Elamur Research⌕
☰
home / papers / 2010 / The Effects of Discourse Connectives Prediction on Implicit Discourse Relation Recognition

The Effects of Discourse Connectives Prediction on Implicit Discourse Relation Recognition

Summary

Without an overt connective such as "but" or "because", recognizing the discourse relation between two text spans is hard. Connectives carry most of the signal: the authors show that mapping gold implicit connectives in PDTB to their most frequent sense yields about 91.8% averaged F-score. They therefore try to recover the missing connective automatically. For each candidate connective, they build two synthetic sentences, one with the connective placed before Arg1 and one with it between Arg1 and Arg2. An n-gram language model trained on unannotated news text scores each sentence, and the candidates with the lowest perplexity are taken as predictions. These predictions feed linear SVMs in two ways:

  • Sense frequency: a 4-dimensional vector holding the training-set sense counts of the top-ranked connective.
  • Presence: binary indicators for the top N predicted connectives, which the authors describe as the unsupervised variant.

On PDTB 2.0, the approach comes close to the feature-rich state of the art and beats it on Comparison. It uses only a handful of features, and because it depends mainly on raw text, it should transfer more easily to other languages and domains.

Key points

  • The evaluation uses four one-vs-rest binary classifiers (Comparison, Contingency, Expansion, Temporal), built as linear SVMs with LIBSVM. Training sets are balanced, and PDTB sections 21–22 serve as the test set.
  • On implicit relations, the best averaged F1 is 35.55, from the presence-of-top-N features, against 40.57 for Pitler et al. (2009a). For Comparison, the sense-frequency approach reaches 26.02 F1 versus 21.96, a gain of more than 4 points.
  • The language models were trained on news corpora (New York, Xin, Ltw) from the North American News collection. 4-gram models gave roughly 0.5% better averaged F-score than 3-gram models. The best values of N mostly fell between 100 and 160.
  • The authors also converted explicit relations into implicit-like examples by deleting the connective. On this data, averaged F1 reached 39.98, compared with about 33.69 that Sporleder and Lascarides (2008) reported on different data.
  • The method needs at most about 200 features, whereas linguistically informed systems use thousands. It also requires no parsers or other heavy NLP tools.

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

13 papers of our library cite it (of the 140673 whose bibliographies we read); 7 of them have a page here, 6 not yet.

Cite

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

@inproceedings{zhou2010effects,
  title = {{The Effects of Discourse Connectives Prediction on Implicit Discourse Relation Recognition}},
  author = {Zhou, Zhi Min and Lan, Man and Niu, Zheng Yu and Xu, Yu and Su, Jian},
  booktitle = {Proceedings of the SIGDIAL 2010 Conference},
  year = {2010},
  url = {https://aclanthology.org/W10-4326/}
}

References

20 references, 11 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. J. Baldridge and A. Lascarides. 2005. Probabilistic head-driven parsing for discourse structure. Proceedings of the Ninth Conference on Computational Natural Language Learning. find
  2. L. Carlson, D. Marcu, and Ma. E. Okurowski. 2001. Building a discourse-tagged corpus in the framework of rhetorical structure theory. Proceedings of the Second SIG dial Workshop on Discourse and Dialogue. find our card
  3. B. Dorr. LCS Verb Database. Technical Report Online Software Database, University of Maryland, College Park, MD,2001. find
  4. R. Girju. 2003. Automatic detection of causal relations for question answering. In ACL 2003 Workshops. find
  5. S. Blair-Goldensohn. 2007. Long-Answer Question Answering and Rhetorical-Semantic Relations. Ph.D. thesis, Columbia Unviersity. find
  6. M. Lapata and A. Lascarides. 2004. Inferring Sentence-internal Temporal Relations. Proceedings of the North American Chapter of the Assocation of Computational Linguistics. find our card
  7. Z.H. Lin, M.Y. Kan and H.T. Ng. 2009. Recognizing Implicit Discourse Relations in the Penn Discourse Treebank. Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing. find our card
  8. D. Marcu and A. Echihabi. 2002. An Unsupervised Approach to Recognizing Discourse Relations. Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics. find our card
  9. E. Pitler, M. Raghupathy, H. Mehta, A. Nenkova, A. Lee, A. Joshi. 2008. find
  10. Easily Identifiable Discourse Relations. Coling 2008: Companion volume: Posters. find
  11. E. Pitler, A. Louis, A. Nenkova. 2009. Automatic sense prediction for implicit discourse relations in text. Proceedings of the 47th Annual Meeting of the Association for Computational Linguistics. find our card
  12. E. Pitler and A. Nenkova. 2009. Using Syntax to Disambiguate Explicit Discourse Connectives in Text. Proceedings of the ACL-IJCNLP 2009 Conference Short Papers. find our card
  13. M. Porter. An algorithm for suffix stripping. In Program, vol. 14, no. 3, pp.130-137, 1980. find
  14. R. Prasad, N. Dinesh, A. Lee, A. Joshi, B. Webber. 2006. Annotating attribution in the Penn Discourse TreeBank. Proceedings of the COLING/ACL Workshop on Sentiment and Subjectivity in Text. find our card
  15. R. Prasad, N. Dinesh, A. Lee, E. Miltsakaki, L. Robaldo, A. Joshi, B. Webber. 2008. The Penn Discourse TreeBank 2.0. Proceedings of LREC’08. find our card
  16. M. Saito, K.Yamamoto, S.Sekine. 2006. Using Phrasal Patterns to Identify Discourse Relations. Proceeding of the HLTCNA Chapter of the ACL. find our card
  17. R. Soricut and D. Marcu. 2003. Sentence Level Discourse Parsing using Syntactic and Lexical Information. Proceedings of the Human Language Technology and North American Association for Computational Linguistics Conference. find our card
  18. C. Sporleder and A. Lascarides. 2008. Using automatically labelled examples to classify rhetorical relations: an assessment. Natural Language Engineering, Volume 14, Issue 03. find
  19. B. Wellner , J. Pustejovsky, C. H. R. S., A. Rumshisky. 2006. Classification of discourse coherence relations: An exploratory study using multiple knowledge sources. Proceedings of the 7th SIGDIAL Workshop on Discourse and Dialogue. find our card
  20. F. Wolf, E. Gibson, A. Fisher, M. Knight. 2005. The Discourse GraphBank: A database of texts annotated with coherence relations. Linguistic Data Consortium. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/papers/zhou-2010-effects-discourse-connectives-prediction-implicit/