Elamur Research
☰
home / papers / 2019 / Entity, Relation, and Event Extraction with Contextualized Span Representations

Entity, Relation, and Event Extraction with Contextualized Span Representations

Written for the Research library (claude-opus-5-5) from the paper text. Original — aclanthology.org/D19-1585.pdf

Summary

Extracting entities, relations and events often requires information beyond a single sentence. Coreference links, for example, can help decide what type a hard-to-label mention should get. This paper introduces DyGIE++, a multi-task information extraction system that extends the span-based DyGIE model. DyGIE++ enumerates candidate spans and encodes them with BERT, feeding each sentence together with neighbouring sentences. It then refines the span vectors through dynamically built graphs (coreference, relation and event propagation) before task-specific classifiers score them. Coreference serves only as an auxiliary task. The authors evaluate on four corpora from different domains: ACE05, SciERC, GENIA and WLPC. They compare a fine-tuned BERT setup with a cheaper variant that trains an LSTM over frozen BERT embeddings.

DyGIE++ sets new best results on nearly every task and subtask. The analysis shows a division of labour: BERT handles context within a sentence and in nearby sentences, while graph propagation captures longer-range links. The graph updates help most in specialized scientific domains. The code is openly released and is designed to be adapted to new tasks and datasets.

Key points

  • DyGIE++ beats prior state of the art on all subtasks except one (ACE05 event argument identification: 55.4 vs 57.2 F1). Relative error reductions range from 0.2% to 27.9%. The largest single gain is SciERC relations, which rise from 41.6 to 48.4 F1.
  • A BERT input window of 3 sentences gave the best scores on all relation and event extraction tasks.
  • Fine-tuning BERT gives modest gains for NER (+0.32 F1 on average) and relations (+1.0 F1) but lowers event extraction scores by 1.6 F1. The BERT+LSTM variant keeps gradients for about 15M parameters, compared with 100M in BERT BASE, so it is lighter and easier to train.
  • Replacing BERT with SciBERT improves the scientific datasets: SciERC entities go from 69.8 to 72.0 F1, SciERC relations from 41.9 to 45.3, and GENIA from 78.4 to 79.5.
  • On the SciERC dev set, coreference propagation fixed 44 entity errors and introduced 21. Most fixes removed overly specific labels, such as replacing Method with Other Scientific Term. Code is available at github.com/dwadden/dygiepp.

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: Coreference resolution

Cited by

254 papers of our library cite it (of the 140673 whose bibliographies we read); 6 of them have a page here, 248 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/d19-1585 · arXiv 1909.03546

@inproceedings{wadden2019entity,
  title = {{Entity, Relation, and Event Extraction with Contextualized Span Representations}},
  author = {Wadden, David and Wennberg, Ulme and Luan, Yi and Hajishirzi, Hannaneh},
  booktitle = {Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP)},
  year = {2019},
  doi = {10.18653/v1/d19-1585},
  eprint = {1909.03546},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.18653/v1/d19-1585}
}

