Elamur Research⌕
☰
home / papers / 2009 / Nested Named Entity Recognition

Nested Named Entity Recognition

Summary

Standard NER systems assume entities never overlap, so they miss entities embedded inside others, such as a location within an organization name or a protein inside a DNA mention. Such cases are common: about 17% of GENIA entities and nearly half of AnCora entities are embedded. Finkel and Manning treat nested NER as constituency parsing. Each sentence becomes a shallow tree in which every entity is a phrase, POS tags are preterminals and a single root spans the sentence, with no other syntax. They then train a discriminative, CRF-based chart parser (CRF-CFG) on these trees, modeling entities and POS tags jointly. They test on biomedical English (GENIA) and on Spanish and Catalan newswire (AnCora) against a flat semi-CRF. On top-level entities the nested model is generally as good or better, winning in three of four settings. On all entities it raises overall F-score by up to 30%, because the flat model cannot output inner entities at all. The work shows that nested annotation can be exploited without hurting standard evaluation, and it argues for keeping nesting in corpora.

Key points

  • Data: GENIA (about 17% embedded entities) and the Spanish and Catalan parts of AnCora (47.1% embedded). AnCora includes strong and weak entities of types person, location, organization, date, number and other.
  • AnCora setup:
    • Sentences longer than 80 words were removed, leaving 15,591 training sentences for Spanish and 14,906 for Catalan.
    • The roughly 250 morphological POS tags were cut to their first two characters, giving 57 tags.
    • A 200-cluster distributional similarity model was trained for each language.
  • Top-level results against the semi-CRF: the nested model was 4.38% better on Spanish and 2.45% worse on Catalan. Joint POS accuracy was slightly higher with nested training (95.93 vs 95.60 for Spanish, 96.62 vs 96.59 for Catalan).
  • Cost: inference scales cubically, O(n³), with the number of words in a sentence. On the full GENIA corpus, the nested model needed about 23 hours to train, compared with 16 hours for the semi-CRF.
  • When the data has no nesting, the model reduces exactly to a semi-CRF with no limit on entity length.

Interesting ideas

  • I1. Recast span structure as a tree and reuse a parser. Encoding entities as constituents lets an existing CRF-CFG parser handle nesting with no new decoding algorithm. The same trick could fit other hierarchical span tasks, such as nested toponyms or compound trade product names.
  • I2. Parent and grandparent label annotation. Adding ancestor labels to each node lets the model learn which entity types tend to appear inside which. This is a cheap way to capture structural regularities in any tree-shaped labeling.
  • I3. Joint modeling as pruning. Restricting each word to POS tags seen in its distributional cluster, and allowing only grammar rules observed in the data, made joint inference faster than modeling entities alone. Constraints taken from the data can make richer models cheaper, not costlier.
  • I4. Cluster-based candidate tag sets. Using unsupervised word clusters to decide which tags a word may take gives broad coverage for unseen words while keeping the search small. This is helpful for morphologically rich languages like Ukrainian.
  • I5. Report two evaluation views. Scoring on top-level entities only and on all entities separates two questions: whether nesting hurts the traditional task, and how much information a flat model loses.

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: Named entity recognition · Part-of-speech tagging

Cited by

91 papers of our library cite it (of the 140884 whose bibliographies we read); 8 of them have a page here, 83 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{finkel2009nested,
  title = {{Nested Named Entity Recognition}},
  author = {Finkel, Jenny Rose and Manning, Christopher D.},
  booktitle = {Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing},
  year = {2009},
  doi = {10.3115/1699510.1699529},
  url = {https://doi.org/10.3115/1699510.1699529}
}

