A Comparative Study of Thematic Progression in Human-written and ChatGPT-generated Texts

Abstract

The wide use of ChatGPT has rose concerns about accuracy of the essays it produced. This paper compared the thematic choices and thematic progression patterns in human-written and ChatGPT-generated texts based on the systemic functional grammar and analyzed their coherence and discourse structure. The founding revealed that ChatGPT uses more repeated marker Themes in topical Theme making complement as Theme, but less interpersonal Themes than human which indicates its lack of interaction with readers and subjectivity of expressing its opinions. It also uses less various types of textual Themes than human who employs richer ways to extend, combine and conclude the clauses. Thematic progressions are much more frequent in human-written essays and linear pattern accounts for the largest percent while constant pattern is more frequent in ChatGPT. Human can adopt different thematic progressions and other thematic structures to make a coherent text but ChatGPT mostly repeat the same type of clause and even produce redundant information or uses inaccurate thematic progression making the text a list of information.