“This dissertation intricately explores…”: ChatGPT's shell noun use in rephrasing dissertation abstracts

Abstract

ChatGPT has emerged as a prominent tool for supporting students' academic writing. However, research on its linguistic attributes in academic writing, especially its use of shell nouns—a crucial linguistic tool reflecting authors' rhetorical awareness in academic writing—remains limited. Our study tries to bridge this gap by evaluating ChatGPT's use of shell nouns when rephrasing dissertation abstracts in the fields of physics and economics. Our corpus comprises 400 abstracts: 200 human-generated and 200 ChatGPT-rephrased. Analysis reveals ChatGPT's overall preference for shell nouns, particularly in Move 5 (the move concerning the discussion of research contributions). Structurally, ChatGPT's versions employ significantly more th-N (cataphoric) shell nouns in both physics and economics while using significantly fewer th-N (anaphoric) and N-be-cl shell nouns in economics. Functionally, ChatGPT's versions utilize significantly more text nouns in both disciplines but fewer evidential nouns and relation nouns in economics. These findings suggest that ChatGPT-rephrased versions are more promotional and repetitive, with reduced authorial visibility. Additionally, ChatGPT relies less on shell nouns to establish cohesion and casual links in social sciences. This comparative study advances the current limited understanding of ChatGPT's linguistic attributes in academic writing and provides pedagogical implications for practitioners integrating ChatGPT into classroom instruction.