Elamur Research⌕
☰
home / papers / 2015 / Two Practical Rhetorical Structure Theory Parsers

Two Practical Rhetorical Structure Theory Parsers

Summary

The paper presents two ready-to-use discourse parsers for Rhetorical Structure Theory (RST). They share one framework and differ only in their syntactic features, which allows a controlled comparison of constituent versus dependency syntax for RST parsing.

The pipeline is modeled on earlier work by Feng and Hirst (2012), as well as by Hernault et al. (2010):

  • A token-level classifier splits text into EDUs.
  • A greedy bottom-up procedure builds the tree by alternating two classifiers: one picks which adjacent units to merge, and the other labels the relation.

The authors make two additions. They re-implement syntactic features, including the Soricut and Marcu dominance relations, on dependency trees. They also introduce features based on coreference links. Parser C uses Stanford CoreNLP constituents, dependencies and coreference. Parser D relies only on dependencies from Malt or the Chen and Manning (2014) parser.

Evaluated end-to-end on the RST corpus, both parsers reach state-of-the-art results. D is about 0.1 F1 behind C but runs roughly 2.5 times faster. The code is released as open-source Scala with a two-line API and a side-by-side visualization tool, making RST parsing practical for downstream applications such as question answering.

Key points

  • With predicted EDUs, the F1 scores were 48.9 for C, 48.8 for D stanford and 47.9 for D malt, compared with 47.3 for Hernault 2010. With manual EDUs, C scored 55.5, while Joty 2013 reached 55.8 and Joty 2014 reached 57.3. Joty 2014 uses a more complex reranking approach.
  • Runtime on the full test set was 88 s for D (about 2.3 s per document), 224 s for C and 605 s for the Feng and Hirst (2012) parser. Times were measured end-to-end on a laptop.
  • Ablation tests with predicted EDUs show that the feature groups complement each other. Removing dependency features costs 1.2 F1, and removing constituent features costs 1.6 F1. Coreference features, used here for the first time in discourse parsing, add 0.3 F1 with gold EDUs.
  • The classifiers are Perceptron for segmentation and relation detection and Logistic Regression for labeling. The parsers use "basic" Stanford dependencies, which performed better than the collapsed variants. Evaluation covers 18 relations with nuclearity.
  • The code is available at github.com/sistanlp/processors and on Maven Central through CoreNLPProcessor and FastNLPProcessor, using annotate(). It also includes a Grails/Dagre-d3 web visualizer.

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 · Coreference resolution

Cited by

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

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.3115/v1/n15-3001

@inproceedings{surdeanu2015two,
  title = {{Two Practical Rhetorical Structure Theory Parsers}},
  author = {Surdeanu, Mihai and Hicks, Tom and Valenzuela-Escárcega, Marco Antonio},
  booktitle = {Proceedings of the 2015 Conference of the North American Chapter of the Association for Computational Linguistics: Demonstrations},
  year = {2015},
  doi = {10.3115/v1/n15-3001},
  url = {https://doi.org/10.3115/v1/n15-3001}
}

References

15 references, 10 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. L. Carlson, D. Marcu, and M. E. Okurowski. 2003. Building a Discourse-Tagged Corpus in the Framework of Rhetorical Structure Theory. In Jan van Kuppevelt and Ronnie Smith, editors, Current Directions in Discourse and Dialogue, pages 85–112. Kluwer Academic Publishers. find our card
  2. D. Chen and C. D. Manning. 2014. A fast and accurate dependency parser using neural networks. In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP). find our card
  3. M.-C. de Marneffe, B. MacCartney, and C. D. Manning. 2006. Generating typed dependency parses from phrase structure parses. In Proceedings of the International Conference on Language Resources and Evaluation (LREC). find our card
  4. V. W. Feng and G. Hirst. 2012. Text-level discourse parsing with rich linguistic features. In Proceedings of the Association for Computational Linguistics. find our card
  5. D. Ferrucci and A. Lally. 2004. UIMA: an architectural approach to unstructured information processing in the corporate research environment. Natural Language Engineering, 10:327–348. 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. find our card
  7. P. Jansen, M. Surdeanu, and P. Clark. 2014. Discourse complements lexical semantics for non-factoid answer reranking. In Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics. find
  8. S. Joty and A. Moschitti. 2014. Discriminative reranking of discourse parses using tree kernels. In Proceedings of the Conference on Empirical Methods in Natural Language Processing. find our card
  9. S. Joty, G. Carenini, R. Ng, and Y. Mehdad. 2013. Combining intra- and multi-sentential rhetorical parsing for document-level discourse analysis. In Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics. find
  10. D. Klein and C. D. Manning. 2003. Accurate unlexicalized parsing. In Proceedings of the 41st Annual Meeting of the Association for Computational Linguistics (ACL). find our card
  11. W. C. Mann and S. A. Thompson. 1988. Rhetorical structure theory: Toward a functional theory of text organization. Text, 8(3):243–281. find our card
  12. C. D. Manning, M. Surdeanu, J. Bauer, J. Finkel, S. J. Bethard, and D. McClosky. 2014. The Stanford CoreNLP natural language processing toolkit. In Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics. find our card
  13. D. Marcu. 2000. The Theory and Practice of Discourse Parsing and Summarization. MIT Press. find
  14. R. Sharp, P. Jansen, M. Surdeanu, and P. Clark. 2015. Spinning straw into gold: Using free text to train monolingual alignment models for non-factoid question answering. In Proceedings of the Conference of the North American Chapter of the Association for Computational Linguistics - Human Language T… find
  15. R. Soricut and D. Marcu. 2003. Sentence level discourse parsing using syntactic and lexical information. In Proceedings of the Human Language Technology and North American Association for Computational Linguistics Conference. find our card

↑ back to top

This page in Markdown →