Intersubjectivity as the distinguishing feature or common ground: A contrastive study between human-written abstracts and LLM-generated abstracts

The vast knowledge and great efficiency of Large Language Models have made it essential to demystify the language written by human beings and that generated by LLMs, particularly their interactional capability. As a pivotal genre in academic discourse, “abstract” serves both the informative and the communicative functions. This study aims to explore intersubjective discourse markers manifested in human-written abstracts and LLM-generated ones as well as their similarities and differences across disciplines. The results show that human writers employ a wider range of stance and engagement markers to facilitate intersubjective positioning. Human-written abstracts exhibit a more sophisticated linguistic realization of intersubjectivity through lexical resources, patterned phrases, and syntactic structures. The correspondence analysis reveals that human writers emphasize disciplinary distinctions, while LLMs adopt a convergent approach to achieving writer-reader interaction among different disciplines. These findings underscore human writers’ superiority in navigating complex writer-reader interaction in abstract writing. Though LLMs have demonstrated some potential in emulating intersubjective communication, their interactional capability falls short of that of human writers. The findings offer significant implications for deepening our understanding of the nature of LLMs and contributing to LLM-assisted EAP studies and EAP teaching across disciplines.