Elamur Research⌕
☰
home / papers / 2015 / Arabic Named Entity Recognition Process using Transducer Cascade and Arabic Wikipedia

Arabic Named Entity Recognition Process using Transducer Cascade and Arabic Wikipedia

Paper · 2015 · Fatma Ben Mesmia, Kais Haddar, Denis Maurel et al.

Summary

The paper addresses recognition of Arabic person ("personality") names, a hard part of Arabic NER. Arabic has no capitalization, agglutinates clitics onto words, drops the Hamza inconsistently, and uses names that double as adjectives or time words. The authors take a rule-based route. They studied a 79,659-token Arabic Wikipedia corpus, exported with Kiwix and covering 19 Arabic countries, and catalogued the forms names take there. They then built first-name and last-name dictionaries, a set of trigger words, and 14 extraction rules. These were encoded as Unitex graphs and chained into a six-transducer cascade with CasSys, in a deliberately chosen order. The cascade ran on a separate 454,959-token Wikipedia test corpus and was checked by hand, reaching recall 0.98, precision 0.94 and F-measure 0.95. The authors report this as comparable to two earlier rule-based systems. The work shows that a free encyclopedic source plus finite-state rules can cover the many shapes of Arabic names, built from ism, kunya, nasab, laqab and nisba.

Key points

  • Corpora: the study corpus has 79,659 tokens and the test corpus 454,959 tokens. Both come from Arabic Wikipedia via Kiwix and cover texts from 19 Arabic countries.
  • Resources:
    • two new dictionaries, one for first names and one for last names, plus the proper-name dictionary of Doumi et al. (2013) available in Unitex;
    • 180 trigger words in eight classes: artistic, civility, military, nobiliary, political, profession, religious and sportive (for example, 47 artistic, 21 civility, 7 military and 22 nobiliary).
  • Rules and cascade: 14 extraction rules became six transducers, generated with CasSys inside the Unitex platform. Each transducer annotates text in "Merge" mode.
  • Cascade order:
    1. trigger-word names first, as the most certain cases;
    2. then first name plus last name;
    3. then Nasab plus Nisba;
    4. then Nasab plus last name;
    5. then exceptional cases;
    6. finally Nasab with an omitted "ben".
  • Evaluation was manual, with recall 0.98, precision 0.94 and F 0.95. Results were compared with Shaalan and Raza (2007) and Elsebai et al. (2009).

Interesting ideas

  • I1. Run the most reliable patterns, those anchored by trigger words, before weaker ones, so later stages work on text where the confident spans are already fixed.
  • I2. Group the cue words that signal an entity into typed classes (military, religious, sportive…), so the cue both finds the name and hints at what kind of person it is.
  • I3. Build reusable sub-graphs, such as a patronymic block, and call them from several higher-level patterns instead of duplicating the logic.
  • I4. Treat names as structured objects with typed parts (given name, patronymic, honorific, epithet, origin) rather than as flat strings.
  • I5. Explicitly model nesting, where a person name sits inside a place, organization or event name such as a stadium named after someone.

For Mova

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

  • I1 (world): Entity detection could resolve title-anchored mentions like "President X" or "Dr. Y" first and let looser name heuristics fill only the remaining gaps, recording which stage decided each entity.
  • I2 (world): Title and role words could feed entity attributes such as occupation or rank directly into the world model, giving an explainable source for each such attribute.
  • I3 (learn): Self-induced repair rules could be stored as composable named fragments, keeping the rule set small, inspectable and easier to test against negative controls.
  • I4 (world): Person entities could carry parsed name components, which would help coreference match "Khalid ibn Faisal" style variants or "Mr. Smith" against "John Smith".
  • I5 (frames): Coreference and frame building could keep the embedded person as a separate linked entity, rather than losing it inside the larger named span.

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 5The 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 worksComputer Vision and Pattern Recognition · 2,869 works

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

Cite

Cite as Fatma Ben Mesmia, Kais Haddar, Denis Maurel and Nathalie Friburger. 2015. Arabic Named Entity Recognition Process using Transducer Cascade and Arabic Wikipedia. Proceedings of the International Conference Recent Advances in Natural Language Processing.

