Intrasentential errors in Chinese as a Second Language (CSL) writing often co-occur non-randomly, yet traditional error analysis has not explained why. We propose that syntactic integration cost, operationalized as Mean Dependency Distance (MDD), is a key mechanism. Planning long-distance dependencies consumes working memory, which in turn triggers multiple errors within a sentence. Using co-occurrence networks from 5718 HSK essays, we employed ordinary least squares (OLS) regression with HC3 robust standard errors and Hurdle models as sensitivity checks. Results show MDD significantly predicts co-occurrence strength in the beginner group (β = 0.512). Advanced learners exhibit a flatter MDD slope (β = −0.271) and a lower overall baseline (β = −0.755), yet their mean MDD is higher (3.28 vs. 3.10). We term this combined pattern conditional stability. This dual mechanism challenges the view that higher proficiency simply implies better local control. Genre significantly affects co-occurrence strength, with the strongest effect in public-issue argumentative essays, but does not moderate the MDD effect. Unlike traditional density, which is sensitive to text length, average degree provides a stable basis for cross-text comparisons. The findings offer a diagnostic framework for writing assessment and automated evaluation.
To date, little research has investigated the combined effects of model texts (MT) and collaborative writing (CW) on L2 learner engagement and text quality. The present study aimed to address this gap by examining how the use of MT affected 104 Vietnamese adult EFL learners’ CW of expository paragraphs, particularly engagement, text quality, and their perceptions of MT. The participants were assigned into two conditions (i.e., MT and non-MT) that included both discussion (15 min) and writing (25 min) stages. Results revealed that MT affected learners’ cognitive engagement, as reflected in language-related episodes (LREs) during the discussion stage. MT also improved rubric-based writing quality in terms of content, organization, vocabulary, and overall performance though no significant effects emerged for CAF measures of text quality. Semantically-engaged talk (indicator of cognitive engagement) positively predicted accuracy, while self-reported engagement was negatively associated with fluency. Additionally, learners valued MT as supportive for L2 writing but voiced concerns about originality, limited feedback, and overreliance on this tool. This study provides empirical evidence on how MT may affect learner engagement in CW, and the findings contribute to advancing ongoing debates about the role of MT in fostering writing development.
Artificial Intelligence (AI) has played an important part in language use and learning for decades. The release of ChatGPT, however, has ushered in an era characterized by both excitement and concern due to its expanded capabilities. To examine what critical and responsible AI-assisted L2 writing could look like, this study explores two concepts in depth: 1) the role of agency in the production of text, and 2) what makes an author in the age of AI. Specifically, the study builds on the premise that the debates around AI offer an opportunity for educators to revisit the foundations of writing: that it is mediated, where meanings emerge from negotiated practices with both the human and nonhuman. Methodologically, twenty-six students from one university in Taiwan participated in this study. The data collected included the students’ compositions and their transparency reports, where they disclosed every interaction with AI and their reflections on this process. The findings show a strong influence of AI on the students’ thinking. In the intense exchanges with AI, the students highlighted the importance of active decision-making, and drawing on their lived experience to strengthen their authorial presence, giving their text a human touch. As a whole, this study demonstrates how a focus on authorship (e.g., cultivating intentionality in communication and developing one’s responsibility as an author) enhances students’ agency in writing. This is not simply about making use of the technological affordances of AI, but to help students reclaim agency by understanding what it means to be an author.
Authorial voice is socially constructed and evolves across different developmental stages. However, a systematic study of voice construction enacted by L2 graduate writers in research genres at their early stage of academic enculturation is still lacking. To fill the gap, we conducted an empirical study to uncover the latent structure of authorial voice construction in graduate students’ writings of research proposal (RP) introduction by a two-stage factor modelling. 17 multi-level voice-related features were first gathered from the relevant literature, and their importance to RP introduction was subsequently confirmed and elucidated by emic interviews with expert writers. These features informed the development of a scale to rate graduate students’ RP introductions. Based on the rated writings, a two-factor structure of voice construction was then identified through exploratory factor analysis (EFA) and validated by exploratory structural equation model (ESEM) in addition to the conventionally used confirmatory factor analysis (CFA). We revealed that voice construction by graduate writers in RP introduction consists of two interrelated factors: language effectiveness and research positioning, while research positioning has a stronger correlation with the holistic impression of voice. Overall, our study enhances the current socio-constructivist conceptualization of authorial voice.
