FairLex

FairLex是一个多语言法律文本处理公平性评估基准，包含四个数据集，覆盖欧洲理事会、美国、瑞士和中国四个司法管辖区，以及英语、德语、法语、意大利语和中文五种语言。数据集关注性别、年龄、地区、语言和法律领域等五个敏感属性，旨在评估预训练语言模型及其微调技术在下游任务中的公平性。

FairLex is a multilingual fairness evaluation benchmark for legal text processing. It comprises four datasets spanning four jurisdictions: the Council of Europe, the United States, Switzerland, and China, as well as five languages: English, German, French, Italian, and Chinese. The benchmark focuses on five sensitive attributes including gender, age, region, language and legal domain, and aims to evaluate the fairness of pretrained language models and their fine-tuning techniques in downstream tasks.