Elamur Research⌕
☰
home / papers / 2025 / Towards the Development of Balanced Synthetic Data for Correcting Grammatical Errors in Arabic: An Approach Based on Error Tagging Model and Synthetic Data Generating Model

Towards the Development of Balanced Synthetic Data for Correcting Grammatical Errors in Arabic: An Approach Based on Error Tagging Model and Synthetic Data Generating Model

Summary

Arabic grammatical error correction (GEC) has little training data. The only parallel corpora are QALB-14 and QALB-15, which are small and dominated by punctuation fixes. Earlier synthetic-data methods were mostly rule-based or random noise, so they covered a narrow range of error types.

The authors split the generation task into two learned models:

  • Error tagging model. A DeBERTav3 multi-label classifier reads a correct sentence and predicts which of 26 ARETA error categories could plausibly occur in it.
  • Synthetic data generation model. An AraT5 seq2seq model receives those tags prepended to the clean sentence and writes a corrupted version with matching errors. This is reverse ("back-translation") generation, from clean to erroneous text.

The tagger reached 94.42% F1 on the QALB test sets. Applied to monolingual text, the pipeline produced about 30.2 million sentence pairs, which are publicly released. An AraT5 corrector trained on this data set a new best F1 of 79.36% on QALB-14 but trailed a rule-based competitor on QALB-15. The work shows a controllable, type-aware way to scale scarce error-annotated data.

Key points

  • The available human-corrected data is small: about 20,430 training samples in QALB-14 and 1,542 in QALB-15.
  • Error tags are 26 ARETA categories. The ARETA tool supplies labels for the multi-label tagger, which uses DeBERTav3.
  • The tagger scored 94.42% F1 on the QALB-14/QALB-15 test sets, which the authors report as state of the art for this task.
  • The released synthetic corpus has 30,219,310 pairs (github.com/Ahlam-alrehili/ArabGEC).
  • GEC results:
    • QALB-14: the model improved on prior results in precision, recall, F1 (79.36%) and F0.5.
    • QALB-15: it lost to Solyman et al.'s augmented data. The authors attribute this partly to their L1-only training text, since QALB-15 mixes native and non-native writers.

Interesting ideas

  • I1. Instead of injecting random noise, first predict which error types fit a given clean sentence, then condition the corruptor on those types; repeating a tag asks for several errors of that kind, giving control over both kind and amount of damage. — For Mova (learn): typed, controllable corruption of clean sentences could produce graded negative controls for evaluating whether the parser and frame layers degrade in explainable, category-specific ways.
  • I2. Train the correction direction in reverse (clean → noisy) on a small parallel set, then run it over a large monolingual corpus to multiply training pairs. — For Mova (tokens): a learned typo/error generator could stress-test the tokenizer, tagger and lemmatizer on realistic misspellings without hand-written noise rules.
  • I3. Use an existing error-annotation tool (ARETA) to label parallel pairs automatically, and treat those labels as supervision for a separate predictor. — For Mova (learn): diffs between a broken and a repaired parse could be auto-labelled by repair type, giving training signal for choosing which self-induced repair rule to apply.
  • I4. When rare classes are swamped in multi-label prediction, a Distribution-Balanced Loss raises the loss weight of underrepresented labels. — For Mova (frames): rare VerbNet roles or uncommon speech acts could get the same reweighting so the classifier does not collapse onto the frequent roles.
  • I5. Report results per test set and tie the gaps to the writer population (native-only training versus a test set that mixes native and learner text). — For Mova (answer): evaluating readers separately on texts from different sources or authors would show where domain mismatch, rather than model weakness, explains failures.

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: Linguistics & languages

Topics: Part-of-speech tagging

Cited by

3 papers of our library cite it (of the 140884 whose bibliographies we read); each has a page here.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2502.05312 · arXiv 2502.05312

@misc{alrehili2025development,
  title = {{Towards the Development of Balanced Synthetic Data for Correcting Grammatical Errors in Arabic: An Approach Based on Error Tagging Model and Synthetic Data Generating Model}},
  author = {Alrehili, Ahlam and Alhothali, Areej},
  year = {2025},
  doi = {10.48550/arxiv.2502.05312},
  eprint = {2502.05312},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2502.05312}
}

