结合生成式对抗网络的神经机器翻译研究

Research on Neural Machine Translation Combined with Generative Adversarial Nets

针对如何进一步提升神经机器翻译精度的问题,提出了一种将生成式对抗网络(GAN)应用于神经机器翻译(NMT)的方法。构建一个条件序列生成对抗网,它包括两个对抗子模型,一个生成器和一个判别器。该生成器旨在生成难以与人类翻译的句子区分开的句子,判别器旨在将生成器生成的句子与人类翻译的句子区分开来。另外,静态句子级BLEU值将会作为强化目标作用于生成器。在训练过程中,动态的判别器和静态的BLEU目标都用于评估生成的句子,并将评估结果反馈给生成器,来指导生成器的学习。实验结果表明,在英-德翻译数据集上,在引入生成式对抗网络后,相比于传统的基于循环神经网络(RNN)的神经机器翻译模型,翻译效果得到了一定的提升。

This paper proposes a method for applying the generative adversarial nets to neural machine translation,to further improve the accuracy of neural machine translation.A con-ditional sequence generative advensarial net is built which comprises of two adversarial sub models,a generator and a discriminator.The generator aims to generate sentences which are hard to be discriminated from human-translated sentences,and the discriminator makes efforts to discriminate the machine-generated sentences from human-translated ones.Additionally,the static sentence-level BLEU is utilized as the reinforced objective for the generator.During training,both the dynamic discriminator and the static BLEU objective are employed to evaluate the generated sentences and feedback the evaluations to guide the learning of the generator.Experimental results show that the proposed model consistently outperforms the traditional RNN translation model on English-German translation tasks.