Elamur Research⌕
☰
home / papers / 2015 / Arib@QALB-2015 Shared Task: A Hybrid Cascade Model for Arabic Spelling Error Detection and Correction

Arib@QALB-2015 Shared Task: A Hybrid Cascade Model for Arabic Spelling Error Detection and Correction

Paper · 2015 · Nouf AlShenaifi, Rehab AlNefie, Maha Al-Yahya et al.

Summary

Arib is a system for detecting and correcting Arabic spelling errors, built for the QALB-2015 shared task. The task covers comments by native speakers (Aljazeera data from QALB-2014) and texts by learners of Arabic (L2). The authors first assembled a large frequency dictionary. They took the QALB corpus, applied its gold corrections to get clean text, and merged it with several other corpora. Word counts were then extracted with the KHAWAS tool. The system runs text through a fixed chain of correctors, each targeting different error types:

  • MADAMIRA's own corrections
  • hand-written rules for common error patterns
  • a Norvig-style probabilistic corrector
  • a Symmetric Delete (FAROO) edit-distance corrector
  • the open-source Ghalatawi tool
  • a rule-based module that restores punctuation

Every added stage raised F1. The full chain reached precision 0.67, recall 0.51 and F1 0.58 on development data. On the native-speaker 2015 test set the best run scored F1 60.23, but learner text was much harder, at about 31–33 F1. The work shows that simple, transparent components chained in order give competitive results.

Key points

  • Dictionary sources: QALB corrected text, KSUCCA (classical Arabic), OSAC, the Al-Sulaiti corpus and the KACST Arabic Corpus. Word frequencies were extracted with KHAWAS.

  • Component order: MADAMIRA → Rule-Based → Probabilistic (Norvig/Bayes) → Distance (FAROO, Damerau-Levenshtein ≤2) → Ghalatawi → Punctuation Recovery.

  • Ablation (F1):

    ConfigurationF1
    MADAMIRA alone0.4722
    + rules0.5178
    + probabilistic0.5305
    + distance0.5768
    + Ghalatawi0.5781

    Recall grew from 0.37 to 0.51, while precision stayed near 0.66.

  • Official test results (M2 Scorer):

    • Alj-test-2015: Arib-1 reached P 64.50, R 56.50, F 60.23.
    • L2-test-2015: Arib-3 reached F 32.68, with recall only about 22–25.

  • Examples of the rules:

    • collapse runs of three or more repeated letters
    • split numbers away from attached words
    • split off a Ta-Marbouta word or the prepositions ElY/AlY when glued to the next word
    • attach a lone W/F conjunction to the following word
    • replace Latin punctuation with Arabic punctuation

Interesting ideas

  • I1. The lexicon was harvested by applying a corpus's gold corrections to its noisy text, which turns error-annotation data into a clean in-domain word list with frequencies.
  • I2. The correctors were added one at a time, with precision, recall and F1 recorded at each step, so each module's contribution is visible and modules that add nothing can be spotted.
  • I3. Symmetric Delete only generates deletions, for both dictionary words and the input, and then looks for matches, which makes edit-distance lookup fast without a full enumeration of candidate edits.
  • I4. The punctuation module keeps every existing mark and only inserts missing ones, a conservative policy that protects precision.
  • I5. The rule set combined patterns mined from data with an expert's summary of typical learner mistakes, a cheap way to cover a population whose errors differ from native writers'.

For Mova

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

  • I1 (tokens): Corrected learner or social-media corpora could be mined the same way, giving the en lemma lexicon frequency priors for informal spellings.
  • I2 (learn): Adopt this cumulative-ablation table as standard practice when new repair rules or witness parsers are added, so each one has to prove its gain.
  • I3 (tokens): Use it as a quick normalizer for typos before tagging, with every proposed fix logged as an explicit, explainable edit.
  • I4 (tree): Any pre-parse text repair could follow the same insert-only rule, so original punctuation that the parser uses for clause boundaries is never destroyed.
  • I5 (answer): Children's answers in school-test readers could get a small expert-written rule list for typical learner misspellings, applied before parsing.

