Elamur Research⌕
☰
home / papers / 2014 / A Hybrid Approach to Features Representation for Fine-grained Arabic Named Entity Recognition

A Hybrid Approach to Features Representation for Fine-grained Arabic Named Entity Recognition

Paper · 2014 · Fahd Alotaibi and Mark Lee

Summary

Most Arabic NER work tags only a few coarse types, such as person, organisation and location, in news text. This limits downstream uses like question answering. The authors address fine-grained Arabic NER with 50 classes grouped under 8 coarse ones.

They built two manually annotated gold corpora and released them publicly. NewsFANE Gold re-labels the ANERcorp news text, and WikiFANE Gold is drawn from Arabic Wikipedia. They compare these with two automatically built Wikipedia corpora, looking at entity density, phrase length and class spread.

A CRF baseline uses the standard features from coarse-grained work. The authors then replace the fixed token window with features taken from a dependency parse, since Arabic word order is fairly free and related words can sit far apart. They also add global evidence from Brown clusters induced on large unlabelled text.

The cluster features clearly raise test F-scores on all four corpora, mainly through better recall. This shows that structure-aware and corpus-level features help where labelled data for many classes is thin.

Key points

  • The new resources are NewsFANE Gold (about 170K tokens, inter-annotator entity F of 91%) and WikiFANE Gold (about 500K tokens, 89%). They follow the two-level taxonomy of 8 coarse and 50 fine classes used for WikiFANE Selective/Whole.
  • The CRF baseline uses the BILOU scheme with these features:
    • a ±2 token window,
    • affixes and token length,
    • gender, number and person,
    • POS and base-phrase chunks,
    • a gazetteer flag.
  • Baseline test F-scores were 61.67 (NewsFANE Gold), 54.04 (WikiFANE Gold), 71.84 (WikiFANE Selective) and 69.15 (WikiFANE Whole).
  • The Brown clustering setup:
    • 1000 clusters, learned on 1.17M–2.1M unlabelled tokens (Arabic Gigaword was added for the news corpus).
    • Features are bit-string prefixes of length 4, 6, 8, 10 and 12.
    • Test F-scores rose to 74.42, 68.64, 80.17 and 73.66.
    • Recall rose by roughly 7–13 points on three of the four corpora.
  • Corpus statistics:
    • WikiFANE Gold is the densest in entities.
    • News text and WikiFANE Whole contain more single-word entity names.
    • Multi-word boundaries are hard to find because of Arabic morphology.

Interesting ideas

  • I1. Build each token's context from its syntactic head, its dependents and their grammatical relations, rather than from a fixed span of neighbours; this reaches distant related words and copes with word-order variation.
  • I2. Use Brown cluster bit-paths cut at several prefix lengths, so one word gets coarse-to-fine group memberships and the model picks the useful granularity.
  • I3. Induce clusters on unlabelled text matched in genre to the target corpus (news with Gigaword, Wikipedia with Wikipedia).
  • I4. Profile corpora before training by entity density, phrase-length distribution and class spread, to explain why scores differ across gold and automatic datasets.
  • I5. Use a two-level type taxonomy that maps cleanly onto coarser schemes (CoNLL/ACE), so fine labels can always be collapsed for comparison.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (tree): Entity and argument features could be read off the UD tree (head lemma, deprel, dependents) rather than surface windows, keeping each decision traceable to arcs.
  • I2 (tokens): Multi-depth cluster prefixes from large unlabelled English text could back off rare or unknown words in the tagger and lemma guesser, in a way that stays transparent.
  • I3 (learn): When building silver data or features for a new domain such as fables or school texts, induce distributional resources from same-genre raw text rather than a generic pool.
  • I4 (learn): Evaluation reports could include such corpus profiles, so that a score change is not mistaken for model progress when it really reflects a data shift.
  • I5 (world): Entity types in the world model could be kept hierarchical, for example person → athlete or politician, letting answers use fine types while staying compatible with coarse benchmarks.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓University of BirminghamBirmingham · United Kingdom25 works · 62,564 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 2Research subjects and languages the work is about.Semitic languages · 2,908 worksNamed entity recognition · 200 works

Cite

Cite as Fahd Alotaibi and Mark Lee. 2014. A Hybrid Approach to Features Representation for Fine-grained Arabic Named Entity Recognition. Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers.

