xDial-Eval

xDial-Eval是一个大规模的多语言开放领域对话评估基准，由新加坡国立大学和腾讯AI实验室等机构合作创建。该数据集基于开源的英语对话评估数据集构建，包含14930个标注回合和8691个标注多回合对话，涵盖英语及其他九种语言。数据集通过高质量的人工标注和商业机器翻译系统进行语言扩展，旨在解决现有对话评估研究主要集中在英语而忽视其他语言的问题。xDial-Eval的应用领域包括对话系统的自动评估和多语言对话质量的量化，旨在通过提供一个全面的多语言基准，推动对话系统评估技术的发展。

xDial-Eval is a large-scale multilingual open-domain dialogue evaluation benchmark co-created by institutions including the National University of Singapore and Tencent AI Lab. Built upon an open-source English dialogue evaluation dataset, this benchmark contains 14,930 annotated dialogue turns and 8,691 annotated multi-turn dialogues, covering English and nine other languages. The dataset was expanded to multiple languages through high-quality manual annotation and commercial machine translation systems, aiming to address the issue that existing dialogue evaluation research primarily focuses on English while neglecting other languages. Application scenarios of xDial-Eval include automatic evaluation of dialogue systems and quantification of multilingual dialogue quality. It aims to promote the development of dialogue system evaluation technologies by providing a comprehensive multilingual benchmark.