Elamur Research⌕
☰
home / papers / 2020 / A Top-down Neural Architecture towards Text-level Parsing of Discourse Rhetorical Structure

A Top-down Neural Architecture towards Text-level Parsing of Discourse Rhetorical Structure

Summary

Earlier text-level parsers of discourse rhetorical structure (DRS) all work bottom-up. They merge adjacent elementary discourse units (EDUs) step by step, so each decision relies mainly on local context. Zhang et al. argue that a top-down strategy suits the task better for three reasons: it can exploit document-wide information, it resembles how readers move from general theme to details, and it matches how the Chinese CDTB was annotated. Their parser treats tree building as recursive ranking of split points, the boundaries between neighbouring EDUs. A BiGRU encoder with self-attention represents each EDU. A convolutional split point encoder represents each boundary. A GRU decoder with an internal stack then splits spans one after another, and at each step Biaffine Attention picks the split point and jointly predicts nuclearity and relation. On the English RST-DT and Chinese CDTB corpora, the model is competitive with or better than strong neural and feature-based bottom-up parsers. Its clearest gains are in nuclearity, which shows top-down parsing is a viable approach at the text level, not only within sentences.

Key points

  • This is the first top-down neural DRS parser that operates at the text level. Liu et al. (2019) and Lin et al. (2019) had previously applied top-down parsing, but only within sentences. The authors stress that this scale-up is hard: RST-DT averages about 2.5 EDUs per sentence but 55.6 EDUs per document.
  • The model has three parts: an EDU encoder (word and POS embeddings, BiGRU, self-attention), a split point encoder that combines the EDUs on either side of each boundary, and a stack-based attention encoder-decoder. Training uses only Negative Log Likelihood Loss.
  • On CDTB, the authors compared against their re-implementation of Sun and Kong (2018) as a bottom-up baseline, out of 933 gold nodes:
    • Top-down got more nodes correct on Bare (795 vs 791), Nuc (535 vs 521) and Rel (497 vs 486).
    • It got fewer correct on Full (426 vs 445), which the authors attribute to the baseline's joint learning framework.
  • The largest gain is on asymmetric nuclearity: NS 42.2 vs 35.4 and SN 33.7 vs 24.5. NN is roughly equal (67.0 vs 67.6).
  • With automatic EDU segmentation and Berkeley parser trees instead of gold input, scores drop considerably on every metric, showing that EDU segmentation errors weigh heavily. Chinese results exceed English, likely because CDTB annotates paragraphs rather than whole documents, which gives smaller trees.

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

23 papers of our library cite it (of the 140673 whose bibliographies we read); 7 of them have a page here, 16 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2020.acl-main.569 · arXiv 2005.02680

@inproceedings{zhang2020top,
  title = {{A Top-down Neural Architecture towards Text-level Parsing of Discourse Rhetorical Structure}},
  author = {Zhang, Longyin and Xing, Yuqing and Kong, Fang and Li, Peifeng and Zhou, Guodong},
  booktitle = {Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics},
  year = {2020},
  doi = {10.18653/v1/2020.acl-main.569},
  eprint = {2005.02680},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.18653/v1/2020.acl-main.569}
}

