以AI對話式學習引導機制融入西拉雅語VR教材對不同程度學生口說學習成效之影響

Effects of Artificial Intelligence–Driven Dialogic Learning Guidance in a Siraya Virtual Reality Platform on Oral Performance Among Students of Different Proficiency Levels

臺灣平埔族歷史、語言與文化嚴重流失，學生學習時面臨口說焦慮、缺乏語境之困境，母語教學亦面臨學生程度差異大之問題，亟需差異化學習教材回應師生需求。據此，本研究旨在發展一套AI對話式學習引導機制融入西拉雅VR教材，並探討其對學生口說學習成效之影響。研究採準實驗設計，以臺南市某國小高年級20位學生為研究對象，並以AI助教介入與否分為AI-SVR組（AI-based Dialogic Learning Guidance + Siraya VR Materials）與SVR組（Siraya VR Materials only）。研究結果顯示，兩組在整體口說成效無顯著差異，但AI-SVR組高分群學生表現顯著優於SVR組，其更能有效內化句型結構，並產出多樣語料；而兩組低分群則未呈現顯著提升。整體而言，本教材能支持不同程度學生之語言學習，AI助教有助提升高成就學生語言內化與輸出，為本土語言教育提供差異化教學模式與教材之參考。本研究根據結果提出若干建議，對少數語言復振具實務價值。

Taiwan’s Pingpu indigenous communities face severe erosion of their history, language, and cultural heritage because of generational shifts and the influence of dominant cultural hegemony. Students engaged in mother tongue instruction often encounter speaking anxiety and a lack of authentic communicative contexts. Furthermore, teachers struggle with major language proficiency discrepancies among students, underscoring the urgent need for differentiated instructional materials. In response to these pedagogical challenges, the present study integrated an artificial intelligence (AI)-based mechanism for dialogic learning guidance into a virtual reality (VR) platform tailored for Siraya language instruction and investigated its effects on students’oral language performance. A quasi-experimental design was implemented with 20 upper-elementary students in Tainan. Participants were assigned to one of two groups: one group received AI-based dialogic learning guidance + Siraya VR materials (the AI-SVR group), and the other received Siraya VR materials only (the SVR group). Although no significant overall differences in oral performance were observed between both groups, high-achieving students in the AI-SVR group significantly outperformed their peers in the SVR group, demonstrating greater internalization of target sentence structures and producing a wider variety of utterances. By contrast, oral performance among low-achieving students in both groups did not improve significantly. These findings suggest that although VR-based materials supported language learning among students of all proficiency levels, the AI dialogic tutor was particularly effective in enhancing language internalization and oral output among high-achievers. This study offers a practical reference for the development of differentiated instructional models and materials within heritage language education and presents valuable insights for endangered language revitalization efforts.