Look Backward and Forward: Self-Knowledge Distillation with Bidirectional Decoder for Neural Machine Translation
This work addresses translation quality for users of neural machine translation systems, but it is incremental as it builds on existing Transformer architectures.
The authors tackled the problem of neural machine translation models focusing on local structure rather than global coherence by proposing a self-knowledge distillation method with a bidirectional decoder, which improved performance over strong Transformer baselines on multiple datasets.
Neural Machine Translation(NMT) models are usually trained via unidirectional decoder which corresponds to optimizing one-step-ahead prediction. However, this kind of unidirectional decoding framework may incline to focus on local structure rather than global coherence. To alleviate this problem, we propose a novel method, Self-Knowledge Distillation with Bidirectional Decoder for Neural Machine Translation(SBD-NMT). We deploy a backward decoder which can act as an effective regularization method to the forward decoder. By leveraging the backward decoder's information about the longer-term future, distilling knowledge learned in the backward decoder can encourage auto-regressive NMT models to plan ahead. Experiments show that our method is significantly better than the strong Transformer baselines on multiple machine translation data sets.