Elamur Research⌕
☰
home / papers / 2016 / OPT: Oslo–Potsdam–Teesside. Pipelining Rules, Rankers, and Classifier Ensembles for Shallow Discourse Parsing

OPT: Oslo–Potsdam–Teesside. Pipelining Rules, Rankers, and Classifier Ensembles for Shallow Discourse Parsing

Summary

The paper presents OPT, a joint Oslo, Potsdam and Teesside entry to the CoNLL 2016 Shared Task on shallow discourse parsing in the PDTB style. The task is to find discourse relations in English text, delimit their two arguments and label their senses. The authors use a conventional multi-stage pipeline built from simple parts:

  • an SVM that disambiguates known explicit connectives,
  • rule-based positing of non-explicit relations between adjacent sentences,
  • an ensemble of classifiers for sense labelling.

The main novelty is in argument identification. Here they reuse a technique from their earlier work on negation and speculation scope: clausal constituents from the parse tree are ranked, and the best candidate is then refined with editing heuristics. OPT reached 27.77 F1 end-to-end on the blind test set, about four points above the previous best system (Wang & Lan, 2015). It did especially well at locating arguments and held up well across the shift in text type. The results suggest that good syntactic analysis directly benefits discourse parsing.

Key points

  • Connective disambiguation uses SVM light with roughly 1.2 million feature types, tuned by ten-fold cross-validation on the training data. The features mix surface n-grams with syntactic features from Pitler & Nenkova (2009) and Wang & Lan (2015).
  • Evaluated on WSJ test data, the rule-based detection of non-explicit relations achieves 93.2 F1, with a precision of 89.9 and a recall of 96.8. AltLex relations are not attempted.
  • Argument candidates are restricted to S, SBAR and SQ constituents. Constituent editing raises the alignment of explicit arguments with constituents from .483 to .813 for Arg1 and from .535 to .840 for Arg2. Ranking features were added only if a Wilcoxon signed-rank test showed a significant gain (p < .05).
  • The sense ensemble combines three predictors:
    • a majority-class predictor,
    • a LIBLINEAR reimplementation of Wang & Lan, which beat the previous year's best WSJ sense score by 0.89 F1,
    • an XGBoost variant.
  • After the submission, the authors tested a variant that adds an LSTM using the provided Word2Vec embeddings. It gave further gains in sense classification.

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

10 papers of our library cite it (of the 140673 whose bibliographies we read); 4 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. DOI 10.18653/v1/k16-2002

@inproceedings{oepen2016opt,
  title = {{OPT: Oslo–Potsdam–Teesside. Pipelining Rules, Rankers, and Classifier Ensembles for Shallow Discourse Parsing}},
  author = {Oepen, Stephan and Read, Jonathon and Scheffler, Tatjana and Sidarenka, Uladzimir and Stede, Manfred and Velldal, Erik and Øvrelid, Lilja},
  booktitle = {Proceedings of the CoNLL-16 shared task},
  year = {2016},
  doi = {10.18653/v1/k16-2002},
  url = {https://doi.org/10.18653/v1/k16-2002}
}

References

14 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Boser, B. E., Guyon, I., & Vapnik, V. (1992). A training algorithm for optimal margin classifiers. In Proceedings of the Fifth Annual ACM Conference on Computational Learning Theory (p. 144 – 152). Pittsburgh, PA, USA. find
  2. Wang, J., & Lan, M. (2015). A refined end-to-end discourse parser. In Proceedings of the 19th Conference on Natural Language Learning (p. 17 – 24). Bejing, China. find our card
  3. Crammer, K., & Singer, Y. (2001). On the algorithmic implementation of multiclass kernel-based vector machines. Journal of Machine Learning Research, 2, 265 – 292. find
  4. Xue, N., Ng, H. T., Pradhan, S., Prasad, R., Bryant, C., & Rutherford, A. (2015). The CoNLL-2015 shared task on shallow discourse parsing. In Proceedings of the 19th Conference on Natural Language Learning: Shared task (p. 1 – 16). Bejing, China. find our card
  5. Fan, R., Chang, K., Hsieh, C., Wang, X., & Lin, C. (2008). LIBLINEAR. A library for large linear classification. Journal of Machine Learning Research, 9, 1871 – 1874. find
  6. Xue, N., Ng, H. T., Pradhan, S., Webber, B., Rutherford, A., Wang, C., & Wang, H. (2016). The CoNLL-2016 shared task on multilingual shallow discourse parsing. In Proceedings of the 20th Conference on Natural Language Learning: Shared task. Berlin, Germany. find our card
  7. Friedman, J. H. (2000). Greedy function approximation. A gradient boosting machine. Annals of Statistics, 29, 1189 – 1232. find
  8. Hochreiter, S., & Schmidhuber, J. (1997). Long short-term memory. Neural Computation, 9(8), 1735 – 1780. find
  9. Joachims, T. (1999). Making large-scale SVM learning practical. In B. Schölkopf, C. Burges, & A. Smola (Eds.), Advances in kernel methods. Support vector learning. Cambridge, MA, USA: MIT Press. find
  10. Joachims, T. (2002). Optimizing search engines using clickthrough data. In Proceedings of the 8th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (p. 133 – 142). Edmonton, Canada. find
  11. Knott, A. (1996). A data-driven methodology for motivating a set of coherence relations. Unpublished doctoral dissertation, University of Edinburgh. find
  12. Lin, Z., Ng, H. T., & Kan, M.-Y. (2014). A PDTB-styled end-to-end discourse parser. Natural Language Engineering, 20(2), 151 – 184. find
  13. Pitler, E., & Nenkova, A. (2009). Using syntax to disambiguate explicit discourse connectives in text. In Proceedings of the 47th Meeting of the Association for Computational Linguistics (p. 13 – 16). Singapore. find our card
  14. Read, J., Velldal, E., Øvrelid, L., & Oepen, S. (2012). UiO1. Constituent-based discriminative ranking for find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/papers/oepen-2016-opt-oslo-potsdam-teesside-pipelining/