GAP

该数据集是一个性别平衡的标注语料库，包含了各种具有歧义的代词-名字组合对，这些组合对覆盖了多种具有歧视性的代词。该数据集的设计目的是为了更准确地测量模型实际存在的偏见。此外，它包含了8,908个模糊的代词-名字组合对。该数据集的任务是进行偏见测量。This dataset is a gender-balanced annotated corpus containing various ambiguous pronoun-name pairs that cover multiple discriminatory pronouns. It is designed to more accurately measure the actual biases present in models. Additionally, it includes 8,908 ambiguous pronoun-name pairs, and its core task is bias measurement.