Knowledge distillation (KD) is a promising technique for model compression in neural machine translation. However, where the knowledge hides in KD is still not clear, which may hinder the development of KD. In this work, we first unravel this mystery from an empirical perspective and show that the knowledge comes from the top-1 predictions of teachers, which also helps us build a potential connection between word- and sequence-level KD. Further, we point out two inherent issues in vanilla word-level KD based on this finding. Firstly, the current objective of KD spreads its focus to whole distributions to learn the knowledge, yet lacks special treatment on the most crucial top-1 information. Secondly, the knowledge is largely covered by the golden information due to the fact that most top-1 predictions of teachers overlap with ground-truth tokens, which further restricts the potential of KD. To address these issues, we propose a new method named Top-1 Information Enhanced Knowledge Distillation (TIE-KD). Specifically, we design a hierarchical ranking loss to enforce the learning of the top-1 information from the teacher. Additionally, we develop an iterative KD procedure to infuse more additional knowledge by distilling on the data without ground-truth targets. Experiments on WMT’14 English-German, WMT’14 English-French and WMT’16 English-Romanian demonstrate that our method can respectively boost Transformer_base students by +1.04, +0.60 and +1.11 BLEU scores and significantly outperforms the vanilla word-level KD baseline. Besides, our method shows higher generalizability on different teacher-student capacity gaps than existing KD techniques.

Neural machine translation (NMT) models are known to be fragile to noisy inputs from automatic speech recognition (ASR) systems. Existing methods are usually tailored for robustness against only homophone errors which account for a small portion of realistic ASR errors. In this paper, we propose an adversarial example generation method based on confusion sets that contain words easily confusable with a target word by ASR to conduct adversarial training for NMT models. Specifically, an adversarial example is generated from the perspective of acoustic relations instead of the traditional uniform or unigram sampling from the confusion sets. Experiments on different test sets with hand-crafted and real-world noise demonstrate the effectiveness of our method over previous methods. Moreover, our approach can achieve improvements on the clean test set.