Elamur Research⌕
☰
home / papers / 2018 / Incremental Decoding and Training Methods for Simultaneous Translation in Neural Machine Translation

Incremental Decoding and Training Methods for Simultaneous Translation in Neural Machine Translation

Paper · 2018 · Fahim Dalvi, Nadir Durrani, Hassan Sajjad et al.

Summary

The paper addresses simultaneous translation, where an NMT system must produce output before the full source sentence is available, as in live lecture or meeting translation. Instead of translating independent chunks, the authors adapt an RNN encoder-decoder so that the encoder and attention are extended word by word through READ steps. A segmentation agent then chooses how many newly generated target words to commit in WRITE steps. Committed words stay fixed, and beam search is re-anchored on the hypothesis that was committed. The proposed STATIC-RW agent first buffers S source tokens and then alternates RW writes with RW reads. S and RW are tuned against user-set limits on BLEU loss and Average Proportion (AP) latency. Experiments on IWSLT data for four language pairs into English show that this simple agent beats the Wait-if-Worse and Wait-if-Diff baselines while also having lower latency. The authors also tried training on partial inputs: it greatly helps chunk-based decoding but harms incremental decoding. The work shows that cheap, tunable agents can rival heavier reinforcement-learning approaches.

Key points

  • Data and models: German-, Arabic-, Czech- and Spanish-English using IWSLT (TED) data, with 2-layer LSTM attention models built on seq2seq-attn (Kim, 2016). Larger out-of-domain models were fine-tuned on in-domain data.
  • Results with small models: at AP < 0.75, STATIC-RW loses under 0.5 BLEU against the Wait-until-end oracle (AP = 1) for Arabic, Czech and Spanish, and about 1.5 BLEU for German. Results hold across test sets 2011–2014.
  • Results with larger models: the gap to the oracle widens. At AP comparable to WIW (above 0.8), STATIC-RW keeps the loss at or below 1.5 BLEU for every pair except German-English.
  • Chunk training: chunk-based training using fast-align alignments improves chunk decoding by up to 12 BLEU.
  • Add-M training: growing source prefixes with N = 6, M = 1 degraded the incremental decoder.
  • Code is released at github.com/fdalvi/seq2seq-attn-stream.

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

🎓Hamad bin Khalifa UniversityDoha · Qatar71 works · 12,942 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 Fahim Dalvi, Nadir Durrani, Hassan Sajjad and Stephan Vogel. 2018. Incremental Decoding and Training Methods for Simultaneous Translation in Neural Machine Translation. Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 2 (Short Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/n18-2079

@inproceedings{dalvi2018incremental,
  title = {{Incremental Decoding and Training Methods for Simultaneous Translation in Neural Machine Translation}},
  author = {Fahim Dalvi, Nadir Durrani, Hassan Sajjad and Stephan Vogel},
  booktitle = {Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 2 (Short Papers)},
  year = {2018},
  doi = {10.18653/v1/n18-2079},
  url = {https://doi.org/10.18653/v1/n18-2079}
}

