Elamur Research⌕
☰
home / papers / 2019 / Predicting Discourse Structure using Distant Supervision from Sentiment

Predicting Discourse Structure using Distant Supervision from Sentiment

Paper · 2019 · Patrick Huber and Giuseppe Carenini

Summary

RST-style discourse parsing has not benefited much from neural methods. The main obstacle is data: human-annotated discourse treebanks hold only a few hundred documents and cover mostly news and instructional text. Huber and Carenini reverse the usual link between discourse and sentiment. Earlier work used discourse trees to improve sentiment classification. Here, document-level sentiment labels are used to induce discourse structure instead.

A MILNet-style multiple-instance network is trained only on review ratings. It gives every EDU a sentiment score and an importance (attention) score. A CKY-style dynamic program then searches for the binary tree whose aggregated scores best reproduce the document's gold sentiment. Applied to Yelp'13, this produces a silver-standard corpus of 100k documents with discourse structure.

The authors train the Two-Stage parser of Wang et al. (2017) on this corpus. It scores below parsers trained and tested on the same domain. However, it beats parsers trained on RST-DT or Instr-DT when the test domain differs from the training domain. This matters because most real applications target domains with no discourse annotation.

Key points

  • The silver corpus has 100k Yelp'13 reviews with structure only, without nuclearity or relation labels. That is about two orders of magnitude larger than RST-DT (385 docs) or Instr-DT (176 docs). EDUs come from the Feng and Hirst (2012) segmenter.
  • The model uses GloVe inputs and a two-level bi-directional GRU with attention, one level over words in an EDU and one over EDUs in the document. Sentiment and attention modules sit on top, and the EDU scores are combined in CKY using avg or max attention-aggregation.
  • Inter-domain results (Table 3):
    • On RST-DT test: Ours(max) reaches 77.24 and Ours(avg) 76.42, against 74.48 for training on Instr-DT.
    • On Instr-DT test: Ours(avg) reaches 74.22, against 73.65 for training on RST-DT.
  • Reference points (Table 3):
    • Intra-domain best: Two-Stage 86.00 on RST-DT and CODRA 82.88 on Instr-DT.
    • Human agreement: 88.30.
    • Hierarchical right-branching baseline: 70.82 and 67.86.
  • Scores grow with more silver data, and Instr-DT saturates earlier. Even 4–8 training documents work reasonably well, since a document with n EDUs yields 2n−1 binary training instances.

Interesting ideas

  • I1. If task X is known to help task Y, cheap labels for Y can distantly supervise X; the authors flip the discourse→sentiment synergy to produce abundant structural data.
  • I2. Latent structure is built by exact search (CKY) constrained to explain a known global label, rather than by a free-running predictor.
  • I3. Multiple-instance learning recovers per-unit scores (EDU sentiment and importance) from document-level labels alone.
  • I4. The silver data is judged on inter-domain transfer, the setting where it is meant to help, and compared against trivial hierarchical-branching baselines.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (learn): Plentiful surface labels (QA answers, test scores) could produce silver structures for scarce layers. One example is choosing the frame or tree analysis that makes the known answer derivable.
  • I2 (world): Event ordering or entity-state graphs could be chosen by exact search that maximizes consistency with a known story outcome. This gives an explainable reason why one structure was picked.
  • I3 (frames): Clause-level polarity and salience learned from review ratings could feed irony and implicature detection. One cue would be a positive clause buried under a negative overall verdict.
  • I4 (learn): Silver-trained parsers should be judged on out-of-domain treebanks and next to structural negative controls. Then gains reflect real transfer rather than in-domain fit.

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

🎓University of British ColumbiaVancouver · Canada46 works · 33,673 citations

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 9,086 worksArtificial Intelligence · 8,889 worksArts and Humanities · 9,296 worksLanguage and Linguistics · 9,206 works

🧭 Topics 2Research subjects and languages the work is about.Sentiment & emotion · 608 worksDiscourse analysis · 394 works

Cite

Cite as Patrick Huber and Giuseppe Carenini. 2019. Predicting Discourse Structure using Distant Supervision from Sentiment. 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/d19-1235

@inproceedings{huber2019predicting,
  title = {{Predicting Discourse Structure using Distant Supervision from Sentiment}},
  author = {Patrick Huber and Giuseppe Carenini},
  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/d19-1235},
  url = {https://doi.org/10.18653/v1/d19-1235}
}

