Newstest2019 Annotation Data

该数据集包含了两个可控长度的翻译模型生成的机器翻译输出的注释。这些翻译涵盖了两个方向：英语到中文和中文到英语，并且每句话都被翻译成了不同的长度，分别是参考长度的80%和50%。注释基于对翻译输出的逐段直接评估，这为详细评估可控长度的翻译模型提供了可能。该数据集规模为6480条注释，涵盖了270个句子、两种语言、两个模型、两种长度和三位标注者的组合。其任务是机器翻译评估。

This dataset contains annotations for machine translation outputs generated by two controllable-length translation models. The translations span two language directions: English to Chinese and Chinese to English, with each source sentence being translated into two versions with lengths set to 80% and 50% of the reference translation’s length respectively. The annotations are derived from direct segment-by-segment evaluations of the translation outputs, enabling comprehensive assessment of controllable-length translation models. Comprising 6480 annotation entries, the dataset covers all combinations of 270 source sentences, two translation directions, two models, two length control levels, and three annotators. The core task supported by this dataset is machine translation evaluation.