Comparing AI-Generated Feedback Modes in EFL Speaking: The Role of Interactivity and Embedded Teacher Expertise
摘要
With the integration of generative artificial intelligence into EFL speaking instruction, AI-generated feedback (AIGF) has diversified in terms of interactivity and the extent of embedded teacher expertise. Although the pedagogical value of AIGF is well established, limited research has examined how these two dimensions jointly shape learners’ speaking development and willingness to communicate (WTC). Grounded in the Interaction Hypothesis, this quasi-experimental mixed-methods study investigated the effects of three AIGF modes differing in interactivity and embedded teacher expertise over a four-week intervention. Sixty-two Chinese university students were assigned to one of three feedback modes: (1) EAP Talk, a low-interactivity, assessment-oriented system integrating teacher-informed feedback through structured scoring and diagnostic guidance; (2) Doubao, a high-interactivity, open-ended dialogic system emphasising learner-initiated expression and meaning negotiation; and (3) Doubao Agent, a moderate-interactivity, pedagogically guided dialogic system balancing instructional scaffolding with conversational flexibility. Pre- and post-tests, together with measures of WTC and technology acceptance, showed that EAP Talk was particularly effective in improving grammatical accuracy and in-class WTC. Weekly structured reflective responses further suggested EAP Talk supported a feedback–revision–repractice pattern, characterised by problem noticing, output revision, and repeated practice. In contrast, Doubao supported idea generation and fluency in a low-pressure dialogic environment, while Doubao Agent demonstrated a partial balance between form-focused guidance and interactional flexibility. Overall, the findings suggest that interactivity alone does not determine effectiveness; rather, learning outcomes depend more critically on the alignment between feedback design, pedagogical structuring, and embedded teacher expertise.