HiTZ/metaphor-llms

该数据集用于研究大型语言模型在隐喻理解方面的性能。数据集包含了多个公开的数据集，用于评估模型在不同数据集、任务和提示配置下的隐喻解释能力。研究结果表明，模型的性能更多受到诸如词汇重叠和句子长度等表面特征的影响，而不是隐喻内容本身。这项工作揭示了当前大型语言模型在处理比喻性语言方面的能力和局限性，并强调了在隐喻解释任务中需要更现实的评估框架。

This dataset is used to study the performance of large language models in metaphor understanding. It includes multiple public datasets for evaluating the models metaphor interpretation capabilities across different datasets, tasks, and prompt configurations. The studys results indicate that the models performance is more influenced by surface features like lexical overlap and sentence length rather than the metaphorical content itself. This work reveals the current capabilities and limitations of large language models in processing figurative language and emphasizes the need for more realistic evaluation frameworks in metaphor interpretation tasks.