Elamur Research⌕
☰
home / papers / 2013 / Combining Intra- and Multi-sentential Rhetorical Parsing for Document-level Discourse Analysis

Combining Intra- and Multi-sentential Rhetorical Parsing for Document-level Discourse Analysis

Paper · 2013 · Shafiq Joty, Giuseppe Carenini, Raymond Ng et al.

Summary

Document-level discourse parsing in the RST framework lags well behind human agreement. Existing parsers such as HILDA predict tree structure and relation labels in separate pipeline steps, build trees greedily, and treat a whole document as one uniform problem. Joty et al. propose a two-stage parser (TSP) to address all three issues. One Conditional Random Field handles parsing inside sentences and a second handles parsing across sentences. Each CRF jointly scores the span and the relation label (with nuclearity) of every candidate constituent. Both stages share one optimal, non-greedy bottom-up parsing algorithm that searches over these probabilities.

The authors test two ways to connect the stages:

  • 1S-1S: parse each sentence separately.
  • Sliding window: parse pairs of adjacent sentences, which catches structures that cross sentence boundaries.

Evaluation covers two genres, RST-DT news and an Instructional how-to corpus. The parser beats HILDA and an ILP-based system by large, statistically significant margins on span, nuclearity and relation F-scores. This narrows the gap to human performance.

Key points

  • On RST-DT with gold segmentation and 18 coarse relations, TSP reaches about 82.5–82.7 span, 68.4 nuclearity and 55.7 relation F-score. HILDA scores 74.68 / 58.99 / 44.32, and human agreement is 88.70 / 77.72 / 65.75.
  • On the Instructional corpus, TSP scores about 79.7–80.9 span, 63.0–63.1 nuclearity and 43.5–43.6 relation. The ILP system of Subba and Di-Eugenio reports 70.35 / 49.47 / 35.44.
  • The sliding-window variant (TSP SW) finds significantly more correct spans than TSP 1-1 but does not improve relation labeling.
  • In a 20-document RST-DT development sample, the two settings show different relation distributions:
    • Attribution and Same-Unit outnumber Joint inside sentences.
    • Joint is more frequent than both across sentences.
  • The intra-sentential model is a chain-structured dynamic CRF. Exhaustive search over a whole document is infeasible because the number of valid trees grows as a Catalan number.

Interesting ideas

  • I1. Split the problem along a natural text boundary. Separate models for sentence-level and document-level decisions can learn different label distributions and use features that only make sense at one level, such as syntactic dominance sets inside sentences or lexical chains across them. The same split could suit any hierarchical structured prediction task.
  • I2. Score structure and label jointly, then decode exactly. Predicting a span and its relation together, then running an optimal search over those probabilities, avoids the error propagation of greedy merge-then-label pipelines.
  • I3. Use overlapping windows to handle leaky boundaries. Parsing two adjacent sentences at a time and reconciling the overlapping results recovers structures that cross sentence boundaries. The authors note this is especially promising for corpora with many such crossings.
  • I4. Fairly replicate a baseline that cannot be rerun. When the competitor's data split was unavailable, the authors averaged results over 5 random splits of matching size. This is a practical way to compare against figures reported only in another paper.
  • I5. Use confusion-matrix error analysis to explain mistakes. The analysis separates two causes of labeling errors: a dominant majority class (Elaboration) and semantically close relations (for example, Cause vs. Explanation). Each cause points to a different fix: ensembles or bagging for class imbalance, and richer semantic features for similar relations.

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 5The 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 worksPhilosophy · 211 works

🧭 Topics 1Research subjects and languages the work is about.Discourse analysis · 394 works

💡 Concepts 1Key ideas and methods explained on our concept pages.Discourse analysis · 30 works

Cite

Cite as Shafiq Joty, Giuseppe Carenini, Raymond Ng and Yashar Mehdad. 2013. Combining Intra- and Multi-sentential Rhetorical Parsing for Document-level Discourse Analysis. Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{joty2013combining,
  title = {{Combining Intra- and Multi-sentential Rhetorical Parsing for Document-level Discourse Analysis}},
  author = {Shafiq Joty, Giuseppe Carenini, Raymond Ng and Yashar Mehdad},
  booktitle = {Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2013},
  url = {https://aclanthology.org/P13-1048/}
}

