In this paper, we present a substantial step in better understanding the SOTA sequence-to-sequence (Seq2Seq) pretraining for neural machine translation (NMT). We focus on studying the impact of the jointly pretrained decoder, which is the main difference between Seq2Seq pretraining and previous encoder-based pretraining approaches for NMT. By carefully designing experiments on three language pairs, we find that Seq2Seq pretraining is a double-edged sword: On one hand, it helps NMT models to produce more diverse translations and reduce adequacy-related translation errors. On the other hand, the discrepancies between Seq2Seq pretraining and NMT finetuning limit the translation quality (i.e., domain discrepancy) and induce the over-estimation issue (i.e., objective discrepancy). Based on these observations, we further propose simple and effective strategies, named in-domain pretraining and input adaptation to remedy the domain and objective discrepancies, respectively. Experimental results on several language pairs show that our approach can consistently improve both translation performance and model robustness upon Seq2Seq pretraining.

Back-translation is a critical component of Unsupervised Neural Machine Translation (UNMT), which generates pseudo parallel data from target monolingual data. A UNMT model is trained on the pseudo parallel data with \bf translated source, and translates \bf natural source sentences in inference. The source discrepancy between training and inference hinders the translation performance of UNMT models. By carefully designing experiments, we identify two representative characteristics of the data gap in source: (1) style gap (i.e., translated vs. natural text style) that leads to poor generalization capability; (2) content gap that induces the model to produce hallucination content biased towards the target language. To narrow the data gap, we propose an online self-training approach, which simultaneously uses the pseudo parallel data {natural source, translated target} to mimic the inference scenario. Experimental results on several widely-used language pairs show that our approach outperforms two strong baselines (XLM and MASS) by remedying the style and content gaps.

We present a simple yet effective self-training approach, named as STAD, for low-resource relation extraction. The approach first classifies the auto-annotated instances into two groups: confident instances and uncertain instances, according to the probabilities predicted by a teacher model. In contrast to most previous studies, which mainly only use the confident instances for self-training, we make use of the uncertain instances. To this end, we propose a method to identify ambiguous but useful instances from the uncertain instances and then divide the relations into candidate-label set and negative-label set for each ambiguous instance. Next, we propose a set-negative training method on the negative-label sets for the ambiguous instances and a positive training method for the confident instances. Finally, a joint-training method is proposed to build the final relation extraction system on all data. Experimental results on two widely used datasets SemEval2010 Task-8 and Re-TACRED with low-resource settings demonstrate that this new self-training approach indeed achieves significant and consistent improvements when comparing to several competitive self-training systems.

This paper describes Tencent AI Lab - Shanghai Jiao Tong University (TAL-SJTU) Low-Resource Translation systems for the WMT22 shared task. We participate in the general translation task on English-Livonian.Our system is based on M2M100 with novel techniques that adapt it to the target language pair.(1) Cross-model word embedding alignment: inspired by cross-lingual word embedding alignment, we successfully transfer a pre-trained word embedding to M2M100, enabling it to support Livonian.(2) Gradual adaptation strategy: we exploit Estonian and Latvian as auxiliary languages for many-to-many translation training and then adapt to English-Livonian.(3) Data augmentation: to enlarge the parallel data for English-Livonian, we construct pseudo-parallel data with Estonian and Latvian as pivot languages.(4) Fine-tuning: to make the most of all available data, we fine-tune the model with the validation set and online back-translation, further boosting the performance.In model evaluation: (1) We find that previous work underestimated the translation performance of Livonian due to inconsistent Unicode normalization, which may cause a discrepancy of up to 14.9 BLEU score.(2) In addition to the standard validation set, we also employ round-trip BLEU to evaluate the models, which we find more appropriate for this task. Finally, our unconstrained system achieves BLEU scores of 17.0 and 30.4 for English to/from Livonian.