References

26 references, 10 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Iz Beltagy, Arman Cohan, and Kyle Lo. 2019. Scibert: Pretrained contextualized embeddings for scientific text. ArXiv, abs/1903.10676. find our card
  2. Samuel R. Bowman, Gabor Angeli, Christopher Potts, and Christopher D. Manning. 2015. A large annotated corpus for learning natural language inference. In EMNLP. find
  3. Fenia Christopoulou, Makoto Miwa, and Sophia Ananiadou. 2018. A walk-based model on entity graphs for relation extraction. In ACL. find
  4. Andrew M Dai and Quoc V Le. 2015. Semi-supervised sequence learning. In NeurIPs. find
  5. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2018. Bert: Pre-training of deep bidirectional transformers for language understanding. In NAACL-HLT. find
  6. Matt Gardner, Joel Grus, Mark Neumann, Oyvind Tafjord, Pradeep Dasigi, Nelson F. Liu, Matthew Peters, Michael Schmitz, and Luke S. Zettlemoyer. 2017. Allennlp: A deep semantic natural language processing platform. find our card
  7. Jin-Dong Kim, Tomoko Ohta, Yuka Tateisi, and Jun’ichi Tsujii. 2003. Genia corpus - a semantically annotated corpus for bio-textmining. Bioinformatics, 19 Suppl 1:i180–2. find
  8. Chaitanya Kulkarni, Wei Xu, Alan Ritter, and Raghu Machiraju. 2018. An annotated corpus for machine reading of instructions in wet lab protocols. In NAACL-HLT. find
  9. Kenton Lee, Luheng He, and Luke S. Zettlemoyer. 2018. Higher-order coreference resolution with coarse-to-fine inference. In NAACL-HLT. find our card
  10. Qi Li and Heng Ji. 2014. Incremental joint extraction of entity mentions and relations. In ACL. find
  11. Yi Luan, Dave Wadden, Luheng He, Amy Shah, Mari Ostendorf, and Hannaneh Hajishirzi. 2019. A general framework for information extraction using dynamic span graphs. In NAACL-HLT. find our card
  12. Makoto Miwa and Mohit Bansal. 2016. End-to-end relation extraction using lstms on sequences and tree structures. In ACL. find our card
  13. Trung Minh Nguyen and Thien Huu Nguyen. 2019. One for all: Neural joint modeling of entities and events. In AAAI. find
  14. Nanyun Peng, Hoifung Poon, Chris Quirk, Kristina Toutanova, and Wen tau Yih. 2017. Cross-sentence n-ary relation extraction with graph lstms. Transactions of the Association for Computational Linguistics, 5:101–115. find
  15. Matthew E. Peters, Waleed Ammar, Chandra Bhagavatula, and Russell Power. 2017. Semi-supervised sequence tagging with bidirectional language models. In ACL. find our card
  16. Matthew E. Peters, Mark Neumann, Mohit Iyyer, Matt Gardner, Christopher Clark, Kenton Lee, and Luke Zettlemoyer. 2018. Deep contextualized word representations. In NAACL. find
  17. Sameer Pradhan, Alessandro Moschitti, Nianwen Xue, Olga Uryupina, and Yuchen Zhang. 2012. Conll-2012 shared task: Modeling multilingual unrestricted coreference in ontonotes. In Joint Conference on EMNLP and CoNLL-Shared Task, pages 1– 40. Association for Computational Linguistics. find our card
  18. Yujie Qian, Enrico Santus, Zhijing Jin, Jiang Guo, and Regina Barzilay. 2018. Graphie: A graph-based framework for information extraction. In NAACL-HLT. find
  19. Pranav Rajpurkar, Jian Zhang, Konstantin Lopyrev, and Percy Liang. 2016. Squad: 100, 000+ questions for machine comprehension of text. In EMNLP. find our card
  20. Erik F Sang and Fien De Meulder. 2003. Introduction to the conll-2003 shared task: Language-independent named entity recognition. In NAACL. find
  21. Lei Sha, Feng Qian, Baobao Chang, and Zhifang Sui. 2018. Jointly extracting event triggers and arguments by dependency-bridge rnn and tensor-based argument interaction. In AAAI. find
  22. Bishan Yang and Tom M. Mitchell. 2016. Joint extraction of events and entities within a document context. In HLT-NAACL. find
  23. Qi Li, Heng Ji, and Liang Huang. 2013. Joint event extraction via structured prediction with global features. In ACL. find
  24. Tongtao Zhang, Heng Ji, and Avirup Sil. 2019. Joint entity and event extraction with generative adversarial imitation learning. Data Intelligence, 1:99–120. find
  25. Yi Luan, Luheng He, Mari Ostendorf, and Hannaneh Hajishirzi. 2018. Multi-task identification of entities, relations, and coreference for scientific knowledge graph construction. In EMNLP. find our card
  26. Yuhao Zhang, Peng Qi, and Christopher D. Manning. 2018. Graph convolution over pruned dependency trees improves relation extraction. In EMNLP. find our card