---
title: "TECHLIMED system description for the Shared Task on"
description: "The paper describes Techlimed's entry in the QALB-2014 shared task on automatic error correction for… 13 references, cited by 2 papers here."
authors: ["Djamel Mostefa", "Omar Asbayou", "Ramzi Abbes"]
year: 2014
venue: "Proceedings of the EMNLP 2014 Workshop on Arabic Natural Language Processing (ANLP)"
source: "https://aclanthology.org/W14-3621/"
doi: "10.3115/v1/w14-3621"
licence: "CC BY-NC-SA 3.0 (ACL Anthology, before 2016)"
licence_url: "https://creativecommons.org/licenses/by-nc-sa/3.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/mostefa-2014-techlimed-system-description-shared-task/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/mostefa-2014-techlimed-system-description-shared-task/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2014](https://md.research.elamur.ai/free/en/papers/2014/page/4/) / TECHLIMED system description for the Shared Task on Automatic Arabic Error Correction

# TECHLIMED system description for the Shared Task on Automatic Arabic Error Correction

**Authors:** [Djamel Mostefa](https://md.research.elamur.ai/free/en/authors/djamel-mostefa/), [Omar Asbayou](https://md.research.elamur.ai/free/en/authors/omar-asbayou/) and [Ramzi Abbes](https://md.research.elamur.ai/free/en/authors/ramzi-abbes/) · **Year:** 2014 · **Venue:** [Proceedings of the EMNLP 2014 Workshop on Arabic Natural Language Processing (ANLP)](https://md.research.elamur.ai/free/en/venues/emnlp-workshop-on-arabic-natural-language-processing-anlp/) ([ANLP 2014](https://md.research.elamur.ai/free/en/venues/emnlp-workshop-on-arabic-natural-language-processing-anlp/2014/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY-NC-SA 3.0 (ACL Anthology, before 2016)](https://creativecommons.org/licenses/by-nc-sa/3.0/)

**Links:** [Publisher's page](https://aclanthology.org/W14-3621/) · [PDF](https://aclanthology.org/W14-3621.pdf) · [DOI 10.3115/v1/w14-3621](https://doi.org/10.3115/v1/w14-3621) · [OpenAlex](https://openalex.org/W2251943172)

**Citations:** 6 citations (OpenAlex, 2026-10-14) · [2](https://md.research.elamur.ai/free/en/papers/mostefa-2014-techlimed-system-description-shared-task/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
The paper describes Techlimed's entry in the QALB-2014 shared task on automatic error correction for Arabic. Arabic is hard to correct automatically because of its rich morphology, missing short vowels, clitics and scarce public resources. The authors compared two strategies on the QALB development data. The first was a lexicon-based approach using off-the-shelf spellcheckers: Microsoft Word, OpenOffice, and Hunspell with Ayaspell dictionaries. The second treated correction as statistical machine translation. They applied the gold edits to the training sentences to get an error→correct parallel corpus, then trained a phrase-based Moses model on surface forms, with GIZA++ alignment and IRSTLM language models built from up to 300M words of news text.

The SMT system clearly won, reaching an F-measure of 58.7% against at most 30.2% for the spellcheckers. Most of the gap came from punctuation, which the spellcheckers ignore. The paper is a useful early baseline showing that a modest annotated parallel corpus can teach a generic MT system the conventions of a correction guideline.

## Key points
- **Data and scoring.** The QALB corpus is drawn from news text. It has about 19.4k training sentences (~1M tokens), a 1,017-sentence dev set and a 968-sentence test set. Scoring uses the M2 scorer, reporting precision, recall and F-measure.
- **Spellchecker baselines on dev (P/R/F, %):**
  - Word: 45.7/16.6/24.3
  - Hunspell+Ayaspell: 51.8/18.8/27.6 (Ayaspell has 52,725 entries and 11,859 affix rules)
  - OpenOffice: 56.1/20.7/30.2
- **SMT setup.** The Moses training data was split into 18k sentences for training and a held-out part for tuning. Three language model sizes were tried (100M, 200M, 300M words), drawn from OSAC, the Adjir corpus and in-house newspaper crawls.
- **SMT results.** Precision stayed at ~70.7–70.8 across sizes, while recall rose from 48.8 to 50.1. The best F-measure was 58.7, with BLEU between 65.45 and 65.98. Running Hunspell on the SMT output gave no extra gain.
- **Punctuation share.** Punctuation accounts for 38.6% of dev annotations (6,266 of 16,231). Commas alone make up 25.4%, and weak punctuation follows few firm rules.

