Elamur Research⌕
☰
home / papers / 2016 / Système hybride pour la reconnaissance des entités nommées arabes à base des CRF (Hybrid Arabic NER system using CRF Model)

Système hybride pour la reconnaissance des entités nommées arabes à base des CRF (Hybrid Arabic NER system using CRF Model)

Paper · 2016 · Emna Hkiri, Souheyl Mallat and Mounir Zrigui

Summary

The paper tackles named entity recognition (persons, locations, organisations) in Arabic. The task is hard because Arabic has no capitalisation, rich morphology, usually unwritten short vowels, and few resources. The authors build a hybrid system with two parts. The first is a rule-based annotator that reproduces GATE's ANNIE pipeline (JAPE rules plus gazetteers). Its gazetteers are greatly enlarged with a bilingual Arabic–English NE lexicon taken from DBpedia linked data. The second part is a CRF++ sequence labeller trained on about 15,000 United Nations sentences, which the rule module annotated automatically, together with ANERcorp. At test time, entities found by the rules are kept as given, and the CRF labels only the remaining text. The CRF draws on rule-derived context features, morphological features, POS tags, gazetteer lookups and punctuation cues. On ANERcorp the system's overall F-measure is slightly above two earlier systems. It was also tested on a manually annotated News Commentary sample. The work shows that cheap linked-data lexicons plus rule-bootstrapped training data can partly make up for scarce annotated resources.

Key points

  • The bilingual NE lexicon built from DBpedia linked datasets has 48,753 Arabic–English pairs: 27,480 persons, 17,237 organisations and 4,036 locations. Its Arabic side was mapped onto ANNIE's much smaller predefined Arabic gazetteers.
  • The CRF++ model was trained on roughly 15,000 sentences from the UN corpus (2005), annotated by the rule module, plus ANERcorp (4,871 sentences, over 150k word tokens). ANERcorp was converted from CoNLL format to XML.
  • The CRF features are:
    • rule-module decisions over a 5-word window
    • verb aspect, mood, voice, number, gender and person
    • proclitic presence (e.g. Fa, Wa, Bi, Li, Sa)
    • POS from the SAPA tool
    • neighbour-word gazetteer membership
    • adjacent-period cues
  • On ANERcorp the overall F-measure is 81.73, against 81.23 for NERAr (Gahbiche et al., 2013) and 76.28 for Benajiba et al. Location F is 89.58, and organisations remain the weakest class (P 85.64 / R 62.50 / F 72.26).
  • A second test set comes from 600 News Commentary 2012 sentences, hand-annotated with 350 persons, 410 locations and 151 organisations. Scoring used GATE's Corpus Quality Assurance plugin.

Interesting ideas

  • I1. Rules first, learner second: high-precision rule output is frozen as correct, and the statistical model only labels spans the rules left untouched, so each component handles what it does best.
  • I2. Rule decisions as learner features: the rule annotator's verdicts in a window around each word become input features, so the model learns when to trust or override the rules.
  • I3. Bootstrapping training data from a rule system: a large unannotated corpus (UN texts) is labelled by rules and mixed with a smaller gold corpus to train the statistical model.
  • I4. Gazetteers grown from linked data: entity lists are filled cheaply by mapping DBpedia entries, with their bilingual pairs, onto existing gazetteer categories instead of building them by hand.
  • I5. Using hybrid output to refine the rules: the authors plan to analyse where the CRF corrects or extends the rule output, so that rule improvements can be automated.

For Mova

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

  • I1 (tokens): A CRF/perceptron tagger in crate en could run after deterministic rules (dates, numbers, known names). Only unresolved tokens would get statistical labels, and every frozen label stays traceable to its rule.
  • I2 (tree): Outputs of rule-based dialect-conversion or attachment heuristics could be fed as features into the perceptron parser. This keeps rule knowledge inside the model and makes it inspectable.
  • I3 (learn): This is a minimal form of silver data. Mova could compare rule-only silver with multi-witness-parser silver, keeping the gold portion as a check against learning rule errors.
  • I4 (world): The entity ontology could import typed names from Wikidata/DBpedia into person/place/organisation classes. Keeping the cross-language pairs would also help translation through the meta-world.
  • I5 (learn): Cases where the learned component consistently disagrees with a rule could seed candidate self-induced repair rules. These would then be validated against negative controls.

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 · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksComputer Vision and Pattern Recognition · 3,486 works

🧭 Topics 2Research subjects and languages the work is about.Semitic languages · 3,066 worksNamed entity recognition · 214 works

Cite

Cite as Emna Hkiri, Souheyl Mallat and Mounir Zrigui. 2016. Système hybride pour la reconnaissance des entités nommées arabes à base des CRF (Hybrid Arabic NER system using CRF Model). Actes de la conférence conjointe JEP-TALN-RECITAL 2016. volume 2 : TALN (Posters).

