无监督神经机器翻译综述

A survey on unsupervised neural machine translation

神经机器翻译模型的学习往往依赖于大规模的平行语料,然而并不是所有语言对都有充足的平行语料。为了解决神经机器翻译中平行语料缺失的问题,研究人员提出了无监督神经机器翻译,通过无监督预训练策略、去噪自编码器、反向翻译和共享潜在表示机制仅依赖于单语语料对翻译任务进行建模。本文从模型主要机制、训练过程出发来分析无监督神经机器翻译,并介绍无监督神经机器翻译现阶段的发展以及存在的挑战。

The training of neural machine translation often relies on large-scale parallel corpora,but not all language pairs have sufficient parallel corpora.In order to alleviate the problem of the lack of parallel corpora in neural machine translation,unsupervised neural machine translation is proposed to model translation relying solely on monolingual corpora with the help of a combination of diverse mechanisms such as unsupervised pre-training strategy,denoising auto-encoder,back-translation,and shared latent representation.This paper analyzes unsupervised neural machine translation from the perspective of model mechanism and training process,and introduces the advances and challenges of unsupervised neural machine translation at the present stage.