Improving the Robustness of Low-Resource Neural Machine Translation with Adversarial Examples

Abstract

AbstractWeak robustness and noise adaptability are major issues for Low-Resource Neural Machine Translation (NMT) models. That is, once some tiny perturbs are added to the input sentence, the model will produce completely different translation with high confidence. Adversarial example is currently a major tool to improve model robustness and how to generate an adversarial examples that can degrade the performance of the model and ensure semantic consistency is a challenging task. In this paper, we adopt reinforcement learning to generate adversarial example for low-resource NMT. Specifically, utilizing the actor-critic algorithm to modify the source sentence, the discriminator and translation model in the environment are used to determine whether the generated adversarial examples maintain semantic consistency and the overall deterioration of the model. Furthermore, we also install a language model reward to measure the fluency of adversarial examples. Experimental results on low-resource translation tasks show that our method highly aggressive to the model while maintaining semantic constraints greatly. Moreover, the model performance is significantly improved after fine-tuning with adversarial examples.KeywordsReinforcement learningAdversarial exampleLow-resource NMT