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. the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 1, Long Papers. Association for Computational Linguistics, Valencia, Spain, pages 1053–1062. find
  2. Ahmed Abdelali, Kareem Darwish, Nadir Durrani, and Hamdy Mubarak. 2016. Farasa: A fast and furious segmenter for arabic. In Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Demonstrations. Association for Computational Linguistics, Sa… our card
  3. Yoon Kim. 2016. Seq2seq-attn. https://github.com/harvardnlp/seq2seq-attn. ↗
  4. Philipp Koehn, Hieu Hoang, Alexandra Birch, Chris Callison-Burch, Marcello Federico, Nicola Bertoldi, Brooke Cowan, Wade Shen, Christine Moran, Richard Zens, Chris Dyer, Ondrej Bojar, Alexandra Constantin, and Evan Herbst. 2007. Moses: Open source toolkit for statistical machine translation. In Proc… find
  5. Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. 2015. Neural machine translation by jointly learning to align and translate. In ICLR. http://arxiv.org/pdf/1409.0473v6.pdf. ↗ our card
  6. Srinivas Bangalore, Vivek Kumar Rangarajan Sridhar, Prakash Kolan, Ladan Golipour, and Aura Jimenez. 2012. Real-time incremental speech-to-speech translation of dialogs. In Proceedings of the 2012 Conference of the North American Chapter of the Association for Computational Linguistics: Human Langua… find
  7. Minh-Thang Luong and Christopher D. Manning. 2015. Stanford Neural Machine Translation Systems for Spoken Language Domains. In Proceedings of the International Workshop on Spoken Language Translation. Da Nang, Vietnam. find
  8. Yusuke Oda, Graham Neubig, Sakriani Sakti, Tomoki Toda, and Satoshi Nakamura. 2014. Optimizing segmentation strategies for simultaneous speech translation. In Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers). Association for Computation… find
  9. Mauro Cettolo, Jan Niehues, Sebastian Stüker, Luisa Bentivogli, and Marcello Federico. 2014. Report on the 11th IWSLT Evaluation Campaign. Proceedings of the International Workshop on Spoken Language Translation, Lake Tahoe, US . our card
  10. Kyunghyun Cho and Masha Esipova. 2016. Can neural machine translation do simultaneous translation? CoRR abs/1606.02012. find
  11. Hassan Sajjad, Fahim Dalvi, Nadir Durrani, Ahmed Abdelali, Yonatan Belinkov, and Stephan Vogel. 2017a. Challenging Language-Dependent Segmentation for Arabic: An Application to Machine Translation and Part-of-Speech Tagging. In Proceedings of the 55th Annual Meeting of the Association for Computatio… our card
  12. Nadir Durrani, Helmut Schmid, and Alexander Fraser. 2011. A Joint Sequence Translation Model with Integrated Reordering. In Proceedings of the Association for Computational Linguistics: Human Language Technologies (ACL-HLT’11). Portland, OR, USA. our card
  13. Nadir Durrani, Helmut Schmid, Alexander Fraser, Philipp Koehn, and Hinrich Schütze. 2015. The Operation Sequence Model – Combining N-Gram-based and Phrase-based Statistical Machine Translation. Computational Linguistics 41(2):157–186. find
  14. Hassan Sajjad, Nadir Durrani, Fahim Dalvi, Yonatan Belinkov, and Stephan Vogel. 2017b. Neural Machine Translation Training in a Multi-Domain Scenario. In Proceedings of the 14th International Workshop on Spoken Language Technology (IWSLT-14). find
  15. Chris Dyer, Victor Chahuneau, and Noah A. Smith. 2013. A simple, fast, and effective reparameterization of ibm model 2. In Proceedings of NAACL’13. find
  16. Harsh Satija and Joelle Pineau. 2016. Simultaneous machine translation using deep reinforcement learning. In Abstraction in Reinforcement Learning Workshop. International Conference on Machine Learning, New York, USA. find
  17. Andreas Eisele and Yu Chen. 2010. MultiUN: A Multilingual Corpus from United Nation Documents. In Proceedings of the Seventh conference on International Language Resources and Evaluation. Valleta, Malta. our card
  18. 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). Association for Computational Linguistics, Berlin, Germany, pages 1… our card
  19. Christian Fügen, Alex Waibel, and Muntsin Kolss. 2007. Simultaneous translation of lectures and speeches. Machine Translation 21(4):209–252. find
  20. 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 Rangarajan Sridhar, Vivek Kumar, John Chen, Srinivas Bangalore, Andrej Ljolje, and Rathinavelu Chengalvarayan. 2013. Segmentation strategies for st… our card
  21. Ilya Sutskever, Oriol Vinyals, and Quoc VV Le. 2014. Sequence to Sequence Learning with Neural Networks. In Advances in neural information processing systems. pages 3104–3112. find
  22. 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. Asian Federation of Natural L… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/dalvi-2018-incremental-decoding-training-methods-simultaneous/