappa = .76; [95% CI: 0.68-0.81]) revealed an 16 17 acceptable level of agreement between the coders. Then, descriptive analysis and 18 19

visualization were applied to the coded data.

20 21

Second, thematic analysis was applied to the transcribed interview data. In

22 23 24 specific, the six-step procedures advocated by Braun and Clarke (2006) was adhered to.Kybernetes 25 26

Two co-researchers were recruited to assist the researcher in the data coding and analysis.

27 28

Initial codes of significant and recurring patterns were extracted from the original data,

29 30 31then they were refined into higher-order themes representing students’ perceptions of 32 33their feedback seeking abilities from ChatGPT in language learning. To ensure the 34 35

trustworthiness of the qualitative analysis, researcher triangulation and peer debriefing

36 37 38were practiced. 39 40 41

Validity

42 POST-PRINT VERSION 43 44 45Results 46 47 48 Quantitative Document Analysis 49

FOR WEBSITE VIEWER ONLY

50 51Toby (Table II) has created a total of 2,537 pieces of prompts in 167 rounds of interaction 52 53 with ChatGPT using 850 minutes; Lisa (Table III) has written 1,648 pieces of prompts 54RESPECT PUBLISHER RIGHTS 55

in 193 rounds of interaction with ChatGPT using 436 minutes; Isabella (Table IV) has

56 57

spent 1,034 minutes in her 70-round interaction with ChatGPT with 1,921 pieces of

Page 13 of 55 Kybernetes

prompts.

Table II. ChatGPT Prompts Created by Toby

7 8

9 10 11 12 13 14 15 16 17 18 19 20 Table III. ChatGPT Prompts Created by Lisa 21

22 23 24 25 26 27 28 29 30 31 32 33

Table IV. ChatGPT Prompts Created by Isabella

34 35

36 37 38 39 40 41 42 43 44 45 46 47 48Against the coding scheme, the coders have categorized all prompts into two 49

FOR WEBSITE VIEWER ONLY

major dimensions of feedback elicitation and refinement. Then the data were further

51 52 53

divided into 11 different subcategories of ChatGPT prompt patterns. With the coded data

54RESPECT PUBLISHER RIGHTS 55 visualized by the temporal order, the development feature of the three participants’ 56 57

feedback seeking abilities were illustrated.

Kybernetes Page 14 of 55

Toby’s development of feedback seeking abilities was steady and balanced (see

Figure 2). His application of basic prompts and direct feedback regeneration was

7 8remarkably reduced as he progressed; his utilization of other feedback elicitation 9 10strategies grew consistently while his command of feedback refinement strategies 11 12

experienced a transition from a rapid heating up to a gradual cooling down.

13 14 15 16 17 18 19 20 21 22 23 24Kybernetes 25 26 27 28 29 30 31 32 33 34 35 36 37 38Figure 2. Prompt Patterns of Toby 39

For Lisa (see Figure 3), her application of basic prompts and direct feedback

40 41 42regeneration remained generally unchanged in the project, while a growing momentum POST-PRINT VERSION 43 44could be observed for the rest feedback seeking strategies. 45 46 47 48 49

FOR WEBSITE VIEWER ONLY

50 51 52 53 54RESPECT PUBLISHER RIGHTS 55 56 57 58

Page 15 of 55 Kybernetes

7 8 9 10 11 12 13 14 15 16 17 18 19 20 21

Figure 3. Prompt Patterns of Lisa

22 23 Finally, Isabella’s development was similar to Toby’s (see Figure 4), only 24Kybernetes 25

relatively less in quantity as she appeared as a slow-paced and detailed prompt writer.

26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 POST-PRINT VERSION 43 44 45

Figure 4. Prompt Patterns of Isabella

46 47 48

Thematic Analysis

FOR WEBSITE VIEWER ONLY

50 51

Through the thematic analysis, the following main themes were identified: 1) perceptions

52 53 of feedback seeking from ChatGPT; 2) significance of active feedback seeking; and 3)RESPECT PUBLISHER RIGHTS 54 55 56factors for the development of feedback seeking abilities. The thematic map of the 57 58interviews is shown in Figure 5. Additionally, each of the participants summarized their

Kybernetes Page 16 of 55

learning experiences in the second interview session.

