Is ChatGPT-Produced Text Authentic? A Contrastive Analysis of Cohesive Markers in Human and AI-Generated Text

Abstract

As artificial intelligence (AI) becomes increasingly prominent in education and writing classrooms, this study compares AI-generated text from ChatGPT with human-generated text to evaluate the authenticity of AI-produced content in using cohesive markers such as linking adverbials (LAs). A corpus of online opinion columns from both sources was compiled to examine the use of LAs as cohesive markers. The analysis reveals variability in LA use in semantic and syntactic categories. Additionally, there are instances of LA overuse among human writers. Evidence suggests that AI prefers syntactically complex LA forms over those used by human writers. The findings provide developers with empirical evidence to enhance this AI tool, reflecting the subtle and complex characteristics of human communication and demonstrate how educators can utilize AI, specifically ChatGPT, for language teaching and learning. Implications and suggestions on the utilization of ChatGPT are also provided.

Ähnliche Arbeiten