Elamur Research
☰
home / papers / 2015 / A Neural Network Approach to Context-Sensitive Generation of Conversational Responses

A Neural Network Approach to Context-Sensitive Generation of Conversational Responses

Written for the Research library (claude-opus-5-5) from the paper text, in our own words: the license "arXiv — non-exclusive licence to distribute" does not allow redistributing the text itself. Original — arxiv.org/pdf/1506.06714

Summary

Open-domain chatbots built from social media data, such as Ritter et al. (2011)'s translation-style system, usually treat a reply as a function of the previous post alone. They ignore earlier turns. Adding that history to phrase-based statistical models makes phrase tables grow and become sparse. The authors address this with continuous embeddings. They extend Mikolov's Recurrent Neural Network Language Model (RLM) so that response generation depends on both the preceding message and the earlier context.

They compare three variants:

  • RLMT simply concatenates all utterances.
  • DCGM-I and DCGM-II encode the context and message as bag-of-words vectors through a feed-forward network, which then biases the decoder's hidden state.

The models learn end to end from unannotated Twitter exchanges. They are used as extra features to rerank outputs from MT-based and IR-based systems. The authors also propose a way to extract multiple references so that BLEU and METEOR evaluation is more reliable.

Context-aware DCGM systems beat both context-free and context-sensitive baselines in automatic and human evaluation. The work is an early step toward neural, fully data-driven dialog generation.

Key points

  • Paper and venue: Published at NAACL-HLT 2015 by authors from Microsoft Research, Université de Montréal and others. The authors believe it is the first neural model for open-domain response generation.
  • Automatic metrics: The best models raise BLEU by as much as 11% (relative) for MT-based reranking and by 24% for IR-based reranking. DCGM+CMM gains up to 0.48 BLEU over MT+CMM and 0.88 BLEU over the Ritter-style baseline.
  • Human evaluation: Five annotators judged each response pair. DCGM-II+CMM was preferred by about 5.3% over IR and 3.1% over MT. DCGM systems also beat non-DCGM context-sensitive systems by 1.5% (MT) and 2.3% (IR).
  • Best system: MT-based DCGM-II+CMM. Its replies average 8.95 tokens, compared with 11.5 tokens for human replies. They are often generic, and longer outputs tend to drift off-topic or contradict themselves.
  • Future work: The authors call for models that capture word order and for direct neural generation.

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

Fields: Other

Topics: Machine translation

Cited by

245 papers of our library cite it (of the 140673 whose bibliographies we read); 19 of them have a page here, 226 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.1506.06714 · arXiv 1506.06714

@misc{sordoni2015neural,
  title = {{A Neural Network Approach to Context-Sensitive Generation of Conversational Responses}},
  author = {Sordoni, Alessandro and Galley, Michel and Auli, Michael and Brockett, Chris and Ji, Yangfeng and Mitchell, Margaret and Nie, Jian-Yun and Gao, Jianfeng and Dolan, Bill},
  year = {2015},
  doi = {10.48550/arxiv.1506.06714},
  eprint = {1506.06714},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.1506.06714}
}