7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24Kybernetes 25 26 27Figure 5. Thematic Map of the Interviews 28 29 30Perception of feedback seeking from ChatGPT 31 32 33When asked about the GAI-enhanced environments, Toby and Lisa stressed the proactive 34 35involvement of learners. As they have experienced, the shift from a feedback receiver to 36 37

seeker was impactful on their “flow of learning”. In the new environment, students’

38 39 40agency, creativity, and proactivity was highly valued. 41 42 POST-PRINT VERSIONPertinent to the role ChatGPT played in L2 writing, Lisa and Isabella agreed upon 43 44

the dual identity as both a feedback provider and explainer. From their viewpoints,

45 46

ChatGPT availed the learners clearer and detailed explanation about the feedback. At the

47 48 49 same time, they believed that such affordance distinguishes ChatGPT from previous

FOR WEBSITE VIEWER ONLY

50 51

feedback providers, as it could compensate for the possible unavailability of the teachers.

52 53

RESPECT PUBLISHER RIGHTSThe three participants reached an agreement in the advantage of ChatGPT as an

54 55 56instant feedback provider. Learners could continuously solicit corrective feedback with 57 58improved quality from the AI system, if they were “well-versed in the language ChatGPT

Page 17 of 55 Kybernetes

spoke, aka, the prompts” (Toby, interview#1). Furthermore, Toby and Isabella argued in

unison that “ChatGPT needs just a few seconds to provide feedback, and every second

7 8would be rewarded with corrections of high quality” (Isabella, interview#2). According 9 10to both Toby and Isabella, who have compared the quality of the corrections from 11 12

multiple sources, e.g., Grammarly or QuillBot, ChatGPT was generally reliable and

13 14 15accurate. 16 17 18

Significance of active feedback seeking

19 20 21

All the three participants agreed that active feedback seeking was beneficial to the

22 23 24

affective and behavioral outcomes of learning. Toby and Isabella emphasized that the

Kybernetes

25 26subsequent processing of feedback gained from frequent and effective feedback seeking. 27 28Isabella underlined the value of the unfocused feedback offered by ChatGPT against the 29 30

focused feedback from teachers that she “could make judgement on writing from a

31 32 33holistic perspective instead of paying attention merely to a certain aspect” (Isabella, 34 35interview#1). 36 37

Regarding the effects of the active feedback seeking from ChatGPT on the

38 39 40

affective domain of learning experiences, all participants kept pace with each other. To

41 42 them, the feedback seeking from ChatGPT helped creating a democratic atmosphere. POST-PRINT VERSION 43 44

Students were mentally at ease asking ChatGPT to regenerate or refine the feedback.

45 46 47

Furthermore, participants felt that they are motivated to actively involve in

48 49feedback seeking. In conventional settings, negative and critical feedback was treated by

FOR WEBSITE VIEWER ONLY

50 51learners as shame. Consequently, students were generally reluctant to ask additional 52 53

questions about the feedback, much less challenge it. The interactive processes to elicit

RESPECT PUBLISHER RIGHTS

54 55 56feedback from ChatGPT, according to Lisa, “works like a black box, in which our shames 57 58are hidden, in a positive way” (Lisa, interview#2).

Kybernetes Page 18 of 55

Finally, for the effects of feedback seeking on L2 writer’s competence, Lisa and

Isabelle pointed to students’ critical thinking. In accordance with their reflections, the

7 8feedback refinement processes required the learners to critically compare the values and 9 10side effects of feedback under consideration. Furthermore, the abilities to accurately 11 12

identify existing issues and critically describe them in writing prompts were vital for

13 14 15effective interaction with ChatGPT. 16 17 18

Factors for the development of feedback seeking abilities

19 20 21

Grounded on the consensus that ChatGPT provided a promoting environment for students

22 23

to develop feedback seeking abilities, all participants accentuate that the experiences of

24Kybernetes 25 26interaction with AI system were most important. ChatGPT, from their viewpoints, is 27 28“shockingly easy-to-use at first glance” (Toby, interview#2), but takes “a huge amount 29 30

of practice and thinking” to “craft the skills and techniques of writing and polishing

31 32 33prompts” (Isabella, interview#2). 34 35Furthermore, all three participants agreed that an open-minded attitude mattered. 36 37

According to Toby, the innovation ChatGPT brought about resulted in “either overhyped

38 39

tunes or total rejections” (Toby, interview#1). As a result, there exists a gap between the

40 41 42 perceived affordance of “ChatGPT as a key to good writing quality” (Lisa, interview#1) POST-PRINT VERSION 43 44

and realistic experiences that “it is a powerful yet demanding tool with diversified and

