Elamur Research⌕
☰
home / papers / 2021 / A description and demonstration of SAFAR framework

A description and demonstration of SAFAR framework

Paper · 2021 · Karim Bouzoubaa, Younes Jaafar, Driss Namly et al.

Summary

Arabic NLP has many separate tools and resources, but they are rarely available in one consistent, extensible environment. Combining them raises software-engineering problems such as interoperability, reuse and portability. The authors present SAFAR, a Java framework for Modern Standard Arabic (MSA) and Moroccan dialect, built over roughly ten years. It has three layers: Tools for technical, pre-processing and ML utilities; Basic for morphology, syntax and semantics; and Applications. Each processing level is defined by a standard API. Behind each API sit existing third-party implementations or in-house ones written when nothing suitable existed. Inputs and outputs use XML, with LMF for lexicons, TEI for corpora and ALECSO guidelines for morphological analyzers. SAFAR now holds more than 50 tools and resources, including a very large stem–lemma lexicon, an enriched Arabic WordNet and Moroccan dialect lexicons and corpora. It can be used through the Java API or a web interface aimed at linguists. Because several implementations share one API, they can be compared directly, which gives the Arabic community a common base for building and benchmarking components.

Key points

  • SAFAR is organized into three content packages (MSA, Dialects, Machine learning models) and offers more than 50 tools and resources. Integrated third-party components include Farasa, Stanford parser, MADAMIRA, BAMA, Alkhalil, and the Khoja, ISRI and Light10 stemmers.
  • The MSA lexicon contains about 7.13 million stems with their lemmas. The authors' enriched Arabic WordNet became the official version for the Global WordNet association.
  • Moroccan dialect resources include:
    • MDED, with roughly 12,000 annotated entries;
    • MRV, with 4.5M normalized words;
    • an orthographic-variant lexicon of about 2.385M entries covering around 54% of MRV;
    • a language-identification corpus of 57k social-media comments labelled MSA, MD or code-switched.
  • Built-in benchmarks compare the integrated stemmers, morphological analyzers and parsers. The ML side uses Weka (a decision-tree POS tagger), an HMM (the lemmatizer) and fastText (dialect spelling normalization).
  • In the API demo, a pipeline runs normalization, stop-word removal, tokenization, stemming and sentiment lookup. Switching from Light10 to Khoja changes only one line of code.

Interesting ideas

  • I1. Each processing level is a fixed interface with several interchangeable backends, so comparing implementations comes almost for free.
  • I2. Lexical resources mirror how words are built: a lemma plus affixes gives a stem, and a stem plus clitics gives a surface word. A huge stem→lemma table is precomputed rather than derived at runtime.
  • I3. The authors stop adding new implementations to a layer once it has matured, but leave the framework open for others to plug in more.
  • I4. Dialect language identification is offered in two forms: a cheap rule-based method using stop-word frequency, and an SVM classifier.
  • I5. Non-standard dialect spellings are mapped to a canonical form word by word, without context, using a learned variant lexicon.

For Mova

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

  • I1 (tree): Put the perceptron parser and gparse behind one parser trait, then run every treebank evaluation as a side-by-side comparison through the same harness.
  • I2 (tokens): Precompute a full inflected-form→lemma/features table for English in crate en, so lemma lookup is an explainable table hit rather than heuristic suffix stripping.
  • I3 (learn): Cap the number of witness parsers for silver data once agreement gains flatten, so effort shifts to weaker stages.
  • I4 (learn): Pair every learned component with a simple rule-based counterpart as a baseline or negative control, so it is clear what the model actually adds.
  • I5 (tokens): Normalize informal or variant English spellings (e.g., in tales or student answers) before tagging, and record the mapping so the step stays explainable.