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

@inproceedings{hkiri2016systeme,
  title = {{Système hybride pour la reconnaissance des entités nommées arabes à base des CRF (Hybrid Arabic NER system using CRF Model)}},
  author = {Emna Hkiri, Souheyl Mallat and Mounir Zrigui},
  booktitle = {Actes de la conférence conjointe JEP-TALN-RECITAL 2016. volume 2 : TALN (Posters)},
  year = {2016},
  url = {https://aclanthology.org/2016.jeptalnrecital-poster.29/}
}

References

22 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. ABDELRAHMAN S., ELARNAOTY M., MAGDY M., FAHMY A. (2010). Integrated machine learning techniques for Arabic named entity recognition. International Journal of Computer Science Issues (IJCSI), 7(4):27–368. find
  2. ABDUL-HAMID A, DARWISH K. (2010). Simplified feature set for Arabic named entity recognition. In Proceedings of the 2010 Named Entities Workshop (NEWS 2010), pages 110–115, Stroudsburg, PA.
  3. AL-JUMAILY H., PALOMA M., ERIK G. (2012). A real time named entity recognition system for Arabic text mining. Language Resources and Evaluation Journal, 46(4):543–563. find
  4. AL-SHALABI R., GHASSAN K., BASHAR A., KAIGHONMEIN K., SALEM A. (2009). Proper noun extracting algorithm for the Arabic language. In International Conference on IT to Celebrate S. Charmonman’s 72nd Birthday, pages 28.1–28.9, Bangkok. find
  5. BENAJIBA Y., DIAB M., ROSSO P. (2008). Arabic named entity recognition: An SVM-based approach. In Proceedings of Arab International Conference on Information Technology (ACIT 2008), pages 16–18, Hammamet BENAJIBA Y., ROSSO P., BENEDÍ J.-M. (2007). ANERsys : An Arabic Named Entity Recognition System Based on Maximum Entropy , CICLing, p. 143-153,. ↗
  6. DAILLE B., FOUROUR N., MORIN E. (2000). Catégorisation des noms propres : une étude en corpus. Cahiers de Grammaire, 25, 115–12 find
  7. ELSEBAI A., FARID M. (2011). Extracting persons names from Arabic newspapers. In Proceedings of the International Conference on Innovations in Information Technology, pages 87– 89, Dubai. find
  8. GAHBICHE S., MAYNARD H., YVON F. (2013). traitement automatique des entités nommées en arabe : détection et traduction. TAL 54(2): 101-132 find
  9. KOULALI M., ABDELOUAFI M. (2012). A contribution to Arabic named entity recognition. In Proceedings of the 10th International Conference on ICT and Knowledge Engineering, pages 46– 52, Morocco. find
  10. KÜÇÜK, D., YAZICI A. (2012). A hybrid named entity recognizer for Turkish. Expert Systems with Applications. 39, 2733-2742. ↗
  11. LAFFERTY J.D., MCCALLUM A., PEREIRA F. (2001). Conditional Random Fields : Probabilistic Models for Segmenting and Labeling Sequence Data. In ICML ’01 : Proceedings of the Eighteenth International Conference on Machine Learning, pages 282–289, San Francisco, CA, USA, 2001. Morgan Kaufmann Publishers Inc. find
  12. MESFAR S. (2007). Named entity recognition for Arabic using syntactic grammars. In Proceedings of the 12th International Conference on Applications of Natural Language to Information Systems (NLDB’07), pages 305–316, Berlin. find
  13. OUDAH M., SHAALAN K. (2012). A pipeline Arabic named entity recognition using a hybrid approach. In Proceedings of the International Conference on Computational Linguistics, pages 2,159–2,176, Mumbai.
  14. PETASIS G., FRANTZ VT., FRANCIS W., GEORGIOS PS., VANGELIS K., CONSTANTINE D.(2001). Using machine learning to maintain rule-based named-entity recognition and classification systems. In Proceedings of the 39th AnnualMeeting of the Association for Computational Linguistics, pages 426–433, Stroudsburg, PA. find
  15. SEON C., KO Y., KIM J., SEO J. (2001). Named Entity Recognition using Machine Learning Methods and Pattern-Selection Rules. In Proceedings of the 6th Natural Language Processing Pacific Rim Symposium. 229-236. find
  16. SHAALAN K., HAFSA R. (2008). Arabic named entity recognition from diverse text types. In Bengt Nordstr ¨om and Aarne Ranta, editors, Advances in Natural Language Processing, volume 5221 of Lecture Notes in Computer Science. Springer, Berlin Heidelberg, pages 440–451. find
  17. TRAN T., PHAM HQ., NGO D., COLLIER N. (2007). Named entity recognition in vietnamese documents. Progress in Informatics, 4:5-13. find
  18. TSAI T., WU S., LEE C., SHIH C., HSU W. (2004). Mencius: A Chinese Named Entity recognizer Using the Maximum Entropy-based Hybrid Model. Computational Linguistics and Chinese Language Processing. 9, 65-82. ↗
  19. TUERLINCKX T. (2004). Lemmatisation de l’arabe non classique ", JADT 2004 : 7es Journées internationales d’Analyse statistique des Données Textuelles, 2004. W LI. find
  20. MCCALLUM A. (2003). Rapid development of hindi named entity recognition using conditional random fields and feature induction. Special Issue of ACM Transactions on Asian Language Information Processing: Rapid Development of Language Capabilities: The Surprise Languages,. find
  21. ZAGHOUANI W. (2012). RENAR: A rule-based Arabic named entity recognition system. ACM Transactions on Asian Language Information Processing (TALIP), 11(1):2:1–2:13. find
  22. ZHOU J., LIANG D., JIAJUN C. (2006). Chinese named entity recognition with a multi-phase model. In Proc. of the Fifth SIGHAN Workshop on Chinese Language Processing, pages 213-216, Sydney, Australia, 2006. Association for Computational Linguistics. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/hkiri-2016-systeme-hybride-pour-la-reconnaissance/