References

30 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Michael Auli, Michel Galley, Chris Quirk, and Geoffrey Zweig. 2013. Joint language and translation modeling with recurrent neural networks. In Proc. of EMNLP, pages 1044–1054. find
  2. Satanjeev Banerjee and Alon Lavie. 2005. METEOR: An automatic metric for MT evaluation with improved correlation with human judgments. In Proc. of ACL Workshop on Intrinsic and Extrinsic Evaluation Measures for Machine Translation and/or Summarization, pages 65–72, Ann Arbor, Jun. find our card
  3. Yoshua Bengio, Rejean Ducharme, and Pascal Vincent. 2003. A neural probabilistic language model. Journ. Mach. Learn. Res., 3:1137–1155. find
  4. Kyunghyun Cho, Bart van Merrienboer, Çaglar Gülçehre, Fethi Bougares, Holger Schwenk, and Yoshua Bengio. 2014. Learning phrase representations using RNN encoder-decoder for statistical machine translation. Proc. of EMNLP. find
  5. Ronan Collobert and Jason Weston. 2008. A unified architecture for natural language processing: Deep neural networks with multitask learning. In Proc. of ICML, pages 160–167. ACM. find
  6. Jacob Devlin, Rabih Zbib, Zhongqiang Huang, Thomas Lamar, Richard Schwartz, and John Makhoul. 2014. Fast and robust neural network joint models for statistical machine translation. In Proc. of ACL. find
  7. John Duchi, Elad Hazan, and Yoram Singer. 2011. Adaptive subgradient methods for online learning and stochastic optimization. Journ. Mach. Learn. Res., 12:2121–2159. find
  8. Jianfeng Gao, Xiaodong He, Wen tau Yih, and Li Deng. 2014a. Learning continuous phrase representations for translation modeling. In Proc. of ACL, pages 699–709. find
  9. Jianfeng Gao, Patrick Pantel, Michael Gamon, Xiaodong He, and Li Deng. 2014b. Modeling interestingness with deep neural networks. In Proc. of EMNLP, pages 2–13. find
  10. Kallirroi Georgila, James Henderson, and Oliver Lemon. 2006. User simulation for spoken dialogue systems: Learning and evaluation. In Proc. of Interspeech/ICSLP. find
  11. Michael Gutmann and Aapo Hyvärinen. 2010. Noise-contrastive estimation: A new estimation principle for unnormalized statistical models. In Proc. of AISTATS, pages 297–304. find
  12. Po-Sen Huang, Xiaodong He, Jianfeng Gao, Li Deng, Alex Acero, and Larry Heck. 2013. Learning deep structured semantic models for web search using clickthrough data. In Proc. of CIKM, pages 2333–2338. find
  13. Nal Kalchbrenner and Phil Blunsom. 2013. Recurrent continuous translation models. Proc. of EMNLP, pages 1700–1709. find
  14. Reinhard Kneser and Hermann Ney. 1995. Improved backing-off for M-gram language modeling. In Proc. of ICASSP, pages 181–184, May. find
  15. 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
  16. Tomas Mikolov and Geoffrey Zweig. 2012. Context Dependent Recurrent Neural Network Language Model. find
  17. Tomas Mikolov, Martin Karafiát, Lukas Burget, Jan Cernocký, and Sanjeev Khudanpur. 2010. Recurrent neural network based language model. In Proc. of INTERSPEECH, pages 1045–1048. find
  18. Franz Josef Och and Hermann Ney. 2004. The alignment template approach to machine translation. Comput. Linguist., 30(4):417–449. find
  19. Franz Josef Och. 2003. Minimum error rate training in statistical machine translation. In Proc. of ACL, pages 160–167. find
  20. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. BLEU: a method for automatic evaluation of machine translation. In Proc. of ACL, pages 311–318. find our card
  21. Razvan Pascanu, Tomas Mikolov, and Yoshua Bengio. 2013. On the difficulty of training recurrent neural networks. Proc. of ICML, pages 1310–1318. find
  22. Alan Ritter, Colin Cherry, and William B. Dolan. 2011. Data-driven response generation in social media. In Proc. of EMNLP, pages 583–593. find
  23. Stephen E Robertson, Steve Walker, Susan Jones, et al. 1995. Okapi at TREC-3. In TREC. find our card
  24. David E. Rumelhart, Geoffrey E. Hinton, and Ronald J. Williams. 1988. Learning representations by back-propagating errors. In James A. Anderson and Edward Rosenfeld, editors, Neurocomputing: Foundations of Research, pages 696–699. MIT Press, Cambridge, MA, USA. find
  25. Yelong Shen, Xiaodong He, Jianfeng Gao, Li Deng, and Grégoire Mesnil. 2014. A latent semantic model with convolutional-pooling structure for information retrieval. In Proc. of CIKM, pages 101–110. find
  26. Amanda Stent and Srinivas Bangalore. 2014. Natural Language Generation in Interactive Systems. Cambridge University Press. find
  27. Ilya Sutskever, Oriol Vinyals, and Quoc V. V Le. 2014. Sequence to sequence learning with neural networks. Proc. of NIPS. find
  28. Marilyn A. Walker, Rashmi Prasad, and Amanda Stent. 2003. A trainable generator for recommendations in multimodal dialog. In Proc. of EUROSPEECH. find
  29. Steve Young, Milica Gašić, Simon Keizer, François Mairesse, Jost Schatzmann, Blaise Thomson, and Kai Yu. 2010. The hidden information state model: A practical framework for pomdp-based spoken dialogue management. Comput. Speech Lang., 24(2):150–174. find
  30. Steve Young. 2002. Talking to machines (statistically speaking). In Proc. of INTERSPEECH. find