Knowledge distillation (KD) is commonly used to construct synthetic data for training non-autoregressive translation (NAT) models. However, there exists a discrepancy on low-frequency words between the distilled and the original data, leading to more errors on predicting low-frequency words. To alleviate the problem, we directly expose the raw data into NAT by leveraging pretraining. By analyzing directed alignments, we found that KD makes low-frequency source words aligned with targets more deterministically but fails to align sufficient low-frequency words from target to source. Accordingly, we propose reverse KD to rejuvenate more alignments for low-frequency target words. To make the most of authentic and synthetic data, we combine these complementary approaches as a new training strategy for further boosting NAT performance. We conduct experiments on five translation benchmarks over two advanced architectures. Results demonstrate that the proposed approach can significantly and universally improve translation quality by reducing translation errors on low-frequency words. Encouragingly, our approach achieves 28.2 and 33.9 BLEU points on the WMT14 English-German and WMT16 Romanian-English datasets, respectively. Our code, data, and trained models are available at \url{https://github.com/longyuewangdcu/RLFW-NAT}.

There have been significant efforts to interpret the encoder of Transformer-based encoder-decoder architectures for neural machine translation (NMT); meanwhile, the decoder remains largely unexamined despite its critical role. During translation, the decoder must predict output tokens by considering both the source-language text from the encoder and the target-language prefix produced in previous steps. In this work, we study how Transformer-based decoders leverage information from the source and target languages -- developing a universal probe task to assess how information is propagated through each module of each decoder layer. We perform extensive experiments on three major translation datasets (WMT En-De, En-Fr, and En-Zh). Our analysis provides insight on when and where decoders leverage different sources. Based on these insights, we demonstrate that the residual feed-forward module in each Transformer decoder layer can be dropped with minimal loss of performance -- a significant reduction in computation and number of parameters, and consequently a significant boost to both training and inference speed.

Self-attention networks (SAN) have attracted a lot of interests due to their high parallelization and strong performance on a variety of NLP tasks, e.g. machine translation. Due to the lack of recurrence structure such as recurrent neural networks (RNN), SAN is ascribed to be weak at learning positional information of words for sequence modeling. However, neither this speculation has been empirically confirmed, nor explanations for their strong performances on machine translation tasks when "lacking positional information" have been explored. To this end, we propose a novel word reordering detection task to quantify how well the word order information learned by SAN and RNN. Specifically, we randomly move one word to another position, and examine whether a trained model can detect both the original and inserted positions. Experimental results reveal that: 1) SAN trained on word reordering detection indeed has difficulty learning the positional information even with the position embedding; and 2) SAN trained on machine translation learns better positional information than its RNN counterpart, in which position embedding plays a critical role. Although recurrence structure make the model more universally-effective on learning word order, learning objectives matter more in the downstream tasks such as machine translation.

Self-attention networks (SANs) have drawn increasing interest due to their high parallelization in computation and flexibility in modeling dependencies. SANs can be further enhanced with multi-head attention by allowing the model to attend to information from different representation subspaces. In this work, we propose novel convolutional self-attention networks, which offer SANs the abilities to 1) strengthen dependencies among neighboring elements, and 2) model the interaction between features extracted by multiple attention heads. Experimental results of machine translation on different language pairs and model settings show that our approach outperforms both the strong Transformer baseline and other existing models on enhancing the locality of SANs. Comparing with prior studies, the proposed model is parameter free in terms of introducing no more parameters.

Recently, the Transformer model that is based solely on attention mechanisms, has advanced the state-of-the-art on various machine translation tasks. However, recent studies reveal that the lack of recurrence hinders its further improvement of translation capacity. In response to this problem, we propose to directly model recurrence for Transformer with an additional recurrence encoder. In addition to the standard recurrent neural network, we introduce a novel attentive recurrent network to leverage the strengths of both attention and recurrent networks. Experimental results on the widely-used WMT14 English-German and WMT17 Chinese-English translation tasks demonstrate the effectiveness of the proposed approach. Our studies also reveal that the proposed model benefits from a short-cut that bridges the source and target sequences with a single recurrent layer, which outperforms its deep counterpart.

With the promising progress of deep neural networks, layer aggregation has been used to fuse information across layers in various fields, such as computer vision and machine translation. However, most of the previous methods combine layers in a static fashion in that their aggregation strategy is independent of specific hidden states. Inspired by recent progress on capsule networks, in this paper we propose to use routing-by-agreement strategies to aggregate layers dynamically. Specifically, the algorithm learns the probability of a part (individual layer representations) assigned to a whole (aggregated representations) in an iterative way and combines parts accordingly. We implement our algorithm on top of the state-of-the-art neural machine translation model TRANSFORMER and conduct experiments on the widely-used WMT14 English-German and WMT17 Chinese-English translation datasets. Experimental results across language pairs show that the proposed approach consistently outperforms the strong baseline model and a representative static aggregation model.

Neural machine translation (NMT) models generally adopt an encoder-decoder architecture for modeling the entire translation process. The encoder summarizes the representation of input sentence from scratch, which is potentially a problem if the sentence is ambiguous. When translating a text, humans often create an initial understanding of the source sentence and then incrementally refine it along the translation on the target side. Starting from this intuition, we propose a novel encoder-refiner-decoder framework, which dynamically refines the source representations based on the generated target-side information at each decoding step. Since the refining operations are time-consuming, we propose a strategy, leveraging the power of reinforcement learning models, to decide when to refine at specific decoding steps. Experimental results on both Chinese-English and English-German translation tasks show that the proposed approach significantly and consistently improves translation performance over the standard encoder-decoder framework. Furthermore, when refining strategy is applied, results still show reasonable improvement over the baseline without much decrease in decoding speed.

Pronouns are frequently omitted in pro-drop languages, such as Chinese, generally leading to significant challenges with respect to the production of complete translations. To date, very little attention has been paid to the dropped pronoun (DP) problem within neural machine translation (NMT). In this work, we propose a novel reconstruction-based approach to alleviating DP translation problems for NMT models. Firstly, DPs within all source sentences are automatically annotated with parallel information extracted from the bilingual training corpus. Next, the annotated source sentence is reconstructed from hidden representations in the NMT model. With auxiliary training objectives, in the terms of reconstruction scores, the parameters associated with the NMT model are guided to produce enhanced hidden representations that are encouraged as much as possible to embed annotated DP information. Experimental results on both Chinese-English and Japanese-English dialogue translation tasks show that the proposed approach significantly and consistently improves translation performance over a strong NMT baseline, which is directly built on the training data annotated with DPs.