Generative AI (GenAI) has transformed L2 writing, producing human-like prose but often impersonal feedback. This study explores the potential of GenAI–human collaborative feedback, focusing on the first author’s experience as a teaching assistant in a Hong Kong public university’s Bachelor of Education (English language track). Grounded in Ecological Languaging Competencies (ELC) and its affordance framework, this study employs an ethnographic approach informed by narrative inquiry and phenomenology. Data were drawn from Zoom tutoring sessions incorporating interview-style questions to investigate participants’ perspectives and experiences, GenAI-student conversation logs, and final assignments in order to analyze two multilingual students’ GenAI–human-mediated L2 writing processes. Findings are organized around three ELC-informed themes: (1) whole-body sense-making and the meshing of first-order languaging and second-order language; (2) individual languaging agency within a distributed ecosystem; and (3) environmental affordances and functional fit. In both cases, GenAI demonstrates consistent limitations in facilitating the situated, embodied, and affectively attuned dimensions of languaging that effective L2 writing entails. This study makes two contributions: it extends ELC’s affordance network to tertiary-level GenAI–human-mediated L2 writing, and it reconceptualizes writerly authorship as a distributed yet agentively orchestrated practice. Moreover, co-agentic GenAI–human feedback foregrounds ecological embeddedness, writerly agency, and ethical GenAI integration.
Traditional second language (L2) writing instruction and assessment frequently emphasize unaided, timed production, a model that no longer fully represents the communicative realities of AI-mediated contexts. This conceptual article aims to reconceptualize the L2 writing construct for educational settings in which generative AI is routinely and legitimately used. The study uses a theory-driven integrative conceptual synthesis. Sources were located through purposive searching of Scopus, ERIC, Web of Science, and Google Scholar, supplemented by citation chaining and journal hand-searching, and screened against stated inclusion criteria across two streams: foundational scholarship on mediated cognition, genre, literacy, and validity, and work on generative AI and writing published from 2020 onward. Forty-seven sources were retained for close analysis, spanning sociocultural learning theory, activity theory, distributed cognition, multiliteracies research, computer-assisted language learning, and language assessment scholarship. Analysis proceeded through manual thematic coding of construct-relevant claims, conducted by the first author and independently reviewed by the second. The resulting orchestration model defines AI-mediated writing as the purposeful coordination of human judgment with machine-generated output under conditions of authorial responsibility. It specifies four interdependent competencies: prompting, critical evaluation, adaptation, and ethical accountability. The analysis shows that traditional dimensions of writing, including coherence, organization, language use, critical thinking, and audience awareness, are not displaced by AI-mediated writing but redistributed across these competencies. The paper also identifies specific challenges for L2 writers, especially the difficulty of evaluating and reshaping fluent AI-generated output in a language still being acquired. The article recommends process-visible assessment designs, genre-specific orchestration tasks, and empirical validation studies that examine construct structure, scoring reliability, and consequential validity.
M. Askari, A. Rahim· Polyglot: Journal of Linguis...· 0 citations
How students and teachers in a Colombian university language program perceived GenAI-supported production was explored and four patterns emerged: GenAI as a rehearsal and revision partner; tension between polished output and language ownership; teacher mediation shifting toward critical language awareness; and the need for explicit ethical and assessment guidance.
Dionelio Jesus Moreno Villalobos· International Journal of AI...· 0 citations
This paper aims to introduce and theorize the ACTS (AI-assisted Collaborative Translanguaging Scaffolding) framework, which is proposed as a middle-range model for regulating human-AI collaboration in proposal writing within the context of English as a Foreign Language (EFL) in Vietnam. Vygotskian Sociocultural Theory, Cognitive Load Theory, and Translanguaging were applied as theoretical frameworks in the study. Based on these frameworks, generative AI is reconceptualized from an autonomous “ghostwriter” into a regulated socio-cognitive partner, which is reported to operate within a “Read-English, Think-Vietnamese, Write-Bilingually” cycle. To explore the plausibility of the proposed framework, illustrative, context-bound evidence from a two-phase Vietnamese pilot was reported, aiming to align with learner experiences rather than to offer full-scale empirical validation. In the first phase, a pre-training survey was administered to a baseline cohort of 112 EFL learners (comprising 71 undergraduates and 41 postgraduates). The analysis of these data indicates the establishment of quantitative and thematic profiles of AI-related anxiety across four competency dimensions, which reveals three convergent clusters: hallucination risk, plagiarism, and critical thinking erosion. Subsequently, the ACTS workflow was applied by a targeted intervention group of 24 postgraduate students to develop their thesis proposals. Findings derived from post-training reflections revealed that a positive psychological shift was reported by 95.8% of the participants. Additionally, an estimated 50% efficiency gain in proposal writing was identified, while a strong commitment to academic integrity was retained. The overall results suggest that these illustrative findings provide preliminary, context-specific support for the ACTS framework, which can be interpreted as a theoretically grounded response to the “Performance Paradox” in AI-mediated EFL academic writing. Based on the outcomes, it is expected that a set of propositions will be generated for future empirical testing in diverse contexts.
