AI-MEDIATED FEEDBACK AND IELTS SPEAKING PERFORMANCE: AN INVESTIGATION INTO STUDENT PROGRESS AND PERCEPTIONS

Abstract

This study examines the potential impact of a human-AI collaborative feedback system on the IELTS speaking abilities of upper-intermediate English majors at a university in Vietnam. An explanatory sequential mixed-methods design was used, with the research focusing on the students’ oral performance over a 4-week formative assessment period. The quantitative findings, derived from a pre-test and post-test conducted with a single group as a quasi-experimental study and from a 15-item Likert-scale questionnaire completed by the participants (N = 76), were supplemented by the qualitative responses to three open-ended survey questions. The quantitative data indicated a statistically significant improvement only in the area of Lexical Resource, whereas the overall band scores did not show a significant change, likely due to a temporary compromise between linguistic complexity and oral spontaneity. The survey results revealed that the learners regarded score breakdowns based on criteria as highly valuable and also placed importance on justifications of their transcripts that were supported by evidence in order to encourage self-regulated learning. At the same time, the qualitative responses emphasized that ongoing involvement by human experts is still necessary for overcoming affective barriers and for ensuring evaluative reliability. The study therefore proposes a Human-in-the-Loop (HITL) framework that combines the accuracy of automation with the guidance of pedagogy and thus complements standalone AI evaluation tools in the context of language assessment.