45 46

unpredictable outcomes” (Isabella, interview#2). To mediate such cognitive dissonance,

47 48 49the participants suggested that deepening one’s understanding of ChatGPT’s mechanism

FOR WEBSITE VIEWER ONLY

50 51and workflow in a real setting of L2 writing was imperative. 52 53

Toby and Isabella provided their take on the concrete learning strategies needed

54RESPECT PUBLISHER RIGHTS 55 56to foster feedback seeking abilities from AI-based systems. Reflected by their 57 58

Page 19 of 55 Kybernetes

experiences, learners’ monitoring and evaluation were decisive for feedback quality.

Toby confirmed that such refinements called for metacognitive abilities:

7 8*“To compare and select from multiple pieces of feedback, one needs to evaluate* 9 comprehensively, and use a whole flow of planning, comparing and controlling from 10a higher order perspective” (Toby, interview#2). 11 12 13Wrap-up of individual learning experiences 14 15In the second interview sessions, each participant summarized their learning 16 17experiences and several post-project perceptions. In specific, their perceptions of 18 19

feedback seeking, recommendations for the best practices and a summary of their

20 21

personal growth were elucidated. The gist of their summaries is presented in Table V.

22 23 24Kybernetes 25 Table V. Individual Summary of Learning Experiences 26

27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58

Kybernetes Page 20 of 55

Discussions

Being a multiple case study, the observations and findings emerged from individual cases

7 8

were triangulated and holistically interpreted. The findings are discussed against existing

9 10

empirical evidence and theoretical perspectives.

11 12 13In answering RQ1, the quantitative document analysis revealed that students with 14 15different language proficiencies and technological competences displayed diversified 16 17

developmental trajectories of feedback seeking abilities. First, students with higher

18 19 20language proficiency (i.e., Toby and Isabella) tended to create content-rich prompts in 21 22relatively fewer rounds of interaction with AI system, while learner with moderate 23 24Kybernetes

language proficiency (i.e., Lisa) resorted to an opposite path. The differences in the

25 26

quantity and length of prompts could be partially explained by the claim that the

27 28 29 utilization and processing of feedback depended on the learners’ language skills 30 31

(Koltovskaia, 2020). Such phenomenon could also be attributed to the interrelation

32 33

between L2 language mindset and FSB (Yao and Zhu, 2022). Second, judging from the

34 35 36time consumption of the participants, Toby and Isabella (representing learners with higher 37 38language proficiency) were relatively slow-paced in action. This echoed with the 39 40

observation of Ranalli (2018) that AWCF would affect mental-effort expenditure on

41 42 POST-PRINT VERSION 43learners with different L2 levels. However, contrary to the claim that AWCF would pose 44 45extra cognitive overload to low-proficiency learners (Bitchener and Ferris, 2012), the 46 47

current study revealed that it was high-proficiency students (i.e., Toby and Isabella) who

48 49 FOR WEBSITE VIEWER ONLYspent more time and efforts analyzing, evaluating and making use the AWCF. Such 50 51 52 contradiction could be attributed to the iterative and interaction mechanism of ChatGPT, 53 54RESPECT PUBLISHER RIGHTS

in which low-proficiency learners could directly ask ChatGPT to regenerate (as evidenced

55 56

in Lisa’s overuse of feedback regeneration in Figure 2) while high-proficiency learners

57 58

Page 21 of 55 Kybernetes

(as evidenced in the dramatic decrease in feedback regeneration of Toby and Isabella)

would stop, think, and write prompts with additionally backgrounds and specific demands

7 8(as evidenced by Toby and Isabella’s prompt writing habits). Additionally, the variances 9 10in time consumption testified the existence of inner feedback, a natural process of 11 12

processing and comparison when the learners was exposed to feedback (Nicol, 2021).

