Speaker
Description
Generative artificial intelligence (AI) tools are increasingly integrated into second language classrooms to provide instant, individualized feedback on spoken performance. However, little empirical evidence exists on whether such technology-mediated feedback benefits learners equitably across proficiency levels, raising concerns that generative AI feedback may inadvertently widen rather than narrow existing achievement gaps. This study examines the effects of generative AI feedback on English speaking performance among students with differing proficiency levels at the Academy of Policy and Development, adopting an equity-oriented lens. A quasi-experimental design was employed with undergraduate students (N = 95) stratified into high-, intermediate-, and low-proficiency groups based on placement test scores. Participants engaged in a structured speaking practice program incorporating generative AI feedback over eight weeks, with oral performance assessed through pre- and post-intervention speaking tests scored on fluency, pronunciation, vocabulary, and grammatical accuracy. Learning engagement was measured through self-report questionnaires and interaction logs capturing the frequency and depth of AI feedback use. Results indicate that while all proficiency groups demonstrated measurable gains in oral performance, the magnitude and nature of improvement varied significantly across groups. Low-proficiency students demonstrated particularly noticeable gains in pronunciation and grammatical accuracy, whereas improvements in fluency and lexical range were more pronounced among intermediate- and high-proficiency learners. Engagement patterns also differed across proficiency groups, with higher-proficiency learners tending to engage more strategically with AI-generated feedback, while lower-proficiency learners showed greater reliance on direct corrective suggestions and more frequent but less elaborated interactions with the AI tools. These differential patterns suggest that access to generative AI feedback alone does not necessarily ensure equitable learning benefits, as learners’ existing linguistic resources may shape their capacity to interpret, evaluate, and act upon AI-generated feedback. These findings contribute to the growing discourse on AI-mediated language learning by highlighting equity implications often overlooked in efficacy-focused research and offer pedagogical recommendations for scaffolding and differentiating generative AI feedback to ensure that learners across proficiency levels can meaningfully engage with and benefit from AI-supported speaking practice.
