Claim Missing Document
Check
Articles

Found 1 Documents
Search
Journal : JoEELE

Human‑AI Feedback for EFL Writing Revision: Accuracy, Agency, and Integrity Amaliah, Suci; Sakkir, Geminastiti
Journal of Excellence in English Language Education Vol 5, No 2, April (2026): Journal of Excellence in English Language Education
Publisher : Program Studi Pendidikan Bahasa Inggris FBS UNM

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.26858/joeele.v5i2, April.84061

Abstract

This mixed‑method study investigates how human-AI collaborative feedback influences the revision of English as a Foreign Language writing. Three research questions examine whether combined feedback improves accuracy and complexity more than AI-only, human-only or minimal feedback; how hybrid feedback shapes learners’ revision agency; and what integrity issues arise. A quasi-experimental design randomly assigned 120 Indonesian undergraduates to four conditions: AI feedback, human feedback, hybrid feedback (AI followed by human) and a control group. Over eight weeks, participants wrote weekly essays; ChatGPT (GPT‑4) provided suggestions on grammar and vocabulary, while experienced lecturers provided holistic comments on organization and argumentation. Writing quality was assessed pre‑ and post‑intervention using an analytic rubric, and participants completed a revision agency questionnaire. Semi‑structured interviews explored perceptions of accuracy, agency and ethical concerns. Quantitative analysis showed that the hybrid group achieved the largest gains, with mean improvements of 20 points in accuracy, 15 points in lexical diversity, and 18 points in organizational coherence. AI-only and human-only groups improved by around 13 points; the control group by six points. An ANOVA confirmed significant differences among groups (p < .05, η² = 0.28). The hybrid condition also yielded the highest agency scores (M = 4.5, SD = 0.3); AI-only feedback promoted autonomy but sometimes overwhelmed learners. Qualitative findings indicated that students valued AI’s immediacy and comprehensive corrections but depended on teachers for higher-order organization and critical thinking. Participants emphasized the need for balanced guidance to avoid over-reliance on AI and to uphold academic honesty. Overall, combining human and AI feedback enhanced writing quality, fostered learner agency and mitigated integrity concerns, yet successful implementation requires ethical training and feedback literacy. These results contribute to EFL feedback research by demonstrating the synergy between AI and human feedback and by highlighting the importance of scaffolding and ethics training for sustainable practice overall.