13 14 15Internally, the weekly ChatGPT prompt patterns reflected the strategical aspect of 16 17 feedback seeking ability development. First, students with different language 18 19

proficiencies kept abreast in exerting feedback elicitation strategies in the interaction with

20 21

ChatGPT. This showcased students’ progress in understanding the conversational and

22 23 24 iterative mechanism of ChatGPT. With the progression in the project, students graduallyKybernetes 25 26

reduced writing minimalistic prompts, instead, additional information were included to

27 28

enrich and clarify the instructions prepared for ChatGPT. This finding echoed the

29 30 31developmental characteristics of students using ChatGPT for text generation in an L2 32 33classroom (Yan, 2023). Second, diversity to use feedback refinement strategies surfaced 34 35

from the quantitative document analysis. Students with higher language proficiencies

36 37 38(i.e., Toby and Isabella) steadily shifted from the style of short and frequent interaction 39 40 with the AI system to a more deliberated and information-rich manner. Conversely, Lisa, 41 42 POST-PRINT VERSION

representing students with moderate language proficiency but higher command of

43 44

technologies, consistently carried out a straightforward way to interact with the AI

45 46 47 system. To some extent, the finding was in tandem with the argument of Havranek and 48 49

Cesnik’s (2001) that proficient learners could make better use of feedback. As feedback

FOR WEBSITE VIEWER ONLY

50 51

refinement strategies functioned on a higher order, such disparities could also be

52 53 54attributed to the variance in metacognitive regulatory skills. In this sense, suchRESPECT PUBLISHER RIGHTS 55 56phenomenon could be explained by the positive correlation between metacognition and 57 58

FSB (Xu, 2021).

Kybernetes Page 22 of 55

In answering RQ2, the synthetization of participants’ perceptions pinpointed the

decisive factors for fostering feedback seeking abilities in a ChatGPT-enhanced

7 8environment. First, the consensus of participants emphasized learners’ proactive 9 10involvement. Such belief was in tandem with the view that contemporary feedback was a 11 12

process of self-generated meaning making (Jensen et al., 2023). Based on such views, we

13 14 15infer that the significance of human efforts, in the era of AI-enhanced learning, should be 16 17 uplifted rather than diminished. Second, the participants stressed the significance of 18 19

metacognitive regulation. In the study, participants effectively combined the internal

20 21

feedback processing and external feedback refinement to make better use of ChatGPT’s

22 23 24 feedback. The observationKybernetescontradicted traditional views that feedback seeking 25 26

encompassed limited metacognitive regulation other than the comparison of value and

27 28

cost of potential feedback information (Ashford et al., 2003). ChatGPT called for higher

29 30 31level of metacognition because students were allowed to state their affective and cognitive 32 33judgments. Such contradiction could be attributed to the new feedback paradigm, in 34 35

which students’ proactivity and agency were held pivotal (Jensen et al., 2023). From

36 37 38another perspective, such phenomenon testified the reported effects of ChatGPT on the 39 40 promotion of self-regulation skills (Wu, Lee, et al., 2023). 41 42 POST-PRINT VERSION 43 44Conclusions 45 46

The study investigated the development of feedback seeking abilities of L2 writers

47 48 49utilizing ChatGPT for corrective feedback. Furthermore, the research identified the key

FOR WEBSITE VIEWER ONLY

50 51factors deciding students’ development of feedback seeking abilities in the new learning 52 53 environment. 54RESPECT PUBLISHER RIGHTS 55 56The study delivered conceptual and pedagogical implications. First, a 57 58sociomaterial perspective towards feedback would trigger researchers to expand and

Page 23 of 55 Kybernetes

redevelop existing conceptualization in the field of instructional feedback (Gravett,

2022). The study revealed that ChatGPT, when actively and strategically utilized, offered

7 8a competitive solution for AWCF in L2 learning. Considering the growth in the 9 10application of AI, such domain-specific and in-depth investigations into the provision and 11 12

utilization of feedback in innovative learning environments would be helpful for the

13 14 15sustained development of educational studies in the new era. Second, educators should 16 17 consider the development of students’ abilities to use state-of-the-art tools. The results 18 19

showed that the quality and effectiveness of ChatGPT-generated feedback were closely

20 21

related to students’ feedback seeking abilities. However, to date, such abilities were

22 23 24 understudied, let alone properly trained. There remained much to be done for educators,Kybernetes 25 26

e.g., altering curriculum and assessment setups, (re)constructing a learning environment

27 28

levering AI affordance and peer collaboration, providing training on AI skills, etc.