References

25 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. O. Biran and O. Rambow. 2011. Identifying Justifications in Written Dialogs by Classifying Text as Argumentative. International Journal of Semantic Computing, 5(4):363–381. find
  2. L. Carlson, D. Marcu, and M. Okurowski. 2002. RST Discourse Treebank (RST-DT) LDC2002T07. Linguistic Data Consortium, Philadelphia. find
  3. V. Feng and G. Hirst. 2012. Text-level Discourse Parsing with Rich Linguistic Features. In Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics, ACL ’12, pages 60–68, Jeju Island, Korea. Association for Computational Linguistics.
  4. S. Fisher and B. Roark. 2007. The Utility of Parse-derived Features for Automatic Discourse Segmentation. In Proceedings of the 45th Annual Meeting of the Association for Computational Linguistics, ACL ’07, pages 488–495, Prague, Czech Republic. Association for Computational Linguistics. find
  5. M. Galley and K. McKeown. 2003. Improving Word Sense Disambiguation in Lexical Chaining. In Proceedings of the 18th International Joint Conference on Artificial Intelligence, IJCAI ’07, pages 1486– 1488, Acapulco, Mexico. find
  6. H. Hernault, H. Prendinger, D. duVerle, and M. Ishizuka. 2010. HILDA: A Discourse Parser Using Support Vector Machine Classification. Dialogue and Discourse, 1(3):1–33.
  7. S. Joty, G. Carenini, and R. T. Ng. 2012. A Novel Discriminative Framework for Sentence-Level Discourse Analysis. In Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, EMNLP-CoNLL ’12, pages 904– 915, Jeju Island,… find
  8. H. LeThanh, G. Abeysinghe, and C. Huyck. 2004. Generating Discourse Structures for Written Texts. In Proceedings of the 20th international conference on Computational Linguistics, COLING ’04, Geneva, Switzerland. Association for Computational Linguistics. find
  9. A. Louis, A. Joshi, and A. 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, SIGDIAL ’10, pages 147–156, Tokyo, Japan. Association for Computational Linguistics. find
  10. W. Mann and S. Thompson. 1988. Rhetorical Structure Theory: Toward a Functional Theory of Text Organization. Text, 8(3):243–281.
  11. D. Marcu and A. Echihabi. 2002. An Unsupervised Approach to Recognizing Discourse Relations. In Proceedings of the 40th Annual Meeting on Association for Computational Linguistics, ACL ’02, pages 368–375. Association for Computational Linguistics.
  12. D. Marcu. 2000a. The Rhetorical Parsing of Unrestricted Texts: A Surface-based Approach. Computational Linguistics, 26:395–448.
  13. D. Marcu. 2000b. The Theory and Practice of Discourse Parsing and Summarization. MIT Press, Cambridge, MA, USA. find
  14. J. Morris and G. Hirst. 1991. Lexical Cohesion Computed by Thesaural Relations as an Indicator of Structure of Text. Computational Linguistics, 17(1):21–48. find
  15. R. Prasad, A. Joshi, N. Dinesh, A. Lee, E. Miltsakaki, and B. Webber. 2005. The Penn Discourse Tree-Bank as a Resource for Natural Language Generation. In Proceedings of the Corpus Linguistics Workshop on Using Corpora for Natural Language Generation, pages 25–32, Birmingham, U.K. find
  16. S. Somasundaran, 2010. Discourse-Level Relations for Opinion Analysis. PhD thesis, University of Pittsburgh. find
  17. R. Soricut and D. Marcu. 2003. Sentence Level Discourse Parsing Using Syntactic and Lexical Information. In Proceedings of the 2003 Conference of the North American Chapter of the Association for Computational Linguistics on Human Language Technology, NAACL-HLT ’03, pages 149– 156, Edmonton, Canada.…
  18. C. Sporleder and M. Lapata. 2005. Discourse Chunking and its Application to Sentence Compression. In Proceedings of the conference on Human Language Technology and Empirical Methods in Natural Language Processing, pages 257–264, Vancouver, British Columbia, Canada. Association for Computational Ling… find
  19. C. Sporleder and A. Lascarides. 2004. Combining Hierarchical Clustering and Machine Learning to Predict High-Level Discourse Structure. In Proceedings of the 20th international conference on Computational Linguistics, COLING ’04, Geneva, Switzerland. Association for Computational Linguistics. find
  20. M. Stede. 2004. The Potsdam Commentary Corpus. In Proceedings of the ACL-04 Workshop on Discourse Annotation, Barcelona. Association for Computational Linguistics. find
  21. R. Subba and B. 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, HLT-NAACL ’09, pages 566–574, Boulder, Colo…
  22. C. Sutton and A. McCallum. 2012. An Introduction to Conditional Random Fields. Foundations and Trends in Machine Learning, 4(4):267–373. find
  23. C. Sutton, A. McCallum, and K. Rohanimanesh. 2007. Dynamic Conditional Random Fields: Factorized Probabilistic Models for Labeling and Segmenting Sequence Data. Journal of Machine Learning Research (JMLR), 8:693–723. find
  24. S. Verberne, L. Boves, N. Oostdijk, and P. Coppen. 2007. Evaluating Discourse-based Answer Extraction for Why-question Answering. In Proceedings of the 30th annual international ACM SIGIR conference on Research and development in information retrieval, pages 735–736, Amsterdam, The Netherlands. ACM. find
  25. N. Vliet and G. Redeker. 2011. Complex Sentences as Leaky Units in Discourse Parsing. In Proceedings of Constraints in Discourse, Agay-Saint Raphael, September. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/joty-2013-combining-intra-multi-sentential-rhetorical/