Elamur Research⌕
☰
home / papers / 2018 / Deep Communicating Agents for Abstractive Summarization

Deep Communicating Agents for Abstractive Summarization

Summary

Neural encoder-decoder models handle short inputs well, but their performance drops when a long document must be condensed into an abstractive summary. Celikyilmaz et al. address this by splitting the encoding work among several agents, each responsible for one paragraph. They adapt CommNet to sequence generation. Each agent first encodes its paragraph with a local bLSTM. It then passes through stacked contextual layers, where it receives the averaged final states of the other agents as a message and updates its own representation.

A single LSTM decoder uses hierarchical attention: first over the words within each agent, then over the agents. A contextual agent attention component discourages the decoder from jumping between agents too often. A multi-agent pointer mechanism lets each agent vote on which out-of-vocabulary words to copy. The model is trained end-to-end with self-critical reinforcement learning.

On CNN/DailyMail and New York Times, communicating encoders beat single-encoder and non-communicating multi-encoder baselines, and human judges rated the summaries as more focused. The work matters because it shows that agent communication is a practical way to represent long inputs.

Key points

  • The architecture combines a local bLSTM encoder, K contextual layers with message passing (messages are averages of the other agents' last hidden states), word-level and agent-level attention, and per-agent generation probabilities for copying.
  • Training uses self-critical RL, together with a semantic coherence loss and intermediate sentence-level rewards intended to reduce semantically repetitive output.
  • In the human evaluation on a CNN/DM subset, the multi-agent model outscored the single agent on focus (3.884 vs 3.694, p < 0.001) and overall quality (3.682 vs 3.558, p < 0.03). It was also preferred head-to-head on overall quality (158 vs 102 wins).
  • An analysis of ROUGE-L scores against how much attention each agent received showed that summaries drawing more evenly on all agents scored higher. The multi-agent model also outperformed the single agent in every attention bin.
  • In qualitative examples, the best multi-agent model (m7) captured key facts that the best single-agent model (m3) missed, and it included fewer unnecessary details.

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: Natural language processing

Cited by

108 papers of our library cite it (of the 140884 whose bibliographies we read); 10 of them have a page here, 98 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/n18-1150

@inproceedings{celikyilmaz2018deep,
  title = {{Deep Communicating Agents for Abstractive Summarization}},
  author = {Celikyilmaz, Asli and Bosselut, Antoine and He, Xiaodong and Choi, Yejin},
  booktitle = {Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers)},
  year = {2018},
  doi = {10.18653/v1/n18-1150},
  url = {https://doi.org/10.18653/v1/n18-1150}
}