Tuyet-Nhung Thi Nguyen, N. Huynh· Journal of Education and Tra...· 0 citations
The brisk integration of Generative Artificial Intelligence (GenAI), particularly Large Language Models (LLMs) like ChatGPT, has fundamentally disrupted educational paradigms, specifically second language acquisition (SLA) and English as a Second Language (ESL) writing instruction. While GenAI tools offer unparalleled scaffolding for grammar, vocabulary generation, and structural organisation, their impact on higher-order cognitive skills remains highly contested. This study evaluates the empirical efficacy of ChatGPT-assisted essay writing in university-level ESL classrooms, focusing specifically on its effects on students’ critical thinking (CT) skills. Using a mixed-methods convergent parallel design over a 15-week academic semester, this study examined $N = 124$ undergraduate ESL students split into an experimental group (utilising ChatGPT as a structured co-writing and dialogue partner) and a control group (utilising conventional process-writing methodologies). Quantitative data were gathered using pre- and post-test assessments scored via a validated Critical Thinking Rubric (evaluating argument analysis, evidence evaluation, alternative perspectives, and bias recognition) along with automated linguistic analysis of student essays. Qualitative data were obtained via semi-structured focus group interviews and student reflective journals. The quantitative findings indicate that while the experimental group demonstrated significant improvements in surface-level linguistic accuracy, lexical density, and structural cohesion, their independent scores in evidence evaluation and independent thesis formulation were statistically lower than those of the control group when ChatGPT assistance was removed in unassisted proctored post-tests ($p < .05$). The qualitative thematic analysis revealed a phenomenon termed "cognitive outsourcing," where students routinely accepted AI-generated assertions without epistemological verification. However, a sub-cohort trained in aggressive prompt engineering and dialectical inquiry demonstrated elevated critical engagement. This paper proposes a “Dialectical AI-Scaffolding Framework” designed to shift ChatGPT from a product-generation tool to an intellectual sparring partner, ensuring that GenAI advances rather than compromises critical thinking in ESL pedagogy.
E. R. B. Raju, E. Namratha· International Journal of Eng...· 0 citations
As generative artificial intelligence (GenAI) becomes increasingly embedded in academic writing, understanding how L2 writers engage with AI-generated feedback and regulate their writing processes is critical. While prior research has focused on learners’ perceptions of GenAI and writing outcomes, little is known about the processes through which learners’ beliefs are enacted in GenAI-assisted writing contexts. Drawing on social cognitive theory and self-regulated learning (SRL) frameworks, this study proposes a process-oriented mediation model in which engagement with GenAI feedback links GenAI writing self-efficacy and writing SRL strategies. Survey data were collected from 564 Chinese non-English-major postgraduate students using GenAI for English academic writing. Structural equation modeling revealed that GenAI writing self-efficacy significantly predicted feedback engagement, which in turn predicted writing SRL strategies. The direct relationship between self-efficacy and SRL strategies became non-significant when engagement was included, indicating full mediation, except for the direct significant relationship between self-efficacy and cognitive strategies. The findings position feedback engagement as a central self-regulatory mechanism in GenAI-assisted writing.
Jiayun Xue, Mei-Yen Chen· Frontiers in Psychology· 0 citations
Amid the ongoing integration of generative artificial intelligence (GenAI) into language education, classroom-based research on how GenAI output is taken up and made meaningful in everyday teaching and learning remains limited. This gap is particularly evident in Chinese-as-an-additional-language (CAL) classrooms in Hong Kong, where ethnic-minority students frequently rely on multilingual and multimodal resources to participate in interaction. Therefore, this study examines how an AI-mediated translanguaging space was constructed interactionally when teachers and students worked with GenAI output during CAL classroom activities. More specifically, it applies a combination of multimodal conversation analysis and interpretative phenomenological analysis to classroom interactions that naturally occurred in 14 CAL lessons, involving one teacher and 15 students, which were video-recorded over two months, alongside GenAI chat logs, fieldnotes, and two semi-structured teacher interviews. The findings indicate that GenAI output became pedagogically consequential through teacher-GenAI-student interaction, specifically through two key interactional phenomena: (1) reorchestrating epistemic authority through group evaluation and (2) enabling creative re-authoring through narrative development. In these ways, GenAI output became a shared object for critical evaluation and creative reworking, contributing to the construction of an AI-mediated translanguaging space in CAL classrooms. Practical implications of the findings are also discussed.
Xinyi Wang, Kevin W. H. Tai, Chin-Hsi Lin· Language Learning & Tech...· 1 citation