Elamur Research⌕
☰
home / papers / 2010 / Using entity features to classify implicit discourse relations

Using entity features to classify implicit discourse relations

Summary

The paper asks whether the way entities are mentioned in two adjacent sentences can reveal the implicit discourse relation between them, meaning a relation with no connective such as "because". Earlier work relied mainly on lexical cues, and entity information had only been mixed with other features. The authors combine three annotation layers over 590 WSJ articles: relation senses from the Penn Discourse Treebank (PDTB), coreference from Ontonotes, and syntax from the Penn Treebank. Each referring expression is described by:

  • grammatical role
  • given/new status
  • syntactic form
  • modification
  • topicalization
  • number

These attributes are counted per sentence. They are also paired across the two sentences, depending on whether the entities corefer, share a head word, or are unrelated. Linear SVMs (LIBLINEAR) handle five one-versus-rest tasks. Entity features beat a random baseline on every task, gaining about 15 F-score points for Contingency and EntRel. Approximate coreference costs little except on Expansion. However, these features trail simple word-based features and add nothing when combined with them. This negative result is useful for anyone designing discourse parsers.

Key points

  • Data split: training uses WSJ sections 0–22, with negatives downsampled to balance the classes. Testing uses sections 23–24, a natural-distribution set of 1002 sentence pairs. The feature space has about 2000 features.
  • Gold-standard entity features (EntGS), F-scores: Comparison 24.18, Contingency 37.57, Temporal 7.58, Expansion 52.42, EntRel 38.03. The word-feature model (WP) scores 27.30, 38.17, 11.09, 48.54 and 38.48. WP+EntGS brings no consistent gain.
  • Approximate coreference (EntApp): treating identical head words as coreferent works nearly as well as gold coreference. The clear exception is Expansion, which drops from 52.42 to 47.82.
  • Feature analysis:
    • Topicalization and same-head ("RELATED") entities mark Comparison.
    • Proper names mark EntRel.
    • Coreference unexpectedly marks Temporal: about 70% of Temporal pairs share an entity, versus roughly 50% for other classes.
  • Pronouns and Contingency: a pronoun in the second sentence signals Contingency. It appears in about 45% of positive pairs versus 32% of negative pairs.

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

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

Cite

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

@inproceedings{louis2010using,
  title = {{Using entity features to classify implicit discourse relations}},
  author = {Louis, Annie and Joshi, Aravind and Prasad, Rashmi and Nenkova, Ani},
  booktitle = {Proceedings of the SIGDIAL 2010 Conference},
  year = {2010},
  url = {https://aclanthology.org/W10-4310/}
}

References

11 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. S. Blair-Goldensohn, K. McKeown, and O. Rambow. 2007. Building and refining rhetorical-semantic relation models. In HLT-NAACL. find
  2. S.H. Corston-Oliver. 1998. Beyond string matching and cue phrases: Improving efficiency and coverage in discourse analysis. In The AAAI Spring Symposium on Intelligent Text Summarization. find
  3. E. Hovy, M. Marcus, M. Palmer, L. Ramshaw, and R. Weischedel. 2006. Ontonotes: the 90% solution. In NAACL-HLT. find
  4. Z. Lin, M. Kan, and H.T. Ng. 2009. Recognizing implicit discourse relations in the Penn Discourse Treebank. In EMNLP. find our card
  5. D. Marcu and A. Echihabi. 2001. An unsupervised approach to recognizing discourse relations. In ACL. find our card
  6. M. Marcus, B. Santorini, and M. Marcinkiewicz. 1994. Building a large annotated corpus of english: The penn treebank. Computational Linguistics. find our card
  7. E. Pitler, A. Louis, and A. Nenkova. 2009. Automatic sense prediction for implicit discourse relations in text. In ACL-IJCNLP. find our card
  8. R. Prasad, N. Dinesh, A. Lee, E. Miltsakaki, L. Robaldo, A. Joshi, and B. Webber. 2008. The penn discourse treebank 2.0. In LREC. find our card
  9. E. Prince. 1992. The zpg letter: subject, definiteness, and information status. In Discourse description: diverse analyses of a fund raising text, pages 295– 325. John Benjamins. find
  10. C. Sporleder and A. Lascarides. 2008. Using automatically labelled examples to classify rhetorical relations: An assessment. Natural Language Engineering, 14:369–416. find
  11. F. Wolf and E. Gibson. 2006. Coherence in natural language: data structures and applications. MIT Press. find

↑ back to top