## Interesting ideas
- **Correction as translation.** Turning an edit-annotated corpus into sentence pairs lets a generic translation learner pick up annotator conventions without hand-written rules.
- **Profile errors by type before blaming a method.** Breaking gold edits down by category showed that the spellcheckers' low recall came mostly from punctuation they never touch.
- **Run tools as forced top-1 decisions.** Interactive tools were made fully automatic by always taking the first suggestion, and the authors saw the right fix often sitting lower in the list.
- **Scale one resource and watch metrics separately.** Tripling language model data left precision flat and only nudged recall, a cheap check for diminishing returns.
- **Test the obvious hybrid explicitly.** Chaining a lexicon checker after the statistical system seemed promising but added nothing, and the authors reported it rather than assuming stacking helps.

## For Mova

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

- **I1** (tokens): noisy input could be normalized by a learned erroneous→clean mapping before tagging, while logging each rewrite so the step stays explainable.
- **I2** (learn): evaluation reports should split errors by relation or phenomenon type, so a weak score is traced to a missing capability rather than general model quality.
- **I3** (tree): keeping k-best parses or lemma candidates and measuring oracle-at-k would show whether later stages could recover what top-1 loses.
- **I4** (learn): when adding silver data, track precision and recall separately per increment to decide when more data stops paying off.
- **I5** (learn): every proposed repair-rule cascade should be ablated against the base system as a control before it is kept.

*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

- [Université de Lyon](https://md.research.elamur.ai/free/en/universities/universite-de-lyon/) — University, Lyon, France · 4 works · 372 citations
- [Centre de Recherche en Linguistique Appliquée](https://md.research.elamur.ai/free/en/labs/centre-de-recherche-en-linguistique-appliquee/) — Research institute, Lyon, France · 3 works · 24 citations

📚 Shelves (6): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works · [🤝 Social Sciences](https://md.research.elamur.ai/free/en/shelves/social-sciences/) · 1,772 works · [🎓 Education](https://md.research.elamur.ai/free/en/shelves/social-sciences/education/) · 203 works

🧭 Topics (1): [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 2,908 works

## Cite

Cite as: Djamel Mostefa, Omar Asbayou and Ramzi Abbes. 2014. TECHLIMED system description for the Shared Task on Automatic Arabic Error Correction. Proceedings of the EMNLP 2014 Workshop on Arabic Natural Language Processing (ANLP).

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

```bibtex
@inproceedings{mostefa2014techlimed,
  title = {{TECHLIMED system description for the Shared Task on Automatic Arabic Error Correction}},
  author = {Djamel Mostefa, Omar Asbayou and Ramzi Abbes},
  booktitle = {Proceedings of the EMNLP 2014 Workshop on Arabic Natural Language Processing (ANLP)},
  year = {2014},
  doi = {10.3115/v1/w14-3621},
  url = {https://doi.org/10.3115/v1/w14-3621}
}
```

- [← previous Rule-and Dictionary-based Solution for Variations in Written Arabic Names in Social Networks, Big Data, Accounting Systems and Large Databases](https://md.research.elamur.ai/free/en/papers/hassanat-2015-rule-dictionary-based-solution-variations/)
- [next → “I Object!” Modeling Latent Pragmatic Effects in Courtroom Dialogues](https://md.research.elamur.ai/free/en/papers/goldwasser-2014-i-object-modeling-latent-pragmatic/)
- [↑ back to papers of 2014](https://md.research.elamur.ai/free/en/papers/2014/page/4/)