29 30 31The study was not without limitations. First, the study took a relatively narrow 32 33stance towards student’s learning behaviors in the new context. The study focused on 34 35

human-AI interaction during the utilization of ChatGPT in L2 classrooms. Consequently,

36 37 38other key factors related to learning process and outcomes, e.g., achievement, 39 40 engagement, motivation, acceptance, etc., were not examined. In the follow-up studies, 41 42 POST-PRINT VERSION

educators and researchers could continue the exploration of the impact of ChatGPT on

43 44

different aspects of pedagogy. Second, the reported findings were based on a three-

45 46 47 participant case study and could not be easily generalizable. Considering the complexity 48 49

and diversity of L2 learning, follow-up studies are encouraged to examine students’

FOR WEBSITE VIEWER ONLY

50 51

feedback seeking with a larger sample size. Third, the study intentionally adopted a

52 53 54student-centered perspective in research design. As a result, the effects of ChatGPT onRESPECT PUBLISHER RIGHTS 55 56teacher’s roles and instructional approaches were not examined. In subsequent studies, 57 58

researchers could posit the research settings in teacher-fronted L2 classrooms, where the

Kybernetes Page 24 of 55

interaction between students and AI systems would be mediated by the teacher-student

and peer interactions. Finally, the study is restricted by its duration. With a relatively short

7 8time span, the study delivered exploratory insights necessary to understand students’ 9 10feedback seeking from GAI tools used in learning. However, due to the limited amount 11 12

of collected data, the study did not apply analytical methods such as inferential statistical

13 14 15analysis, which would offer more evidence regarding the developmental trajectories of 16 17 students’ feedback seeking abilities. In subsequent studies, researchers are encouraged to 18 19

conduct longitudinal studies for the deepened understanding of feedback seeking with

20 21

AI-enhanced learning tools.

22 23 24Kybernetes 25 26

Acknowledgement

27 28

The author would like to thank the handling editor and all three anonymous reviewers

29 30 31who have offered constructive feedback for the paper’s quality improvement. 32 33Declaration of Interest 34 35

The author has no competing interests to declare.

36 37 38Funding 39 40 This research was funded by ****, grant number ****. 41 42 POST-PRINT VERSION 43 44References 45 Adams, G. (2019), “A narrative study of the experience of feedback on a professional doctorate: ‘a kind 46 47 of flowing conversation’”, Studies in Continuing Education, Routledge, Vol. 41 No. 2, pp. 191– 48 49

FOR WEBSITE VIEWER ONLY 206, doi: 10.1080/0158037X.2018.1526782.

50 51 Ashford, S.J., Blatt, R. and VandeWalle, D. (2003), “Reflections on the Looking Glass: A Review of 52 53 Research on Feedback-Seeking Behavior in Organizations”, Journal of Management, Vol. 29 54RESPECT PUBLISHER RIGHTS 55 No. 6, pp. 773–799, doi: 10.1016/S0149-2063(03)00079-5. 56 57 58

Page 25 of 55

7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58

Kybernetes

Bang, Y., Cahyawijaya, S., Lee, N., Dai, W., Su, D., Wilie, B., Lovenia, H., et al. (2023), “A Multitask,

Multilingual, Multimodal Evaluation of ChatGPT on Reasoning, Hallucination, and

Interactivity”, arXiv, doi: 10.48550/arXiv.2302.04023.

Barrot, J.S. (2021), “Using automated written corrective feedback in the writing classrooms: effects on L2

writing accuracy”, Computer Assisted Language Learning, Routledge, pp. 1–24, doi:

10.1080/09588221.2021.1936071. Barrot, J.S. (2023), “Using ChatGPT for second language writing: Pitfalls and potentials”, Assessing

Writing, Vol. 57, p. 100745, doi: 10.1016/j.asw.2023.100745.

Birt, L., Scott, S., Cavers, D., Campbell, C. and Walter, F. (2016), “Member Checking: A Tool to

Enhance Trustworthiness or Merely a Nod to Validation?”, Qualitative Health Research, SAGE

Publications Inc, Vol. 26 No. 13, pp. 1802–1811, doi: 10.1177/1049732316654870.

Bitchener, J. and Ferris, D. (2012),

Kybernetes

Written Corrective Feedback in Second Language Acquisition and

Writing, Routledge, New York ; London.

Braun, V. and Clarke, V. (2006), “Using thematic analysis in psychology”, Qualitative Research in

Psychology, Hodder Arnold, United Kingdom, Vol. 3 No. 2, pp. 77–101, doi:

10.1191/1478088706qp063oa. Creswell, J.W. and Plano Clark, V.L. (2018), Designing and Conducting Mixed Methods Research, Third

Edition., SAGE, Los Angeles.

Duff, P. (2010), Case Study Research in Applied Linguistics, Transferred to digital printing., Routledge,

New York.

