Elamur Research⌕
☰
home / papers / 2010 / A PDTB-Styled End-to-End Discourse Parser

A PDTB-Styled End-to-End Discourse Parser

Paper · 2010 · Ziheng Lin, Hwee Tou Ng and Min-Yen Kan

Summary

Discourse parsing in the PDTB framework had been handled piecemeal. Separate systems disambiguated connectives, located argument heads, or guessed Implicit relations. Lin, Ng and Kan combine these pieces into what they present as the first complete PDTB-style parser, which turns raw text into a discourse structure. The algorithm follows the order in which PDTB annotators worked:

  1. Explicit relations. Decide whether each connective occurrence has a discourse function, find its Arg1 and Arg2, and assign a Level 2 type.
  2. Remaining sentence pairs. Label each unlinked adjacent pair inside a paragraph as an Implicit/AltLex type, EntRel or NoRel.
  3. Attribution. Test the clauses inside relations as possible attribution spans.

The new components are richer lexico-syntactic connective features, a two-stage argument labeler and an attribution span labeler. Every component is scored under three settings: gold parses without error propagation, gold parses with propagated errors, and full automation. End-to-end F1 under partial matching is 46.80% with gold parses and 38.18% fully automatic. The gold-parse figure is close to a reported RST parser (44.3%). The authors argue that such output can support summarization and answering why-questions.

Key points

  • The pipeline has five parts: a connective classifier; an argument labeler (an Arg1 position classifier, SS vs PS, followed by an argument extractor); a classifier that types Explicit relations; a classifier for relations that are not Explicit; and a labeler that marks attribution spans. It is trained on WSJ Sec. 02-21 and tested on Sec. 23.
  • New connective features combine the connective with its previous/next word and their POS tags. They also include the path from the connective to the root and a compressed path that merges repeated adjacent labels.
  • Explicit classifier results:
    • F1 is 86.77 under gold settings, against 86.00 for a connective-only baseline and 84.00 human agreement on Level 2.
    • F1 is 80.61 fully automatic.
  • Non-explicit classifier results: 39.63 micro F1 component-wise against a majority-EntRel baseline of about 21; 25.46 fully automatic.
  • Argument extractor and attribution labeler:
    • Arg2 partial F1 reaches 99.13, and relation-level F1 is 86.24 partial vs 53.85 exact. Most exact-match misses come from annotators trimming spans to a minimal extent.
    • The attribution labeler scores 79.68 partial and 65.95 exact.

Interesting ideas

  • I1. Scoring each stage with gold input, then with upstream errors, then fully automatically separates a component's own weakness from damage inherited from earlier stages.
  • I2. Reporting both partial and exact span matches shows when a system finds the right material but misses boundary conventions.
  • I3. The parser first decides where the first argument lies (same sentence or previous sentence) and only then extracts its span, which splits a hard search into two simpler decisions.
  • I4. The parser handles overt connectives first and then sends every still-unlinked adjacent sentence pair to a fallback classifier, which may also answer "entity link only" or "no relation".
  • I5. The authors suggest causal relations can help answer why-questions and contrast relations can help spot updates in summaries.

For Mova

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

  • I1 (learn): Mova's evaluation could report every pipeline step under these three regimes, showing whether a bad answer comes from the step itself or from earlier ones.
  • I2 (learn): when comparing silver data from witness parsers, scoring overlap and exact boundaries separately would keep boundary disputes from hiding real agreement.
  • I3 (frames): linking clauses or resolving coreference could first classify the antecedent's location and then search only there.
  • I4 (world): building event order and causality could take explicit links from connectives first and then mark leftover sentence pairs as merely sharing entities or unrelated.
  • I5 (answer): fable and tale QA could follow typed cause links between clauses to answer "why", with the linking relation as the explanation.

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

🎓National University of SingaporeSingapore · Singapore84 works · 32,204 citations

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 9,085 worksArtificial Intelligence · 8,888 worksArts and Humanities · 9,296 worksLanguage and Linguistics · 9,206 works

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

Cite

Cite as Ziheng Lin, Hwee Tou Ng and Min-Yen Kan. 2010. A PDTB-Styled End-to-End Discourse Parser. Natural Language Engineering 20 (02), 151 - 184, 2014.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.1011.0835 · arXiv 1011.0835

@article{lin2010pdtb,
  title = {{A PDTB-Styled End-to-End Discourse Parser}},
  author = {Ziheng Lin, Hwee Tou Ng and Min-Yen Kan},
  journal = {Natural Language Engineering 20 (02), 151 - 184, 2014},
  year = {2010},
  doi = {10.48550/arxiv.1011.0835},
  eprint = {1011.0835},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.1011.0835}
}

