WMT-22 General MT
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/google/wmt-mqm-human-evaluation
下载链接
链接失效反馈官方服务：
资源简介：
该数据集包含了使用多维度质量度量（MQM）标注的机器翻译提交内容，旨在评估翻译质量。每个样本包含一个错误跨度，其平均长度根据不同的语言对在9到13个字符之间变化。该数据集主要关注充分性错误和小错误。规模上，每个语言对（包括中文-英文、英文-德语、英文-俄语）包含1000个实例。该数据集的任务是用于机器翻译（MT）的评估和后编辑工作。
This dataset comprises machine translation submissions annotated using multi-dimensional quality metrics (MQM) for translation quality assessment. Each sample contains an error span, with an average length ranging from 9 to 13 characters depending on the specific language pair. This dataset primarily focuses on adequacy errors and minor errors. Regarding its scale, each language pair (including Chinese-English, English-German, and English-Russian) contains 1000 instances. The task supported by this dataset is machine translation (MT) evaluation and post-editing.
提供机构：
WMT
