Elamur Research⌕
☰
home / papers / 2015 / Closing the Gap: Domain Adaptation from Explicit to Implicit Discourse Relations

Closing the Gap: Domain Adaptation from Explicit to Implicit Discourse Relations

Summary

Implicit discourse relations, which lack a connective such as "however", are costly to annotate. Explicitly signaled relations, by contrast, can be labeled automatically from their connectives. However, classifiers trained on such explicit data transfer poorly to implicit relations. Ji, Zhang and Eisenstein treat this as a domain adaptation problem, with explicit relations as the source domain and implicit ones as the target. They combine two techniques. First, a marginalized denoising autoencoder (mDA) learns dense shared features and is kept tractable by reconstructing only pivot features. Second, a resampling step draws explicit examples with replacement so that their label proportions match the implicit distribution. This step requires only that target distribution as minimal supervision, and it keeps only instances that are cosine-similar to some target example. The authors test on the Penn Discourse Treebank using binary linear SVMs for the four top-level relation classes, with no implicit training labels. Average F1 rises from about 31% to 39.5%, close to the 41.3% supervised ceiling. Cheap explicit data can therefore nearly replace manual implicit annotation.

Key points

  • Data: PDTB explicit relations from sections 02-20 and 23-24 serve as training data, labeled through the majority connective-to-relation mapping of Prasad et al. (2007). Implicit relations from sections 21-22 form the test set. The four classes are Temporal, Comparison, Expansion and Contingency.
  • Training directly on explicit examples with the FULL feature set gives 31.04 average F1, versus 41.32 for in-domain supervision, a transfer loss of 10.28 points.
  • Using only κ high-frequency PIVOT features gives a stronger baseline of 32.74. Resampling alone reaches 35.44, mDA alone 36.69, and both together 38.62.
  • Adding about 1,000 CNN.com news articles, with adjacent sentences used as Arg1/Arg2, lifts the score to 39.46. This is a gain of 8.42 points and removes over 80% of the transfer loss. Using even more extra data brought no further gain.
  • Replacing the gold PDTB argument spans with adjacent sentences lowers the score to 38.52, which shows that argument identification matters.

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: Linguistics & languages

Topics: Discourse analysis

Cited by

14 papers of our library cite it (of the 140673 whose bibliographies we read); 7 of them have a page here, 7 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/d15-1264

@inproceedings{ji2015closing,
  title = {{Closing the Gap: Domain Adaptation from Explicit to Implicit Discourse Relations}},
  author = {Ji, Yangfeng and Zhang, Gongbo and Eisenstein, Jacob},
  booktitle = {Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing},
  year = {2015},
  doi = {10.18653/v1/d15-1264},
  url = {https://doi.org/10.18653/v1/d15-1264}
}