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

@inproceedings{alotaibi2014hybrid,
  title = {{A Hybrid Approach to Features Representation for Fine-grained Arabic Named Entity Recognition}},
  author = {Fahd Alotaibi and Mark Lee},
  booktitle = {Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers},
  year = {2014},
  url = {https://aclanthology.org/C14-1093/}
}

References

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

  1. Samir AbdelRahman, Mohamed Elarnaoty, Marwa Magdy, and Aly Fahmy. 2010. Integrated machine learning techniques for arabic named entity recognition. IJCSI International Journal of Computer Science, 7(4):27–36. find
  2. Ahmed Abdul-Hamid and Kareem Darwish. 2010. Simplified feature set for arabic named entity recognition. In Proceedings of the 2010 Named Entities Workshop, pages 110–115, Uppsala, Sweden. Association for Computational Linguistics.
  3. Fahd Alotaibi and Mark Lee. 2013. Automatically developing a fine-grained arabic named entity corpus and gazetteer by utilizing wikipedia. In Proceedings of the Sixth International Joint Conference on Natural Language Processing, pages 392–400, Nagoya, Japan, October. Asian Federation of Natural Language Processing.
  4. Maha Althobaiti, Udo Kruschwitz, and Massimo Poesio. 2013. A semi-supervised learning approach to arabic named entity recognition. In Proceedings of the International Conference Recent Advances in Natural Language Processing RANLP 2013, pages 32–40, Hissar, Bulgaria, September. INCOMA Ltd. Shoumen, BUL-GARIA.
  5. Yassine Benajiba and Paolo Rosso. 2007. Anersys 2.0: Conquering the ner task for the arabic language by combining the maximum entropy with pos-tag information. In Proceedings of the Workshop on Natural Language-Independent Engineering, 3rd Indian Int. Conf. on Artificial Intelligence, IICAI-2007, Pune, India. find
  6. Yassine Benajiba and Paolo Rosso. 2008. Arabic named entity recognition using conditional random fields. In Proceedings of the Workshop on HLT & NLP Within the Arabic World. Arabic Language and Local Languages Processing: Status Updates and Prospects, 6th International Conference on Language Resources and Evaluation, volume 8, pages 26–31, Marrakech, Morocco. LREC-2008. find
  7. Yassine Benajiba, Paolo Rosso, and José Miguel Benedı́ruiz. 2007. Anersys: An arabic named entity recognition system based on maximum entropy. In Alexander Gelbukh, editor, Computational Linguistics and Intelligent Text Processing, volume 4394 of Lecture Notes in Computer Science, pages 143–153. Springer Berlin / Heidelberg. find
  8. Yassine Benajiba, Mona Diab, and Paolo Rosso. 2008a. Arabic named entity recognition: An svm-based approach. In Proceedings of 2008 Arab International Conference on Information Technology (ACIT), pages 16– 18, Amman, Jordan. Association of Arab Universities. ↗
  9. Yassine Benajiba, Mona Diab, and Paolo Rosso. 2008b. Arabic named entity recognition using optimized feature sets. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, pages 284–293, Honolulu, Hawaii. Association for Computational Linguistics.
  10. Yassine Benajiba, Mona Diab, and Paolo Rosso. 2009. Using language independent and language specific features to enhance arabic named entity recognition. IEEE Transactions on Audio, Speech and Language Processing. Special Issue on Processing Morphologically Rich Languages, 17(5). find
  11. Y. Benajiba, I. Zitouni, M. Diab, and P. Rosso. 2010. Arabic named entity recognition: Using features extracted from noisy data. In Proceedings of the ACL 2010 Conference Short Papers, pages 281–285, Uppsala, Sweden. Association for Computational Linguistics.
  12. Peter F Brown, Peter V Desouza, Robert L Mercer, Vincent J Della Pietra, and Jenifer C Lai. 1992. Class-based n-gram models of natural language. Computational linguistics, 18(4):467–479. find