## References

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

1. Ramzi Abbès, Joseph Dichy, and Mohamed Hassoun. 2004. [The architecture of a standard arabic lexical database: some figures, ratios and categories from the Diinar](https://md.research.elamur.ai/free/en/papers/abbes-2004-architecture-standard-arabic-lexical-database/). 1 source program. In Proceedings of the Workshop on Computational Approaches to Arabic Script-based Languages, pages 15–22. Association for Computational Linguistics, 2004.
2. Mariam Aboelezz. 2009. Latinised arabic and connections to bilingual ability. In Papers from the Lancaster University Postgraduate Conference in Linguistics and Language Teaching, 2009. [find](https://scholar.google.com/scholar?q=Mariam+Aboelezz%2E+2009%2E+Latinised+arabic+and+connections+to+bilingual+ability%2E+In+Papers+from+the+Lancaster+University+Postgraduate+Conference+in+Linguistics+and+Language+Teaching%2C+2009%2E)
3. Ahmed Abdelali. 2005. [http://aracorpus.e3rab.com/Ayaspell](http://aracorpus.e3rab.com/Ayaspell) Arabic dictionary project, http://ayaspell.sourceforge.net 2008. [↗](http://aracorpus.e3rab.com/Ayaspell)
4. Daniel Dahlmeier and Hwee Tou Ng. 2012. 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, pages 568–572. Association for Computational Linguistics, 2012. [find](https://scholar.google.com/scholar?q=Daniel+Dahlmeier+and+Hwee+Tou+Ng%2E+2012%2E+Better+evaluation+for+grammatical+error+correction%2E+In+Proceedings+of+the+2012+Conference+of+the+North+American+Chapter+of+the+Association+for+Computational+Lin)
5. Marcello Federico, Nicola Bertoldi, and Mauro Cettolo. 2008. Irstlm: an open source toolkit for handling large scale language models. In Interspeech, pages 1618–1621, 2008. [find](https://scholar.google.com/scholar?q=Marcello+Federico%2C+Nicola+Bertoldi%2C+and+Mauro+Cettolo%2E+2008%2E+Irstlm%3A+an+open+source+toolkit+for+handling+large+scale+language+models%2E+In+Interspeech%2C+pages+1618%E2%80%931621%2C+2008%2E)
6. Hunspell, 2007. [http://hunspell.sourceforge.net/](http://hunspell.sourceforge.net/) [↗](http://hunspell.sourceforge.net/)
7. Philipp Koehn, Hieu Hoang, Alexandra Birch, Chris Callison-Burch, Marcello Federico, Nicola Bertoldi, Brooke Cowan, Wade Shen, Christine Moran, Richard Zens, et al. 2007. Moses: Open source toolkit for statistical machine translation. In Proceedings of the 45th Annual Meeting of the ACL on Interactive Poster and Demonstration Sessions, pages 177–180. Association for Computational Linguistics, 2007. [find](https://scholar.google.com/scholar?q=Philipp+Koehn%2C+Hieu+Hoang%2C+Alexandra+Birch%2C+Chris+Callison%2DBurch%2C+Marcello+Federico%2C+Nicola+Bertoldi%2C+Brooke+Cowan%2C+Wade+Shen%2C+Christine+Moran%2C+Richard+Zens%2C+et+al%2E+2007%2E+Moses%3A+Open+source+toolkit+fo)
8. Vladimir Levenshtein. 1966. Binary codes capable of correcting deletions, insertions and reversals. In Soviet physics doklady, volume 10, page 707, 1966. [find](https://scholar.google.com/scholar?q=Vladimir+Levenshtein%2E+1966%2E+Binary+codes+capable+of+correcting+deletions%2C+insertions+and+reversals%2E+In+Soviet+physics+doklady%2C+volume+10%2C+page+707%2C+1966%2E)
9. Behrang Mohit, Alla Rozovskaya, Nizar Habash, Wajdi Zaghouani, and Ossama Obeid. 2014. [The First QALB Shared Task on Automatic Text Correction for Arabic](https://md.research.elamur.ai/free/en/papers/mohit-2014-first-qalb-shared-task-automatic/). In Proceedings of EMNLP Workshop on Arabic Natural Language Processing, Doha, Qatar, October 2014.
10. Franz Josef Och and Hermann Ney. 2003. A systematic comparison of various statistical alignment models. Computational Linguistics, 29(1):19–51, 2003. [↗](https://doi.org/10.1162/089120103321337421)
11. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. [Bleu: a method for automatic evaluation of machine translation](https://md.research.elamur.ai/free/en/papers/papineni-2002-bleu-method-automatic-evaluation-machine/). In Proceedings of the 40th annual meeting on association for computational linguistics, pages 311–318. Association for Computational Linguistics, 2002.
12. Motaz K Saad and Wesam Ashour. 2010 Osac: Open source arabic corpora. In 6th ArchEng Int. Symposiums, EEECS, volume 10, 2010. [↗](https://doi.org/10.13140/2.1.4664.9288)
13. Wajdi Zaghouani, Behrang Mohit, Nizar Habash, Ossama Obeid, Nadi Tomeh, Alla Rozovskaya, Noura Farra, Sarah Alkuhlani, and Kemal Oflazer. 2014. [Large scale arabic error annotation: Guidelines and framework](https://md.research.elamur.ai/free/en/papers/zaghouani-2014-large-scale-arabic-error-annotation/). In Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC’14), Reykjavik, Iceland, May 2014. European Language Resources Association (ELRA). [↗](https://doi.org/10.13140/rg.2.1.3273.6169)

*Updated 1 November 2026*