Elamur Research⌕
☰
home / papers / 2018 / Simpler but More Accurate Semantic Dependency Parsing

Simpler but More Accurate Semantic Dependency Parsing

Summary

Semantic dependency schemes represent meaning-oriented relations between words as directed acyclic graphs rather than trees. This lets a word such as a shared subject attach to several heads. Most parsers, however, are built to output trees. Starting from the biaffine BiLSTM parser that Dozat and Manning (2017) designed for syntax, the authors make only small changes so it can produce graphs:

  • The MST decoding step is removed.
  • Softmax is replaced with sigmoid cross-entropy, so each possible edge is scored independently.
  • A separate labeler assigns a relation to each predicted edge, and the two losses are balanced by an interpolation weight.

They evaluate on the three SemEval 2014/2015 formalisms built on the Penn Treebank: DM, PAS and PSD. Testing covers in-domain WSJ and out-of-domain Brown data. The basic model already surpasses the more complex multitask system of Peng et al. (2017). Adding character-level and lemma embeddings widens that lead. Ablation studies suggest that careful tuning of sizes and dropout matters more than elaborate architecture. The result argues that graph-based parsers adapt naturally to different dependency structures.

Key points

  • The basic system beats the previous state of the art (PTS17 Freda3) by 0.6% labeled F1. The +Char +Lemma version beats it by 1.9%, reaching 86.3 average out-of-domain LF1 and 89.5 in-domain.
  • In-domain LF1 for the best model is 93.7 on DM, 93.9 on PAS and 81.0 on PSD. PAS gains little from lemma or character embeddings, possibly because it is close to a ceiling.
  • The setup uses a 3-layer BiLSTM with 600 units, 100-dimensional GloVe vectors projected to 125 dimensions, and a loss interpolation weight λ = .025. Training runs at roughly 300 sequences per second and parsing at about 1,000 per second on an nVidia Titan X.
  • The authors trained twenty DM models per variant. Removing the hidden FNN layers or using a diagonal tensor in the edge classifier significantly lowers accuracy. Choosing bilinear versus biaffine classifiers makes little difference.

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: Natural language processing

Topics: Dependency parsing

Cited by

112 papers of our library cite it (of the 140884 whose bibliographies we read); 10 of them have a page here, 102 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/p18-2077

@inproceedings{dozat2018simpler,
  title = {{Simpler but More Accurate Semantic Dependency Parsing}},
  author = {Dozat, Timothy and Manning, Christopher D.},
  booktitle = {Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers)},
  year = {2018},
  doi = {10.18653/v1/p18-2077},
  url = {https://doi.org/10.18653/v1/p18-2077}
}

