Elamur Research⌕
☰
home / papers / 2026 / Corruption-Based Data Augmentation for Arabic Essay Scoring: A Preliminary Study on the Organization Trait

Corruption-Based Data Augmentation for Arabic Essay Scoring: A Preliminary Study on the Organization Trait

Paper · 2026 · May Saed Bashendy and Tamer Elsayed

Summary

Arabic automated essay scoring (AES) is held back by small, unbalanced annotated datasets, and few of them have trait-level scores. The authors propose COrE, an augmentation method for the organization trait. It takes well-organized essays (organization score of 4 or higher) and creates weaker copies by swapping sentence pairs. Each copy gets a lower score from a deterministic formula. The penalty grows with the number of swaps relative to the maximum possible, and with the total distance between swapped sentences relative to the largest distance the essay length allows. Each sentence can be swapped at most once. Swapping stops once the score lands within a tolerance of a target value.

To mimic a low-resource setting, the experiments train on a tiny balanced seed set drawn from TAQAE (620 essays, 4 prompts). They compare naïve oversampling, COrE, and COrE combined with oversampling, each fine-tuned on AraBERTv2 and CAMeLBERT-mix. COrE beats the no-augmentation baseline by roughly 9–17%. The study suggests that augmentation aimed at one trait can partly make up for missing data in low-resource languages, and the synthetic essays are publicly released.

Key points

  • TAQAE has 620 Arabic essays over 4 prompts:
    • 425 essays (P1 explanatory, P2 persuasive) come from the TAQEEM 2025 shared task.
    • 195 essays (P3, P4) come from QCAW, using the QAES annotations.
    • Organization is scored on a 1–5 scale, as one of seven annotated traits.
  • The seed training set has at most 5 essays per organization score level, so at most 25 per prompt. Some levels have fewer because of class imbalance.
  • COrE builds on CASE (Yoo et al., 2025) and fixes two of its weaknesses:
    • CASE ignores how severe a disruption is, while COrE weights swaps by distance.
    • CASE allows repeated swaps that can restore the original order, while COrE does not.
  • With both AraBERTv2 and CAMeLBERT-mix, COrE improves on the no-augmentation baseline by 9–17%.
  • Limitations:
    • Only one trait is studied.
    • The method cannot produce top-score essays.
    • Results on P3/P4 are unstable because their test sets are small, so the reliable gains come mainly from P1 and P2.

Interesting ideas

  • I1. The label of a synthetic example is computed from how hard it was damaged (count and distance of perturbations), not assigned by hand, which yields a graded range of quality rather than a simple good/bad split.
  • I2. Each sentence may join at most one swap, so a later perturbation can never silently undo an earlier one and the intended damage is guaranteed.
  • I3. Disruption is divided by the theoretical maximum for that essay length (pairing first with last, second with second-to-last, and so on), which makes scores comparable across texts of different sizes.
  • I4. A target score plus a tolerance margin stops the corruption loop early, letting the generator hit chosen difficulty levels instead of random ones.
  • I5. The augmentation is checked against naïve duplication with the same balanced class counts, which shows whether the gain comes from new information or just from more examples.

For Mova

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

  • I1 (learn): Silver data for event-order or discourse-coherence checks could be built by shuffling clauses of trusted texts, with the expected degradation score derived from the shuffle itself.
  • I2 (learn): Perturbation-based negative controls should carry the same guard, so a "corrupted" test item is never secretly identical to the original.
  • I3 (world): A timeline-reconstruction metric could normalize displacement against the worst possible reordering for that number of events, so short and long fables are judged on one scale.
  • I4 (answer): School-test and word-problem variants could be produced at controlled difficulty bands by perturbing until a scoring function falls within a set margin of the target.
  • I5 (learn): Any silver-data boost should be compared against a same-size oversampling control, to confirm that the improvement is not just a data-volume effect.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 9,085 worksArtificial Intelligence · 8,888 worksArts and Humanities · 9,296 worksLanguage and Linguistics · 9,206 works

🧭 Topics 2Research subjects and languages the work is about.Semitic languages · 1,485 worksGrammatical error correction & learner language · 80 works

Cite

Cite as May Saed Bashendy and Tamer Elsayed. 2026. Corruption-Based Data Augmentation for Arabic Essay Scoring: A Preliminary Study on the Organization Trait. Proceedings of the Fifteenth Language Resources and Evaluation Conference.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/5kt6mmyaumav

@inproceedings{bashendy2026corruption,
  title = {{Corruption-Based Data Augmentation for Arabic Essay Scoring: A Preliminary Study on the Organization Trait}},
  author = {May Saed Bashendy and Tamer Elsayed},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation Conference},
  year = {2026},
  doi = {10.63317/5kt6mmyaumav},
  url = {https://doi.org/10.63317/5kt6mmyaumav}
}