  13. Marie Candito and Benoı̂t Crabbé. 2009. Improving generative statistical parsing with semi-supervised word clustering. In Proceedings of the 11th International Conference on Parsing Technologies, pages 138–141. Association for Computational Linguistics. find
  14. Mona Diab. 2009. Second generation amira tools for arabic processing: Fast and robust tokenization, pos tagging, and base phrase chunking. In 2nd International Conference on Arabic Language Resources and Tools. find
  15. Benjamin Farber, Dayne Freitag, Nizar Habash, and Owen Rambow. 2008. Improving ner in arabic using a morphological tagger. In Proceedings of the Sixth International Language Resources and Evaluation (LREC’08), pages 2509–2514, Marrakech, Morocco. European Language Resources Association (ELRA).
  16. George Hripcsak and Adam S Rothschild. 2005. Agreement, the f-measure, and reliability in information retrieval. Journal of the American Medical Informatics Association, 12(3):296–298. ↗
  17. Terry Koo, Xavier Carreras, and Michael Collins. 2008. Simple semi-supervised dependency parsing.
  18. Percy Liang. 2005. Semi-supervised learning for natural language. Ph.D. thesis, Massachusetts Institute of Technology. find
  19. Xiao Ling and Daniel S Weld. 2012. Fine-grained entity recognition. In Proceedings of the 26th Conference on Artificial Intelligence (AAAI). find
  20. Yuval Marton, Nizar Habash, and Owen Rambow. 2013. Dependency parsing of modern standard arabic with lexical and inflectional features. Computational Linguistics, 39(1):161–194. ↗
  21. Scott Miller, Jethran Guinness, and Alex Zamanian. 2004. Name tagging with word clusters and discriminative training. In HLT-NAACL, volume 4, pages 337–342. Citeseer.
  22. Alexis Mitchell, Stephanie Strassel, Shudong Huang, and Ramez Zakhary. 2005. Ace 2004 multilingual training corpus [ldc2005t09], March 15. [accessed 20 December 2013]. find
  23. Robert Parker, David Graff, Ke Chen, Junbo Kong, and Kazuaki Maeda. 2011. Arabic gigaword fifth edition [ldc2011t11], October 21. [accessed 20 December 2013]. find
  24. Lev Ratinov and Dan Roth. 2009. Design challenges and misconceptions in named entity recognition. In Proceedings of the Thirteenth Conference on Computational Natural Language Learning, pages 147–155, Boulder, Colorado. Association for Computational Linguistics.
  25. Khaled Shaalan and Mai Oudah. 2013. A hybrid approach to arabic named entity recognition. Journal of Information Science. ↗
  26. Maksim Tkachenko, Andrey Simanovsky, and St Petersburg. 2012. Named entity recognition: Exploring features. In Proceedings of KONVENS, pages 118–127. find
  27. Joseph Turian, Lev Ratinov, Yoshua Bengio, and Dan Roth. 2009. A preliminary evaluation of word representations for named-entity recognition. In NIPS Workshop on Grammar Induction, Representation of Language and Language Learning. find
  28. Joseph Turian, Lev Ratinov, and Yoshua Bengio. 2010. Word representations: a simple and general method for semi-supervised learning. In Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics, pages 384–394. Association for Computational Linguistics.
  29. Christopher Walker, Stephanie Strassel, Julie Medero, and Kazuaki Maeda. 2006. Ace 2005 multilingual training corpus [ldc2006t06], February 15. [accessed 20 December 2013]. find
  30. Ziqi Zhang. 2013. Named entity recognition: challenges in document annotation, gazetteer construction and disambiguation. Ph.D. thesis. find
  31. Imed Zitouni and Radu Florian. 2008. Mention detection crossing the language barrier. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, pages 600–609. Association for Computational Linguistics. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/alotaibi-2014-hybrid-approach-features-representation-fine/