科技增強語言學習觀點下的對話互動：AI聊書同伴與真人教師在口語鷹架、回饋機制與認知歷程之比較

Dialogue Interaction from the Perspective of TELL: A Comparison of Oral Scaffolding, Feedback Mechanisms, and Cognitive Processes Between AI Book Companions and Human Teachers

隨著生成式人工智慧與大型語言模型的快速發展，科技增強語言學習已從單向練習轉向具備高度互動性的對話式學習。其中，運用人工智慧作為學習同伴進行「聊書」活動，成為促進學生口語表達與閱讀反思的重要策略。然而，人工智慧學習同伴與真人教師在對話引導特質上的差異，及其對學生認知歷程的影響，仍待深入探究。本研究採用Chin提出的「以提問為基礎的對話」分析框架，比較兩者在互動結構、回饋策略與學生認知層次上的差異。研究對象為37名國小高年級學生，進行為期四個月的實徵研究。研究結果顯示：一、在互動與回饋策略上，人工智慧同伴展現高度結構化特徵，傾向使用「肯定兼直接講授」與「回饋—提問」（F-I）循環，能提供穩定的語言鷹架並維持對話流暢度，營造出低壓力的對話情境；真人教師偏好「啟發式引導」與「聚焦式提問」，展現較高的教學彈性；二、人工智慧同伴雖能提升學生的語用複雜度及參與動機，但學生多停留在基礎認知層次；真人教師雖互動頻率較低，卻能激發學生進行推論、評估等高階批判性思考。本研究據此提出「人機協作的語言教學模式」：建議將人工智慧同伴定位為「語言流暢度與基礎概念」的練習同伴；真人教師則專注於「深度認知與批判思考」的引導。

With the rapid development of generative artificial intelligence and large language models, technology-enhanced language learning has shifted from one-way practice toward highly interactive dialogic learning. In this context, using artificial intelligence as a learning companion to engage students in “book talk” activities has become an important strategy for promoting students’ oral expression and reading reflection. However, the differences between AI learning companions and human teachers in their dialogic guidance characteristics, as well as their influences on students’ cognitive processes, remain underexplored. This study adopts Chin’s framework of questioning based discourse to compare the differences between the two in terms of interaction structure, feedback strategies, and students’ cognitive levels. The participants were 37 upper-grade elementary school students who took part in a four-month empirical study. The findings reveal two major results. First, in terms of interaction and feedback strategies, the AI companion demonstrated highly structured characteristics. It tended to use “affirmation with direct instruction” and a “feedback-questioning (F-I)” cycle, providing stable language scaffolding, maintaining conversational fluency, and creating a low-pressure dialogic environment. In contrast, human teachers preferred heuristic guidance and focused questioning, showing greater instructional flexibility. Second, regarding students’ cognitive performance, although the AI companion helped enhance students’ pragmatic complexity and participation motivation, students tended to remain at basic cognitive levels. Human teachers, despite lower interaction frequency, were more capable of stimulating students’ higher-order critical thinking, such as inference and evaluation. Based on these findings, this study proposes a human-AI collaborative language teaching model. It suggests positioning AI companions as practice partners for language fluency and basic concept development, while human teachers focus on guiding deep cognition and critical thinking.