Elamur Research⌕
☰
home / papers / 2014 / Automatic Creation of Arabic Named Entity Annotated Corpus Using Wikipedia

Automatic Creation of Arabic Named Entity Annotated Corpus Using Wikipedia

Paper · 2014 · Maha Althobaiti, Udo Kruschwitz and Massimo Poesio

Summary

Arabic lacks large hand-labelled corpora for named entity recognition. Its rich morphology and the absence of capital letters also make NER harder. The authors build a NE-tagged corpus, WDC, from the 26 March 2013 Arabic Wikipedia dump with almost no manual effort.

First, articles are sorted into entity types by a classifier trained on a small hand-labelled set. Links in article text then inherit the type of the page they point to. Unlinked mentions are found by matching a list of alternative names built from redirects, anchor texts and inter-language links. These surface forms already cover inflected variants, so no morphological analyser or POS tagger is needed. A capitalisation-probability measure removes ambiguous name phrases, and noisy sentences are dropped.

A CRF tagger trained on the resulting corpus of about 6 million tokens beats taggers trained on earlier Wikipedia-derived Arabic corpora. It also holds up well against a model trained on manually annotated newswire, especially on open-domain text and tweets. This shows that cheap, structure-derived silver data can substitute for expensive gold annotation.

Key points

  • The article classifier was trained on 4,000 manually labelled Arabic Wikipedia articles.
    • These use the ACE (2008) taxonomy plus a Product class, eight coarse types in all, mapped to CoNLL categories.
    • Features are TF-IDF over filtered tokens, with conjunctions, prepositions and pronouns split off.
  • The tagger is a CRF (CRFsuite) using:
    • neighbouring words in raw and stemmed form;
    • 1–4 character prefixes and suffixes;
    • a light stem;
    • a gazetteer of about 5,000 entries from Benajiba et al. (2008).
  • WDC beats WikiFANE(whole) and WikiFANE(selective) on ANERcorp by 8–12 F-measure points, with the gain coming mainly from recall.
  • Against the ANERcorp-Model, the WDC-Model:
    • wins by roughly 48% on a held-out Wikipedia slice of about 14k tokens;
    • loses by about 13 points on the ANERcorp test, where 69% of test entities were seen in training;
    • performs comparably on NEWS (about 15k tokens);
    • wins on TWEETS (1,423 tweets, about 26k tokens).
  • Merging WDC with ANERcorp gives scores between the two single-corpus models on ANERcorp and NEWS, with no significant gain on TWEETS.

Interesting ideas

  • I1. Redirect titles and link labels serve as a ready-made lexicon of surface variants for each entity, which sidesteps morphological analysis when spotting mentions
  • I2. A link label automatically receives the type of the page it points to, turning existing hyperlinks into free supervision
  • I3. A scored filter (here, how name-like a phrase is) discards ambiguous candidate names before annotation, trading some coverage for precision
  • I4. The authors explain the gap between in-domain and cross-domain results by measuring how many test entities were already seen in training (69%)
  • I5. Much of the cross-domain error comes from differing annotation conventions (e.g., whether a head word like "Prefecture" belongs to the name) rather than from model quality

For Mova

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

  • I1 (world): Mova could harvest Wikipedia/Wikidata aliases into the entity layer of its knowledge graph, so that variant names in a text resolve to one world entity without hand-written alias rules.
  • I2 (learn): link-typed Wikipedia sentences could act as an extra witness when building silver entity annotations, to be checked against the witness parsers' outputs.
  • I3 (learn): silver data could carry a per-alias ambiguity score, so that common-word aliases are dropped instead of polluting training.
  • I4 (learn): reporting the overlap of seen entities or lemmas next to each test score would help separate real generalisation from memorisation, much like a negative control.
  • I5 (tree): when converting between treebank dialects or comparing parsers, conventions for multiword names should be normalised before scoring, so that convention mismatches are not counted as parse errors.

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 EssexColchester · United Kingdom34 works · 3,537 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 Maha Althobaiti, Udo Kruschwitz and Massimo Poesio. 2014. Automatic Creation of Arabic Named Entity Annotated Corpus Using Wikipedia. Proceedings of the Student Research Workshop at the 14th Conference of the European Chapter of the Association for Computational Linguistics.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.3115/v1/e14-3012

@inproceedings{althobaiti2014automatic,
  title = {{Automatic Creation of Arabic Named Entity Annotated Corpus Using Wikipedia}},
  author = {Maha Althobaiti, Udo Kruschwitz and Massimo Poesio},
  booktitle = {Proceedings of the Student Research Workshop at the 14th Conference of the European Chapter of the Association for Computational Linguistics},
  year = {2014},
  doi = {10.3115/v1/e14-3012},
  url = {https://doi.org/10.3115/v1/e14-3012}
}