Self-training has proven effective for improving NMT performance by augmenting model training with synthetic parallel data. The common practice is to construct synthetic data based on a randomly sampled subset of large-scale monolingual data, which we empirically show is sub-optimal. In this work, we propose to improve the sampling procedure by selecting the most informative monolingual sentences to complement the parallel data. To this end, we compute the uncertainty of monolingual sentences using the bilingual dictionary extracted from the parallel data. Intuitively, monolingual sentences with lower uncertainty generally correspond to easy-to-translate patterns which may not provide additional gains. Accordingly, we design an uncertainty-based sampling strategy to efficiently exploit the monolingual data for self-training, in which monolingual sentences with higher uncertainty would be sampled with higher probability. Experimental results on large-scale WMT English⇒German and English⇒Chinese datasets demonstrate the effectiveness of the proposed approach. Extensive analyses suggest that emphasizing the learning on uncertain monolingual sentences by our approach does improve the translation quality of high-uncertainty sentences and also benefits the prediction of low-frequency words at the target side.

Non-Autoregressive machine Translation (NAT) models have demonstrated significant inference speedup but suffer from inferior translation accuracy. The common practice to tackle the problem is transferring the Autoregressive machine Translation (AT) knowledge to NAT models, e.g., with knowledge distillation. In this work, we hypothesize and empirically verify that AT and NAT encoders capture different linguistic properties of source sentences. Therefore, we propose to adopt multi-task learning to transfer the AT knowledge to NAT models through encoder sharing. Specifically, we take the AT model as an auxiliary task to enhance NAT model performance. Experimental results on WMT14 En-De and WMT16 En-Ro datasets show that the proposed Multi-Task NAT achieves significant improvements over the baseline NAT models. Furthermore, the performance on large-scale WMT19 and WMT20 En-De datasets confirm the consistency of our proposed method. In addition, experimental results demonstrate that our Multi-Task NAT is complementary to knowledge distillation, the standard knowledge transfer method for NAT.

This paper describes Tencent Translation systems for the WMT21 shared task. We participate in the news translation task on three language pairs: Chinese-English, English-Chinese and German-English. Our systems are built on various Transformer models with novel techniques adapted from our recent research work. First, we combine different data augmentation methods including back-translation, forward-translation and right-to-left training to enlarge the training data. We also apply language coverage bias, data rejuvenation and uncertainty-based sampling approaches to select content-relevant and high-quality data from large parallel and monolingual corpora. Expect for in-domain fine-tuning, we also propose a fine-grained “one model one domain” approach to model characteristics of different news genres at fine-tuning and decoding stages. Besides, we use greed-based ensemble algorithm and transductive ensemble method to further boost our systems. Based on our success in the last WMT, we continuously employed advanced techniques such as large batch training, data selection and data filtering. Finally, our constrained Chinese-English system achieves 33.4 case-sensitive BLEU score, which is the highest among all submissions. The German-English system is ranked at second place accordingly.

This paper describes the Tencent AI Lab submission of the WMT2021 shared task on biomedical translation in eight language directions: English-German, English-French, English-Spanish and English-Russian. We utilized different Transformer architectures, pretraining and back-translation strategies to improve translation quality. Concretely, we explore mBART (Liu et al., 2020) to demonstrate the effectiveness of the pretraining strategy. Our submissions (Tencent AI Lab Machine Translation, TMT) in German/French/Spanish⇒English are ranked 1st respectively according to the official evaluation results in terms of BLEU scores.

This paper describes Tencent Neural Machine Translation systems for the WMT 2020 news translation tasks. We participate in the shared news translation task on English ↔ Chinese and English → German language pairs. Our systems are built on deep Transformer and several data augmentation methods. We propose a boosted in-domain finetuning method to improve single models. Ensemble is used to combine single models and we propose an iterative transductive ensemble method which can further improve the translation performance based on the ensemble results. We achieve a BLEU score of 36.8 and the highest chrF score of 0.648 on Chinese → English task.

This paper describes the Tencent AI Lab’s submission of the WMT 2020 shared task on chat translation in English-German. Our neural machine translation (NMT) systems are built on sentence-level, document-level, non-autoregressive (NAT) and pretrained models. We integrate a number of advanced techniques into our systems, including data selection, back/forward translation, larger batch learning, model ensemble, finetuning as well as system combination. Specifically, we proposed a hybrid data selection method to select high-quality and in-domain sentences from out-of-domain data. To better capture the source contexts, we exploit to augment NAT models with evolved cross-attention. Furthermore, we explore to transfer general knowledge from four different pre-training language models to the downstream translation task. In general, we present extensive experimental results for this new translation task. Among all the participants, our German-to-English primary system is ranked the second in terms of BLEU scores.