References

21 references, 9 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Lynn Carlson, Daniel Marcu, and Mary Ellen Okurowski. Building a discourse-tagged corpus in the framework of Rhetorical Structure Theory. In Proceedings of the Second SIGdial Workshop on Discourse and Dialogue, Morristown, NJ, USA, 2001. ↗
  2. Nikhil Dinesh, Alan Lee, Eleni Miltsakaki, Rashmi Prasad, Aravind Joshi, and Bonnie Webber. Attribution and the (non)-alignment of syntactic and discourse arguments of connectives. In Proceedings of the ACL Workshop on Frontiers in Corpus Annotation II: Pie in the Sky, Ann Arbor, MI, USA, 2005. find
  3. David duVerle and Helmut Prendinger. A novel discourse parser based on Support Vector Machine classification. In Proceedings of the Joint Conference of the 47th Annual Meeting of the ACL and the 4th International Joint Conference on Natural Language Processing of the AFNLP, Singapore, 2009.
  4. Robert Elwell and Jason Baldridge. Discourse connective argument identification with connective specific rankers. In Proceedings of the IEEE International Conference on Semantic Computing, Washington, DC, USA, 2008. find
  5. Jerry R. Hobbs. Literature and cognition. In CSLI Lecture Notes Number 21. CSLI Publications, 1990. find
  6. Alistair Knott and Ted Sanders. The classification of coherence relations and their linguistic markers: An exploration of two languages. Journal of Pragmatics, 30(2): 135–175, 1998. ↗
  7. Alex Lascarides and Nicholas Asher. Temporal interpretation, discourse relations and commonsense entailment. Linguistics and Philosophy, 16(5):437–493, 1993. find
  8. Ziheng Lin, Min-Yen Kan, and Hwee Tou Ng. Recognizing implicit discourse relations in the Penn Discourse Treebank. In Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing, Singapore, 2009.
  9. William C. Mann and Sandra A. Thompson. Rhetorical Structure Theory: Toward a functional theory of text organization. Text, 8(3):243–281, 1988. ↗
  10. Daniel Marcu. The Rhetorical Parsing, Summarization, and Generation of Natural Language Texts. PhD thesis, University of Toronto, 1997. find
  11. Eleni Miltsakaki, Rashmi Prasad, Aravind Joshi, and Bonnie Webber. The Penn Discourse Treebank. In Proceedings of the 4th International Conference on Language Resources and Evaluation, Lisbon, Portugal, 2004.
  12. Eleni Miltsakaki, Nikhil Dinesh, Rashmi Prasad, Aravind Joshi, and Bonnie Webber. Experiments on sense annotations and sense disambiguation of discourse connectives. In Proceedings of the Fourth Workshop on Treebanks and Linguistic Theories (TLT2005), Barcelona, Spain, 2005. find
  13. PDTB-Group. The Penn Discourse Treebank 2.0 Annotation Manual. The PDTB Research Group, 2007. find
  14. Emily Pitler and Ani Nenkova. Using syntax to disambiguate explicit discourse connectives in text. In Proceedings of the ACL-IJCNLP 2009 Conference Short Papers, Singapore, 2009.
  15. Emily Pitler, Annie Louis, and Ani Nenkova. Automatic sense prediction for implicit discourse relations in text. In Proceedings of the Joint Conference of the 47th Annual Meeting of the ACL and the 4th International Joint Conference on Natural Language Processing of the AFNLP, Singapore, 2009.
  16. Rashmi Prasad, Nikhil Dinesh, Alan Lee, Eleni Miltsakaki, Livio Robaldo, Aravind Joshi, and Bonnie Webber. The Penn Discourse Treebank 2.0. In Proceedings of the 6th International Conference on Language Resources and Evaluation (LREC 2008), 2008.
  17. Peter Rossen Skadhauge and Daniel Hardt. Syntactic identification of attribution in the RST Treebank. In Proceedings of the Recent Advances in Natural Language Processing, Borovets, Bulgaria, 2005. find
  18. Radu Soricut and Daniel Marcu. Sentence level discourse parsing using syntactic and lexical information. In Proceedings of the Human Language Technology Conference of the North American Chapter of the Association for Computational Linguistics, Edmonton, Canada, 2003.
  19. Bonnie Webber. D-LTAG: Extending lexicalized TAG to discourse. Cognitive Science, 28(5):751–779, 2004. ↗
  20. Ben Wellner. Sequence Models and Ranking Methods for Discourse Parsing. PhD thesis, Brandeis University, 2009. find
  21. Ben Wellner and James Pustejovsky. Automatically identifying the arguments of discourse connectives. In Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, Prague, Czech Republic, 2007. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/lin-2010-pdtb-styled-end-end-discourse/