Grammatical Categoricity and Contextual Variability: A Corpus-Based Study of AI-Generated and Human-Written English in Academic Prose

The rapid integration of generative artificial intelligence (AI) into academic writing has prompted urgent questions about the linguistic character of machine-produced text and its relationship to established norms of scholarly English. This study investigates whether AI-generated academic prose exhibits distinctive grammatical and lexical patterns when compared with human-written texts, and whether such patterns vary according to disciplinary context. Drawing on a corpus of 120 academic abstracts (60 generated by a large language model, 60 authored by human scholars) across three disciplines, linguistics, literary studies, and the natural sciences, the study employs a mixed-methods design informed by Systemic Functional Linguistics (SFL) and register theory. Quantitative analysis examines syntactic complexity, lexical diversity, nominalisation density, and modality distribution; qualitative analysis explores thematic progression and hedging strategies through close textual examination. Findings indicate that AI-generated texts display significantly higher nominalisation density and more uniform modality profiles, while human texts exhibit greater interpersonal variability and discipline-specific hedging. The study argues that AI-generated English constitutes a recognisable register rather than a neutral imitation of human academic style, with implications for academic integrity, writing pedagogy, and the sociology of scholarly communication. It concludes by proposing a descriptive framework for identifying register markers in machine-generated academic discourse.