MiTTenS

MiTTenS是由谷歌研究院创建的一个专注于评估翻译中性别错误的数据集，包含26种语言，共计2252条数据。该数据集通过手工制作、合成生成和多领域自然文本等方式构建，旨在解决翻译系统中的性别偏见问题。数据集的应用领域主要集中在机器翻译和基础模型的性别偏见评估，以期提高翻译质量和公平性。MiTTenS is a dataset created by Google Research, which focuses on evaluating gender errors in translation. It covers 26 languages and contains a total of 2252 instances. Constructed via manual curation, synthetic generation, and multi-domain natural text sources, this dataset aims to address gender bias in translation systems. Its application scenarios primarily focus on gender bias evaluation for machine translation and foundational models, with the goal of improving translation quality and fairness.