Humans use figurative language / analogy to communicate their ideas (legal text is a great example of this!). Can models employ the same figurative language? Is it sensible?
If you are inspired by this idea, you can reach out to the authors for collaboration or cite it:
@misc{heineman-how-do-llms-2025,
author = {Heineman, David},
title = {How do LLMs use figurative language?},
year = {2025},
url = {https://hypogenic.ai/ideahub/idea/IveG63VYqUv3sWPvk1sc}
}Please sign in to comment on this idea.
I find this line of thinking very interesting. In fact, I believe that models are quite adept at using metaphors, but humans currently seem unable to accurately assess the correctness of these metaphors. I have recently been working on a project on concise summary generation, aiming to transform professional medical literature abstracts into simplified text summaries to improve accessibility of specialized medical knowledge for the general public. In experiments with LLMs, I noticed that for certain specialized medical terms or mechanisms, the models tend to employ vivid metaphors. However, determining whether these metaphors accurately convey the intended knowledge without introducing ambiguity is particularly crucial in the medical domain.