arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2111.04079cs.CL
收藏

方差感知机器翻译测试集

Variance-Aware Machine Translation Test Sets

University of Macau(澳门大学);Harbin Institute of Technology (Shenzhen)(哈尔滨工业大学(深圳))

Runzhe Zhan, Xuebo Liu, Derek F. Wong, Lidia S. Chao

摘要:

我们发布了70个用于机器翻译(MT)评估的小而有区别的测试集,称为方差感知测试集(VAT),涵盖了从WMT 16到WMT 20比赛的35个翻译方向。通过一种新的方差感知过滤方法自动创建VAT,该方法过滤当前MT测试集的无差别测试实例,而无需任何人工劳动。实验结果表明,在主流语言对和测试集上,VAT在与人类判断的相关性方面优于原始WMT测试集。对VAT属性的进一步分析揭示了具有挑战性的语言特征(例如,低频词和专有名词的翻译)的竞争机器翻译系统,为构建未来的机器翻译测试集提供指导。测试集和准备方差感知MT测试集的代码可在https://github.com/NLP2CT/Variance-Aware-MT-Test-Sets上免费获得。

英文摘要:

We release 70 small and discriminative test sets for machine translation (MT) evaluation called variance-aware test sets (VAT), covering 35 translation directions from WMT16 to WMT20 competitions. VAT is automatically created by a novel variance-aware filtering method that filters the indiscriminative test instances of the current MT test sets without any human labor. Experimental results show that VAT outperforms the original WMT test sets in terms of the correlation with human judgement across mainstream language pairs and test sets. Further analysis on the properties of VAT reveals the challenging linguistic features (e.g., translation of low-frequency words and proper nouns) for competitive MT systems, providing guidance for constructing future MT test sets. The test sets and the code for preparing variance-aware MT test sets are freely available at https://github.com/NLP2CT/Variance-Aware-MT-Test-Sets .

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