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

🎓Mohammed V UniversityRabat · Morocco8 works · 103 citations🎓Ecole Mohammadia d'IngénieursRabat · Morocco4 works · 32 citations🎓Cadi Ayyad UniversityMarrakesh · Morocco3 works · 13 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 1Research subjects and languages the work is about.Semitic languages · 2,908 works

Cite

Cite as Karim Bouzoubaa, Younes Jaafar, Driss Namly, Ridouane Tachicart, Rachida Tajmout, Hakima Khamar, Hamid Jaafar, Lhoussain Aouragh and Abdellah Yousfi. 2021. A description and demonstration of SAFAR framework. Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: System Demonstrations.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2021.eacl-demos.16

@inproceedings{bouzoubaa2021description,
  title = {{A description and demonstration of SAFAR framework}},
  author = {Karim Bouzoubaa, Younes Jaafar, Driss Namly, Ridouane Tachicart, Rachida Tajmout, Hakima Khamar, Hamid Jaafar, Lhoussain Aouragh and Abdellah Yousfi},
  booktitle = {Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: System Demonstrations},
  year = {2021},
  doi = {10.18653/v1/2021.eacl-demos.16},
  url = {https://doi.org/10.18653/v1/2021.eacl-demos.16}
}

References

25 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Abdelali, A., Darwish, K., Durrani, N., and Mubarak, H. 2016. Farasa: A fast and furious segmenter for Arabic. In Proceedings of the 2016 conference of the North American chapter of the association for computational linguistics: Demonstrations, pp. 11-16.
  2. Abouenour, L., Bouzoubaa, K., and Rosso, P. 2013. On the evaluation and improvement of Arabic WordNet coverage and usability. Language Resources and Evaluation, vol. 47, n° 13, pp. 891-917. http://arabic.emi.ac.ma:8080/SW_V3/6 132 ↗
  3. Algasaier, H. The ISRI Arabic Stemmer. 2005. http://www.nltk.org/_modules/nltk/stem/isri.html (accessed February 1, 2015). ↗
  4. Althobaiti, M., Kruschwitz, U., and Poesio, M. 2014. AraNLP: a Java-Based Library for the Processing of Arabic Text. In Proceedings of the 9th Language Resources and Evaluation Conference (LREC’14), Reykjavik, Iceland.
  5. Boudchiche, M., Mazroui, A., Bebah, M. O. A. O., Lakhouaja, A., and Boudlal, A. 2017. AlKhalil Morpho Sys 2: A robust Arabic morphosyntactic analyzer. Journal of King Saud University-Computer and Information Sciences 29, no. 2: 141-146. find
  6. Boudlal, A., Lakhouaja, A., Mazroui, A., Meziane, A., Bebah, M. O. A. O., and Shoul, M. 2010. Alkhalil Morpho Sys1: A Morphosyntactic analysis System for Arabic texts. Proceedings of the 11th International Arab Conference on Information Technology (ACIT’10). Benghazi. 1-6. find
  7. Buckwalter, T. 2002. Buckwalter Arabic Morphological Analyzer Version 1.0." Linguistic Data Consortium. find
  8. Green, S., and Manning, C. D. 2010. Better Arabic parsing: baselines, evaluations, and analysis. The 23rd International Conference on Computational Linguistics (COLING '10). Beijing: Association for Computational Linguistics. 394-402.
  9. Jaafar, Y. and Bouzoubaa, K. 2014. Benchmark of Arabic morphological analyzers: Challenges and Solutions. 9th International Conference on Intelligent Systems: Theories and Applications (SITA'14), Rabat, Jaafar, Y. and Bouzoubaa, K. 2015. Arabic Natural Language Processing from Software Engineering to Complex Pipelines Cicling Cairo, Egypt. find
  10. Jaafar, Y., Namly, D., Bouzoubaa, K., Yousfi, A. 2016. Enhancing Arabic Stemming Process Using Resources and Benchmarking Tool. King Saud University - Computer and Information Sciences (JKSU-CIS). find
  11. Jaafar, Y., and Bouzoubaa, K. 2017. A New Tool for Benchmarking and Assessing Arabic Syntactic Parsers. 6th International Conference on Arabic Language Processing CITALA 2017, Fes, Morocco Fes, Morocco. find
