Akademik Veri Yönetim Sistemi
17th CHINA TO ADRIATIC TURKISH WORLD INTERNATIONAL SCIENTIFIC RESEARCH CONGRESS, Van, Türkiye, 5 - 07 Aralık 2025, ss.233-244, (Tam Metin Bildiri)
The integration of Large Language Models (LLMs) into academia has formed a "grey zone" of authorship between humans and machines. The purpose of this study is to examine this "space in between" through a comparison of corpus studies of stance, hedging, and lexical bundles in English Language Teaching (ELT) writing. The research is based on a specialized corpus consisting of 90 texts, which have been divided into Human Corpus (HC), AI-Generated Corpus (AIC), and AI-Likely Corpus (ALC) categories, with which the study demonstrates that AI-assisted writing produces a novel ‘synthetic’ register. The research demonstrates that a core difference between the texts is the presence of authors. In human-authored texts the presence of a self was used to signal a root modality (ability) and a personal recount of a story. In contrast, AI-likely authored texts self-mention authors without real stance and instead used the self-mention for structural purposes and thus had a "hallucination of voice." Moreover, hybrid texts signal an "epistemic shift" toward a vague possibility of improvement and a "listicle effect" which describes a disproportionate use of fixed and rigid enumerative structures with frames of justification. These results point to a concerning homogenization of student discourse, suggesting that assessment frameworks must be revised to prioritize the functional nuances of organic human authorship over the algorithmic structural rigidity of AI-mediated texts.
Keywords: AI writing; AI-likely text, corpus linguistics; hedging; lexical bundles; ELT discourse