References

43 references, 12 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Shamil Chollampatt and Hwee Tou Ng. A multilayer convolutional encoder-decoder neural network for grammatical error correction. In Proceedings of the AAAI conference on artificial intelligence, volume 32, 2018. find
  2. Marcin Junczys-Dowmunt, Roman Grundkiewicz, Shubha Guha, and Kenneth Heafield. Approaching neural grammatical error correction as a low-resource machine translation task. arXiv preprint arXiv:1804.05940, 2018. ↗
  3. Khaled Shaalan, Sanjeera Siddiqui, Manar Alkhatib, and Azza Abdel Monem. Challenges in arabic natural language processing. In Computational linguistics, speech and image processing for arabic language, pages 59–83. World Scientific, 2019. find
  4. Behrang Mohit, Alla Rozovskaya, Nizar Habash, Wajdi Zaghouani, and Ossama Obeid. The first qalb shared task on automatic text correction for arabic. In Proceedings of the EMNLP 2014 Workshop on Arabic Natural Language Processing (ANLP), pages 39–47, 2014. find
  5. Alla Rozovskaya, Houda Bouamor, Nizar Habash, Wajdi Zaghouani, Ossama Obeid, and Behrang Mohit. The second qalb shared task on automatic text correction for arabic. In Proceedings of the Second workshop on Arabic natural language processing, pages 26–35, 2015. find
  6. Serena Jeblee, Houda Bouamor, Wajdi Zaghouani, and Kemal Oflazer. Cmuq@ qalb-2014: An smt-based system for automatic arabic error correction. In Proceedings of the EMNLP 2014 Workshop on Arabic Natural Language Processing (ANLP), pages 137–142, 2014. find
  7. Ossama Obeid, Wajdi Zaghouani, Behrang Mohit, Nizar Habash, Kemal Oflazer, and Nadi Tomeh. A web-based annotation framework for large-scale text correction. In The Companion Volume of the Proceedings of IJCNLP 2013: System Demonstrations, pages 1–4, 2013. find
  8. Aiman Solyman, Wang Zhenyu, Tao Qian, Arafat Abdulgader Mohammed Elhag, Muhammad Toseef, and Zeinab Aleibeid. Synthetic data with neural machine translation for automatic correction in arabic grammar. Egyptian Informatics Journal, 22(3):303–315, 2021. find
  9. Aiman Solyman, Zhenyu Wang, Qian Tao, Arafat Abdulgader Mohammed Elhag, Rui Zhang, and Zeinab Mahmoud. Automatic arabic grammatical error correction based on expectation-maximization routing and target-bidirectional agreement. Knowledge-Based Systems, 241:108180, 2022. find
  10. Aiman Solyman, Marco Zappatore, Wang Zhenyu, Zeinab Mahmoud, Ali Alfatemi, Ashraf Osman Ibrahim, and Lubna Abdelkareim Gabralla. Optimizing the impact of data augmentation for low-resource grammatical error correction. Journal of King Saud University-Computer and Information Sciences, 35(6):101572, 2023. find
  11. Sang Yun Kwon, Gagan Bhatia, El Moatez Billah Nagoud, and Muhammad Abdul-Mageed. Chatgpt for arabic grammatical error correction. arXiv preprint arXiv:2308.04492, 2023. ↗
  12. Riadh Belkebir and Nizar Habash. Automatic error type annotation for arabic. arXiv preprint arXiv:2109.08068, 2021. ↗ our card
  13. Sarah AlOyaynaa and Yasser Kotb. Arabic grammatical error detection using transformers-based pretrained language models. In ITM Web of Conferences, volume 56, page 04009. EDP Sciences, 2023. find
  14. Nora Madi and Hend S Al-Khalifa. A7’ ta: Data on a monolingual arabic parallel corpus for grammar checking. Data in brief, 22:237, 2019. find
  15. Wissam Antoun, Fady Baly, and Hazem Hajj. Arabert: Transformer-based model for arabic language understanding. In LREC 2020 Workshop Language Resources and Evaluation Conference 11–16 May 2020, page 9, 2020. find our card
  16. Fatemah Husain and Ozlem Uzuner. Leveraging offensive language for sarcasm and sentiment detection in arabic. In Proceedings of the Sixth Arabic Natural Language Processing Workshop, pages 364–369, 2021. find our card
  17. Muhammad Abdul-Mageed, AbdelRahim Elmadany, and El Moatez Billah Nagoudi. Arbert & marbert: Deep bidirectional transformers for arabic. arXiv preprint arXiv:2101.01785, 2020. ↗ our card
  18. Go Inoue, Bashar Alhafni, Nurpeiis Baimukan, Houda Bouamor, and Nizar Habash. The interplay of variant, size, and task type in arabic pre-trained language models. arXiv preprint arXiv:2103.06678, 2021. ↗ our card
  19. Pengcheng He, Xiaodong Liu, Jianfeng Gao, and Weizhu Chen. Deberta: Decoding-enhanced bert with disentangled attention. arXiv preprint arXiv:2006.03654, 2020. ↗ our card
  20. Pengcheng He, Jianfeng Gao, and Weizhu Chen. Debertav3: Improving deberta using electra-style pre-training with gradient-disentangled embedding sharing. arXiv preprint arXiv:2111.09543, 2021. ↗ our card
  21. Ibrahim Abu El-Khair. 1.5 billion words arabic corpus. arXiv preprint arXiv:1611.04033, 2016. ↗