This paper describes the Tencent AI Lab submission of the WMT2020 shared task on biomedical translation in four language directions: German<->English, English<->German, Chinese<->English and English<->Chinese. We implement our system with model ensemble technique on different transformer architectures (Deep, Hybrid, Big, Large Transformers). To enlarge the in-domain bilingual corpus, we use back-translation of monolingual in-domain data in the target language as additional in-domain training data. Our systems in German->English and English->German are ranked 1st and 3rd respectively according to the official evaluation results in terms of BLEU scores.

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.

Large-scale training datasets lie at the core of the recent success of neural machine translation (NMT) models. However, the complex patterns and potential noises in the large-scale data make training NMT models difficult. In this work, we explore to identify the inactive training examples which contribute less to the model performance, and show that the existence of inactive examples depends on the data distribution. We further introduce data rejuvenation to improve the training of NMT models on large-scale datasets by exploiting inactive examples. The proposed framework consists of three phases. First, we train an identification model on the original training data, and use it to distinguish inactive examples and active examples by their sentence-level output probabilities. Then, we train a rejuvenation model on the active examples, which is used to re-label the inactive examples with forward- translation. Finally, the rejuvenated examples and the active examples are combined to train the final NMT model. Experimental results on WMT14 English-German and English-French datasets show that the proposed data rejuvenation consistently and significantly improves performance for several strong NMT models. Extensive analyses reveal that our approach stabilizes and accelerates the training process of NMT models, resulting in final models with better generalization capability.

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 modeling 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 models 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.

Multi-head attention is appealing for its ability to jointly extract different types of information from multiple representation subspaces. Concerning the information aggregation, a common practice is to use a concatenation followed by a linear transformation, which may not fully exploit the expressiveness of multi-head attention. In this work, we propose to improve the information aggregation for multi-head attention with a more powerful routing-by-agreement algorithm. Specifically, the routing algorithm iteratively updates the proportion of how much a part (i.e. the distinct information learned from a specific subspace) should be assigned to a whole (i.e. the final output representation), based on the agreement between parts and wholes. Experimental results on linguistic probing tasks and machine translation tasks prove the superiority of the advanced information aggregation over the standard linear transformation.

In this work, we present novel approaches to exploit sentential context for neural machine translation (NMT). Specifically, we show that a shallow sentential context extracted from the top encoder layer only, can improve translation performance via contextualizing the encoding representations of individual words. Next, we introduce a deep sentential context, which aggregates the sentential context representations from all of the internal layers of the encoder to form a more comprehensive context representation. Experimental results on the WMT14 English-German and English-French benchmarks show that our model consistently improves performance over the strong Transformer model, demonstrating the necessity and effectiveness of exploiting sentential context for NMT.

Current state-of-the-art neural machine translation (NMT) uses a deep multi-head self-attention network with no explicit phrase information. However, prior work on statistical machine translation has shown that extending the basic translation unit from words to phrases has produced substantial improvements, suggesting the possibility of improving NMT performance from explicit modeling of phrases. In this work, we present multi-granularity self-attention (Mg-Sa): a neural network that combines multi-head self-attention and phrase modeling. Specifically, we train several attention heads to attend to phrases in either n-gram or syntactic formalisms. Moreover, we exploit interactions among phrases to enhance the strength of structure modeling – a commonly-cited weakness of self-attention. Experimental results on WMT14 English-to-German and NIST Chinese-to-English translation tasks show the proposed approach consistently improves performance. Targeted linguistic analysis reveal that Mg-Sa indeed captures useful phrase information at various levels of granularities.

