AI Can Interpret Complex MetaphoresPublished on Sat Aug 12 2023 by Dustin Van Tate Testa Life Is A Highway | JD Hancock on Flickr
New research has found that large language models (LLMs) have the ability to interpret complex, novel metaphors. LLMs are artificial intelligence systems that have been trained on vast amounts of text data to perform various language processing tasks. While they have shown impressive performance in areas such as language comprehension and reasoning, their ability to understand creative human abilities, such as interpreting metaphors, has been in question.
In this study, researchers assessed the performance of GPT-4, a state-of-the-art LLM, in interpreting literary metaphors drawn from Serbian poetry and translated into English. Despite not being exposed to these specific metaphors during training, GPT-4 consistently provided detailed and insightful interpretations that were rated as superior to those provided by a group of college students.
The findings suggest that GPT-4 has acquired an emergent ability to interpret complex novel metaphors. This is significant because the interpretation of metaphors is considered one of the highest cognitive abilities in humans, involving sophisticated analogical reasoning. By demonstrating this ability, GPT-4 brings us closer to achieving human-like language understanding in artificial intelligence.
However, it is important to note that while GPT-4 can generate interpretations of novel metaphors, it does not necessarily have the ability to create genuinely new metaphors. The metaphors it generates are limited to those that humans have already formed and planted in texts. The study highlights the distinction between generating interpretations of metaphors and generating entirely new metaphors.
This research contributes to our understanding of the capabilities of LLMs and their progress towards human-like language abilities. While there is still much to learn about the inner workings of these models, the results showcase their potential in understanding and processing complex aspects of language, such as metaphors.