A Corpus-Based Discourse Analysis of Interactional Metadiscourse in Human-Written and AI-Assisted Essays

Abstract

The increasing integration of Artificial Intelligence (AI) tools in academic writing has transformed how students compose and revise essays. This study investigates the use of interactional metadiscourse in human-written and AI-generated essays using a corpus-based discourse-analytic approach. Drawing on Hyland’s (2005) interpersonal model of metadiscourse, the study examines five interactional categories: hedges, boosters, attitude markers, self-mentions, and engagement markers. Two specialised corpora comprising human-written and AI-generated argumentative essays were compiled and analysed using AntConc software. The findings revealed significant differences in the frequency and distribution of interactional metadiscourse across the two corpora. Human-written essays demonstrated a greater use of hedges, attitude markers, self-mentions, and engagement markers, reflecting stronger authorial presence, richer interpersonal interaction, and heightened audience awareness. In contrast, AI-generated essays exhibited a more formal, objective, and impersonal rhetorical style characterised by limited writer visibility and minimal reader engagement. Although AI-generated texts displayed grammatical accuracy and coherent organization, they lacked the rhetorical subtlety and interpersonal sophistication characteristic of human-authored discourse. The study concludes that interactional metadiscourse serves as an important linguistic indicator for distinguishing between human-written and AI-generated academic texts. The findings contribute to the growing body of research on AI-mediated academic discourse and offer important pedagogical implications for the responsible integration of AI technologies in academic writing instruction.