Elamur Research⌕
☰
home / papers / 2019 / Simpler and Faster Learning of Adaptive Policies for Simultaneous Translation

Simpler and Faster Learning of Adaptive Policies for Simultaneous Translation

Paper · 2019 · Baigong Zheng, Renjie Zheng, Mingbo Ma et al.

Summary

Simultaneous MT must decide, word by word, whether to wait for more source input (READ) or emit output (WRITE). Fixed schedules like wait-k are simple, but they force the model to guess upcoming content. Adaptive policies avoid this, but they have usually been trained with reinforcement learning, which is slow and unstable. The authors recast policy learning as plain supervised learning. A pre-trained Transformer NMT model builds oracle action sequences from parallel text: it emits a WRITE once the gold next target word ranks within the top r predictions given the source prefix, and otherwise reads another word. Sequences are kept only if their Average Lagging (AL) is low and they consume the full source before the final WRITE. A small recurrent policy then learns to imitate these sequences, using encoder and decoder states plus cross-attention. At test time, a probability threshold ρ adjusts latency without retraining. On WMT15 German↔English, the policy beats WIW/WID, RL and test-time wait-k at comparable latency, and the NMT model never needs retraining.

Key points

  • Data: models are trained on the WMT15 DE↔EN parallel corpora, validated on newstest-2013 and tested on newstest-2015; BPE segmentation, training pairs limited to under 50 tokens.
  • Setup: the underlying model is Transformer-base in OpenNMT (PyTorch), with an added source-side <eos> token. The policy is a single GRU layer, and the best oracle settings were r = 50 with AL filter α = 3.
  • Latency control: sweeping ρ (0.4–0.65) lets one policy model cover several latency regimes. With beam search, it scores higher BLEU than retrained wait-k models at low AL.
  • Training cost: policy training takes roughly 11–12 GPU-hours on 1 GPU, versus 630–966 hours on 8 GPUs for retrained wait-k models. Generating the oracle sequences takes about 2 hours in parallel.
  • Stability: averaged over four seeds, SL learning curves converge faster and more smoothly than RL training.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🏢BaiduBeijing · China40 works · 14,627 citations🎓Oregon State UniversityCorvallis · United States6 works · 886 citations

📚 Shelves 4 Computer Science · 9,086 works Artificial Intelligence · 8,889 works Arts and Humanities · 9,296 works Language and Linguistics · 9,206 works

🧭 Topics 1 Machine translation · 958 works

Cite

Cite as Baigong Zheng, Renjie Zheng, Mingbo Ma and Liang Huang. 2019. Simpler and Faster Learning of Adaptive Policies for Simultaneous Translation. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/d19-1137

@inproceedings{zheng2019simpler,
  title = {{Simpler and Faster Learning of Adaptive Policies for Simultaneous Translation}},
  author = {Baigong Zheng, Renjie Zheng, Mingbo Ma and Liang Huang},
  booktitle = {Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP)},
  year = {2019},
  doi = {10.18653/v1/d19-1137},
  url = {https://doi.org/10.18653/v1/d19-1137}
}