References

32 references, 11 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Alfred V Aho and Jeffrey D Ullman. 1972. The theory of parsing, translation, and compiling, volume 1. Prentice Hall. find
  2. Daniel Gildea and Daniel Jurafsky. 2002. Automatic labeling of semantic roles. Computational linguistics 28(3):245–288. find our card
  3. Mariana SC Almeida and André FT Martins. 2015. Lisbon: Evaluating turbosemanticparser on multiple languages and out-of-domain data. In Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015). pages 970–973. find
  4. Jan Hajic, Eva Hajicová, Jarmila Panevová, Petr Sgall, Ondrej Bojar, Silvie Cinková, Eva Fucı́ková, Marie Mikulová, Petr Pajas, Jan Popelka, et al. 2012. Announcing prague czech-english dependency treebank 2.0. In LREC. pages 3153–3160. find
  5. Miguel Ballesteros, Yoav Goldberg, Chris Dyer, and Noah A Smith. 2016. Training with exploration improves a greedy stack-LSTM parser. Proceedings of the conference on empirical methods in natural language processing . find
  6. Young-Bum Kim, Karl Stratos, and Ruhi Sarikaya. 2016. Frustratingly easy neural domain adaptation. In Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers. pages 387–396. find
  7. Marie Candito, Bruno Guillaume, Guy Perrier, and Djamé Seddah. 2017. Enhanced ud dependencies with neutralized diathesis alternation. In Depling 2017-Fourth International Conference on Dependency Linguistics. find our card
  8. Erich Leo Lehmann, HJM D’Abrera, et al. 1975. Nonparametrics. Holden-Day. find
  9. Jinho D Choi and Andrew McCallum. 2013. Transition-based dependency parsing with selectional branching. In Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). volume 1, pages 1052–1062. find
  10. Xuezhe Ma, Zecong Hu, Jingzhou Liu, Nanyun Peng, Graham Neubig, and Eduard Hovy. 2018. Stack-pointer networks for dependency parsing. ACL . find
  11. André FT Martins, Noah A Smith, Pedro MQ Aguiar, and Mário AT Figueiredo. 2011. Dual decomposition with many overlapping components. In Proceedings of the Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics, pages 238–249. find
  12. Hal Daumé III. 2007. Frustratingly easy domain adaptation. find our card
  13. Marie-Catherine De Marneffe, Bill MacCartney, Christopher D Manning, et al. 2006. Generating typed dependency parses from phrase structure parses. In Proceedings of LREC. volume 6, pages 449–454. find our card
  14. Yusuke Miyao and Jun’ichi Tsujii. 2004. Deep linguistic analysis for the accurate identification of predicate-argument relations. In Proceedings of the 20th international conference on Computational Linguistics. Association for Computational Linguistics, page 1392. find
  15. Timothy Dozat and Christopher D. Manning. 2017. Deep biaffine attention for neural dependency parsing. find our card
  16. Joakim Nivre. 2003. An efficient algorithm for projective dependency parsing. In Proceedings of the 8th International Workshop on Parsing Technologies (IWPT. Citeseer. find our card
  17. Timothy Dozat, Peng Qi, and Christopher D Manning. 2017. Stanford’s graph-based neural dependency parser at the conll 2017 shared task. CoNLL pages 20–30. find our card
  18. Joakim Nivre, Marie-Catherine de Marneffe, Filip Ginter, Yoav Goldberg, Jan Hajic, Christopher D Manning, Ryan McDonald, Slav Petrov, Sampo Pyysalo, Natalia Silveira, et al. 2016. Universal dependencies v1: A multilingual treebank collection. In Proceedings of the 10th International Conference on La… find our card
  19. Yantao Du, Fan Zhang, Xun Zhang, Weiwei Sun, and Xiaojun Wan. 2015. Peking: Building semantic dependency graphs with a hybrid parser. In Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015). pages 927–931. find
  20. Chris Dyer, Miguel Ballesteros, Wang Ling, Austin Matthews, and Noah A Smith. 2015. Transition-based dependency parsing with stack long short-term memory. Proceedings of the conference on empirical methods in natural language processing . find our card
  21. Stephan Oepen, Marco Kuhlmann, Yusuke Miyao, Daniel Zeman, Silvie Cinková, Dan Flickinger, Jan Hajic, and Zdenka Uresova. 2015. find
  22. Semeval 2015 task 18: Broad-coverage semantic dependency parsing. In Proceedings of the 9th International Workshop on Semantic Evaluation. pages 915–926. find
  23. Dan Flickinger, Yi Zhang, and Valia Kordoni. 2012. Deepbank. a dynamically annotated treebank of the wall street journal. In Proceedings of the 11th International Workshop on Treebanks and Linguistic Theories. pages 85–96. find
  24. Stephan Oepen, Marco Kuhlmann, Yusuke Miyao, Daniel Zeman, Dan Flickinger, Jan Hajic, Angelina Ivanova, and Yi Zhang. 2014. find
  25. Semeval 2014 task 8: Broad-coverage semantic dependency parsing. In Proceedings of the 8th International Workshop on Semantic Evaluation. pages 63–72. find
  26. Stephan Oepen and Jan Tore Lønning. 2006. Discriminant-based mrs banking. In Proceedings of the 5th International Conference on Language Resources and Evaluation. pages 1250–1255. find our card
  27. Hao Peng, Sam Thomson, and Noah A Smith. 2017. Deep multitask learning for semantic dependency parsing. In ACL. volume 1, pages 2037–2048. find
  28. Jeffrey Pennington, Richard Socher, and Christopher D Manning. 2014. Glove: Global vectors for word representation. Proceedings of the Empiricial Methods in Natural Language Processing (EMNLP 2014) 12. find
  29. Siva Reddy, Oscar Täckström, Slav Petrov, Mark Steedman, and Mirella Lapata. 2017. Universal semantic parsing. arXiv preprint arXiv:1702.03196 . find
  30. Sebastian Schuster and Christopher D. Manning. 2016. Enhanced English Universal Dependencies: An improved representation for natural language understanding tasks. In Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC 2016). find our card
  31. Sebastian Schuster, Éric Villemonte de la Clergerie, Marie Candito, Benoı̂t Sagot, Christopher D. Manning, and Djamé Seddah. 2017. Paris and Stanford at EPE 2017: Downstream evaluation of graph-based dependency representations. In Proceedings of the 2017 Shared Task on Extrinsic Parser Evaluation… find
  32. Yuxuan Wang, Wanxiang Che, Jiang Guo, and Ting Liu. 2018. A neural transition-based approach for semantic dependency graph parsing. AAAI . find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/dozat-2018-simpler-but-more-accurate-semantic/