Elamur Research⌕
☰
home / papers / 2019 / Rethinking Cooperative Rationalization: Introspective Extraction and Complement Control

Rethinking Cooperative Rationalization: Introspective Extraction and Complement Control

Paper · 2019 · Mo Yu, Shiyu Chang, Yang Zhang et al.

Summary

Selective rationalization trains a generator to highlight a small part of the input text and a predictor that classifies using only that highlighted part. This paper identifies two weaknesses of the cooperative setup.

  • No access to the outcome. The generator does not know which outcome it should justify, so training can settle into poorly generalizing solutions.
  • No control over the unselected text. Nothing checks what is left out of the rationale. This allows "degeneration": the two players agree on a trivial code, such as punctuation or word position, that signals the label without explaining it.

The authors propose two fixes. The first is an introspective generator, which receives the label predicted from the full text as an extra input. The second is a three-player game that adds an adversarial complement predictor. This third player tries to recover the label from the words outside the rationale, and the generator tries to defeat it.

They formalize three properties of a good rationale: sufficiency, comprehensiveness and compactness. They show theoretically that the game's equilibrium yields rationales with these properties. Experiments use beer-review sentiment and relation classification. The combined model keeps predictive accuracy high and produces more complete, non-degenerate rationales than Lei et al. (2016). Human studies support this result.

Key points

  • Comprehensiveness condition. The label entropy given the complement must exceed the label entropy given the rationale by some margin. The authors argue that only this condition rules out degeneration.
  • Model variants. The paper compares three frameworks:
    • the cooperative baseline of Lei et al. (2016);
    • the same baseline with an added complement predictor (minimax);
    • the full introspective three-player model (Intros+minimax).
  • Human evaluation on beer reviews. Annotators saw only the unselected text. The full model drove their label guesses closer to chance and led to more "UNK" answers. Some differences were significant at p < 0.005 (t-test).
  • Relation classification. Lei et al. (2016) sometimes failed to cover both entities. Dataset biases led the proposed model to occasionally keep extra correlated words, such as "working".
  • Code and data are released at github.com/Gorov/three_player_for_emnlp.

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

🏢IBMArmonk · United States85 works · 66,118 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

Cite

Cite as Mo Yu, Shiyu Chang, Yang Zhang and Tommi Jaakkola. 2019. Rethinking Cooperative Rationalization: Introspective Extraction and Complement Control. 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/d19-1420

@inproceedings{yu2019rethinking,
  title = {{Rethinking Cooperative Rationalization: Introspective Extraction and Complement Control}},
  author = {Mo Yu, Shiyu Chang, Yang Zhang and Tommi Jaakkola},
  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/d19-1420},
  url = {https://doi.org/10.18653/v1/d19-1420}
}

