Elamur Research⌕
☰
home / papers / 2013 / Named Entity Recognition using Cross-lingual Resources: Arabic as an Example

Named Entity Recognition using Cross-lingual Resources: Arabic as an Example

Paper · 2013 · Kareem Darwish

Summary

Arabic NER has no capitalization cue and relatively small knowledge bases such as Arabic Wikipedia, so recognizers mostly memorize names seen in training and miss new ones. Darwish addresses this by projecting each Arabic word into English and reading evidence from the English side. The links come from Wikipedia cross-language links and a large true-cased MT phrase table. Three cues are drawn from English: capitalization, weighted by confidence rather than a yes/no flag; transliteration mining, which treats a word whose translation is a transliteration as a likely name; and type information from English DBpedia. These features are added to a CRF++ baseline with character, stem and gazetteer features. The system is tested on ANERCORP and on two new test sets, recent news and tweets. On ANERCORP, F-measure is 4.1% (relative) above the best previously published result. On the new sets the gains are 17.1% (news) and 20.5% (tweets), driven mainly by recall. The work matters because a resource-poor language can borrow discriminative signals from a richer one without any parallel annotated text.

Key points

  • Baseline: CRF++ with default settings. Features are the leading and trailing 1–4 letters, membership in the Benajiba et al. gazetteer (under 5,000 entries), light-stemmed word forms, and the neighbouring words in raw and stemmed form.
  • Data: ANERCORP (about 150k tokens) with the 80/20 split of Abdul-Hamid and Darwish (2010). The new NEWS and TWEETS test sets cover politics, economics, health, sports, entertainment and science. On these new sets, baseline recall is below 50% for every entity type.
  • ANERCORP, all features: overall F 84.3, a 5.5% relative gain over the baseline. Recall rises 10.7% and precision barely changes.
  • NEWS: F 63.9 (+17.1%), with recall +27.7% and precision +1.6%.
  • TWEETS: F 39.9 (+20.5%) and recall +35.9%, but precision falls 6.2%.
  • Feature behaviour: DBpedia was the only cross-lingual feature that improved both precision and recall, except precision on tweets.

Interesting ideas

  • I1. Borrowing a missing cue through translation: when a language lacks a useful surface signal, translate the token and read the signal (here capitalization) on the other side. This works for any feature that exists in only one language.
  • I2. Confidence-weighted features instead of binary flags: Arabic names are often ordinary words, so the proportion of translations that are capitalized is more informative than a single yes/no bit. Ambiguous lexical evidence in general benefits from graded rather than hard features.
  • I3. Transliteration as a proxy for name-hood: if the most likely translation sounds like the source, the word is probably a name. This cheap test needs few training examples and works even when neither language has capitalization.
  • I4. Out-of-period and out-of-genre test sets: the new NEWS and TWEETS sets showed that strong standard scores largely reflected memorization, and the error analysis traced the precision drop to dialect words, shortened names and nicknames.

For Mova

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

  • I1 (tokens): when the tagger is unsure whether a lowercase or noisy-case token is a proper noun, it could check a cross-lingual or knowledge-base link and log that as an explicit, traceable piece of evidence.
  • I2 (world): entity typing could store a graded score, such as how often a string links to a Wikidata entity versus a common sense, so that later steps can explain why they chose a name reading.
  • I3 (back): during translation through the meta-world, a transliteration-like rendering could flag that a node should be treated as a named entity rather than a concept to be translated.
  • I4 (learn): evaluation could include deliberately shifted test sets (newer texts, other genres) as a control, so that gains on in-domain data are not mistaken for generalization.

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, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 works

🧭 Topics 3Research subjects and languages the work is about.Semitic languages · 3,066 worksMultilingual & cross-lingual NLP · 882 worksNamed entity recognition · 214 works

Cite

Cite as Kareem Darwish. 2013. Named Entity Recognition using Cross-lingual Resources: Arabic as an Example. Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

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

@inproceedings{darwish2013named,
  title = {{Named Entity Recognition using Cross-lingual Resources: Arabic as an Example}},
  author = {Kareem Darwish},
  booktitle = {Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2013},
  url = {https://aclanthology.org/P13-1153/}
}

