| 摘要: | 隨著人工智慧（AI）技術的快速發展，許多研究已經證實其在教育中的優勢，特別是在英語學習領域。常見的應用技術包括影像辨識、語音辨識與文字生成。過去研究指出，影像辨識能協助學生將英語學習延伸至真實情境，然而，學生往往難以在真實情境中即時查閱單字或撰寫句子；語音辨識可提供即時口說回饋，增進口語表現，但多數研究僅分別探討 text-to-speech (TTS) 與 speech-to-text recognition (STR)，缺乏整合性研究；文字生成則能產生練習題與訂正回饋，協助學生修正錯誤並支援教師與家長給予建議，但相關研究多聚焦於學生本身，較少兼顧親、師、生三方的互動。同時，許多研究建議在使用AI的過程中應加入人為監控。
為解決上述問題，本研究開發EduCommAI，融合情境學習、親師生溝通回饋機制、親子共學與口說回饋模組。系統可透過影像辨識分析學生拍攝生活情境照片，生成具情境關聯之多樣練習題，強化學習與生活的連結；回饋模組依據學生表現自動生成草稿，供教師與家長修訂，進而提供個別化建議並促進親師生溝通；親子共學則鼓勵家長將日常經驗轉化為學習素材，提升親子互動與學習深度；另由口說回饋模組即時指出錯誤並給予修正建議，以提升學生英語口語能力。此外，本研究引入Human-in-the-Loop(HITL) 設計理念，學生會檢查影像辨識結果是否符合其上傳圖片，而教師與家長則會檢核回饋草稿內容，以確保其準確性。
本研究國中學生為實驗對象，分為三組進行為期十週的教學實驗。第一階與第二階段分別探討聚焦探討AI生成回饋與親子共學對親師生互動與學習的影響。第一階段研究結果顯示，接受教師與家長親筆回饋的實驗組二，其學習成就顯著優於接受AI生成回饋的實驗組一。雖然實驗組一所獲回饋在內容複雜度上較高，但情感支持度明顯低於實驗組二。此結果突顯人工撰寫回饋在情感支持層面的優勢，即便家長僅提供簡短訊息，仍能透過關懷與鼓勵有效提升學習動機與表現。教師則可專注於提供具專業性的學習指導，實現人機協作下的角色分工，共同促進學習成效。此外，兩組皆使用口說回饋糾正模組，在口說表現上均優於未使用該模組的控制組。
第二階段結果顯示，實驗組一與實驗組二在後測筆試成績無顯著差異，顯示親子共學模組有助於提升學習表現，縮小第一階段的成就差距。進一步分析亦顯示，後測筆試成績與親子共學的完成次數與活動時長皆呈顯著正相關，表示參與次數越多、時間越長，學習成效越佳。此外，兩組實驗組在第二階段持續使用口說回饋糾正模組，口說成就皆顯著優於控制組。
最後，透過問卷與訪談發現，親師生對本系統與學習活動持正向態度。最終本研究發現在親師合作下，透過教師提供專業建議、家長給予情感支持的明確分工，可有效提升學生的學習動機，進而促進其學習成就。
;With the rapid development of artificial intelligence (AI), English learning has shown new trends, with image recognition, speech recognition, and text generation as core applications. Image recognition embeds learning in authentic contexts, speech recognition provides real-time oral feedback, and text generation supports learning through corrective exercises and parent guidance. However, few studies explore how AI fosters interaction among teachers, students, and parents. Human-in-the-loop (HITL) highlights collaboration between humans and AI, ensuring accuracy and maintaining human-centered care. In this study, HITL was applied in two ways: students verified whether system-generated vocabulary matched their uploaded images, and teachers and parents reviewed feedback drafts for accuracy. These processes added professional guidance and emotional support, strengthening family–school collaboration.
To address these issues, this study developed EduCommAI, a system integrating situated learning, interactive feedback, parent-child co-learning, and a speaking correction module. Image recognition generated context-based practice items, verified by students before use. The feedback module produced drafts refined by teachers and parents to provide personalized suggestions and improve communication. The co-learning module encouraged parents to transform daily experiences into learning materials, while the oral feedback module offered real-time error detection to improve speaking ability.
A ten-week experiment with junior high school students compared three groups. In Stage 1, the group receiving teacher- and parent-written feedback outperformed the AI-feedback group, showing the advantage of emotional support, even from short parental messages. Both groups using the oral feedback module achieved higher speaking performance than the control group. In Stage 2, no significant difference appeared in test scores between groups, indicating that the co-learning module improved achievement and reduced the Stage 1 gap. Participation frequency and time correlated positively with post-test scores, and both groups continued to show stronger speaking performance than the control group.
Questionnaires and interviews further revealed positive attitudes from teachers, parents, and students. The findings suggest that clear role division, with teachers providing professional guidance and parents offering emotional support, effectively enhances student motivation and learning achievement through human–AI collaboration. |
