Metaphors and Analogies in the Context of Large Language Models

Metaphors are figures of speech that connect new concepts with well-known ones and they help scientists look at familiar things in a different way, and thanks to that new discoveries are made. Automated detection of metaphors is a challenging task because words are context-specific and different from their literal definition. The article takes this special set of metaphors as its case study. The authors conducted two experiments to discover if large language models (LLMs) are able to detect them and explain what they mean. It was concluded that the task of metaphor-identification requires comprehensive analysis of word phrases. Besides, LLMs should be trained on datasets containing extralinguistic information about the world in general. The task of explaining metaphors explanation seemed to be less difficult. Two language models managed this task, however, further research is required to improve their nonliteral reasoning capabilities to interpret figurative phrases with different meanings.