References

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

  1. Beatrice Alex, Barry Haddow, and Claire Grover. 2007. Recognising nested named entities in biomedical text. In BioNLP Workshop at ACL 2007, pages 65–72. find
  2. Galen Andrew. 2006. A hybrid markov/semi-markov conditional random field for sequence segmentation. In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP 2006). find
  3. A. Borthwick. 1999. A Maximum Entropy Approach to Named Entity Recognition. Ph.D. thesis, New York University. find
  4. Kate Byrne. 2007. Nested named entity recognition in historical archive text. In ICSC ’07: Proceedings of the International Conference on Semantic Computing, pages 589– 596. find
  5. Alexander Clark. 2003. Combining distributional and morphological information for part of speech induction. In Proceedings of the tenth Annual Meeting of the European Association for Computational Linguistics (EACL), pages 59–66. find our card
  6. Nigel Collier, J. Kim, Y. Tateisi, T. Ohta, and Y. Tsuruoka, editors. 2004. Proceedings of the International Joint Workshop on NLP in Biomedicine and its Applications. find
  7. J. R. Curran and S. Clark. 1999. Language independent NER using a maximum entropy tagger. In CoNLL 1999, pages 164–167. find
  8. Jenny Finkel, Shipra Dingare, Christopher Manning, Malvina Nissim, Beatrice Alex, and Claire Grover. 2005. Exploring the boundaries: Gene and protein identification in biomedical text. In BMC Bioinformatics 6 (Suppl. 1). find
  9. Jenny Rose Finkel, Alex Kleeman, and Christopher D. Manning. 2008. Efficient, feature-based conditional random field parsing. In ACL/HLT-2008. find our card
  10. Baohua Gu. 2006. Recognizing nested named entities in GE-NIA corpus. In BioNLP Workshop at HLT-NAACL 2006, pages 112–113. find
  11. Jiřı́ Hana and Hana Hanová. 2002. Manual for morphological annotation. Technical Report TR-2002-14, UK MFF CKL. find
  12. Jun’ichi Kazama, Takaki Makino, Yoshihiro Ohta, and Jun’ichi Tsujii. 2002. Tuning support vector machines for biomedical named entity recognition. In Proceedings of the Workshop on Natural Language Processing in the Biomedical Domain (ACL 2002). find
  13. John Lafferty, Andrew McCallum, and Fernando Pereira. 2001. Conditional Random Fields: Probabilistic models for segmenting and labeling sequence data. In ICML 2001, pages 282–289. Morgan Kaufmann, San Francisco, CA. find
  14. Mitchell P. Marcus, Beatrice Santorini, and Mary Ann Marcinkiewicz. 1993. Building a large annotated corpus of English: The Penn Treebank. Computational Linguistics, 19(2):313–330. find our card
  15. L. Márquez, L. Padrè, M. Surdeanu, and L. Villarejo. 2007a. UPC: Experiments with joint learning within semeval task 9. In Proceedings of the 4th International Workshop on Semantic Evaluations (SemEval-2007). find
  16. L. Márquez, L. Villarejo, M.A. Martı́, and M. Taulè. 2007b. Semeval-2007 task 09: Multilevel semantic annotation of Catalan and Spanish. In Proceedings of the 4th International Workshop on Semantic Evaluations (SemEval-2007). find
  17. M.A. Martı́, M. Taulè, M. Bertran, and L. Márquez. 2007. Ancora: Multilingual and multilevel annotated corpora. MS , Universitat de Barcelona. find
  18. Ryan McDonald, Koby Crammer, and Fernando Pereira. 2005. Flexible text segmentation with structured multilabel classification. In HLT ’05: Proceedings of the conference on Human Language Technology and Empirical Methods in Natural Language Processing, pages 987– 994. find
  19. Tomoko Ohta, Yuka Tateisi, and Jin-Dong Kim. 2002. The GENIA corpus: an annotated research abstract corpus in molecular biology domain. In Proceedings of the second international conference on Human Language Technology Research, pages 82–86. find
  20. Erik F. Tjong Kim Sang and Fien De Meulder. 2003. Introduction to the conll-2003 shared task: Language-independent named entity recognition. In Proceedings of CoNLL-2003. find
  21. Sunita Sarawagi and William W. Cohen. 2004. Semi-markov conditional random fields for information extraction. In In Advances in Neural Information Processing Systems 17, pages 1185–1192. find
  22. Dan Shen, Jie Zhang, Guodong Zhou, Jian Su, and Chew-Lim Tan. 2003. Effective adaptation of a hidden markov model-based named entity recognizer for biomedical domain. In Proceedings of the ACL 2003 workshop on Natural language processing in biomedicine. Association for Computational Linguistics (ACL… find
  23. Yoshimasa Tsuruoka and Jun’ichi Tsujii. 2003. Boosting precision and recall of dictionary-based protein name recognition. In Proceedings of the ACL-03 Workshop on Natural Language Processing in Biomedicine, pages 41– 48. find
  24. Yoshimasa Tsuruoka, Yuka Tateishi, Jin-Dong Kim, Tomoko Ohta, John McNaught, Sophia Ananiadou, and Jun’ichi Tsujii. 2005. Developing a robust part-of-speech tagger for biomedical text. In Advances in Informatics - 10th Panhellenic Conference on Informatics, LNCS 3746, pages 382–392. find
  25. Jie Zhang, Dan Shen, Guodong Zhou, Jian Su, and Chew-Lim Tan. 2004. Enhancing HMM-based biomedical named entity recognition by studying special phenomena. Journal of Biomedical Informatics, 37(6):411–422. find
  26. GuoDong Zhou and Jian Su. 2004. Exploring deep knowledge resources in biomedical name recognition. In Joint Workshop on Natural Language Processing in Biomedicine and Its Applications at Coling 2004. find
  27. Guodong Zhou, Jie Zhang, Jian Su, Dan Shen, and Chewlim Tan. 2004. Recognizing names in biomedical texts: a machine learning approach. Bioinformatics, 20(7):1178– 1190. find
  28. Guodong Zhou. 2006. Recognizing names in biomedical texts using mutual information independence model and SVM plus sigmoid. International Journal of Medical Informatics, 75:456–467. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/finkel-2009-nested-named-entity-recognition/