MT-GenEval

MT-GenEval是一个用于评估机器翻译中性别准确性的基准数据集，由亚马逊AI实验室创建。该数据集包含从英语翻译到八种广泛使用语言的真实、性别平衡的反事实数据，特别关注性别在翻译中的准确性。数据集通过从维基百科提取英文句子，并由注释者手动创建反事实版本，确保性别平衡。MT-GenEval旨在支持性别翻译准确性的研究，并提高对模型在此任务上表现的了解，特别适用于评估和改进机器翻译系统中的性别处理能力。

MT-GenEval is a benchmark dataset for evaluating gender accuracy in machine translation, created by Amazon AI Labs. This dataset contains real, gender-balanced counterfactual data translated from English into eight widely used languages, with a particular focus on the accuracy of gender handling in translation. It is constructed by extracting English sentences from Wikipedia and having annotators manually create counterfactual versions to ensure gender balance. MT-GenEval is designed to support research on gender translation accuracy, improve understanding of model performance on this task, and is particularly applicable for evaluating and enhancing gender processing capabilities in machine translation systems.