References

22 references, 9 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. 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. Association for Computational Linguistics.
  2. Fahd Alotaibi and Mark Lee. 2012. Mapping Arabic Wikipedia into the Named Entities Taxonomy. In COLING (Posters), pages 43–52.
  3. Fahd Alotaibi and Mark Lee. 2013. Automatically Developing a Fine-grained Arabic Named Entity Corpus and Gazetteer by utilizing Wikipedia. In IJC-NLP.
  4. Maha Althobaiti, Udo Kruschwitz, and Massimo Poesio. 2013. A Semi-supervised Learning Approach to Arabic Named Entity Recognition. In RANLP, pages 32–40.
  5. Joohui An, Seungwoo Lee, and Gary Geunbae Lee. 2003. Automatic acquisition of named entity tagged corpus from world wide web. In Proceedings of the 41st Annual Meeting on Association for Computational Linguistics-Volume 2, pages 165–168. Association for Computational Linguistics. find
  6. Yassine Benajiba, Paolo Rosso, and José Miguel Benedı́ruiz. 2007. Anersys: An Arabic Named Entity Recognition System based on Maximum Entropy. In Computational Linguistics and Intelligent Text Processing, pages 143–153. Springer. find
  7. Yassine Benajiba, Mona Diab, and Paolo Rosso. 2008. Arabic Named Entity Recognition using optimized feature sets. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, pages 284–293. Association for Computational Linguistics.
  8. Daniel M Bikel, Scott Miller, Richard Schwartz, and Ralph Weischedel. 1997. Nymble: a high-performance learning name-finder. In Proceedings of the fifth conference on Applied natural language processing, pages 194–201. Association for Computational Linguistics. ↗
  9. Nancy Chinchor, Erica Brown, Lisa Ferro, and Patty Robinson. 1999. 1999 Named Entity Recognition Task Definition. MITRE and SAIC. find
  10. Wisam Dakka and Silviu Cucerzan. 2008. Augmenting Wikipedia with Named Entity Tags. In IJCNLP, pages 545–552. find
  11. Kareem Darwish. 2013. Named Entity Recognition using Cross-lingual Resources: Arabic as an Example. In ACL.
  12. Junichi Kazama and Kentaro Torisawa. 2007. Exploiting Wikipedia as external knowledge for named entity recognition. In Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 698– 707.
  13. L.S. Larkey, L. Ballesteros, and M.E. Connell. 2002. Improving stemming for Arabic information retrieval: light stemming and co-occurrence analysis. In Annual ACM Conference on Research and Development in Information Retrieval: Proceedings of the 25 th annual international ACM SIGIR conference on Research and development in information retrieval, volume 11, pages 275–282. find
  14. Andrew McCallum and Wei Li. 2003. Early results for named entity recognition with conditional random fields, feature induction and web-enhanced lexicons. In Proceedings of the seventh conference on Natural language learning at HLT-NAACL 2003-Volume 4, pages 188–191. Association for Computational Linguistics.
  15. Peter Mika, Massimiliano Ciaramita, Hugo Zaragoza, and Jordi Atserias. 2008. Learning to Tag and Tagging to Learn: A Case Study on Wikipedia. volume 23, pages 26–33. find
  16. Joel Nothman, James R Curran, and Tara Murphy. 2008. Transforming Wikipedia into Named Entity training data. In Proceedings of the Australian Language Technology Workshop, pages 124–132. find
  17. Joel Nothman, Nicky Ringland, Will Radford, Tara Murphy, and James R Curran. 2013. Learning multilingual Named Entity Recognition from Wikipedia. Artificial Intelligence, 194:151–175. ↗
  18. Alexander E Richman and Patrick Schone. 2008. Mining Wiki Resources for Multilingual Named Entity Recognition. In ACL, pages 1–9. find
  19. Nicky Ringland, Joel Nothman, Tara Murphy, and James R Curran. 2009. Classifying articles in English and German Wikipedia. In Australasian Language Technology Association Workshop 2009, page 20. find
  20. Satoshi Sekine et al. 1998. NYU: Description of the Japanese NE system used for MET-2. In Proc. of the Seventh Message Understanding Conference (MUC-7), volume 17. find
  21. Sam Tardif, James R. Curran, and Tara Murphy. 2009. Improved Text Categorisation for Wikipedia Named Entities. In Proceedings of the Australasian Language Technology Association Workshop, pages 104–108. find
  22. Torsten Zesch, Iryna Gurevych, and Max Mühlhäuser. 2007. Analyzing and accessing Wikipedia as a lexical semantic resource. pages 197–205. Tuebingen, Germany: Gunter Narr, Tübingen. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/althobaiti-2014-automatic-creation-arabic-named-entity/