Fang, T., Yang, S., Lan, K., Wong, D.F., Hu, J., Chao, L.S. and Zhang, Y. (2023), “Is ChatGPT a Highly

POST-PRINT VERSION

Fluent Grammatical Error Correction System? A Comprehensive Evaluation”, arXiv, 4 April,

doi: 10.48550/arXiv.2304.01746.

Gravett, K. (2022), “Feedback literacies as sociomaterial practice”, Critical Studies in Education,

Routledge, Vol. 63 No. 2, pp. 261–274, doi: 10.1080/17508487.2020.1747099.

FOR WEBSITE VIEWER ONLY

Hattie, J. and Timperley, H. (2007), “The Power of Feedback”, Review of Educational Research,

American Educational Research Association, Vol. 77 No. 1, pp. 81–112, doi:

RESPECT PUBLISHER RIGHTS

10.3102/003465430298487.

7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58

Kybernetes Page 26 of 55

Havranek, G. and Cesnik, H. (2001), “Factors affecting the success of corrective feedback”:, EUROSLA

Yearbook, John Benjamins Publishing Company, Vol. 1 No. 1, pp. 99–122, doi:

10.1075/eurosla.1.10hav. Hyland, K. and Hyland, F. (2019), “Contexts and Issues in Feedback on L2 Writing”, in Hyland, F. (Ed.),

Feedback in Second Language Writing: Contexts and Issues, 2nd ed., Cambridge University

Press, Cambridge, pp. 1–22, doi: 10.1017/9781108635547.003.

Jensen, L.X., Bearman, M. and Boud, D. (2023), “Feedback encounters: towards a framework for

analysing and understanding feedback processes”, Assessment & Evaluation in Higher

Education, Routledge, Vol. 48 No. 1, pp. 121–134, doi: 10.1080/02602938.2022.2059446.

Kaptelinin, V. and Nardi, B. (2018), “Activity Theory as a Framework for Human-Technology

Interaction Research”, Mind, Culture, and Activity, Routledge, Vol. 25 No. 1, pp. 3–5, doi:

10.1080/10749039.2017.1393089.

Kybernetes

Kessler, M. (2023), “Written corrective feedback in an online community: A typology of English

language learners’ requests and interlocutors’ responses”, Computers and Composition, Vol. 67,

p. 102752, doi: 10.1016/j.compcom.2023.102752. Koltovskaia, S. (2020), “Student engagement with automated written corrective feedback (AWCF)

provided by Grammarly: A multiple case study”, Assessing Writing, Vol. 44, p. 100450, doi:

10.1016/j.asw.2020.100450. Leenknecht, M., Hompus, P. and Van Der Schaaf, M. (2019), “Feedback seeking behaviour in higher

education: the association with students’ goal orientation and deep learning approach”,

Assessment & Evaluation in Higher Education, Routledge, Vol. 44 No. 7, pp. 1069–1078, doi:

POST-PRINT VERSION

10.1080/02602938.2019.1571161. Long, M. (1996), “The role of the linguistic environment in second language acquisition”, in Ritchie, W.

and Bhatia, T. (Eds.), Handbook of Second Language Acquisition, Academic Press, San Diego,

CA, pp. 413–468.

FOR WEBSITE VIEWER ONLY

Malecka, B., Boud, D. and Carless, D. (2022), “Eliciting, processing and enacting feedback: mechanisms

for embedding student feedback literacy within the curriculum”, Teaching in Higher Education,

RESPECT PUBLISHER RIGHTS

Routledge, Vol. 27 No. 7, pp. 908–922, doi: 10.1080/13562517.2020.1754784.

Page 27 of 55

7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58

Kybernetes

Mizumoto, A. and Eguchi, M. (2023), “Exploring the potential of using an AI language model for

automated essay scoring”, Research Methods in Applied Linguistics, Vol. 2 No. 2, p. 100050,

doi: 10.1016/j.rmal.2023.100050.

Ngo, T.T.-N., Chen, H.H.-J. and Lai, K.K.-W. (2022), “The effectiveness of automated writing evaluation

in EFL/ESL writing: a three-level meta-analysis”, Interactive Learning Environments,

Routledge, pp. 1–18, doi: 10.1080/10494820.2022.2096642.

