抄録
This study explores the capability of the large language model GPT-4 in generating metaphors and attempts to guide it through the Conceptual Blending Theory. Experimental results indicate that without guidance, GPT-4 tends to produce metaphors with convergence, lacking logicality and creativity. Conversely, the Conceptual Blending Theory can effectively steer GPT-4 towards generating more logical and creative metaphors. By accessing and analyzing information related to the Conceptual Blending Theory, GPT-4 is able to comprehend and apply concepts such as input spaces, generic space, and blended space, thereby devising diverse metaphors. Differences exist in the generation of metaphors between Chinese and Japanese, yet commonalities also emerge, suggesting a consensus in medical knowledge across cultural boundaries. Although some of the metaphors generated by GPT-4 appear far-fetched, they demonstrate a certain level of creative potential. This research underscores that by incorporating cognitive linguistic theories and guidance strategies, the metaphor-generating capability of large language models can be enhanced, opening up new possibilities for their application in fields such as medical and cross-cultural communication.