References

26 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. A. Abdul-Hamid and K. Darwish. 2010. Simplified Feature Set for Arabic Named Entity Recognition. Proceedings of the 2010 Named Entities Workshop, ACL 2010, pages 110115.
  2. Mohammed Attia, Antonio Toral, Lamia Tounsi, Monica Monachini, and Josef van Genabith. 2010. An automatically built named entity lexicon for Arabic. In: LREC 2010 - 7th conference on International Language Resources and Evaluation, 17-23 May 2010, Valletta, Malta.
  3. Y. Benajiba, M. Diab, and P. Rosso. 2008. Arabic Named Entity Recognition using Optimized Feature Sets. Proceedings of the 2008 Conference on Empirical Methods in Natural Language Processing, pages 284293, Honolulu, October 2008.
  4. Y. Benajiba and P. Rosso. 2008. Arabic Named Entity Recognition using Conditional Random Fields. In Proc. of Workshop on HLT & NLP within the Arabic World, LREC08. find
  5. Y. Benajiba, P. Rosso and J. M. Benedi. 2007. ANERsys: An Arabic Named Entity Recognition system based on Maximum Entropy. In Proc. of CICLing-2007, Springer-Verlag, LNCS(4394), pp.143-153 find
  6. Y. Benajiba and P. Rosso. 2007. ANERsys 2.0: Conquering the NER task for the Arabic language by combining the Maximum Entropy with POS-tag information. In Proc. of Workshop on Natural Language-Independent Engineering, IICAI-2007. find
  7. Christian Bizer, Jens Lehmann, Georgi Kobilarov, Sren Auer, Christian Becker, Richard Cyganiak, Sebastian Hellmann. 2009. DBpedia A Crystallization Point for the Web of Data. Journal of Web Semantics: Science, Services and Agents on the World Wide Web, Issue 7, Pages 154165, 2009. ↗
  8. D. Burkett, S. Petrov, J. Blitzer, D. Klein. 2010. Learning Better Monolingual Models with Unannotated Bilingual Text. Proceedings of the Fourteenth Conference on Computational Natural Language Learning, pages 46–54. find
  9. A. El Kahki, K. Darwish, A. Saad El Din, M. Abd El-Wahab and A. Hefny. 2011. Improved Transliteration Mining Using Graph Reinforcement. EMNLP-2011. find
  10. B. Farber, D. Freitag, N. Habash, and O. Rambow. 2008. Improving NER in Arabic Using a Morphological Tagger. In Proc. of LREC08.
  11. K. Ganchev, J. Gillenwater, and B. Taskar. 2009. Dependency grammar induction via bitext projection constraints. In ACL-2009.
  12. Spence Green and John DeNero. 2012. A Class-Based Agreement Model for Generating Accurately Inflected Translations. In ACL-2012. find
  13. Ulf Hermjakob, Kevin Knight, and Hal Daum III. 2008. Name translation in statistical machine translation: Learning when to transliterate. ACL-08: HLT, Pages 389-397. find
  14. F. Huang. 2005. Multilingual Named Entity Extraction and Translation from Text and Speech. Ph.D. Thesis. Pittsburgh: Carnegie Mellon University. Philipp Koehn, Hieu Hoang, Alexandra Birch, Chris Callison-Burch, Marcello Federico, Nicola Bertoldi, Brooke Cowan, Wade Shen, Christine Moran, Richard Zens, Chris Dyer, Ondrej Bojar, Alexandra Constantin, Evan Herbst, Moses: Open Source Toolkit for Statistical Machine Translation, Annual Meeting of the Association for Computational Linguistics (ACL), demonstration session, Prague, Czech Republic, June 2007. ↗
  15. J. Lafferty, A. McCallum, and F. Pereira. 2001. Conditional random fields: Probabilistic models for segmenting and labeling sequence data, In Proc. of ICML, pp.282-289, 2001. find
  16. Leah S. Larkey, Lisa Ballesteros, and Margaret E. Connell. 2002. Improving stemming for Arabic information retrieval: light stemming and co-occurrence analysis. SIGIR-2002. find
  17. J. Mayfield, P. McNamee, and C. Piatko. 2003.Named Entity Recognition using Hundreds of Thousands of Features. HLT-NAACL 2003-Volume 4, 2003. find
  18. A. McCallum and W. Li. 2003. Early Results for Named Entity Recognition with Conditional Random Fields, Features Induction and Web-Enhanced Lexicons. In Proc. Conference on Computational Natural Language Learning. find
  19. P. McNamee and J. Mayfield. 2002. Entity extraction without language-specific. Proceedings of CoNLL, .2002 find
  20. Behrang Mohit, Nathan Schneider, Rishav Bhowmick, Kemal Oflazer, Noah A. Smith. 2012. Recall-oriented learning of named entities in Arabic Wikipedia. In Proceedings of the 13th Conference of the European Chapter of the Association for Computational Linguistics (EACL 2012), pp. 162-173. 2012.
  21. D. Nadeau and S. Sekine. 2009. A Survey of Named Entity Recognition and Classification. Named Entities: Recognition, Classification and Use, ed. S. Sekine and E. Ranchhod, John Benjamins Publishing Company. ↗
  22. Alexander E. Richman and Patrick Schone. 2008. Mining wiki resources for multilingual named entity recognition. Proceedings of the 46th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies. 2008. find
  23. K. Shaalan and H. Raza. 2007. Person Name Entity Recognition for Arabic. Proceedings of the 5th Workshop on Important Unresolved Matters, pages 1724, Prague, Czech Republic, June 2007.
  24. L. Shi, R. Mihalcea, M. Tian. 2010. Cross Language Text Classification by Model Translation and Semi-supervised Learning. Proceedings of the Conference on Empirical Methods in Natural Language Processing, EMNLP 2010. find
  25. Raghavendra Udupa, Anton Bakalov, and Abhijit Bhole. 2009. They Are Out There, If You Know Where to Look: Mining Transliterations of OOV Query Terms for Cross-Language Information Retrieval. Advances in Information Retrieval. Pages: 437-448. find
  26. D. Yarowsky and G. Ngai. 2001. Inducing Multilingual POS Taggers and NP Bracketers via Robust Projection across Aligned Corpora. In NAACL-2001. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/darwish-2013-named-entity-recognition-cross-lingual/