Multilingual neural machine translation has witnessed remarkable progress in recent years. However, the long-tailed distribution of multilingual corpora poses a challenge of Pareto optimization, i.e., optimizing for some languages may come at the cost of degrading the performance of others. Existing balancing training strategies are equivalent to a series of Pareto optimal solutions, which trade off on a Pareto frontierIn Pareto optimization, Pareto optimal solutions refer to solutions in which none of the objectives can be improved without sacrificing at least one of the other objectives. The set of all Pareto optimal solutions forms a Pareto frontier..In this work, we propose a new training framework, Pareto Mutual Distillation (Pareto-MD), towards pushing the Pareto frontier outwards rather than making trade-offs. Specifically, Pareto-MD collaboratively trains two Pareto optimal solutions that favor different languages and allows them to learn from the strengths of each other via knowledge distillation. Furthermore, we introduce a novel strategy to enable stronger communication between Pareto optimal solutions and broaden the applicability of our approach. Experimental results on the widely-used WMT and TED datasets show that our method significantly pushes the Pareto frontier and outperforms baselines by up to +2.46 BLEUOur code will be released upon acceptance..

Although all-in-one-model multilingual neural machine translation (MNMT) has achieved remarkable progress, the convergence inconsistency in the joint training is ignored, i.e., different language pairs reaching convergence in different epochs. This leads to the trained MNMT model over-fitting low-resource language translations while under-fitting high-resource ones. In this paper, we propose a novel training strategy named LSSD (LanguageSpecific Self-Distillation), which can alleviate the convergence inconsistency and help MNMT models achieve the best performance on each language pair simultaneously. Specifically, LSSD picks up language-specific best checkpoints for each language pair to teach the current model on the fly. Furthermore, we systematically explore three sample-level manipulations of knowledge transferring. Experimental results on three datasets show that LSSD obtains consistent improvements towards all language pairs and achieves the state-of-the-art.

Non-autoregressive translation (NAT) models, which eliminate the sequential dependencies within the target sentence, have achieved remarkable inference speed, but suffer from inferior translation quality. Towards exploring the underlying causes, we carry out a thorough preliminary study on the attention mechanism, which demonstrates the serious weakness in capturing localness compared with conventional autoregressive translation (AT). In response to this problem, we propose to improve the localness of NAT models by explicitly introducing the information about surrounding words. Specifically, temporal convolutions are incorporated into both encoder and decoder sides to obtain localness-aware representations. Extensive experiments on several typical translation datasets show that the proposed method can achieve consistent and significant improvements over strong NAT baselines. Further analyses on the WMT14 En-De translation task reveal that compared with baselines, our approach accelerates the convergence in training and can achieve equivalent performance with a reduction of 70% training steps.

Non-autoregressive neural machine translation, which decomposes the dependence on previous target tokens from the inputs of the decoder, has achieved impressive inference speedup but at the cost of inferior accuracy. Previous works employ iterative decoding to improve the translation by applying multiple refinement iterations. However, a serious drawback is that these approaches expose the serious weakness in recognizing the erroneous translation pieces. In this paper, we propose an architecture named RewriteNAT to explicitly learn to rewrite the erroneous translation pieces. Specifically, RewriteNAT utilizes a locator module to locate the erroneous ones, which are then revised into the correct ones by a revisor module. Towards keeping the consistency of data distribution with iterative decoding, an iterative training strategy is employed to further improve the capacity of rewriting. Extensive experiments conducted on several widely-used benchmarks show that RewriteNAT can achieve better performance while significantly reducing decoding time, compared with previous iterative decoding strategies. In particular, RewriteNAT can obtain competitive results with autoregressive translation on WMT14 En-De, En-Fr and WMT16 Ro-En translation benchmarks.

Self-attention networks (SANs) with selective mechanism has produced substantial improvements in various NLP tasks by concentrating on a subset of input words. However, the underlying reasons for their strong performance have not been well explained. In this paper, we bridge the gap by assessing the strengths of selective SANs (SSANs), which are implemented with a flexible and universal Gumbel-Softmax. Experimental results on several representative NLP tasks, including natural language inference, semantic role labelling, and machine translation, show that SSANs consistently outperform the standard SANs. Through well-designed probing experiments, we empirically validate that the improvement of SSANs can be attributed in part to mitigating two commonly-cited weaknesses of SANs: word order encoding and structure modeling. Specifically, the selective mechanism improves SANs by paying more attention to content words that contribute to the meaning of the sentence.

Although end-to-end neural machine translation (NMT) has achieved remarkable progress in the recent years, the idea of adopting multi-pass decoding mechanism into conventional NMT is not well explored. In this paper, we propose a novel architecture called adaptive multi-pass decoder, which introduces a flexible multi-pass polishing mechanism to extend the capacity of NMT via reinforcement learning. More specifically, we adopt an extra policy network to automatically choose a suitable and effective number of decoding passes, according to the complexity of source sentences and the quality of the generated translations. Extensive experiments on Chinese-English translation demonstrate the effectiveness of our proposed adaptive multi-pass decoder upon the conventional NMT with a significant improvement about 1.55 BLEU.