From Semantic Frames to Metaphor: A Data-Driven Generative Approach

This research introduces a novel methodology for metaphor generation, combining frame semantics, semantic role labeling, and advanced language models to enhance creativity, semantic coherence, and contextual relevance. Using resource such as FrameNet the approach integrates frame-based contextualization and paraphrasing techniques into a unified pipeline. Extensive experiments on datasets such as the VUA Metaphor Corpus, Metaphor Knowledge Graph and MOH-X demonstrate significant improvements over baseline models in metrics such as BLEU, ROUGE, and human evaluations. Case studies highlight the critical role of frame conditioning in generating diverse domain-specific metaphors. Case studies further underscore the ability of the model to produce novel and meaningful metaphors in domains such as business, education, and politics. This work establishes a robust framework for metaphor generation, with potential applications in creative AI, education, and natural language understanding, while identifying challenges such as scalability and ambiguity handling for future exploration.