References

36 references, 12 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. 2015. Neural machine translation by jointly learning to align and translate. In ICLR. find our card
  2. Huimin Chen, Maosong Sun, Cunchao Tu, Yankai Lin, and Zhiyuan Liu. 2016. Neural sentiment classification with user and product attention. In EMNLP. find
  3. Jianpeng Cheng and Mirella Lapata. 2016. Neural summarization by extracting sentences and words. In ACL. find our card
  4. Sumit Chopra, Michael Auli, and Alexander M. Rush. 2016. Abstractive sentence summarization with attentive recurrent neural networks. In NAACL. find our card
  5. Abhishek Das, Satwik Kottur, Jos M. F. Moura, and Dhruv Batra Stefan Lee. 2017. Learning cooperative visual dialog agents with deep reinforcement learning. In CVPR. find
  6. Greg Durrett, Taylor Berg-Kirkpatrick, and Dan Klein. 2016. Learning-based single-document summarization with compression and anaphoricity constraints. In ACL. find our card
  7. Angela Fan, David Grangier, and Michael Auli. 2017. Controllable abstractive summarization. In arXiv:1711.05217v1. find our card
  8. J. Foerster, G. Farquhar, T. Afouras, N. Nardelli, and S. Whiteson. 2017. Counterfactual multi-agent policy gradients. In arXiv:1705.08926. — [arXiv:1705.08926](https://arxiv.org/abs/1705.08926) find
  9. Jakob N. Foerster, Yannis M. Assael, Nando de Freitas, and Shimon Whiteson. 2016. Learning to communicate with deep multi-agent reinforcement learning. In NIPS. find
  10. Karl Moritz Hermann, Tomas Kocisky, Edward Grefenstette, Lasse Espeholt, Will Kay, Mustafa Suleyman, and Phil Blunsom. 2015. Teaching machines to read and comprehend. In NIPS. find our card
  11. Kai Hong, Michel Marcus, and Ani Nenkova. 2015. System combination for multi-document summarization. In EMNLP. find
  12. Kai Hong and Ani Nenkova. 2014. Improving the estimation of word importance for news multi-document summarization. In Extended technical report. find
  13. Natasha Jaques, Shixiang Gu, Dzmitry Bahdanau, Jose Miguel Hernandez-Lobato, Richard E. Turner, and Douglas Eck. 2017. Sequence tutor: Conservative fine-tuning of sequence generation models with kl-control. In ICML. find
  14. Angeliki Lazaridou, Nghia The Pham, and Marco Baroni. 2016. Towards multi-agent communication-based language learning. In arXiv:1605.07133. — [arXiv:1605.07133](https://arxiv.org/abs/1605.07133) find
  15. Junyi Jessy Li, Kapil Thandani, and Amanda Stent. 2016. The role of discourse units in near-extractive summarization. In Proceedings of the 17th Annual Meeting of the Special Interest Group on Discourse and Dialog. find
  16. Minh-Thang Luong, Hieu Pham, and Christopher D. Manning. 2015. Effective approaches to attention-based neural machine translation. In EMNLP. find
  17. Christopher D. Manning, Mihai Surdeanu, John Bauer, Jenny Finkel, Steven J. Bethard, and David Mc-Closky. 2014. The Stanford CoreNLP natural language processing toolkit. In Proceedings of 52nd Annual Meeting of the Association for Computational Linguistics: System Demonstrations. find our card
  18. I. Mordatch and P. Abbeel. 2017. Emergence of grounded compositional language in multi-agent populations. In arXiv:1703.04908. — [arXiv:1703.04908](https://arxiv.org/abs/1703.04908) find
  19. Oriol Vinyals, Meire Fortunato, and Navdeep Jaitly. 2015a. Pointer networks. In NIPS. find
  20. Oriol Vinyals, Lukasz Kaiser, Terry Koo, Slav Petrov, Ilya Sutskever, and Geoffrey Hinton. 2015b. Grammar as a foreign language. In NIPS. find
  21. Yonghui Wu, Mike Schuster, Zhifeng Chen, Quoc V Le, Mohammad Norouzi, Wolfgang Macherey, Maxim Krikun, Yuan Cao, Qin Gao, Klaus Macherey, et al. 2016. Google’s neural machine translation system: Bridging the gap between human and machine translation. arXiv preprint arXiv:1609.08144 . find
  22. Ramesh Nallapati, Feifei Zhai, and Bowen Zhou. 2017. Summarunner: A recurrent neural network based sequence model for extractive summarization of documents. In Association for the Advancement of Artificial Intelligence. find our card
  23. Zichao Yang, Diyi Yang, Chris Dyer, Xiaodong He, Alex Smola, and Eduard Hovy. 2016. Hierarchical attention networks for document classification. In NAACL. find our card
  24. Ramakanth Pasunuru and Mohit Bansal. 2017. Reinforced video captioning with entailment rewards. In EMNLP. find
  25. Wenyuan Zeng, Wenjie Luo, Sanja Fidler, and Raquel Urtasun. 2016. Efcient summarization with read-again and copy mechanism. In arXiv preprint arXiv:1611.03382. — [arXiv:1611.03382](https://arxiv.org/abs/1611.03382) find
  26. Romain Paulus, Caiming Xiong, and Richard Socher. 2018. A deep reinforced model for abstractive summarization. ICLR . find our card
  27. Jeffrey Pennington, Richard Socher, and Christopher D. Manning. 2014. Glove: Global vectors for word representation. In Empirical Methods in Natural Language Processing. find
  28. Xiangyang Zhou, Daxiang Dong, Hua Wu, Shiqi Zhao, D Yu R Yan, Xuan Liu, and H Tian. 2016. Multiview response selection for human-computer conversation. In EMNLP. find
  29. Steven J Rennie, Etienne Marcheret, Youssef Mroueh, Jarret Ross, and Vaibhava Goel. 2016. Self-critical sequence training for image captioning. arXiv preprint arXiv:1612.00563 . find our card
  30. Alexander M. Rush, Sumit Chopra, and Jason Weston. 2015. A neural attention model for abstractive sentence summarization. In EMNLP. find
  31. Evan Sandhaus. 2008. New york times annotated corpus. In Linguistic Data Consortium, Philadelphia. find
  32. Abigale See, Peter J. Liu, and Christopher Manning. 2017. Gettothepoint: Summarization with pointer-generatornetworks. In ACL. find
  33. Sainbayar Sukhbaatar, Arthur Szlam, and Rob Fergus. 2016. Learning multiagent communication with back-propagation. In NIPS. find
  34. Ilya Sutskever, Oriol Vinyals, and Quoc V Le. 2014. Sequence to sequence learning with neural networks. In Advances in neural information processing systems. find
  35. Jiwei Tan, Xiaojun Wan, and Jianguo Xiao. 2017. Abstractive document summarization with a graph-based attentional neural model. In ACL. find our card
  36. O. Vinyals, T. Ewalds, S. Bartunov, P. Georgiev, A. S. Vezhnevets, M. Yeo, A. Makhzani, H. Kuttler, J. Agapiou, and J. et al. Schrittwieser. 2017. Starcraft ii: A new challenge for reinforcement learning. In arXiv preprint arXiv:1708.04782. — [arXiv:1708.04782](https://arxiv.org/abs/1708.04782) find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/celikyilmaz-2018-deep-communicating-agents-abstractive-summarization/