WinoMTDE
收藏资源简介:
WinoMTDE数据集是由柏林工业大学的研究人员创建的,旨在评估德语机器翻译系统中的性别偏见。该数据集包含288个德语句子,这些句子根据Winograd模式构建,每个句子都有一个明确性别的主题以及另一个相反性别的主题。数据集根据德国劳动统计局的数据进行平衡,以性别和刻板印象为标准,分为两个子集:WinoMTDEpro和WinoMTDEanti。该数据集可用于评估机器翻译系统在处理性别和职业刻板印象方面的性能。
The WinoMTDE dataset was developed by researchers at the Technical University of Berlin with the aim of assessing gender bias in German machine translation systems. It consists of 288 German sentences built upon Winograd schemas, each of which features a theme with a clearly defined gender and another theme with the opposite gender. The dataset is balanced in terms of gender and occupational stereotypes using data from Germany's Federal Labor Statistical Office, and is divided into two subsets: WinoMTDEpro and WinoMTDEanti. This dataset can be employed to evaluate the performance of machine translation systems when handling gender and occupational stereotypes.

- 1Evaluating Gender Bias in German Machine Translation柏林工业大学 · 2025年



