Retrieving Sequential Information for Non-Autoregressive Neural Machine Translation
This work addresses translation quality issues in NAT models for machine translation tasks, representing an incremental improvement over existing methods.
The paper tackles the problem of over-translation and under-translation errors in Non-Autoregressive Transformer (NAT) models, especially for long sentences, by proposing two methods to retrieve target sequential information, resulting in significant BLEU score improvements without slowing decoding speed and achieving performance comparable to autoregressive Transformers with speedup.
Non-Autoregressive Transformer (NAT) aims to accelerate the Transformer model through discarding the autoregressive mechanism and generating target words independently, which fails to exploit the target sequential information. Over-translation and under-translation errors often occur for the above reason, especially in the long sentence translation scenario. In this paper, we propose two approaches to retrieve the target sequential information for NAT to enhance its translation ability while preserving the fast-decoding property. Firstly, we propose a sequence-level training method based on a novel reinforcement algorithm for NAT (Reinforce-NAT) to reduce the variance and stabilize the training procedure. Secondly, we propose an innovative Transformer decoder named FS-decoder to fuse the target sequential information into the top layer of the decoder. Experimental results on three translation tasks show that the Reinforce-NAT surpasses the baseline NAT system by a significant margin on BLEU without decelerating the decoding speed and the FS-decoder achieves comparable translation performance to the autoregressive Transformer with considerable speedup.