Ai generated research texts are less flexible than human writing
Does AI-Generated Scientific Text Follow Human Argumentation Patterns? A CARS-Based Comparison of Research Article Introductions
Computation and Language
Summary
Large language models now help write scientific papers, but this study looked at how their writing compares to human writing in research article introductions. Instead of focusing on word choice or style, the authors studied the structure of arguments, following a model called CARS. They found that human writers are more creative and flexible in how they build their case, while AI-generated texts follow a more rigid, uniform pattern. Giving AI models explicit instructions about the structure made their writing even less flexible.
What this means in practice
- •For scientific publishers: Detect AI-generated scientific texts by their more rigid and uniform argument structure compared to human-written introductions.
- •For academic writing software makers: Improve AI writing assistants by incorporating flexible argumentation patterns that better mimic human research article introductions.
Authors
Abdelrahman Sadallah, Narjes Sheikh Asadi, Lonneke van der Plas
Abstract
Large language models are moving from helping write up research to helping do it, which makes it important to know how the scientific text they produce differs from human writing. Work on this question has stayed mostly at the surface, using lexical and stylistic cues that light paraphrasing erases. We look instead at rhetorical structure, the sequence of argumentative moves through which a text makes its case. We study research-article introductions under Swales' CARS model, and compare original introductions from published linguistics articles with generated counterparts of the same papers. We find that human-written introductions are more flexible in which moves they use and in what order, while the generated ones are more uniform. Giving the models the CARS definitions makes them more rigid.