Predicting the Presence of Discourse Connectives

Summary

Text generation and summarization systems must decide whether to mark the relation between two adjacent clauses with an explicit connective (e.g., because) or leave it for the reader to infer. Too little marking hurts comprehension, and too much makes text sound redundant. Patterson and Kehler treat this as a binary prediction task. Given two clauses, their coherence relation and a candidate connective, the model predicts whether the author actually used that connective. The data come from the Penn Discourse Treebank (PDTB), restricted to adjacent relations of the eight most frequent senses. For implicit relations, the annotators' suggested connective serves as the candidate. The authors train a separate logistic regression classifier for each relation type, using relation-level, argument-level and discourse-level features. The model clearly beats a majority-per-relation baseline. Its most useful cues come from dependencies between neighbouring coherence relations. A Mechanical Turk study shows that humans find the task hard and often disagree. This suggests that connective use is frequently optional, which may cap the performance any system can reach.

Key points

Data: PDTB over the WSJ corpus. After filtering, it contains 10,039 explicit and 11,690 implicit relations. Sections 2–22 are used for training (18,218 tokens), 23–24 for testing and 0–1 for development.
Relation preferences: ASYNCHRONOUS, CONJUNCTION, CONTRAST and SYNCHRONOUS are usually explicit. INSTANTIATION, REASON, RESTATEMENT and RESULT are mostly implicit; for example, RESTATEMENT is implicit 96% of the time.
Model performance: Accuracy is 86.6%, compared with a 77.0% baseline that uses only the relation sense. Discourse-level features, which capture dependencies between consecutive relations, are the most informative.
Human study: On 100 items, Mechanical Turk judges reached 68% accuracy by majority vote and 60.4% when individual responses are averaged. The model scored 84% on the same sample.
Optionality evidence: Judges tended to add connectives where the WSJ text had none. Both the judges and the model were significantly less confident on items they misclassified, which points to cases where either choice is acceptable.

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

22 papers of our library cite it (of the 140673 whose bibliographies we read); 11 of them have a page here, 11 not yet.

Cite

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

@inproceedings{patterson2013predicting,
  title = {{Predicting the Presence of Discourse Connectives}},
  author = {Patterson, Gary and Kehler, Andrew},
  booktitle = {Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing},
  year = {2013},
  doi = {10.18653/v1/d13-1094},
  url = {https://doi.org/10.18653/v1/d13-1094}
}

References

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

Nicholas Asher and Alex Lascarides. 2003. Logics of conversation. Cambridge University Press. find
Fatemeh Torabi Asr and Vera Demberg. 2012a. Implicitness of discourse relations. In Proceedings of the 24th International Conference on Computational Linguistics, pages 2669–2684. find our card
Fatemeh Torabi Asr and Vera Demberg. 2012b. Measuring the strength of linguistic cues for discourse relations. In Proceedings of the Workshop on Advances in Discourse Analysis and its Computational Aspects (ADACA), pages 33–42. find our card
Michael Elhadad and Kathleen R McKeown. 1990. Generating connectives. In Proceedings of the 13th Conference on Computational Linguistics-Volume 3, pages 97–101. find
Jerry R Hobbs. 1979. Coherence and coreference. Cognitive science, 3(1):67–90. find
Andrew Kehler. 2002. Coherence, reference, and the theory of grammar. CSLI Publications, Stanford. find
Ziheng Lin, Min-Yen Kan, and Hwee Tou Ng. 2009. Recognizing implicit discourse relations in the Penn Discourse Treebank. In Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing: Volume 1, pages 343–351. find our card
William C Mann and Sandra A Thompson. 1988. Rhetorical structure theory: Toward a functional theory of text organization. Text, 8(3):243–281. find our card
Mitchell P Marcus, Mary Ann Marcinkiewicz, and Beatrice Santorini. 1993. Building a large annotated corpus of English: The Penn Treebank. Computational Linguistics, 19(2):313–330. find our card
Emily Pitler, Mridhula Raghupathy, Hena Mehta, Ani Nenkova, Alan Lee, and Aravind K Joshi. 2008. Easily identifiable discourse relations. In Proceedings of the 22nd International Conference on Computational Linguistics. find our card
Emily Pitler, Annie Louis, and Ani Nenkova. 2009. 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: Volume 2, pages… find our card
Rashmi Prasad, Nikhil Dinesh, Alan Lee, Eleni Miltsakaki, Livio Robaldo, Aravind Joshi, and Bonnie Webber. 2008. The Penn Discourse Treebank 2.0. In Proceedings of the 6th International Conference on Language Resources and Evaluation. find our card
Caroline Sporleder and Alex Lascarides. 2008. Using automatically labelled examples to classify rhetorical relations: An assessment. Natural Language Engineering, 14(3):369–416. find
Zhi-Min Zhou, Yu Xu, Zheng-Yu Niu, Man Lan, Jian Su, and Chew Lim Tan. 2010. Predicting discourse connectives for implicit discourse relation recognition. In Proceedings of the 23rd International Conference on Computational Linguistics: Posters, pages 1507– 1514. find our card

↑ back to top

📄 Predicting the Presence of Discourse Connectives