LLMs underuse obligation language compared to humans, study finds

A new research paper published on arXiv explores the "deontic gap" in large language models (LLMs), specifically their underutilization of modal verbs indicating obligation and necessity. The study found that LLM-generated text uses positive deontic modals like "must," "should," and "have to" less frequently than contemporary human writing. This usage pattern aligns more closely with formal written English from historical corpora, suggesting LLMs reflect their training data rather than the more immediate, interpersonal modal usage common in current informal human communication. AI

IMPACT LLM communication may lack the nuance of human obligation, impacting their use in persuasive or interpersonal contexts.

RANK_REASON Research paper published on arXiv detailing findings about LLM language usage. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• Google Books Ngram corpus
• Have To
• large-language models
• Must
• Need To

• paper
• safety

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