arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2106.04060cs.CL
收藏

面向资源丰富型机器翻译的自监督与监督联合训练

Self-supervised and Supervised Joint Training for Resource-rich Machine Translation

Google Research, Google LLC(谷歌有限责任公司谷歌研究院);Language Technologies Institute, Carnegie Mellon University(卡内基梅隆大学语言技术研究所)

Yong Cheng, Wei Wang, Lu Jiang, Wolfgang Macherey

摘要:

文本表示的自监督预训练已成功应用于低资源神经机器翻译(NMT)。然而,它通常无法在资源丰富的NMT上取得显着的收益。在本文中,我们提出了一种联合训练方法,$F_2$ -XEnDec,结合自监督和监督学习优化NMT模型。为了利用互补的自监督信号进行监督学习,NMT模型通过一个称为交叉编码器-解码器的新过程在单语和平行句子混合的例子上进行训练。两个资源丰富的翻译基准,WMT'14英语-德语和WMT'14英语-法语,实验表明,我们的方法实现了实质性的改进,几个强大的基线方法,并获得了新的艺术状态的46.19 BLEU英语-法语时,纳入回译。结果还表明,我们的方法能够提高模型对输入扰动的鲁棒性,例如经常出现在社交媒体上的代码切换噪声。

英文摘要:

Self-supervised pre-training of text representations has been successfully applied to low-resource Neural Machine Translation (NMT). However, it usually fails to achieve notable gains on resource-rich NMT. In this paper, we propose a joint training approach, $F_2$-XEnDec, to combine self-supervised and supervised learning to optimize NMT models. To exploit complementary self-supervised signals for supervised learning, NMT models are trained on examples that are interbred from monolingual and parallel sentences through a new process called crossover encoder-decoder. Experiments on two resource-rich translation benchmarks, WMT'14 English-German and WMT'14 English-French, demonstrate that our approach achieves substantial improvements over several strong baseline methods and obtains a new state of the art of 46.19 BLEU on English-French when incorporating back translation. Results also show that our approach is capable of improving model robustness to input perturbations such as code-switching noise which frequently appears on social media.

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。

↑