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

🎓King Saud UniversityRiyadh · Saudi Arabia37 works · 527 citations

 Fields, topics and concepts

📚 Shelves 6The 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 worksSocial Sciences · 1,772 worksEducation · 203 works

🧭 Topics 1Research subjects and languages the work is about.Semitic languages · 2,908 works

Cite

Cite as Nouf AlShenaifi, Rehab AlNefie, Maha Al-Yahya and Hend Al-Khalifa. 2015. Arib@QALB-2015 Shared Task: A Hybrid Cascade Model for Arabic Spelling Error Detection and Correction. Proceedings of the Second Workshop on Arabic Natural Language Processing.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/w15-3214

@inproceedings{alshenaifi2015arib,
  title = {{Arib@QALB-2015 Shared Task: A Hybrid Cascade Model for Arabic Spelling Error Detection and Correction}},
  author = {Nouf AlShenaifi, Rehab AlNefie, Maha Al-Yahya and Hend Al-Khalifa},
  booktitle = {Proceedings of the Second Workshop on Arabic Natural Language Processing},
  year = {2015},
  doi = {10.18653/v1/w15-3214},
  url = {https://doi.org/10.18653/v1/w15-3214}
}

References

29 references, 11 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. B. Mohit, A. Rozovskaya, N. Habash, W. Zaghouani, and O. Obeid., “The First QALB Shared Task on Automatic Text Correction for Arabic,” in Proceedings of EMNLP Workshop on Arabic Natural Language Processing, Doha, Qatar, 2014.
  2. W. Zaghouani, B. Mohit, N. Habash, O. Obeid, N. Tomeh, A. Rozovskaya, N. Farra and S. Alkuhlani, and K. Oflazer., “Large Scale Arabic Error Annotation: Guidelines and Framework,” in Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC'14), 2014.
  3. W. Zaghouani, N. Habash, H. Bouamor, A. Rozovskaya, B. Mohit, A. Heider, and K. Oflazer., “Correction Annotation for Non-Native Arabic Texts: Guidelines and Corpus,” in Proceedings of the Ninth Linguistic Annotation Workshop, Association for Computational Linguistics, 2015, pp. 129–139.
  4. Al-Fedaghi, S. and Amin, A., “Automatic correction of spelling errors in Arabic,” J-Univ. KUWAIT Sci., vol. 19, no. 2, p. 175, 1992. find
  5. K. Shaalan, Amin Allam, and Abdullah Gomah, “Towards automatic spell checking for Arabic,” 2003. find
  6. Rachidi T., M. Bouzoubaa, L. ElMortaji, B. Boussouab, and A. Bensaid, “Arabic user search Query correction and expansion,” in Proc. of COPSTIC’03, Rabat December 11–13, 2003. find
  7. Mona Diab, Mohammed Attia, and Mohamed Al-Badrashiny, “GWU-HASP: Hybrid Arabic Spelling and Punctuation Corrector,” in EMNLP, Doha, Qatar, 2014.
  8. Attia, Mohammed, Pavel Pecina, and Younes Samih, “Improved Spelling Error Detection and Correction for Arabic,” in COLING, 2012.
  9. B. Mohit, “QALB: Qatar Arabic language bank,” Qatar Found. Annu. Res. Forum Proc., p. ICTP 032, Nov. 2013. find
  10. Muaidi H, “Extraction Of Arabic Word Roots: An Ap- proach Based on Computational Model and Multi-ackpropagation Neural Networks.,” PhD Thesis, De Mont- fort University, UK, 2008. find
  11. Y. Hassan, M. Aly, and A. Atiya, “Arabic Spelling Correction using Supervised Learning,” ArXiv14098309 Cs, Sep. 2014.
  12. Taha Zerrouki, Khaled Alhowaity, and Amar Balla, “Auto-correction of arabic common errors for large text corpus,” in EMNLP, Doha, Qatar, 2014. find