Multivariate approaches to modeling lexical richness and second language (L2) writing proficiency have expanded rapidly, yet this work has focused overwhelmingly on L2 English, and questions about how best to operationalize lexical items in languages with richer morphology remain underexplored. This study builds on existing work by introducing human judgments of productive vocabulary and language use scores for a corpus of L2 Spanish, evaluating different lexical item operationalizations, and examining the relationship between Vocabulary and Language Use scores and indices of lexical richness. The results indicated that optimal operationalizations varied across indices, with lemmas that include full verbal information best capturing lexical diversity, and raw words best capturing most bigram association measures. In linear mixed-effects models, lexical diversity was the strongest predictor of Vocabulary and Language Use scores, while a smaller number of bigram association measures were also significant predictors. These findings suggest that it is beneficial to consider using lexical item operationalizations that reflect the morphology of the target language, and that lexical diversity is a strong predictor Vocabulary and Language Use in L2 Spanish. The study concludes by discussing implications for L2 Spanish instruction and the development of automated productive vocabulary and language use assessment tools.
Emotion labor—the ongoing management of one's emotional experiences and expressions—is central to effective L2 writing instruction and teachers' wellbeing. However, longitudinal qualitative research on L2 writing teachers' emotion labor remains limited, particularly in English-medium instruction (EMI) contexts, and its theorization within L2 writing studies remains underdeveloped. Addressing these gaps, this study employs poetic autoethnography to explore my emotional experiences and emotion labor while teaching L2 writing in a Thai EMI university from 2020 to 2025. Drawing on artifacts from multiple L2 writing courses, I composed 100 poems and present seven that foreground salient emotional experiences, including confusion, fear, indignation, passion, and compassion. These poems are interpreted through a proposed ecological model of L2 writing teachers' emotion labor, comprising four interrelated components: (A) L2 writing–related events and words, (B) teacher emotions, (C) emotion labor practices, and (D) effects on the self and the immediate instructional context, which recursively shape subsequent L2 teaching-related events and pedagogical decisions. The findings show how poetic autoethnography, guided by an ecological model, can generate fine-grained, contextualized insights into L2 writing teachers' emotion labor, thereby extending both theoretical and methodological approaches in L2 writing research.
Writing teacher expertise is a crucial yet under-explored area in L2 writing scholarship. To deepen our understanding of this emerging concept and inform instructional practices, the current qualitative case study investigates how three Chinese university teachers enact their expertise in digital multimodal composing (DMC) within L2 writing classrooms, as well as the factors shaping this expertise. Data were collected from multiple sources, including teachers’ narrative frames, semi-structured interviews, and classroom documents. The findings reveal that L2 teachers’ DMC expertise functions as a multifaceted and complex system, characterized by the interrelation of beliefs, knowledge, and skills. Specifically, teachers held nuanced beliefs about DMC’s role in complementing traditional writing amid institutional constraints, demonstrated DMC-specific pedagogical content knowledge (e.g., multimodal design principles), and developed adaptive skills (e.g., responsive problem-solving and reflection) for implementing and refining DMC tasks. Furthermore, this expertise emerges from a situated negotiation between individual factors (e.g., teacher identities and agency) and contextual forces (e.g., institutional requirements and exam-driven curricula). As an early attempt to explore the enactment of DMC expertise among L2 teachers and the underlying factors influencing this process, this study provides valuable insights that enrich our understanding of teacher DMC expertise and inform writing teacher education.