References

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

  1. Chloé Braud, Maximin Coavoux, and Anders Søgaard. 2017. Cross-lingual RST discourse parsing. arXiv preprint arXiv:1701.02946. find our card
  2. Chloé Braud, Barbara Plank, and Anders Søgaard. 2016. Multi-view and multi-task training of RST discourse parsers. In Proceedings of COLING 2016, pages 1903–1913. find
  3. Lynn Carlson and Daniel Marcu. 2001. Discourse tagging reference manual. ISI Technical Report ISI-TR-545, 54:56. find
  4. Kyunghyun Cho, Bart van Merrienboer, Caglar Gulcehre, Dzmitry Bahdanau, Fethi Bougares, Holger Schwenk, and Yoshua Bengio. 2014. Learning phrase representations using RNN encoder–decoder for statistical machine translation. In Proceedings of EMNLP 2014, pages 1724–1734. find
  5. Eunsol Choi, Hannah Rashkin, Luke Zettlemoyer, and Yejin Choi. 2016. Document-level sentiment inference with social, faction, and discourse context. In Proceedings of ACL 2016, pages 333–343. find
  6. Qianyin Dai, Longyin Zhang, and Fang Kong. 2019. Event temporal relation identification based on dependency and discourse relation. find
  7. Timothy Dozat and Christopher D. Manning. 2017. Deep biaffine attention for neural dependency parsing. In Proceedings of ICLR 2017. find our card
  8. Vanessa Wei Feng and Graeme Hirst. 2014. A linear-time bottom-up discourse parser with constraints and post-editing. In Proceedings of ACL 2014, pages 511–521. find our card
  9. Naman Goyal and Jacob Eisenstein. 2016. A joint model of rhetorical discourse structure and summarization. In Proceedings of the Workshop on Structured Prediction for NLP, pages 25–34. find
  10. Michael Heilman and Kenji Sagae. 2015. Fast rhetorical structure theory discourse parsing. arXiv preprint arXiv:1505.02425. find our card
  11. Hugo Hernault, Helmut Prendinger, Mitsuru Ishizuka, et al. 2010. Hilda: A discourse parser using support vector machine classification. Dialogue & Discourse, 1(3). find our card
  12. Yangfeng Ji and Jacob Eisenstein. 2014. Representation learning for text-level discourse parsing. In Proceedings of ACL 2014, pages 13–24. find our card
  13. Yangfeng Ji and Noah A. Smith. 2017. Neural discourse structure for text categorization. In Proceedings of ACL 2017, pages 996–1005. find
  14. Shafiq Joty, Giuseppe Carenini, Raymond Ng, and Yashar Mehdad. 2013. Combining intra-and multi-sentential rhetorical parsing for document-level discourse analysis. In Proceedings of ACL 2013, pages 486–496. find
  15. Jing Li, Aixin Sun, and Shafiq Joty. 2018. Segbot: A generic neural text segmentation model with pointer network. In IJCAI, pages 4166–4172. find
  16. Jiwei Li, Rumeng Li, and Eduard Hovy. 2014a. Recursive deep models for discourse parsing. In Proceedings of EMNLP 2014, pages 2061–2069. find our card
  17. Qi Li, Tianshi Li, and Baobao Chang. 2016. Discourse parsing with attention-based hierarchical neural networks. In Proceedings of EMNLP 2016, pages 362– 371. find our card
  18. Sujian Li, Liang Wang, Ziqiang Cao, and Wenjie Li. 2014b. Text-level discourse dependency parsing. In Proceedings of ACL 2014, pages 25–35. find
  19. Yancui Li, wenhe Feng, jing Sun, Fang Kong, and Guodong Zhou. 2014c. Building chinese discourse corpus with connective-driven dependency tree structure. In Proceedings of EMNLP 2014, pages 2105–2114. find
  20. Yancui Li, Wenhe Feng, Guodong Zhou, and Kunhua Zhu. 2013. Research of Chinese clause identificiton based on comma. Acta Scientiarum Naturalium Universitatis Pekinensis, 49(1):7–14. find
  21. Xiang Lin, Shafiq Joty, Prathyusha Jwalapuram, and M Saiful Bari. 2019. A unified linear-time framework for sentence-level discourse parsing. In Proceedings of ACL 2019, pages 4190–4200. find our card
  22. Linlin Liu, Xiang Lin, Shafiq Joty, Simeng Han, and Lidong Bing. 2019. Hierarchical pointer net parsing. In Proceedings of EMNLP 2019, pages 1006–1016. find
  23. William Mann and Sandra Thompson. 1988. Rhetorical structure theory: Toward a functional theory of text organization. Text, 8(3):243–281. find our card
  24. Daniel Marcu. 2000. The Theory and Practice of Discourse Parsing and Summarization. MIT Press. find
  25. Mathieu Morey, Philippe Muller, and Nicholas Asher. 2018. A dependency perspective on RST discourse parsing and evaluation. Computational Linguistics, pages 198–235. find our card
  26. Jeffrey Pennington, Richard Socher, and Christopher Manning. 2014. Glove: Global vectors for word representation. In Proceedings of EMNLP 2014, pages 1532–1543. find
  27. Rashmi Prasad, Nikhil Dinesh, Alan Lee, Eleni Milt-sakaki, Livio Robaldo, Aravind Joshi, and Bonnie Webber. 2008. The penn discourse treebank 2.0. In LREC 2008. find our card
  28. Yuanyuan Qiu, Hongzheng Li, Shen Li, Yingdi Jiang, Renfen Hu, and Lijiao Yang. 2018. Revisiting correlations between intrinsic and extrinsic evaluations of word embeddings. In CCL & NLP-NABD 2017, pages 209–221. Springer. find
  29. Cheng Sheng, Fang Kong, and Guodong Zhou. 2017. Towards better Chinese zero pronoun resolution from discourse perspective. In Processings of NLPCC 2017, pages 406–418. find
  30. Cheng Sun and Fang Kong. 2018. A transition-based framework for Chinese discourse structure parsing. Journal of Chinese Information Processing, 32(12):26–34. find
  31. Yizhong Wang, Sujian Li, and Houfeng Wang. 2017. A two-stage parsing method for text-level discourse analysis. In Proceedings of ACL 2017: short paper, pages 184–188. find our card
  32. 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

↑ back to top