Decoding Metaphors: A Multi-Model Approach Using Abstract Meaning Representation and FrameNet

This research explores the integration of Large Language Models, Meaning Representation and FrameNet to enhance the identification of metaphorical language. Given the inherent ambiguity and complexity of metaphors, this study focuses on developing a robust framework to address the challenges in detecting metaphorical expressions. This research proposes a systematic approach that determines whether a given sentence contains a metaphor by comparing its actual abstract meaning representation graph with an expected literal representation, leveraging the semantic insights provided by FrameNet and the contextual embeddings generated by large language models. This research validates the proposed methodology against multiple metaphor-rich datasets, demonstrating significant improvements in metaphor identification accuracy. Research findings reveal that the synergistic use of Meaning Representation, FrameNet, and Large Language Models can effectively capture the nuances of metaphorical language, paving the way for advancements in computational semantics and practical applications in areas such as sentiment analysis and automated content generation. This study contributes to the ongoing discourse on metaphor identification by providing empirical evidence of the effectiveness of integrated models in understanding figurative language.