  22. Imad Zeroual, Dirk Goldhahn, Thomas Eckart, and Abdelhak Lakhouaja. Osian: Open source international arabic news corpus-preparation and integration into the clarin-infrastructure. In Proceedings of the fourth arabic natural language processing workshop, pages 175–182, 2019. find
  23. Pedro Javier Ortiz Suárez, Laurent Romary, and Benoît Sagot. A monolingual approach to contextualized word embeddings for mid-resource languages. arXiv preprint arXiv:2006.06202, 2020. ↗ our card
  24. Pedro Javier Ortiz Suárez, Benoît Sagot, and Laurent Romary. Asynchronous pipeline for processing huge corpora on medium to low resource infrastructures. In 7th Workshop on the Challenges in the Management of Large Corpora (CMLC-7). Leibniz-Institut für Deutsche Sprache, 2019. find
  25. Yukun Zhu. Aligning books and movies: Towards story-like visual explanations by watching movies and reading books. arXiv preprint arXiv:1506.06724, 2015. ↗ our card
  26. Aaron Gokaslan, Vanya Cohen, E Pavlick, and S Tellex. Openwebtext corpus, 2019. URL http://Skylion007.github. io/OpenWebTextCorpus, page 9, 2019. ↗
  27. Trieu H Trinh and Quoc V Le. A simple method for commonsense reasoning. arXiv preprint arXiv:1806.02847, 2018. ↗
  28. Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. Exploring the limits of transfer learning with a unified text-to-text transformer. Journal of machine learning research, 21(140):1–67, 2020. find
  29. L Xue. mt5: A massively multilingual pre-trained text-to-text transformer. arXiv preprint arXiv:2010.11934, 2020. ↗
  30. Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, and Colin Raffel. Byt5: Towards a token-free future with pre-trained byte-to-byte models. Transactions of the Association for Computational Linguistics, 10:291–306, 2022. find
  31. El Moatez Billah Nagoudi, AbdelRahim Elmadany, and Muhammad Abdul-Mageed. Arat5: Text-to-text transformers for arabic language generation. arXiv preprint arXiv:2109.12068, 2021. ↗ our card
  32. Abdullah Alfaifi and Eric Atwell. Arabic learner corpora (alc): a taxonomy of coding errors. In The 8th International Computing Conference in Arabic, 2012. find
  33. S Farwaneh and M Tamimi. Arabic learners written corpus: A resource for research and learning. retrieved 2 september, 2012, from the the university of arizona. the Center for Educational Resources in Culture, Language and Literacy web site: http://l2arabiccorpus.cercll. arizona. edu, 2012. ↗
  34. Nizar Habash and Owen Rambow. Arabic tokenization, part-of-speech tagging and morphological disambiguation in one fell swoop. In Proceedings of the 43rd annual meeting of the association for computational linguistics (ACL’05), pages 573–580, 2005. find our card
  35. Nizar Habash, Owen Rambow, and Ryan Roth. Mada+ tokan: A toolkit for arabic tokenization, diacritization, morphological disambiguation, pos tagging, stemming and lemmatization. In Proceedings of the 2nd international conference on Arabic language resources and tools (MEDAR), Cairo, Egypt, volume 41, page 62, 2009. find
  36. Nizar Habash and David Palfreyman. Zaebuc: An annotated arabic-english bilingual writer corpus. In Proceedings of the Thirteenth Language Resources and Evaluation Conference, pages 79–88, 2022. find
  37. Yi Huang, Buse Giledereli, Abdullatif Köksal, Arzucan Özgür, and Elif Ozkirimli. Balancing methods for multi-label text classification with long-tailed class distribution. arXiv preprint arXiv:2109.04712, 2021. ↗
  38. Xi-Zhu Wu and Zhi-Hua Zhou. A unified view of multi-label performance measures. In international conference on machine learning, pages 3780–3788. PMLR, 2017. find
  39. Yiming Yang. An evaluation of statistical approaches to text categorization. Information retrieval, 1:69–90, 1999. find
  40. Francisco Herrera, Francisco Charte, Antonio J Rivera, María J Del Jesus, Francisco Herrera, Francisco Charte, Antonio J Rivera, and María J del Jesus. Multilabel classification. Springer, 2016. find
  41. Bishan Yang, Jian-Tao Sun, Tengjiao Wang, and Zheng Chen. Effective multi-label active learning for text classification. In Proceedings of the 15th ACM SIGKDD international conference on Knowledge discovery and data mining, pages 917–926, 2009. find
  42. Daniel Dahlmeier and Hwee Tou 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, pages 568–572, 2012. find
  43. Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, Rémi Louf, Morgan Funtowicz, et al. Transformers: State-of-the-art natural language processing. In Proceedings of the 2020 conference on empirical methods in natural language processing: system demonstrations, pages 38–45, 2020. find our card

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/alrehili-2025-towards-development-balanced-synthetic-data/