References

37 references, 26 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Muhammad Abdul-Mageed and Lyle Ungar. 2017. Emonet: Fine-grained emotion detection with gated recurrent neural networks. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 718–728.
  2. Stefanos Angelidis and Mirella Lapata. 2018. Multiple instance learning networks for fine-grained sentiment analysis. Transactions of the Association for Computational Linguistics, 6:17–31. ↗
  3. Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. 2014. Neural machine translation by jointly learning to align and translate. arXiv preprint arXiv:1409.0473. ↗
  4. Parminder Bhatia, Yangfeng Ji, and Jacob Eisenstein. 2015. Better document-level sentiment analysis from rst discourse parsing. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 2212–2218. ↗
  5. Chloé Braud, Maximin Coavoux, and Anders Søgaard. 2017. Cross-lingual rst discourse parsing. arXiv preprint arXiv:1701.02946. ↗
  6. Chloé Braud, Barbara Plank, and Anders Søgaard. 2016. Multi-view and multi-task training of rst discourse parsers. In Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 1903–1913.
  7. Lynn Carlson, Mary Ellen Okurowski, and Daniel Marcu. 2002. RST discourse treebank. Linguistic Data Consortium, University of Pennsylvania. find
  8. Qiming Diao, Minghui Qiu, Chao-Yuan Wu, Alexander J Smola, Jing Jiang, and Chong Wang. 2014. Jointly modeling aspects, ratings and sentiments for movie recommendation (jmars). In Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining, pages 193– 202. ACM. find
  9. Vanessa Wei Feng and Graeme Hirst. 2012. Text-level discourse parsing with rich linguistic features. In Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 60–68.
  10. Vanessa Wei Feng and Graeme Hirst. 2014. A linear-time bottom-up discourse parser with constraints and post-editing. In Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), volume 1, pages 511–521.
  11. Elisa Ferracane, Greg Durrett, Junyi Jessy Li, and Katrin Erk. 2019. Evaluating discourse in structured text representations. arXiv preprint arXiv:1906.01472. ↗
  12. Shima Gerani, Yashar Mehdad, Giuseppe Carenini, Raymond T Ng, and Bita Nejat. 2014. Abstractive summarization of product reviews using discourse structure. In Proceedings of the 2014 conference on empirical methods in natural language processing (EMNLP), pages 1602–1613.
  13. Hugo Hernault, Helmut Prendinger, Mitsuru Ishizuka, et al. 2010. Hilda: A discourse parser using support vector machine classification. Dialogue & Discourse, 1(3). ↗
  14. Alexander Hogenboom, Flavius Frasincar, Franciska De Jong, and Uzay Kaymak. 2015. Using rhetorical structure in sentiment analysis. Commun. ACM, 58(7):69–77. ↗
  15. Yangfeng Ji and Jacob Eisenstein. 2014. Representation learning for text-level discourse parsing. In Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), volume 1, pages 13–24.
  16. Yangfeng Ji and Noah A Smith. 2017. Neural discourse structure for text categorization. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 996–1005. ↗
  17. Shafiq Joty, Giuseppe Carenini, and Raymond T Ng. 2015. Codra: A novel discriminative framework for rhetorical analysis. Computational Linguistics, 41(3). ↗
  18. Dan Jurafsky and James H Martin. 2014. Speech and language processing, volume 3. Pearson London. find
  19. Jim Keeler and David E Rumelhart. 1992. A self-organizing integrated segmentation and recognition neural net. In Advances in neural information processing systems, pages 496–503. find
  20. Qi Li, Tianshi Li, and Baobao Chang. 2016. Discourse parsing with attention-based hierarchical neural networks. In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 362–371.
  21. Yang Liu and Mirella Lapata. 2017. Learning contextually informed representations for linear-time discourse parsing. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 1289–1298.
  22. Yang Liu and Mirella Lapata. 2018. Learning structured text representations. Transactions of the Association for Computational Linguistics, 6:63–75. ↗
  23. William C Mann and Sandra A Thompson. 1988. Rhetorical structure theory: Toward a functional theory of text organization. Text-Interdisciplinary Journal for the Study of Discourse, 8(3):243–281. ↗
  24. Micol Marchetti-Bowick and Nathanael Chambers. 2012. Learning for microblogs with distant supervision: Political forecasting with twitter. In Proceedings of the 13th Conference of the European Chapter of the Association for Computational Linguistics, pages 603–612. find
  25. Mathieu Morey, Philippe Muller, and Nicholas Asher. 2017. How much progress have we made on rst discourse parsing? a replication study of recent results on the rst-dt. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 1319–1324.
  26. Bita Nejat, Giuseppe Carenini, and Raymond Ng. 2017. Exploring joint neural model for sentence level discourse parsing and sentiment analysis. In Proceedings of the 18th Annual SIGdial Meeting on Discourse and Dialogue, pages 289–298.
  27. Nan Yu, Meishan Zhang, and Guohong Fu. 2018. Transition-based neural rst parsing with implicit syntax features. In Proceedings of the 27th International Conference on Computational Linguistics, pages 559–570. find
  28. Matthew D Zeiler. 2012. Adadelta: an adaptive learning rate method. arXiv preprint arXiv:1212.5701. ↗
  29. Amir Zeldes. 2017. The GUM corpus: Creating multilayer resources in the classroom. Language Resources and Evaluation, 51(3):581–612. ↗
  30. Xiang Zhang, Junbo Zhao, and Yann LeCun. 2015. Character-level convolutional networks for text classification. In Advances in neural information processing systems, pages 649–657. ↗
  31. Jeffrey Pennington, Richard Socher, and Christopher D. Manning. 2014. Glove: Global vectors for word representation. In Empirical Methods in Natural Language Processing (EMNLP), pages 1532– 1543. ↗
  32. Rashmi Prasad, Nikhil Dinesh, Alan Lee, Eleni Miltsakaki, Livio Robaldo, Aravind Joshi, and Bonnie Webber. 2008. The penn discourse treebank 2.0. LREC.
  33. Rajen Subba and Barbara Di Eugenio. 2009. An effective discourse parser that uses rich linguistic information. In Proceedings of Human Language Technologies: The 2009 Annual Conference of the North American Chapter of the Association for Computational Linguistics, pages 566–574. Association for Computational Linguistics.
  34. Jeniya Tabassum, Alan Ritter, and Wei Xu. 2016. Tweetime: A minimally supervised method for recognizing and normalizing time expressions in twitter. In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 307–318. find
  35. Duyu Tang, Bing Qin, and Ting Liu. 2015. Document modeling with gated recurrent neural network for sentiment classification. In Proceedings of the 2015 conference on empirical methods in natural language processing, pages 1422–1432.
  36. Yizhong Wang, Sujian Li, and Houfeng Wang. 2017. A two-stage parsing method for text-level discourse analysis. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 184–188. ↗
  37. Zichao Yang, Diyi Yang, Chris Dyer, Xiaodong He, Alex Smola, and Eduard Hovy. 2016. Hierarchical attention networks for document classification. In Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 1480–1489.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/huber-2019-predicting-discourse-structure-distant-supervision/