Elamur Research⌕
☰
home / papers / 2015 / Shallow Discourse Parsing Using Constituent Parsing Tree

Shallow Discourse Parsing Using Constituent Parsing Tree

Summary

This paper presents a shallow discourse parser built for the closed track of the CoNLL-2015 shared task. It identifies discourse relations in text, along with their arguments and senses, using only organizer-provided resources such as the Penn Discourse Treebank (PDTB). Following the annotation-mimicking design of Lin et al. (2014), the authors split the problem into a chain of classification steps, each handled by a maximum entropy model.

The explicit branch finds connectives, labels their Arg1/Arg2 spans using a constituent-tree method adapted from Kong et al. (2014), and assigns each connective its most frequent sense. The non-explicit branch filters candidate pairs of adjacent sentences, ignores the rare 'AltLex' type, and treats 'EntRel' as one more implicit sense.

Connective detection worked reasonably well, but argument extraction was weak, and these errors propagated through the pipeline, giving very low end-to-end scores. The paper is mainly useful as a baseline and as a case study of how errors compound in sequential discourse parsing.

Key points

  • Connective candidates are found by a level-order traversal of the parse tree. This recovers 13,911 of 14,722 explicit connectives in the training set. The connective classifier then reaches 93.87% accuracy and 90.1% F1 on dev.
  • The authors re-implemented the argument pruning of Kong et al. (2014), but it gave only 37.5% (Arg1) and 51.3% (Arg2) recall. Adding the rest of the sentence after the connective as an extra Arg2 candidate raised Arg2 recall to 93.1%.
  • On gold-pruned candidates, the argument labeler reached F1 of 0.7645 for Arg1 and 0.8812 for Arg2.
  • In the training data, non-explicit relations are 73.85% 'Implicit', 23.2% 'EntRel' and 2.94% 'AltLex'. Filtering leaves 8,728 non-explicit relations.
  • Official results:
    • Explicit connective F1 was 0.8168 on blind test, 0.7867 on test and 0.8609 on dev.
    • Joint Arg1+Arg2 extraction F1 was only about 0.05.
    • Overall parser F1 was 0.0358 on blind test, 0.0443 on test and 0.0655 on dev.

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

7 papers of our library cite it (of the 140673 whose bibliographies we read); 6 of them have a page here, 1 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/k15-2005

@inproceedings{chen2015shallow,
  title = {{Shallow Discourse Parsing Using Constituent Parsing Tree}},
  author = {Chen, Changge and Wang, Peilu and Zhao, Hai},
  booktitle = {Proceedings of the Nineteenth Conference on Computational Natural Language Learning - Shared Task},
  year = {2015},
  doi = {10.18653/v1/k15-2005},
  url = {https://doi.org/10.18653/v1/k15-2005}
}

References

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

  1. Zhongye Jia, Peilu Wang, and Hai Zhao. 2013. Grammatical error correction as multiclass classification with single model. pages 74–81, Sofia, Bulgaria, August. find
  2. Fang Kong, Hwee Tou Ng, and Guodong Zhou. 2014. A constituent-based approach to argument labeling with joint inference in discourse parsing. In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing, pages 68–77, Doha, Qatar, October. find our card
  3. Ziheng Lin, Hwee Tou Ng, and Min-Yen Kan. 2014. A PDTB-styled end-to-end discourse parser. Natural Language Engineering, 20:151–184, April. find
  4. Emily Pitler and Ani Nenkova. 2009. Using syntax to disambiguate explicit discourse connectives 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, pages 13–16, Suntec, Singapore… find our card
  5. 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, pages 683–691, S… find our card
  6. Rashmi Prasad, Dinesh Nikhil, Alan Lee, Eleni Miltsakaki, Livio Robaldo, Aravind K Joshi, and Bonnie L Webber. 2008. The Penn discourse treebank 2.0. In The International Conference on Language Resources and Evaluation, Marrakech, Morocco, May. find our card
  7. Peilu Wang, Zhongye Jia, and Hai Zhao. 2014. Grammatical error detection and correction using a single maximum entropy model. pages 74–82, Baltimore, Maryland, USA, July. find
  8. Hai Zhao and Chunyu Kit. 2008. Parsing syntactic and semantic dependencies with two single-stage maximum entropy models. In Proceedings of the Twelfth Conference on Computational Natural Language Learning, pages 203–207, Manchester, August. find
  9. Hai Zhao, Wenliang Chen, Jun’ichi Kazama, Kiyotaka Uchimoto, and Kentaro Torisawa. 2009. Multilingual dependency learning: Exploiting rich features for tagging syntactic and semantic dependencies. In Proceedings of the Thirteenth Conference on Computational Natural Language Learning: Shared Task, pa… find
  10. 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, pages 1507– 1514, Beijing, China, August. find our card

↑ back to top

This page in Markdown →