Improving a Strong Neural Parser with Conjunction-Specific Features

What it is about

Parsers are accurate overall but attach the conj relation poorly, even in simple sentences. The authors add features to the BIST transition parser (arc-hybrid, with a Kiperwasser and Goldberg scorer) that measure the symmetry of the two conjuncts: whether their head words share a lemma, suffix, capitalization or sentiment, and whether their POS match. The features are used only where the candidate attachment is a conjunction. On the Penn Treebank with Stanford dependencies, labelled attachment rises from 91.9 to 92.1 LAS, and the F1 on the conj relation with its head rises from 82.5 to 83.6. Most of the newly correct conj attachments involve a conjunction feature, while few of the cases the baseline gets right alone do. A reviewer's suggestion to add character-level morphology gave no meaningful change.

Ideas

• I1. en/src/parse.rs feature template: add conjunct-symmetry features (lemma match, suffix match, capitalization match, UPOS match) for candidate conj attachments in the arc-standard perceptron; metric: conj Rel+Att F1 and LAS on EWT test.
• I2. gparse biaffine scorer: add the same pair features as an input to the arc score for conj candidates; metric: conj F1 on EWT dev, two seeds, against the current model.
• I3. treeagree: a gate rule that counts conj attachments between words of different UPOS (a symmetry violation) per 1k sentences in the silver corpus; use it as a check, not a fix.

For Mova

The chapter is 24. Coordination and subordination. The code is en/src/parse.rs (the feature set of the perceptron) and gparse (the arc scorer), with treeagree as the check. The paper's gain is about one point on conj F1, so measure on EWT test with the conj counts before claiming any LAS change.

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.18653/v1/e17-2055

@inproceedings{ficler2017improving,
  title = {{Improving a Strong Neural Parser with Conjunction-Specific Features}},
  author = {Ficler, Jessica and Goldberg, Yoav},
  booktitle = {Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 2, Short Papers},
  year = {2017},
  doi = {10.18653/v1/e17-2055},
  url = {https://doi.org/10.18653/v1/e17-2055}
}

References

14 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Daniel Andor, Chris Alberti, David Weiss, Aliaksei Severyn, Alessandro Presta, Kuzman Ganchev, Slav Petrov, and Michael Collins. 2016. Globally normalized transition-based neural networks. In Proc. of ACL. find
2. Deirdre Hogan. 2007. Coordinate noun phrase disambiguation in a generative parsing model. In Proceedings of Association for Computational Linguistics, pages 680–687. find
3. Miguel Ballesteros, Chris Dyer, and Noah A. Smith. 2015. Improved transition-based parsing by modeling characters instead of words with LSTMs. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 349–359, Lisbon, Portugal, September. Association for Computational Linguistics. find
4. Mark Johnson, Stuart Geman, Stephen Canon, Zhiyi Chi, and Stefan Riezler. 1999. Estimators for stochastic ”unification-based” grammars. In Proc. of ACL, pages 535–541, College Park, Maryland, USA, June. Association for Computational Linguistics. find
5. Miguel Ballesteros, Yoav Goldberg, Chris Dyer, and Noah A Smith. 2016. Training with exploration improves a greedy stack-LSTM parser. In proceedings of Short Papers EMNLP. find
6. Eliyahu Kiperwasser and Yoav Goldberg. 2016. Simple and accurate dependency parsing using bidirectional LSTM feature representations. Transactions of the Association for Computational Linguistics, 4:313–327. find our card
7. Steven Bird. 2006. NLTK: the natural language toolkit. In Proceedings of the COLING/ACL on Interactive presentation sessions, pages 69–72. Association for Computational Linguistics. find
8. Marco Kuhlmann, Carlos Gómez-Rodrı́guez, and Giorgio Satta. 2011. Dynamic programming algorithms for transition-based dependency parsers. In Proc. of ACL, pages 673–682. Association for Computational Linguistics. find
9. Jeffrey Oliver Breen. 2012. Mining twitter for airline consumer sentiment. Practical text mining and statistical analysis for non-structured text data applications, 133. find
10. Adhiguna Kuncoro, Miguel Ballesteros, Lingpeng Kong, Chris Dyer, and Noah A. Smith. 2016. Distilling an ensemble of greedy dependency parsers into one MST parser. In Proc. of EMNLP, pages 1744–1753, Austin, Texas, November. Association for Computational Linguistics. find
11. 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
12. George A. Miller. 1995. WordNet: a lexical database for english. Communications of the ACM, 38(11):39–41. ↗ our card
13. Masashi Shimbo and Kazuo Hara. 2007. A discriminative learning model for coordinate conjunctions. In Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning (EMNLP-CoNLL), pages 610–619, Prague, Czech Republic, June. Association for Computational Linguistics. find
14. David Weiss, Chris Alberti, Michael Collins, and Slav Petrov. 2015. Structured training for neural network transition-based parsing. In Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 323–333, Beijing, China, July. Association for Computational Linguistics. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/aclbulk-2017-ficler-improving-strong-neural-parser-conjunction/