Generative artificial intelligence (AI) tools are reshaping how voice is understood, taught, and evaluated in second language (L2) writing classrooms. In a recent contribution to the Journal of Second Language Writing, Sandstead and Kibler (2025) propose a model of voice that distinguishes human-written and AI-generated texts through the opposing constructs of authenticity and amalgamation, indexed to differing degrees of writer agency. While their model offers a thoughtful representation of teachers’ conceptions of voice, we argue that it is limited in its ability to support broader theoretical and methodological discussion across contexts. Drawing on recent studies of L2 writers’ writing practices, we show that AI-mediated writing often involves complex, iterative forms of human engagement that cannot be captured by distinctions between human and AI texts or by product-focused notions of authenticity. We further argue that reassigning amalgamation to AI discourse departs from its original conceptualization applicable to all writing. While recognizing the value of teacher knowledge in their own historical and institutional contexts, we call for a more systematic theory of voice and propose an alternative conceptualization of voice as an emergent, reader-mediated effect shaped by varying configurations of human agency, responsibility, and rhetorical control in increasingly AI-mediated writing processes.
This article responds to Kashiha's (2025) recent study comparing instructor feedback and ChatGPT feedback in second-language academic writing. Extending his arguments, this article analyzes epistemic strategies and delivery methods in instructor and ChatGPT feedback by reframing human-AI differences as outcomes of feedback ecologies rather than fixed evaluator types. Its primary contribution lies in shifting attention from comparing feedback sources to calibrating the conditions that shape feedback and governing how evaluative responsibility remains visible in AI-rich classrooms. To operationalize this shift, it outlines stance-aware routines that develop evaluative agency and protect pedagogical accountability: auditing what kind of authority feedback claims, requiring decision traces through criterion-based acceptance/rejection of suggestions with textual evidence, and teaching configuration literacy to elicit prioritized, rubric-aligned AI feedback while reducing template drift. It further develops Kashiha's complementary blended feedback model through a governance-oriented perspective that foregrounds evaluative accountability, stance-aware calibration, decision-trace practices, and configuration literacy as mechanisms for preserving learner agency and instructor authority in AI-mediated L2 writing environments. It concludes that effective blended feedback should be framed as responsibility-sharing, with instructors retaining accountability and students practicing criteriabased judgment in how feedback is interpreted and taken up.
Generative Artificial Intelligence (GenAI) tools offer new possibilities for operationalizing second language (L2) writing revision, a process that requires relatively high levels of self-regulation (SR). However, research on the mechanisms underlying effective revision has been limited by a lack of conceptual clarity and the absence of validated instruments for measuring SR strategies in this context. To address this gap, the present study developed and validated a scale to measure SR in student-GenAI collaborative revision (SGAICR) of L2 writing. This instrument is referred to as the SR-SGAICR scale. The study was conducted in four phases: Phase 1 involved interviews to identify and conceptualize the SR strategy types in SGAICR. Phase 2 focused on item development and evaluation of the scale's content validity. Phase 3 involved exploratory testing with 296 participants to determine the initial factor structure. In Phase 4, the factor structure was confirmed with 510 participants, and the scale's convergent, discriminant, and predictive validity was assessed. Nine strategy types of SR-SGAICR were established and validated: planning, evaluating GenAI's feedback and suggestions, emotional management, self-motivation, modifying prompts, clarification-seeking, note-taking, environmental structuring, and social support-seeking. The scale provides a foundation for future research and interventions to optimize the educational benefits of SGAICR.
