Elamur Research⌕
☰
home / papers / 2016 / Adapting Event Embedding for Implicit Discourse Relation Recognition

Adapting Event Embedding for Implicit Discourse Relation Recognition

Summary

This CoNLL 2016 shared task submission addresses sense classification of discourse relations in the Penn Discourse Treebank, with a focus on implicit relations, where no connective signals the sense. Instead of hand-crafted features, the authors represent each argument with pre-trained word vectors (Google News word2vec) together with a new type of representation: event embeddings. An event is a (verb, dependency) pair. Events that share a coreferent entity are linked into chains. These chains are treated like sentences and used to train a skip-gram model on a Wikipedia dump, with no manual annotation required. A multi-layer perceptron then combines the word and event vectors to predict the sense. Explicit relations are handled separately by a linear SVM with syntactic and connective features.

On the development set, adding event vectors raised implicit accuracy from 0.315 with words alone to 0.369. This is roughly on par with the strongest CoNLL 2015 system. The result suggests that event-level abstractions add useful semantic signal and slow overfitting in neural discourse classifiers.

Key points

  • Event embeddings:
    • Events were extracted with Stanford CoreNLP, using dependency parsing and coreference resolution.
    • Embeddings were trained with skip-gram plus negative sampling (5 negatives per event), rare-event pruning, subsampling and a dynamic window.
  • Implicit classifier:
    • The model is an MLP with three hidden layers.
    • Two parallel layers each merge the event and word vectors for one argument, and a third layer joins the two arguments.
    • A softmax layer outputs the sense.
  • Implicit results:
    • Development accuracy was 0.25 for the word-pair baseline, 0.315 with word vectors and 0.369 with word+event vectors.
    • Unofficial F1 was 0.3818 (dev), 0.3435 (test) and 0.3365 (blind).
    • The official TIRA blind score was 0.291, because an older model trained on all labels was used.
    • Dropping infrequent classes from training raised dev F1 from 0.36 to 0.38.
  • Explicit SVM: the classifier chooses among 14 senses and reached F1 of 0.8968 (dev) and 0.8796 (test).
  • Common confusions:
    • Expansion.Instantiation vs. Expansion.Restatement.
    • Contingency.Cause.Result vs. Contingency.Cause.Reason.
    • Many other classes vs. Expansion.Conjunction.

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

5 papers of our library cite it (of the 140673 whose bibliographies we read); 3 of them have a page here, 2 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/k16-2019

@inproceedings{pacheco2016adapting,
  title = {{Adapting Event Embedding for Implicit Discourse Relation Recognition}},
  author = {Pacheco, Maria Leonor and Lee, I-Ta and Zhang, Xiao and Zehady, Abdullah Khan and Daga, Pranjal and Jin, Di and Parolia, Ayush and Goldwasser, Dan},
  booktitle = {Proceedings of the CoNLL-16 shared task},
  year = {2016},
  doi = {10.18653/v1/k16-2019},
  url = {https://doi.org/10.18653/v1/k16-2019}
}

References

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

  1. Martin Potthast, Tim Gollub, Francisco Rangel, Paolo Rosso, Efstathios Stamatatos, and Benno Stein. 2014. Improving the Reproducibility of PAN’s Shared Tasks: Plagiarism Detection, Author Identification, and Author Profiling. In Evangelos Kanoulas, Mihai Lupu, Paul Clough, Mark Sander-son, Mark Hall… find
  2. Nathanael Chambers and Dan Jurafsky. 2009. Unsupervised learning of narrative schemas and their participants. 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 - Volume 2, ACL… find
  3. Yoav Goldberg and Omer Levy. 2014. word2vec explained: deriving mikolov et al.’s negative-sampling word-embedding method. arXiv preprint arXiv:1402.3722. — [arXiv:1402.3722](https://arxiv.org/abs/1402.3722) find
  4. Rashmi Prasad, Nikhil Dinesh, Alan Lee, Eleni Milt-sakaki, Livio Robaldo, Aravind Joshi, and Bonnie Webber. 2008. The penn discourse treebank 2.0. In In Proceedings of LREC. find our card
  5. 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 - Volume 1, EMNLP ’09, pages 343–351, Stroudsburg, PA, USA. Association for Com… find our card
  6. Xin Rong. 2014. word2vec parameter learning explained. arXiv preprint arXiv:1411.2738. — [arXiv:1411.2738](https://arxiv.org/abs/1411.2738) find
  7. Jianxiang Wang and Man Lan. 2015. A refined end-to-end discourse parser. In Proceedings of the Nineteenth Conference on Computational Natural Language Learning - Shared Task, pages 17–24, Beijing, China, July. Association for Computational Linguistics. find our card
  8. Christopher D. Manning, Mihai Surdeanu, John Bauer, Jenny Finkel, Steven J. Bethard, and David Mc-Closky. 2014. The Stanford CoreNLP natural language processing toolkit. In Association for Computational Linguistics (ACL) System Demonstrations, pages 55–60. find our card
  9. Nianwen Xue, Hwee Tou Ng, Sameer Pradhan, Rashmi Prasad, Christopher Bryant, and Attapol Rutherford. 2015. The conll-2015 shared task on shallow discourse parsing. In Proceedings of the Nineteenth Conference on Computational Natural Language Learning - Shared Task, pages 1–16, Beijing, China, July.… find our card
  10. Tomas Mikolov, Kai Chen, Greg Corrado, and Jeffrey Dean. 2013a. Efficient estimation of word representations in vector space. CoRR, abs/1301.3781. find
  11. Tomas Mikolov, Wen-tau Yih, and Geoffrey Zweig. 2013b. Linguistic regularities in continuous space word representations. In Proc. of NAACL. find
  12. Nianwen Xue, Hwee Tou Ng, Sameer Pradhan, Bonnie Webber, Attapol Rutherford, Chuan Wang, and Hongmin Wang. 2016. The conll-2016 shared task on multilingual shallow discourse parsing. In Proceedings of the Twentieth Conference on Computational Natural Language Learning - Shared Task, Berlin, Germany,… find our card
  13. Ashutosh Modi and Ivan Titov. 2014. Inducing neural models of script knowledge. In Proceedings of the Eighteenth Conference on Computational Natural Language Learning, pages 49–57, Ann Arbor, Michigan, June. Association for Computational Linguistics. find
  14. Karl Pichotta and Raymond J. Mooney. 2016. Learning statistical scripts with LSTM recurrent neural networks. In Proceedings of the 30th AAAI Conference on Artificial Intelligence (AAAI-16), Phoenix, Arizona. find
  15. Emily Pitler and Ani Nenkova. 2009. Using syntax to disambiguate explicit discourse connectives in text. In Proceedings of the ACL-IJCNLP 2009 Conference Short Papers, pages 13–16. Association for Computational Linguistics. find our card
  16. 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 - Volum… find our card

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/pacheco-2016-adapting-event-embedding-implicit-discourse/