Previous studies have proved that cross-lingual knowledge distillation can significantly improve the performance of pre-trained models for cross-lingual similarity matching tasks. However, the student model needs to be large in this operation. Otherwise, its performance will drop sharply, thus making it impractical to be deployed to memory-limited devices. To address this issue, we delve into cross-lingual knowledge distillation and propose a multi-stage distillation framework for constructing a small-size but high-performance cross-lingual model. In our framework, contrastive learning, bottleneck, and parameter recurrent strategies are delicately combined to prevent performance from being compromised during the compression process. The experimental results demonstrate that our method can compress the size of XLM-R and MiniLM by more than 50%, while the performance is only reduced by about 1%.

Encoder pre-training is promising in end-to-end Speech Translation (ST), given the fact that speech-to-translation data is scarce. But ST encoders are not simple instances of Automatic Speech Recognition (ASR) or Machine Translation (MT) encoders. For example, we find that ASR encoders lack the global context representation, which is necessary for translation, whereas MT encoders are not designed to deal with long but locally attentive acoustic sequences. In this work, we propose a Stacked Acoustic-and-Textual Encoding (SATE) method for speech translation. Our encoder begins with processing the acoustic sequence as usual, but later behaves more like an MT encoder for a global representation of the input sequence. In this way, it is straightforward to incorporate the pre-trained models into the system. Also, we develop an adaptor module to alleviate the representation inconsistency between the pre-trained ASR encoder and MT encoder, and develop a multi-teacher knowledge distillation method to preserve the pre-training knowledge. Experimental results on the LibriSpeech En-Fr and MuST-C En-De ST tasks show that our method achieves state-of-the-art BLEU scores of 18.3 and 25.2. To our knowledge, we are the first to develop an end-to-end ST system that achieves comparable or even better BLEU performance than the cascaded ST counterpart when large-scale ASR and MT data is available.

This paper describes TenTrans’ submission to WMT21 Multilingual Low-Resource Translation shared task for the Romance language pairs. This task focuses on improving translation quality from Catalan to Occitan, Romanian and Italian, with the assistance of related high-resource languages. We mainly utilize back-translation, pivot-based methods, multilingual models, pre-trained model fine-tuning, and in-domain knowledge transfer to improve the translation quality. On the test set, our best-submitted system achieves an average of 43.45 case-sensitive BLEU scores across all low-resource pairs. Our data, code, and pre-trained models used in this work are available in TenTrans evaluation examples.

This paper describes TenTrans large-scale multilingual machine translation system for WMT 2021. We participate in the Small Track 2 in five South East Asian languages, thirty directions: Javanese, Indonesian, Malay, Tagalog, Tamil, English. We mainly utilized forward/back-translation, in-domain data selection, knowledge distillation, and gradual fine-tuning from the pre-trained model FLORES-101. We find that forward/back-translation significantly improves the translation results, data selection and gradual fine-tuning are particularly effective during adapting domain, while knowledge distillation brings slight performance improvement. Also, model averaging is used to further improve the translation performance based on these systems. Our final system achieves an average BLEU score of 28.89 across thirty directions on the test set.

The paper describes the TenTrans’s submissions to the WMT 2021 Efficiency Shared Task. We explore training a variety of smaller compact transformer models using the teacher-student setup. Our model is trained by our self-developed open-source multilingual training platform TenTrans-Py. We also release an open-source high-performance inference toolkit for transformer models and the code is written in C++ completely. All additional optimizations are built on top of the inference engine including attention caching, kernel fusion, early-stop, and several other optimizations. In our submissions, the fastest system can translate more than 22,000 tokens per second with a single Tesla P4 while maintaining 38.36 BLEU on En-De newstest2019. Our trained models and more details are available in TenTrans-Decoding competition examples.

Large amounts of data has made neural machine translation (NMT) a big success in recent years. But it is still a challenge if we train these models on small-scale corpora. In this case, the way of using data appears to be more important. Here, we investigate the effective use of training data for low-resource NMT. In particular, we propose a dynamic curriculum learning (DCL) method to reorder training samples in training. Unlike previous work, we do not use a static scoring function for reordering. Instead, the order of training samples is dynamically determined in two ways - loss decline and model competence. This eases training by highlighting easy samples that the current model has enough competence to learn. We test our DCL method in a Transformer-based system. Experimental results show that DCL outperforms several strong baselines on three low-resource machine translation benchmarks and different sized data of WMT’16 En-De.

Pre-trained language models like BERT have proven to be highly performant. However, they are often computationally expensive in many practical scenarios, for such heavy models can hardly be readily implemented with limited resources. To improve their efficiency with an assured model performance, we propose a novel speed-tunable FastBERT with adaptive inference time. The speed at inference can be flexibly adjusted under varying demands, while redundant calculation of samples is avoided. Moreover, this model adopts a unique self-distillation mechanism at fine-tuning, further enabling a greater computational efficacy with minimal loss in performance. Our model achieves promising results in twelve English and Chinese datasets. It is able to speed up by a wide range from 1 to 12 times than BERT if given different speedup thresholds to make a speed-performance tradeoff.

This paper proposes a new pre-training method, called Code-Switching Pre-training (CSP for short) for Neural Machine Translation (NMT). Unlike traditional pre-training method which randomly masks some fragments of the input sentence, the proposed CSP randomly replaces some words in the source sentence with their translation words in the target language. Specifically, we firstly perform lexicon induction with unsupervised word embedding mapping between the source and target languages, and then randomly replace some words in the input sentence with their translation words according to the extracted translation lexicons. CSP adopts the encoder-decoder framework: its encoder takes the code-mixed sentence as input, and its decoder predicts the replaced fragment of the input sentence. In this way, CSP is able to pre-train the NMT model by explicitly making the most of the alignment information extracted from the source and target monolingual corpus. Additionally, we relieve the pretrain-finetune discrepancy caused by the artificial symbols like [mask]. To verify the effectiveness of the proposed method, we conduct extensive experiments on unsupervised and supervised NMT. Experimental results show that CSP achieves significant improvements over baselines without pre-training or with other pre-training methods.

Existing works, including ELMO and BERT, have revealed the importance of pre-training for NLP tasks. While there does not exist a single pre-training model that works best in all cases, it is of necessity to develop a framework that is able to deploy various pre-training models efficiently. For this purpose, we propose an assemble-on-demand pre-training toolkit, namely Universal Encoder Representations (UER). UER is loosely coupled, and encapsulated with rich modules. By assembling modules on demand, users can either reproduce a state-of-the-art pre-training model or develop a pre-training model that remains unexplored. With UER, we have built a model zoo, which contains pre-trained models based on different corpora, encoders, and targets (objectives). With proper pre-trained models, we could achieve new state-of-the-art results on a range of downstream datasets.