Elamur Research⌕
☰
home / papers / 2014 / A Constituent-Based Approach to Argument Labeling with Joint Inference in Discourse Parsing

A Constituent-Based Approach to Argument Labeling with Joint Inference in Discourse Parsing

Summary

Within the Penn Discourse Treebank (PDTB), each explicit connective links two text arguments, Arg1 and Arg2. Recovering the full spans of these arguments is the weakest step in PDTB-style discourse parsing. The strongest earlier system (Lin et al., 2014) scored only 53.85% F-measure with gold parses and connectives, far below human agreement.

The authors combine strengths of the two earlier paradigms, subtree extraction and linear tagging. Their model classifies parse-tree constituents as Arg1, Arg2 or NULL, then joins constituents with the same label into one span. A recursive pruning step keeps as candidates the siblings of the connective node and of each of its ancestors, which removes roughly 82% of tree nodes. The immediately preceding sentence is added as an extra pseudo-constituent. This lets one model cover both same-sentence and previous-sentence Arg1 cases. Integer linear programming (ILP) then enforces global consistency between argument decisions.

On PDTB, the approach outperforms both Lin et al. (2014) and Ghosh et al. (2012). Inserted into the NUS end-to-end parser, it clearly improves overall results, most of all with automatic parses.

Key points

  • Comparison with Lin et al. (2014) (training on sections 02–21, testing on section 23, without joint inference):
    • Arg1&2 F1 is 55.69 vs 53.85 with gold parses and no error propagation.
    • It is 48.89 vs 40.37 with automatic parses and error propagation.
    • Adding ILP joint inference raises these to 58.18 and 50.73.
  • Pruning recall:
    • With gold parse trees, pruning keeps 89.56% of Arg1 and 92.98% of Arg2 spans.
    • With Charniak parser output, it keeps 80.59% and 89.87%.
  • Comparison with the linear-tagging system of Ghosh et al. (2012):
    • Arg1 F1 is 65.68 under exact match and 67.48 under chunking-style evaluation, vs 59.39.
    • Arg2 F1 is 84.50 and 88.08, vs 79.48.
  • Sources of performance change:
    • Automatic parses lower F1 by about 5.5%.
    • Connective-identification errors cost under 2%.
    • Joint inference adds about 2% on automatic parses.
    • Combining Charniak and Berkeley parses adds about 1.5% more.
  • End-to-end effect: Replacing the argument labeler in the NUS PDTB-style end-to-end parser raises overall F-measure by about 1.8% with gold trees and 6.7% with automatic trees, reaching 34.80 and 27.39 respectively.

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

17 papers of our library cite it (of the 140673 whose bibliographies we read); 4 of them have a page here, 13 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.3115/v1/d14-1008

@inproceedings{kong2014constituent,
  title = {{A Constituent-Based Approach to Argument Labeling with Joint Inference in Discourse Parsing}},
  author = {Kong, Fang and Ng, Hwee Tou and Zhou, Guodong},
  booktitle = {Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP)},
  year = {2014},
  doi = {10.3115/v1/d14-1008},
  url = {https://doi.org/10.3115/v1/d14-1008}
}

References

20 references, 8 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Eugene Charniak. 2000. A maximum-entropy-inspired parser. In Proceedings of the First Meeting of the North American Chapter of the Association for Computational Linguistics, pages 132–139. find our card
  2. Nikhil Dinesh, Alan Lee, Eleni Miltsakaki, Rashmi Prasad, Aravind Joshi, and Bonnie Webber. 2005. Attribution and the (non-)alignment of syntactic and discourse arguments of connectives. In Proceedings of the Workshop on Frontiers in Corpus Annotation II: Pie in the Sky, pages 29–36. find
  3. Robert Elwell and Jason Baldridge. 2008. Discourse connective argument identification with connective specific rankers. In Second IEEE International Conference on Semantic Computing, pages 198–205. find
  4. Sucheta Ghosh, Richard Johansson, Giuseppe Riccardi, and Sara Tonelli. 2011. Shallow discourse parsing with conditional random fields. In Proceedings of the 5th International Joint Conference on Natural Language Processing, pages 1071–1079. find our card
  5. Sucheta Ghosh, Giuseppe Riccardi, and Richard Johansson. 2012. Global features for shallow discourse parsing. In Proceedings of the 13th Annual Meeting of the Special Interest Group on Discourse and Dialogue, pages 150–159. find our card
  6. Sucheta Ghosh. 2012. End-to-End Discourse Parsing using Cascaded Structured Prediction. Ph.D. thesis, University of Trento. find
  7. Alistair Knott. 1996. A Data-Driven Methodology for Motivating a Set of Coherence Relations. Ph.D. thesis, University of Edinburgh. find
  8. Ziheng Lin, Hwee Tou Ng, and Min-Yen Kan. 2014. A PDTB-styled end-to-end discourse parser. Natural Language Engineering, 20(2):151–184. find
  9. Mitchell P. Marcus, Beatrice Santorini, and Mary Ann Marcinkiewicz. 1993. Building a large annotated corpus of English: The Penn Treebank. Computational Linguistics, 19(2):313–330. find our card
  10. Eleni Miltsakaki, Rashmi Prasad, Aravind Joshi, and Bonnie Webber. 2004. The Penn Discourse Treebank. In Proceedings of the Fourth International Conference on Language Resources and Evaluation, pages 2237–2240. find our card
  11. Slav Petrov, Leon Barrett, Romain Thibaux, and Dan Klein. 2006. Learning accurate, compact, and interpretable tree annotation. In Proceedings of the 21st International Conference on Computational Linguistics and the 44th Annual Meeting of the Association for Computational Linguistics, pages 433– 440… find our card
  12. 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 LREC 2008 Conference, pages 2961–2968. find our card
  13. Rashmi Prasad, Aravind Joshi, and Bonnie Webber. 2010. Exploiting scope for shallow discourse parsing. In Proceedings of the Seventh International Conference on Language Resources and Evaluation. find our card
  14. Vasin Punyakanok, Dan Roth, and Wen-tau Yih. 2008. The important of syntactic parsing and inference in semantic role labeling. Computational Linguistics, 34(2):257–287. find
  15. Bonnie Webber. 2004. D-LTAG: extending lexicalized TAG to discourse. Cognitive Science, 28(5):751– 779. find
  16. Ben Wellner and James Pustejovsky. 2007. Automatically identifying the arguments of discourse connectives. In Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 92–101. find
  17. Nianwen Xue and Martha Palmer. 2004. Calibrating features for semantic role labeling. In Proceedings of 2004 Conference on Empirical Methods in Natural Language Processing, pages 88–94. find
  18. Hui Zhang, Min Zhang, Chew Lim Tan, and Haizhou Li. 2009. K-best combination of syntactic parsers. In Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing, pages 1552–1560. find
  19. Addendum to Paper approved by SIGDAT board in October 2015 find
  20. This addendum concerns EMNLP 2014 paper Fang Kong;; Hwee Tou Ng;; Guodong Zhou. A Constituent-Based Approach to Argument Labeling with Joint Inference in Discourse 68 - 77. Due to experimental setting and evaluation problems in the code written by the first auth… find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/papers/kong-2014-constituent-based-approach-argument-labeling/