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

@inproceedings{mesmia2015arabic,
  title = {{Arabic Named Entity Recognition Process using Transducer Cascade and Arabic Wikipedia}},
  author = {Fatma Ben Mesmia, Kais Haddar, Denis Maurel and Nathalie Friburger},
  booktitle = {Proceedings of the International Conference Recent Advances in Natural Language Processing},
  year = {2015},
  url = {https://aclanthology.org/R15-1007/}
}

References

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

  1. Aboaoga M. and Aziz MJA. 2013. Arabic person names recognition by using a rule based approach. Journal of Computer Science, 922–927. ↗
  2. Ciravegna F. and Lavelli A. 1999. « Full text parsing using cascades of rules: An information extraction perspective », In Proceedings of EACL'99, Bergen, Norway, 102-109. find
  3. Elsebai A., Meziane F. and BelKredim FZ. 2009. A rule based Persons names Arabic extraction system. Communications of the IBIMA, 11: 53–59. find
  4. Elsebai A. 2008. Arabic Proper Names Recognition Using Heuristics. Proceeding of the 9 th Annual Post Graduate Symposium on the Convergence of Telecommunications, Networking and Broadcasting (PGNET), ISBN: 978-1-902560-19-9. find
  5. Ben Mesmia F., Friburger N., Haddar K. and Maurel D. 2015. Transducer cascade for an automatic recognition of Arabic Named Entities in order to establish links to free resources. Will appear in IEEE-proceedings issue from CICLING’15. find
  6. Ben Mesmia F., Friburger N., Haddar K. and Maurel D. 2015. Construction d’une cascade de transducteurs pour la reconnaissance des dates à partir d’un corpus Wikipédia. Colloque pour les Étudiants Chercheurs en Traitement Automatique du Langage naturel et ses applications, 8-11. find
  7. Fehri H., Haddar K., Hamadou A. B. 2011. Recognition and Translation of Arabic Named Entities with NooJ Using a New Representation Model, in M. Constant, A. Maletti, A. Savary (eds), FSMNLP, 9th International Workshop, ACL, Blois, France, 134-142.
  8. Friburger N., Maurel D. 2004. Finite-state transducer cascades to extract named entities in texts, Theoretical Computer Science, volume 313, 94-104. find
  9. Doumi, N., Lehireche, A., Maurel, D., and Ali Cherif, M. (2013a). La conception d'un jeu de ressources libres pour le TAL arabe sous Unitex.Paper presented at the TRADETAL2013, Colloque international en Traductologie et TAL, Oran - Algeria, 5-6 may. find
  10. Maurel D., Friburger N., Eshkol I. 2009. « Who are you, you who speak? Transducer cascades for information retrieval ». In Proceedings of 4th Language & Technology Conference: Human Language Technologies as a Challenge for Computer Science and Linguistics, Poznan, Poland, 220-223. find
  11. Maurel D., Friburger N., Antoine J.-Y., Eshkol-Taravella I. and Nouvel D. 2011. Cascades de transducteurs autour de la reconnaissance des entités nommées. Traitement automatique des langues, 52(1) : 69-961. find
  12. Mesfar S. 2007. Named Entity Recognition for Arabic Using Syntactic Grammars. Proceedings of the 12th International Conference on Application of Natural Language to Information Systems. Berlin, Heidelberg, 305-316. find
  13. Shaalan K. and Raza H. 2007. Person name entity recognition for Arabic. In: Proceedings of the 5th workshop on important unresolved matters, 17-24.
  14. Shaalan K. and Oudah M. 2014. A hybrid approach to Arabic named entity recognition. Journal of Information Science, 40(1): 67–87. ↗
  15. Shaalan K. 2014. A Survey of Arabic Named Entity Recognition and Classification. Computational Linguistics, 40 (2) 469-510. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/mesmia-2015-arabic-named-entity-recognition-process/