  13. H. Mubarak and K. Darwish, “Automatic Correction of Arabic Text: a Cascaded Approach,” ANLP 2014, p. 132, 2014.
  14. A. R. N. Habash and R. E. N. F. W. Salloum, “The Columbia System in the QALB-2014 Shared Task on Arabic Error Correction,” ANLP 2014, p. 160, 2014.
  15. Michael Nawar and Moheb Ragheb, “Fast and Robust Arabic Error Correction System,” in EMNLP, Doha, Qatar, 2014. ↗
  16. H. Muaidi and R. Al-Tarawneh, “Towards Arabic Spell-Checker Based on N-Grams Scores,” Int. J. Comput. Appl., vol. 53, no. 3, pp. 12–16, Sep. 2012. ↗
  17. B. Haddad and M. Yaseen, “Detection and Correction of Non-Words in Arabic: A Hybrid Approach,” Int. J. Comput. Process. Lang., vol. 20, no. 04, pp. 237–257, Dec. 2007. ↗
  18. Alrabiah, M., Al-Salman A. and Atwell, E., “The design and construction of the 50 million words KSUCCA King Saud University Corpus of Classical Arabic,” in Proceedings of the Second Workshop on Arabic Corpus Linguistics (WACL-2), Lancaster University, UK, 2013. find
  19. Saad, M. and Ashour, W., “OSAC: Open Source Arabic Corpora,” in Proceedings of the 6th International Symposium on Electrical and Electronics Engineering and Computer Science (EEECS’10), European University of Lefke, Cyprus, pp. 118-123, 2010. ↗
  20. Al-Sulaiti, Latifa; Atwell, Eric. “The design of a corpus of contemporary Arabic,” International Journal of Corpus Linguistics, vol. 11, pp. 135-171, 2006. ↗
  21. Al-Thubaity, A., “A 700M+ Arabic corpus: KACST Arabic corpus design and construction,” Language Resources and Evaluation, pp. 1-31, 2014. ↗
  22. “Ghawwas: An open source system for Arabic corpora processing.”[Online]. Available: http://sourceforge.net/projects/ghawwasv4/. [Accessed: 1-June-2015]. ↗
  23. A. Pasha, M. Al-Badrashiny, M. T. Diab, A. E. Kholy, R. Eskander, N. Habash, M. Pooleery, O. Rambow, and R. Roth, “MADAMIRA: A Fast, Comprehensive Tool for Morphological Analysis and Disambiguation of Arabic,” in Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC-2014), Reykjavik, Iceland, May 26-31, 2014, 2014, pp. 1094– 1101.
  24. “How to Write a Spelling Corrector.” [Online]. Available: http://norvig.com/spell-correct.html. [Accessed: 28-May-2015]. ↗
  25. T. Segaran and J. Hammerbacher, Beautiful Data: The Stories Behind Elegant Data Solutions. O’Reilly Media, Inc., 2009. find
  26. “1000x Faster Spelling Correction algorithm | FAROO Blog.” [Online]. Available: http://blog.faroo.com/2012/06/07/improved-edit-distance-based-spelling-correction/. [Accessed: 28-May-2015]. ↗
  27. “يﻱوﻭاﺍطﻁلﻝغﻍ:حﺡيﻱحﺡصﺹتﺕلﻝاﺍ يﻱئﺉاﺍقﻕلﻝتﺕلﻝاﺍ يﻱبﺏرﺭعﻉلﻝاﺍ Ghalatawi:Arabic AutoCorrect.” [Online]. Available: http://ghalatawi.sourceforge.net/. [Accessed: 28-May-2015]. ↗
  28. D. Daniel, and H. Ng., “Better evaluation for grammatical error correction,” in Proceedings of the 2012 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. Association for Computational Linguistics, 2012. find
  29. A. Rozovskaya, H. Bouamor N. Habash, W. Zaghouani, O. Obeid, and B. Mohit., “The Second QALB Shared Task on Automatic Text Correction for Arabic,” in Proceedings of ACL Workshop on Arabic Natural Language Processing, Beijing, China, 2015

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/alshenaifi-2015-arib-qalb-2015-shared-task/