NLP papers increasingly use "rather than" phrasing, potentially due to AI assistance

A recent study published on arXiv has identified a significant increase in the use of "antithesis" phrasing, particularly the construction "rather than," in natural language processing (NLP) papers. This trend is more pronounced in papers from 2026 compared to 2019, and even more so in papers generated with the assistance of AI models like GPT. Annotators found that while the "rather than" construction in older papers was rarely annoying, a substantial portion of recent papers contained uses that were perceived as irritating, often by presenting the rejected alternative unfavorably. Researchers suggest this may be a side effect of post-training methods that reward disavowal in single responses without considering the broader cost across a text. AI

IMPACT Highlights a potential stylistic artifact of current LLM writing assistance that may negatively impact scientific communication.

RANK_REASON Academic paper analyzing linguistic trends in NLP research. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• arXiv
• Association for Computational Linguistics
• generative pre-trained transformer
• Hugging Face
• natural language processing

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →