The article details the creation of V-FLUTE, a comprehensive dataset aimed at enhancing the understanding of figurative language by AI models. It draws from existing multimodal datasets covering metaphors, similes, idioms, sarcasm, and humor. By utilizing expert human annotators in the transformation process, V-FLUTE establishes a high-quality benchmark combining visual elements with linguistic captions to assess how well AI can interpret these complex expressions. The dataset's construction ensures a unified format, facilitating a thorough evaluation of AI capabilities in handling figurative content.
In developing V-FLUTE, we merge existing figurative datasets with human-AI collaboration to create a benchmark that enables AI models to understand visual entailment and figurative language.
The V-FLUTE dataset bridges the gap between visual input and linguistic subtleties, focusing on metaphors, similes, idioms, sarcasm, and humor in a structured format.
Collection
[
|
...
]