Nicol, D. (2021), “The power of internal feedback: exploiting natural comparison processes”, Assessment

& Evaluation in Higher Education, Routledge, Vol. 46 No. 5, pp. 756–778, doi:

10.1080/02602938.2020.1823314. Papi, M., Rios, A., Pelt, H. and Ozdemir, E. (2019), “Feedback-Seeking Behavior in Language Learning:

Basic Components and Motivational Antecedents”, The Modern Language Journal, Vol. 103

No. 1, pp. 205–226, doi: 10.1111/modl.12538.

Kybernetes

Ranalli, J. (2018), “Automated written corrective feedback: how well can students make use of it?”,

Computer Assisted Language Learning, Routledge, Vol. 31 No. 7, pp. 653–674, doi:

10.1080/09588221.2018.1428994. Stake, R.E. (1995), The Art of Case Study Research, Sage Publications, Thousand Oaks.

Stake, R.E. (2005), Multiple Case Study Analysis, Illustrated edition., The Guilford Press, New York.

White, J., Fu, Q., Hays, S., Sandborn, M., Olea, C., Gilbert, H., Elnashar, A., et al. (2023), “A Prompt

Pattern Catalog to Enhance Prompt Engineering with ChatGPT”, arXiv, 21 February, doi:

10.48550/arXiv.2302.11382. Wiliam, D. (2018), “Feedback: At the Heart of – But Definitely Not All of – Formative Assessment”, in

POST-PRINT VERSION

Lipnevich, A.A. and Smith, J.K. (Eds.), The Cambridge Handbook of Instructional Feedback,

Cambridge University Press, Cambridge, pp. 3–28, doi: 10.1017/9781316832134.003.

Winstone, N.E., Hepper, E.G. and Nash, R.A. (2021), “Individual differences in self-reported use of

assessment feedback: the mediating role of feedback beliefs”, Educational Psychology,

FOR WEBSITE VIEWER ONLY

Routledge, Vol. 41 No. 7, pp. 844–862, doi: 10.1080/01443410.2019.1693510.

Wu, H., Wang, W., Wan, Y., Jiao, W. and Lyu, M. (2023), “ChatGPT or Grammarly? Evaluating

RESPECT PUBLISHER RIGHTS

ChatGPT on Grammatical Error Correction Benchmark”, arXiv, 14 March, doi:

10.48550/arXiv.2303.13648.

7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58

Kybernetes Page 28 of 55

Wu, T.-T., Lee, H.-Y., Li, P.-H., Huang, C.-N. and Huang, Y.-M. (2023), “Promoting Self-Regulation

Progress and Knowledge Construction in Blended Learning via ChatGPT-Based Learning Aid”,

Journal of Educational Computing Research, SAGE Publications Inc, p. 07356331231191125,

doi: 10.1177/07356331231191125.

Xu, J. (2021), “Chinese University Students’ L2 Writing Feedback Orientation and Self-Regulated

Learning Writing Strategies in Online Teaching During COVID-19”, The Asia-Pacific

Education Researcher, Vol. 30 No. 6, pp. 563–574, doi: 10.1007/s40299-021-00586-6.

Xu, J. and Wang, Y. (2023), “Individual differences in L2 writing feedback-seeking behaviors: The

predictive roles of various motivational constructs”, Assessing Writing, Vol. 56, p. 100698, doi:

10.1016/j.asw.2023.100698. Yan, D. (2023), “Impact of ChatGPT on learners in a L2 writing practicum: An exploratory

investigation”, Education and Information Technologies

Kybernetes

, Vol. 28 No. 11, pp. 13943–13967, doi:

10.1007/s10639-023-11742-4. Yao, Y. and Zhu, X. (2022), “Modeling chinese high school EFL writers’ language mindsets, goal

orientations, and feedback-seeking behavior: the interaction effect between language mindsets

and writing competence”, Reading and Writing, doi: 10.1007/s11145-022-10339-6.

Yin, R.K. (2013), Case Study Research: Design and Methods, Fifth edition., SAGE Publications, Inc, Los

Angeles.

Zhang, S. (2021), “Review of automated writing evaluation systems”, Journal of China Computer-

Assisted Language Learning, De Gruyter, Vol. 1 No. 1, pp. 170–176, doi: 10.1515/jccall-2021-

POST-PRINT VERSION

Zhang, Z. and Hyland, K. (2023), “Student engagement with peer feedback in l2 writing: Insights from

reflective journaling and revising practices”, Assessing Writing, Vol. 58, p. 100784, doi:

10.1016/j.asw.2023.100784.

FOR WEBSITE VIEWER ONLY RESPECT PUBLISHER RIGHTS