References

24 references, 13 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Shai Ben-David, John Blitzer, Koby Crammer, Fernando Pereira, et al. 2007. Analysis of representations for domain adaptation. In Neural Information Processing Systems (NIPS), Vancouver. find
  2. Or Biran and Kathleen McKeown. 2013. Aggregated word pair features for implicit discourse relation disambiguation. In Proceedings of the Association for Computational Linguistics (ACL), pages 69–73, Sophia, Bulgaria. find our card
  3. John Blitzer, Ryan McDonald, and Fernando Pereira. 2006. Domain adaptation with structural correspondence learning. In Proceedings of Empirical Methods for Natural Language Processing (EMNLP), pages 120–128. find
  4. Minmin Chen, Z. Xu, Killian Weinberger, and Fei Sha. 2012. Marginalized denoising autoencoders for domain adaptation. In Proceedings of the International Conference on Machine Learning (ICML). find
  5. Rong-En Fan, Kai-Wei Chang, Cho-Jui Hsieh, Xiang-Rui Wang, and Chih-Jen Lin. 2008. LIBLINEAR: A library for large linear classification. The Journal of Machine Learning Research, 9:1871–1874. find
  6. Xavier Glorot, Antoine Bordes, and Yoshua Bengio. 2011. Domain adaptation for large-scale sentiment classification: A deep learning approach. In Proceedings of the International Conference on Machine Learning (ICML), Seattle, WA. find
  7. H Paul Grice. 1975. Logic and Conversation. In P. Cole and J. L. Morgan, editors, Syntax and Semantics Volume 3: Speech Acts, pages 41–58. Academic Press. find
  8. Jing Jiang and ChengXiang Zhai. 2007. Instance weighting for domain adaptation in nlp. In Proceedings of the Association for Computational Linguistics (ACL), Prague. find
  9. Mahesh Joshi, William W Cohen, Mark Dredze, and Carolyn P Rosé. 2012. Multi-domain learning: when do domains matter? In Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 1302–1312. Association for Computat… find
  10. Man Lan, Yu Xu, and Zhengyu Niu. 2013. Leveraging Synthetic Discourse Data via Multi-task Learning for Implicit Discourse Relation Recognition. In Proceedings of the Association for Computational Linguistics (ACL), pages 476–485, Sophia, Bulgaria. find our card
  11. Ziheng Lin, Min-Yen Kan, and Hwee Tou Ng. 2009. Recognizing implicit discourse relations in the penn discourse treebank. In Proceedings of Empirical Methods for Natural Language Processing (EMNLP), pages 343–351, Singapore. find our card
  12. Ziheng Lin, Hwee Tou Ng, and Min-Yen Kan. 2011. Automatically Evaluating Text Coherence Using Discourse Relations. In Proceedings of the Association for Computational Linguistics (ACL), pages 997–1006, Portland, OR. find our card
  13. Annie Louis, Aravind Joshi, and Ani Nenkova. 2010. Discourse indicators for content selection in summarization. In Proceedings of the 11th Annual Meeting of the Special Interest Group on Discourse and Dialogue, pages 147–156. Association for Computational Linguistics. find
  14. 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, pages 55–60. find our card
  15. Daniel Marcu and Abdessamad Echihabi. 2003. An unsupervised approach to recognizing discourse relations. In Proceedings of the Association for Computational Linguistics (ACL), pages 368–375. find our card
  16. Joonsuk Park and Claire Cardie. 2012. Improving Implicit Discourse Relation Recognition Through Feature Set Optimization. In Proceedings of the 13th Annual Meeting of the Special Interest Group on Discourse and Dialogue, pages 108–112, Seoul, South Korea, July. Association for Computational Linguist… find our card
  17. Emily Pitler, Mridhula Raghupathy, Hena Mehta, Ani Nenkova, Alan Lee, and Aravind Joshi. 2008. Easily identifiable discourse relations. In Proceedings of the International Conference on Computational Linguistics (COLING), pages 87–90, Manchester, UK. find our card
  18. Emily Pitler, Annie Louis, and Ani Nenkova. 2009. Automatic Sense Prediction for Implicit Discourse Relations in Text. In Proceedings of the Association for Computational Linguistics (ACL), Suntec, Singapore. find our card
  19. Rashmi Prasad, Eleni Miltsakaki, Nikhil Dinesh, Alan Lee, Aravind Joshi, Livio Robaldo, and Bonnie L Webber. 2007. The Penn Discourse Treebank 2.0 annotation manual. The PDTB Research Group. find
  20. Rashmi Prasad, Nikhil Dinesh, Alan Lee, Eleni Miltsakaki, Livio Robaldo, Aravind Joshi, and Bonnie Webber. 2008. The Penn Discourse Treebank 2.0. In Proceedings of LREC. find our card
  21. Attapol Rutherford and Nianwen Xue. 2015. Improving the Inference of Implicit Discourse Relations via Classifying Explicit Discourse Connectives. In Proceedings of the North American Chapter of the Association for Computational Linguistics (NAACL), pages 799–808, Denver, CO, May–June. find our card
  22. Swapna Somasundaran, Galileo Namata, Janyce Wiebe, and Lise Getoor. 2009. Supervised and unsupervised methods in employing discourse relations for improving opinion polarity classification. In Proceedings of Empirical Methods for Natural Language Processing (EMNLP), Singapore. find our card
  23. Caroline Sporleder and Alex Lascarides. 2008. Using automatically labelled examples to classify rhetorical relations: An assessment. Natural Language Engineering, 14(3):369–416. find
  24. Yasuhisa Yoshida, Jun Suzuki, Tsutomu Hirao, and Masaaki Nagata. 2014. Dependency-based Discourse Parser for Single-Document Summarization. In Proceedings of Empirical Methods for Natural Language Processing (EMNLP). find our card

↑ back to top