VALSE

该数据集名为VALSE，主要用于测试模型对否定的理解能力，尤其是在视觉与语言模型如CLIP的背景下。然而，作为衡量语言理解能力的基准，该数据集存在一定的局限性，尤其是在处理否定方面的能力。该任务旨在探究多模态模型对否定的理解。This dataset, named VALSE, is primarily designed to test models' ability to understand negation, particularly in the context of vision-language models such as CLIP. However, as a benchmark for measuring language understanding capabilities, this dataset has certain limitations, especially in its ability to handle negation. This task aims to explore multimodal models' understanding of negation.