HiTZ/es_eu-flask

该数据集是为论文《Towards Reliable Multilingual Judge Models: An Empirical Study》构建的基准测试数据集，用于评估多语言裁判模型的可靠性。它源自FLASK数据集的英文分区，并提供了四种语言变体：flask_es（所有字段翻译成西班牙语）、flask_eu（所有字段翻译成巴斯克语）、flask_io_es（仅需评估的模型输入和输出翻译为西班牙语，其余字段保持英文）、flask_io_eu（仅需评估的模型输入和输出翻译为巴斯克语，其余字段保持英文）。这些变体旨在支持多语言环境下的模型性能分析。

This benchmark dataset is constructed for the paper *Towards Reliable Multilingual Judge Models: An Empirical Study*, and is used to evaluate the reliability of multilingual judge models. It is derived from the English partition of the FLASK dataset, and offers four language variants: flask_es (all fields translated into Spanish), flask_eu (all fields translated into Basque), flask_io_es (only the model inputs and outputs requiring evaluation are translated into Spanish, while the remaining fields remain in English), and flask_io_eu (only the model inputs and outputs requiring evaluation are translated into Basque, while the remaining fields remain in English). These variants are designed to support model performance analysis in multilingual environments.