Zero pronouns (ZPs) are frequently omitted in pro-drop languages, but should be recalled in non-pro-drop languages. This discourse phenomenon poses a significant challenge for machine translation (MT) when translating texts from pro-drop to non-pro-drop languages. In this paper, we propose a unified and discourse-aware ZP translation approach for neural MT models. Specifically, we jointly learn to predict and translate ZPs in an end-to-end manner, allowing both components to interact with each other. In addition, we employ hierarchical neural networks to exploit discourse-level context, which is beneficial for ZP prediction and thus translation. Experimental results on both Chinese-English and Japanese-English data show that our approach significantly and accumulatively improves both translation performance and ZP prediction accuracy over not only baseline but also previous works using external ZP prediction models. Extensive analyses confirm that the performance improvement comes from the alleviation of different kinds of errors especially caused by subjective ZPs.

Although neural machine translation (NMT) has advanced the state-of-the-art on various language pairs, the interpretability of NMT remains unsatisfactory. In this work, we propose to address this gap by focusing on understanding the input-output behavior of NMT models. Specifically, we measure the word importance by attributing the NMT output to every input word through a gradient-based method. We validate the approach on a couple of perturbation operations, language pairs, and model architectures, demonstrating its superiority on identifying input words with higher influence on translation performance. Encouragingly, the calculated importance can serve as indicators of input words that are under-translated by NMT models. Furthermore, our analysis reveals that words of certain syntactic categories have higher importance while the categories vary across language pairs, which can inspire better design principles of NMT architectures for multi-lingual translation.

Recent studies have shown that a hybrid of self-attention networks (SANs) and recurrent neural networks RNNs outperforms both individual architectures, while not much is known about why the hybrid models work. With the belief that modeling hierarchical structure is an essential complementary between SANs and RNNs, we propose to further enhance the strength of hybrid models with an advanced variant of RNNs – Ordered Neurons LSTM (ON-LSTM), which introduces a syntax-oriented inductive bias to perform tree-like composition. Experimental results on the benchmark machine translation task show that the proposed approach outperforms both individual architectures and a standard hybrid model. Further analyses on targeted linguistic evaluation and logical inference tasks demonstrate that the proposed approach indeed benefits from a better modeling of hierarchical structure.

Although self-attention networks (SANs) have advanced the state-of-the-art on various NLP tasks, one criticism of SANs is their ability of encoding positions of input words (Shaw et al., 2018). In this work, we propose to augment SANs with structural position representations to model the latent structure of the input sentence, which is complementary to the standard sequential positional representations. Specifically, we use dependency tree to represent the grammatical structure of a sentence, and propose two strategies to encode the positional relationships among words in the dependency tree. Experimental results on NIST Chinese-to-English and WMT14 English-to-German translation tasks show that the proposed approach consistently boosts performance over both the absolute and relative sequential position representations.

Advanced neural machine translation (NMT) models generally implement encoder and decoder as multiple layers, which allows systems to model complex functions and capture complicated linguistic structures. However, only the top layers of encoder and decoder are leveraged in the subsequent process, which misses the opportunity to exploit the useful information embedded in other layers. In this work, we propose to simultaneously expose all of these signals with layer aggregation and multi-layer attention mechanisms. In addition, we introduce an auxiliary regularization term to encourage different layers to capture diverse information. Experimental results on widely-used WMT14 English-German and WMT17 Chinese-English translation data demonstrate the effectiveness and universality of the proposed approach.

Phrases play an important role in natural language understanding and machine translation (Sag et al., 2002; Villavicencio et al., 2005). However, it is difficult to integrate them into current neural machine translation (NMT) which reads and generates sentences word by word. In this work, we propose a method to translate phrases in NMT by integrating a phrase memory storing target phrases from a phrase-based statistical machine translation (SMT) system into the encoder-decoder architecture of NMT. At each decoding step, the phrase memory is first re-written by the SMT model, which dynamically generates relevant target phrases with contextual information provided by the NMT model. Then the proposed model reads the phrase memory to make probability estimations for all phrases in the phrase memory. If phrase generation is carried on, the NMT decoder selects an appropriate phrase from the memory to perform phrase translation and updates its decoding state by consuming the words in the selected phrase. Otherwise, the NMT decoder generates a word from the vocabulary as the general NMT decoder does. Experiment results on the Chinese to English translation show that the proposed model achieves significant improvements over the baseline on various test sets.