ANALYZING LEARNER INTERLANGUAGE DISCOURSE WITH AI: FACILITATING NOTICING IN THE EFL CLASSROOM

Abstract

Schmidt’s (1990, 2001) Noticing Hypothesis proposes that conscious awareness of gaps in output is necessary for second language development. This classroom-based action research study operationalizes that principle in spoken EFL by integrating structured discourse analysis with AI-supported feedback. Twenty-two Japanese university students recorded five-minute English conversations at the beginning and midpoint of a seven-week applied linguistics course. Using predefined analytic definitions, students manually analyzed transcripts for macro-structure, speech-act “interacts,” exchange patterns, and embedded story sub-genres before comparing their analyses with AI-generated categorizations using standardized ChatGPT prompts. Quantitative summaries served as descriptive indicators of change, while structured reflections provided primary evidence of noticing. Across the instructional cycle, conversations showed a modest increase in total word production (6,951 → 7,287 words overall), increased follow-up information and clarification moves, and more frequent and more deliberately structured storytelling (7 of 22 → 20 of 22 students). Reflections indicated heightened awareness of pre-closings, turn balance, limited elaboration, and narrative sequencing. A post-course questionnaire (N = 22) further indicated that learners perceived AI as a useful analytic mediator, though requiring critical oversight. Rather than claiming causal effectiveness, the study documents how AI-mediated discourse analysis can make conversational architecture visible and support structured noticing within communicative classrooms. The findings highlight the pedagogical potential and practical limitations of integrating generative AI into discourse-focused EFL instruction.