References

22 references, 8 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. ends. Without this token, test-time wait-𝑘 generates either very short target sentences or many punctuations for small 𝑘’s. find
  2. Ashkan Alinejad, Maryam Siahbani, and Anoop Sarkar. 2018. Prediction improves simultaneous neural machine translation. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 3022–3027. find
  3. Naveen Arivazhagan, Colin Cherry, Wolfgang Macherey, Chung-Cheng Chiu, Semih Yavuz, Ruoming Pang, Wei Li, and Colin Raffel. 2019. Monotonic infinite lookback attention for simultaneous machine translation. In Proceedings of the 57th Conference of the Association for Computational Linguistics, ACL 20… our card
  4. Srinivas Bangalore, Vivek Kumar Rangarajan Sridhar, Prakash Kolan, Ladan Golipour, and Aura Jimenez. 2012. Real-time incremental speech-to-speech translation of dialogs. In Proc. of NAACL-HLT. find
  5. Kyunghyun Cho and Masha Esipova. 2016. Can neural machine translation do simultaneous translation? arXiv preprint arXiv:1606.02012. find
  6. Fahim Dalvi, Nadir Durrani, Hassan Sajjad, and Stephan Vogel. 2018. Incremental decoding and training methods for simultaneous translation in neural machine translation. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Langua… our card
  7. Christian Fügen, Alex Waibel, and Muntsin Kolss. 2007. Simultaneous translation of lectures and speeches. Machine translation, 21(4):209–252. find
  8. Alex Graves, Abdel-rahman Mohamed, and Geoffrey Hinton. 2013. Speech recognition with deep recurrent neural networks. In 2013 IEEE international conference on acoustics, speech and signal processing, pages 6645–6649. IEEE. find
  9. Alvin Grissom II, He He, Jordan Boyd-Graber, John Morgan, and Hal Daumé III. 2014. Don’t until the final verb wait: Reinforcement learning for simultaneous machine translation. In Proceedings of the 2014 Conference on empirical methods in natural language processing (EMNLP), pages 1342–1352. find
  10. Jiatao Gu, Graham Neubig, Kyunghyun Cho, and Victor O. K. Li. 2017. Learning to translate in real-time with neural machine translation. In Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics, EACL 2017, pages 1053–1062. our card
  11. He He, Alvin Grissom II, John Morgan, Jordan Boyd-Graber, and Hal Daumé III. 2015. Syntax-based rewriting for simultaneous machine translation. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 55– 64. find
  12. Navdeep Jaitly, David Sussillo, Quoc V Le, Oriol Vinyals, Ilya Sutskever, and Samy Bengio. 2016. An online sequence-to-sequence model using partial conditioning. In Advances in Neural Information Processing Systems, pages 5067–5075. find
  13. Guillaume Klein, Yoon Kim, Yuntian Deng, Jean Senellart, and Alexander M Rush. 2017. Opennmt: Open-source toolkit for neural machine translation. arXiv preprint arXiv:1701.02810. our card
  14. Mingbo Ma, Liang Huang, Hao Xiong, Renjie Zheng, Kaibo Liu, Baigong Zheng, Chuanqiang Zhang, Zhongjun He, Hairong Liu, Xing Li, Hua Wu, and Haifeng Wang. 2019. STACL: Simultaneous translation with implicit anticipation and controllable latency using prefix-to-prefix framework. In Proceedings of the… our card
  15. Jan Niehues, Ngoc-Quan Pham, Thanh-Le Ha, Matthias Sperber, and Alex Waibel. 2018. Low-latency neural speech translation. In Interspeech 2018, 19th Annual Conference of the International Speech Communication Association., pages 1293–1297. find
  16. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. Bleu: a method for automatic evaluation of machine translation. In Proceedings of ACL, pages 311–318, Philadephia, USA. our card
  17. Rico Sennrich, Barry Haddow, and Alexandra Birch. 2016. Neural machine translation of rare words with subword units. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1715–1725. our card
  18. Vivek Kumar Rangarajan Sridhar, John Chen, Srinivas Bangalore, Andrej Ljolje, and Rathinavelu Chengalvarayan. 2013. Segmentation strategies for streaming speech translation. In Proceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguistics: Human La… find
  19. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Advances in neural information processing systems, pages 5998–6008. our card
  20. Mahsa Yarmohammadi, Vivek Kumar Rangarajan Sridhar, Srinivas Bangalore, and Baskaran Sankaran. 2013. Incremental segmentation and decoding strategies for simultaneous translation. In Proceedings of the Sixth International Joint Conference on Natural Language Processing. find
  21. Baigong Zheng, Renjie Zheng, Mingbo Ma, and Liang Huang. 2019a. Simultaneous translation with flexible policy via restricted imitation learning. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 5816– 5822. find
  22. Renjie Zheng, Mingbo Ma, Baigong Zheng, and Liang Huang. 2019b. Speculative beam search for simultaneous translation. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and 9th International Joint Conference on Natural Language Processing. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zheng-2019-simpler-faster-learning-adaptive-policies/