The Deontic Gap: Large Language Models and the Modal Language of Obligation
Researchers have found that large language models (LLMs) tend to use deontic modals like 'must', 'should', and 'have to' less frequently than humans do in informal digital contexts. This is evident across multiple datasets and corpora, including a historical comparison with the Google Books Ngram corpus from 1920-2022. The study suggests that LLMs reflect their training on formal written resources rather than contemporary human language patterns.
Researchers have found that large language models (LLMs) tend to use deontic modals like 'must', 'should', and 'have to' less frequently than humans do in informal digital contexts. This is evident across multiple datasets and corpora, including a historical comparison with the Google Books Ngram corpus from 1920-2022. The study suggests that LLMs reflect their training on formal written resources rather than contemporary human language patterns.
---
Why it matters: This research matters to AI engineers because it highlights the limitations of current language models in capturing nuanced aspects of human communication, such as interpersonal obligation and persuasive writing.
Source: https://arxiv.org/abs/2608.18144
This article was originally published at: https://arxiv.org/abs/2608.18144