References

30 references, 11 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Lucian Busoniu, Robert Babuska, and Bart De Schutter. 2006. Multi-agent reinforcement learning: A survey. In 2006 9th International Conference on Control, Automation, Robotics and Vision, pages 1– 6. IEEE. find
  2. Jianbo Chen, Le Song, Martin Wainwright, and Michael Jordan. 2018a. Learning to explain: An information-theoretic perspective on model interpretation. In International Conference on Machine Learning, pages 882–891. find
  3. Jianbo Chen, Le Song, Martin J Wainwright, and Michael I Jordan. 2018b. L-Shapley and C-Shapley: Efficient model interpretation for structured data. arXiv preprint arXiv:1808.02610. — [arXiv:1808.02610](https://arxiv.org/abs/1808.02610) find
  4. Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. 2014. Generative adversarial nets. In Advances in neural information processing systems, pages 2672–2680. find
  5. Iris Hendrickx, Su Nam Kim, Zornitsa Kozareva, Preslav Nakov, Diarmuid Ó Séaghdha, Sebastian Padó, Marco Pennacchiotti, Lorenza Romano, and Stan Szpakowicz. 2009. Semeval-2010 task 8: Multi-way classification of semantic relations between pairs of nominals. In Proceedings of the Workshop on Seman… our card
  6. Justin Johnson, Bharath Hariharan, Laurens van der Maaten, Judy Hoffman, Li Fei-Fei, C Lawrence Zitnick, and Ross Girshick. 2017. Inferring and executing programs for visual reasoning. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 2989–2998. find
  7. David Alvarez-Melis and Tommi S Jaakkola. 2018. Towards robust interpretability with self-explaining neural networks. arXiv preprint arXiv:1806.07538. — [arXiv:1806.07538](https://arxiv.org/abs/1806.07538) find
  8. Igor Kononenko et al. 2010. An efficient explanation of individual classifications using game theory. Journal of Machine Learning Research, 11(Jan):1–18. find
  9. Jacob Andreas, Marcus Rohrbach, Trevor Darrell, and Dan Klein. 2016a. Learning to compose neural networks for question answering. In Proceedings of NAACL-HLT, pages 1545–1554. our card
  10. Guang-He Lee, David Alvarez-Melis, and Tommi S Jaakkola. 2018. Game-theoretic interpretability for temporal modeling. arXiv preprint arXiv:1807.00130. — [arXiv:1807.00130](https://arxiv.org/abs/1807.00130) find
  11. Jacob Andreas, Marcus Rohrbach, Trevor Darrell, and Dan Klein. 2016b. Neural module networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 39–48. our card
  12. Tao Lei, Regina Barzilay, and Tommi Jaakkola. 2016. Rationalizing neural predictions. In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 107–117. our card
  13. Kristjan Arumae and Fei Liu. 2019. Guiding extractive summarization with question-answering rewards. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 2566– 2577. find
  14. Jiwei Li, Xinlei Chen, Eduard Hovy, and Dan Jurafsky. 2016a. Visualizing and understanding neural models in NLP. In Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 681–691. our card
  15. Joost Bastings, Wilker Aziz, and Ivan Titov. 2019. Interpretable neural predictions with differentiable binary variables. arXiv preprint arXiv:1905.08160. our card
  16. Jiwei Li, Will Monroe, and Dan Jurafsky. 2016b. Understanding neural networks through representation erasure. arXiv preprint arXiv:1612.08220. find
  17. Scott M Lundberg and Su-In Lee. 2017. A unified approach to interpreting model predictions. In Advances in Neural Information Processing Systems, pages 4765–4774. our card
  18. Yuan Zhang, Regina Barzilay, and Tommi Jaakkola. 2017. Aspect-augmented adversarial networks for domain adaptation. Transactions of the Association for Computational Linguistics, 5:515–528. find
  19. Julian McAuley, Jure Leskovec, and Dan Jurafsky. 2012. Learning attitudes and attributes from multi-aspect reviews. In 2012 IEEE 12th International Conference on Data Mining, pages 1020–1025. IEEE. find
  20. Mingmin Zhao, Shichao Yue, Dina Katabi, Tommi S Jaakkola, and Matt T Bianchi. 2017. Learning sleep stages from radio signals: A conditional adversarial architecture. In Proceedings of the 34th International Conference on Machine Learning-Volume 70, pages 4100–4109. JMLR. org. find
  21. Thien Huu Nguyen and Ralph Grishman. 2015. Combining neural networks and log-linear models to improve relation extraction. arXiv preprint arXiv:1511.05926. — [arXiv:1511.05926](https://arxiv.org/abs/1511.05926) find
  22. Marco Tulio Ribeiro, Sameer Singh, and Carlos Guestrin. 2016. Why should I trust you?: Explaining the predictions of any classifier. In Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining, pages 1135–1144. ACM. our card
  23. David Silver, Thomas Hubert, Julian Schrittwieser, Ioannis Antonoglou, Matthew Lai, Arthur Guez, Marc Lanctot, Laurent Sifre, Dharshan Kumaran, Thore Graepel, et al. 2017. Mastering chess and shogi by self-play with a general reinforcement learning algorithm. arXiv preprint arXiv:1712.01815. our card
  24. Karen Simonyan, Andrea Vedaldi, and Andrew Zisserman. 2013. Deep inside convolutional networks: Visualising image classification models and saliency maps. arXiv preprint arXiv:1312.6034. our card
  25. Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. 2014. Dropout: a simple way to prevent neural networks from overfitting. The Journal of Machine Learning Research, 15(1):1929–1958. find
  26. Mukund Sundararajan, Ankur Taly, and Qiqi Yan. 2017. Axiomatic attribution for deep networks. In Proceedings of the 34th International Conference on Machine Learning-Volume 70, pages 3319–3328. JMLR. org. find
  27. Ronald J. Williams. 1992. Simple statistical gradient-following algorithms for connectionist reinforcement learning. Machine Learning. find
  28. Adam Yala, Constance Lehman, Tal Schuster, Tally Portnoi, and Regina Barzilay. 2019. A deep learning mammography-based model for improved breast cancer risk prediction. Radiology, page 182716. find
  29. Mo Yu, Shiyu Chang, and Tommi S Jaakkola. 2018. Learning corresponded rationales for text matching. find
  30. Omar Zaidan, Jason Eisner, and Christine Piatko. 2007. Using annotator rationales to improve machine learning for text categorization. In Human language technologies 2007: The conference of the North American chapter of the association for computational linguistics; proceedings of the main conferenc… our card

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/yu-2019-rethinking-cooperative-rationalization-introspective-extraction/