This teaching report examines how place-based writing mediates students’ perceptions of authenticity in L2 writing. In many university EFL contexts, writing instruction still relies on decontextualised prompts that limit students’ sense of purpose and audience. To address this issue, this eight-week place-based writing project invited students to investigate their campus environment and compose English interview reports about university community members. Students conducted field interviews, drafted reports, discussed peers’ work, and revised through a rubric-supported process. Analysis suggests that authenticity was not inherent in the task but developed through engagement with place. Drawing on Agnew’s tripartite conceptualisation of place, the study reconceptualises authenticity as a layered experiential process across location, locale, and sense of place. The findings indicate that place-based writing can operationalise authenticity for university students by linking writing to lived experience, social context, and personal meaning, with implications for socially situated, authentic L2 writing in large EFL classrooms.
This qualitative study examines how false-positive AI detection disrupts authorship among 12 university-based TESOL scholars in Thailand whom Turnitin, GPTZero, and iThenticate wrongly flagged. Amidst intensifying Scopus-indexed publication pressures and systemic institutional deficits, the research explores how algorithmic misjudgement affects scholars negotiating highstakes, resource-constrained, non-Anglophone environments. Through semi-structured interviews, reflective journals, and narrative accounts examined through inductive content analysis, three critical findings emerged. First, algorithmic governance functions as epistemic erasure. Opaque detection systems invalidate knowledge claims while precluding authorial contestation. Second, linguistic bias disproportionately flags non-native English writing as artificial, transforming multilingual scholars' legitimate use of tools into evidence of misconduct. Third, participants experienced crises in scholarly identity, questioning acceptable authorship boundaries when digital assistance norms remain undefined. These findings indicate that the boundary between generative and assistive AI use remains conceptually contested and is poorly addressed by current editorial frameworks. Rather than viewing misclassification as a technical error, this study frames it as an institutional failure that replaces human judgement with automated suspicion. It calls for publishing practices grounded in transparency, author dialogue, and recognition of authorship as a contextual human practice, particularly urgent where linguistic diversity intersects with constrained institutional resources and geopolitical hierarchies in global knowledge production.
This quasi-experimental study evaluated whether question-only, dialogic feedback from an AI Socratic assistant can support L2 argumentative writing. Seventy-four first-year EFL undergraduates in Taiwan completed pretest and posttest essays rated on stance clarity, evidence use, and counterargument integration. The treatment group engaged in multi-turn AI questioning during revision, while the control group used conventional peer/teacher feedback. ANCOVA showed no statistically detectable group difference in stance under the present two-point Claims rubric, but significant treatment advantages for evidence use and counterargument integration. Students also reported high perceived usefulness, ease of use, and writing self-efficacy. Robustness checks supported these results. Findings extend L2 writing research by suggesting that dialogic, question-driven feedback may support evidence use and counterargument integration, two higher-order rhetorical moves central to academic argumentation, while stance formulation may require explicit genre-based instruction and more fine-grained assessment. Pedagogically, the study suggests combining early genre analysis, AI-supported revision, and scaffold-fading toward source-based assessment. Limitations include intact classes, short duration, platform specificity, and the limited granularity of the Claims rubric. Future research should examine durability, transfer, discipline-specific effects, and more sensitive measures of stance development.
This longitudinal narrative inquiry examines how a high-achieving Chinese EFL student constructed ecological learning spaces across her three-year undergraduate academic writing trajectory. Drawing on affordance theory, ecological learning space theory, and assemblage theory, the study traces how the student's evolving goals fundamentally shaped her perception and actualization of affordances within hybrid physical-digital environments. Analysis of narrative interviews and extensive archival materials reveals seven actualization strategies through which she transformed perceived affordances into personalized learning spaces: direct utilization, strategic selection, creative adaptation, critical utilization, transformative action, resource coupling, and iterative application. Findings demonstrate that successful learning space construction involves strategic assemblage of institutional, pedagogical, social, technological, and material affordances that generate emergent properties exceeding their individual components. By examining the upper bounds of affordance orchestration in a resource-rich institutional ecology, this study advances ecological-assemblage theories of learner agency while offering a critical framework-attentive to both agentic mechanisms and the structural conditions that enable them-for understanding and scaffolding L2 writing development across Chinese and international EFL contexts.
