Lexical Bundle Structure and Function of ChatGPT-Generated Essay: Corpus-Based Study of Advanced Learners in Indonesia

Abstract

Numerous studies have primarily focused on investigating the linguistic structures of academic writing, such as theses, dissertations, and articles from reputable journals authored by individuals ranging from novices to experts. However, few studies have examined the linguistic structures of texts written by AI, particularly those produced by ChatGPT. This study aims to analyze the lexical bundle structure and function variations of 10 argumentative essays generated by ChatGPT, which were composed based on prompts provided by advanced learners in Indonesia, specifically graduate students of applied linguistics from Yogyakarta University. The objective is to explore the lexical bundle structure and function of ChatGPT-generated essays to understand how the AI employs various linguistic forms and constructs relationships between sentences. For data collection, this study utilized the corpus tool AntConc's N-gram feature to refine the four-word lexical bundles from the collected essays. The analysis used the frameworks of Hyland (2008b) and Biber et al. (1999). The result showed the Noun Phrase + pattern as the most frequent lexical bundle structure found in ChatGPT-generated essays. This structure has been used primarily to assert ideas and positions on the discussed topics (Stance Features), marking that ChatGPT has a strong ability in presenting and elaborating ideas. However, the least frequently found lexical bundle function is text construction that can link one idea to others to create the connection. Therefore, this study suggested that ChatGPT-generated essays can assist in effectively conveying the writer's perspectives but need attention or development in creating cohesive links between ideas in written discourse.

Ähnliche Arbeiten