培養自主學習者：大學EFL教學整合人工智慧強化教學與反思實踐

Fostering Autonomous Learners: Integrating AI-Powered Instruction with Reflective Practice in EFL University Contexts

本研究探討人工智慧強化英語寫作教學（AI-Powered English Writing Teaching, APEWT）結合反思促進機制（Reflective Thinking Promoting, RTP）對於英語作為外語（English as a Foreign Language, EFL）的大學生之寫作能⼒與自律學習發展成效。⽣成式人工智慧橫空出世，對於寫作輔助⽽⾔是前所未有的契機，但也產⽣過度依賴和自主學習能⼒減弱的憂慮。本研究以自律學習模型結合⼈⼯智慧強化英語寫作教學，⽬的在於促進寫作技巧與後設認知的發展。本研究採⽤單組前測後測的準實驗設計，並搭配質性分析，對台灣某間大學英文系45名⼀年級學⽣進⾏⼗八週教學。參與者利⽤ChatGPT參與了四種類型的寫作作業，並藉由結構化規範進⾏批判評估和反思練習。研究資料包括IELTS Writing Task 1前後測、反思日誌、半結構式訪談、寫作樣本。量性分析顯⽰，平均寫作分數僅有輕微且無統計顯著性的提升（前測均值M = 4.58，後測均值M = 4.73），但達到CEFR B2級別或以上者比例從31.1%上升至44.4%。然⽽，質性分析揭⽰出學⽣在AI應⽤與⾃律學習⾏為⽅⾯呈現出具意義的發展模式。學⽣在Zimmerman的三個⾃律學習階段中表現出明顯的進步：在前想階段，從迴避任務演變為以⽬標為導向的規劃，在實施階段發展⾃我監控能⼒，在反省階段建⽴適應性評估策略。交叉驗證的結果顯⽰：在適當的鷹架和反思機制下，RTP-APEWT能促進寫作能⼒和⾃律學習。然⽽，這種教學法並非⼀體適⽤，加上對維持⾃主能⼒的顧慮，在突顯出平衡整合不同教學策略有所必要。本研究提供了以實證為基礎的模型，將⼈⼯智慧融入英語寫作教學，同時保持學習者的⾃主性，對解決台灣2030年雙語政策所缺乏的產出技能提供了實⽤的借鑑。

This study investigates the effectiveness of AI-Powered English Writing Teaching (APEWT) incorporating a Reflective Thinking Promoting (RTP) mechanism on EFL university students’ writing competence and self-regulated learning development. With the emergence of generative AI tools offering unprecedented opportunities for writing support, concerns about over-dependence and diminished autonomous learning capabilities have arisen. This research addresses these challenges by integrating a self-regulated learning model with AI-powered writing instruction to foster both writing skills and metacognitive development. Employing a quasi-experimental single-group pretest-posttest design combined with qualitative analysis, the study examined 45 first-year English majors at a Taiwanese university over an 18-week intervention. Participants engaged in four genre-based writing assignments while utilizing ChatGPT through structured protocols that emphasized critical evaluation and reflective practice. Data collection included IELTS Writing Task 1 assessments, reflective journals, semi-structured interviews, and writing samples. Quantitative analysis revealed modest, nonsignificant gains in mean writing scores (pretest M = 4.58, posttest M = 4.73), though the proportion of participants achieving CEFR B2 level or above increased from 31.1% to 44.4%. However, qualitative analysis revealed meaningful developmental patterns in students’ AI utilization and self-regulated learning behaviors. Students demonstrated progression across Zimmerman’s three SRL phases, evolving from task-avoidance to goal-oriented planning in the forethought phase, developing self-monitoring capabilities in the performance phase, and establishing adaptive evaluation strategies in the self-reflection phase. The triangulated findings indicate that the RTP-APEWT approach observably promotes both writing competence and self-regulated learning when implemented with appropriate scaffolding and reflective mechanisms. However, individual differences in response to the intervention and concerns about maintaining autonomous capabilities highlight the need for balanced integration. This study contributes an evidence-based model for incorporating generative AI into writing instruction while preserving learner autonomy, offering practical implications for addressing the productive skills gap identified in Taiwan’s 2030 Bilingual Policy.