語言教育中的AI整合最佳化：評估架構之建構

Optimizing AI Integration: A Framework for Evaluating AI Use in Language Education

本研究探討大型語言模型與生成式人工智慧在語言教育中的應用評估。現有研究多著重比較人工智慧與非人工智慧學習成效，較少檢視其限制與風險，且教師在課程設計前缺乏評估指引。本文先說明大型語言模型與生成式人工智慧的運作及幻覺問題，但不納入資料分析型之機器學習。進一步結合布魯姆分類學、可供性理論與逆向設計，提出人工智慧情境評估框架，包含兩項評量規準，以評估人工智慧在學習任務中的教學效果與適切性。初步測試顯示具良好評分者一致性，對教學設計具實務價值，並可跨領域應用。

The emergence of Large Language Model (LLM) chatbots and Generative Artificial Intelligence (GenAI) has triggered a rapid expansion of research on their use in education, especially in language learning contexts. Most studies compare AI-supported and non-AI-supported learning and tend to report positive outcomes, while fewer critically examine limitations, risks, or unintended consequences. There is also limited guidance for educators and learning designers on how to evaluate AI tools before implementation, particularly during curriculum planning and material design, where pedagogical decisions are most consequential. This study addresses this gap by first outlining key concepts underlying LLMs and GenAI, including how they function and the problem of hallucinations in generated outputs. It focuses specifically on Generative and LLM-based systems as instructional tools, excluding machine learning applications used for data analysis. Because the field is still emerging, both peer-reviewed and selected non-peer-reviewed sources are used. The study develops a conceptual model grounded in Bloom’s Taxonomy, Affordance Theory, and Backwards Design to examine how AI integration can align with established learning theory in instructional design. Based on this, the authors propose an AI Context Assessment Framework with two components: an AI Context Evaluation Rubric and an AI Context Assessment Rubric, which jointly evaluate both the pedagogical effects and the suitability of AI use in specific learning tasks. The framework draws on Bloom’s Digital Taxonomy and an affordance perspective informed by the 6 + 1 Traits of Writing. Pilot testing shows strong inter-rater reliability, suggesting practical value for educators and designers. Although developed in TESOL, TELL, and EMI contexts, the framework is intended to be transferable across disciplines and supports more structured, theory-aligned use of AI in education.