  12. Jaafar, Y., Nasri, M., and Bouzoubaa, K. 2018. Semantic Analysis of Arabic Texts within SAFAR Framework. In proceedings of the 5th International IEEE Congress on Information Science and Technology (CIST'18), Marrakech, Morocco. find
  13. Jaafar, Y., and Bouzoubaa, K. (2018). A Survey and Comparative Study of Arabic NLP Architectures. In: Shaalan K., Hassanien A., and Tolba F. 2018. (eds) Intelligent Natural Language Processing: Trends and Applications. Studies in Computational Intelligence, volume 740. Springer, Cham. Khoja, S. Khoja stemmer. 2002. http://zeus.cs.pacificu.edu/shereen/research.htm#ste mming (accessed February 1, 2015). ↗
  14. Larkey, L. S., Ballesteros, L., and Connell, M. E. 2007. Light Stemming for Arabic Information Retrieval. In Arabic computational morphology: knowledge-based and empirical methods, 221-243. Springer Netherlands. ↗
  15. Lou B., and Syd, B. 2008. TEI P5: Guidelines for electronic text encoding and interchange". TEI Consortium. Loukili,T., and Bouzoubaa, K. 2011. Structuration et Standardisation des ressources linguistiques de l'Arabe - cas de l'alphabet, préfixes et suffixes, Journées Doctorales en Technologies de l'Information et Communication, Tangier, Morocco, 7/ 2011. find
  16. Saad, M. K., and Ashour, W. M. 2010. Arabic morphological tools for text mining. 6th International Conference on Electrical and Computer Systems (EECS’10). Lefke, North Cyprus. find
  17. Namly, D., Bouzoubaa, K., Tajmout, R., Tahir, Y., and Khamar, H. 2015. A Complex Arabic stop-words list design. The Second National Doctoral Symposium On Arabic Language Engineering (JDILA'2015) ENSA of Fez USMBA. find
  18. Namly, D., Regragui, Y., and Bouzoubaa, K. 2016. Interoperable Arabic language resources building and exploitation in SAFAR platform. In Proceeding of the 13th ACS/IEEE International Conference on Computer Systems and Applications (AICCSA’16), Agadir, Morocco. find
  19. Namly, D., Bouzoubaa, K., El Jihad, A., and Aouragh, S. L. (2020). Improving Arabic Lemmatization Through a Lemmas Database and a Machine-Learning Technique. In Recent Advances in NLP: The Case of Arabic Language, pp. 81-100. Springer, Cham. ↗
  20. Pasha, A., Al-Badrashiny, M., Diab, M., El Kholy, A., Eskander, R., Habash, N., Pooleery, M., Rambow, O., and Roth R. M. 2014. MADAMIRA: A Fast, Comprehensive Tool for Morphological Analysis and Disambiguation of Arabic. In Proceedings of the 9th Language Resources and Evaluation Conference (LREC’14), Reykjavik, Iceland.
  21. Gülşen, E. 2014. ITU Turkish NLP Web Service. in European Chapter of the Association for Computational Linguistics, Sweden Tachicart, R., Bouzoubaa, K., and Jaafar, H. 2014. Building a Moroccan dialect electronic Dictionary (MDED). In Proceedings of the 5th International Conference on Arabic Language Processing (CITALA'14), Oujda, Morocco. 7 133 find
  22. Tachicart, R., and Bouzoubaa, K. 2019. Towards Automatic Normalization of the Moroccan Dialectal Arabic User Generated Text. In Arabic Language Processing: From Theory to Practice, Springer International Publishing, 2019, pp. 264-275. find
  23. Tachicart, R., Bouzoubaa, K., Aouragh, S. L., and Jaafar, H. 2018. Automatic Identification of Moroccan Colloquial Arabic. Arabic Language Processing: From Theory to Practice, Springer International Publishing, Cham, vol. 782, pp. 201-214. find
  24. Tnaji, K., Bouzoubaa, K., and Aouragh, S.L. 2021, A light Arabic POS Tagger using a hybrid approach. In the international conference on digital technologies and applications, January 29-30, 2021. Zerrouki, T. Tashaphyne 0.2. 2012. https://pypi.python.org/pypi/Tashaphyne. Retrieved April 14, 2016. ↗
  25. Zhang, Y., Li, C., Barzilay, R., and Darwish, K. 2015. Randomized greedy inference for joint segmentation, POS tagging and dependency parsing. In Proceedings of the 2015 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp. 42-52. 8 134 find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/bouzoubaa-2021-description-demonstration-safar-framework/