Abstract
This study presents a corpus-based multi-dimensional analysis of metadiscourse markers in the academic writing of second language (L2) English graduate students, comparing Science, Technology, Engineering, and Mathematics (STEM) with Humanities disciplines. Utilizing a custom-built corpus of 200 master's theses (totaling approximately 1.2 million words) written by L2 English graduate students, we analyzed the frequency, distribution, and pragmatic functions of interactive and interactional metadiscourse markers based on Hyland's (2005) interpersonal model. Quantitative analysis revealed statistically significant differences in metadiscourse preferences between the two disciplinary domains. STEM writers relied heavily on interactive markers, particularly transitions and frame markers, to establish linear, logical cohesion. Conversely, Humanities writers employed a significantly higher density of interactional markers, such as hedges, attitude markers, and self-mentions, to construct persuasive, dialogic authorial voices. Qualitative analysis of concordance lines further illuminated how L2 writers negotiate disciplinary norms, showing that STEM students occasionally struggled with hedge calibration, while Humanities students sometimes overused self-mentions. These findings offer critical pedagogical implications for English for Academic Purposes (EAP) instruction, emphasizing the need for discipline-specific metadiscourse training to help L2 graduate students navigate the distinct rhetorical expectations of their scholarly communities.