References

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

  1. Abdelhamid M. Ahmed, Xiao Zhang, Lameya M. Rezk, and Wajdi Zaghouani. 2024. Building an Annotated L1 Arabic/L2 English Bilingual Writer Corpus: The Qatari Corpus of Argumentative Writing (QCAW). Corpus-based Studies across Humanities, 1(1):183–215. ↗
  2. Mansour Alghamdi, Mohamed Alkanhal, Mohamed Al-Badrashiny, Abdulaziz Al-Qabbany, Ali Areshey, and Abdulaziz Alharbi. 2014. A Hybrid Automatic Scoring System for Arabic Essays. AI Communications, 27(2):103–111. ↗
  3. Aqil M. Azmi, Maram F. Al-Jouie, and Muhammad Hussain. 2019. AAEE – Automated Evaluation of Students’ Essays in Arabic Language. Information Processing Management, 56(5):1736– 1752. ↗
  4. May Bashendy, Salam Albatarni, Sohaila Eltanbouly, Walid Massoud, Houda Bouamor, and Tamer Elsayed. 2025. TAQEEM 2025: Overview of the First Shared Task for Arabic Quality Evaluation of Essays in Multi-dimensions. In Proceedings of the Third Arabic Natural Language Processing Conference (ArabicNLP 2025), China. find
  5. May Bashendy, Salam Albatarni, Sohaila Eltanbouly, Eman Zahran, Hamdo Elhuseyin, Tamer Elsayed, Walid Massoud, and Houda Bouamor. 2024. QAES: First Publicly-Available Trait-Specific Annotations for Automated Scoring of Arabic Essays. In Proceedings of the Second Arabic Natural Language Processing Conference, pages 337–351, Bangkok, Thailand. Association for Computational Linguistics. find
  6. Domenic V Cicchetti. 1981. Testing the Normal Approximation and Minimal Sample Size Requirements of Weighted Kappa When the Number of Categories is Large. Applied psychological measurement, 5(1):101–104. ↗
  7. Jacob Cohen. 1968. Weighted kappa: Nominal Scale Agreement Provision for Scaled Disagreement or Partial Credit. Psychological Bulletin, 70(4):213–220. ↗
  8. S.A. Crossley, Y. Tian, P. Baffour, A. Franklin, M. Benner, and U. Boser. 2024. A Large-Scale Corpus for Assessing Written Argumentation: PERSUADE 2.0. Assessing Writing, 61:100865. ↗
  9. Bosheng Ding, Chengwei Qin, Ruochen Zhao, Tianze Luo, Xinze Li, Guizhen Chen, Wenhan Xia, Junjie Hu, Anh Tuan Luu, and Shafiq Joty. 2024. Data Augmentation Using Large Language Models: Data Perspectives, Learning Paradigms and Challenges. In Findings of the Association for Computational Linguistics: ACL 2024, pages 1679–1705, Bangkok, Thailand. Association for Computational Linguistics. ↗
  10. Ahmed Adel ElSabagh, Shahira Shaaban Azab, and Hesham Ahmed Hefny. 2025. A Comprehensive Survey on Arabic Text Augmentation: Approaches, Challenges, and Applications. Neural Computing and Applications, pages 1–34. ↗
  11. Rayed Ghazawi and Edwin Simpson. 2024. Automated Essay Scoring in Arabic: A Dataset and Analysis of a BERT-based system. arXiv preprint arXiv:2407.11212. ↗
  12. Rayed Ghazawi and Edwin Simpson. 2025. How Well Can LLMs Grade Essays in Arabic? Computers and Education: Artificial Intelligence, 9:100449. ↗
  13. Kshitij Gupta. 2023. Data Augmentation for Automated Essay Scoring using Transformer Models. In International Conference on Artificial Intelligence and Smart Communication (AISC), pages 853–857. IEEE. find
  14. Nizar Habash and David Palfreyman. 2022. ZAE-BUC: An Annotated Arabic-English Bilingual Writer Corpus. In Proceedings of the Thirteenth Language Resources and Evaluation Conference, pages 79–88, Marseille, France. European Language Resources Association. find
  15. Bohan Li, Yutai Hou, and Wanxiang Che. 2022. Data Augmentation Approaches in Natural Language Processing: A Survey. AI Open, 3:71–90. ↗
  16. Ruibo Liu, Jerry Wei, Fangyu Liu, Chenglei Si, Yanzhe Zhang, Jinmeng Rao, Steven Zheng, Daiyi Peng, Diyi Yang, Denny Zhou, et al. 2024. Best Practices and Lessons Learned on Synthetic Data. arXiv preprint arXiv:2404.07503. ↗
  17. Sandeep Mathias and Pushpak Bhattacharyya. 2018. ASAP++: Enriching the ASAP Automated Essay Grading Dataset with Essay Attribute Scores. In Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), Miyazaki, Japan. European Language Resources Association (ELRA). find
  18. Chatrine Qwaider, Bashar Alhafni, Kirill Chirkunov, Nizar Habash, and Ted Briscoe. 2025. Enhancing Arabic Automated Essay Scoring with Synthetic Data and Error Injection. In Proceedings of the 20th Workshop on Innovative Use of NLP for Building Educational Applications (BEA 2025), pages 549–563, Vienna, Austria. Association for Computational Linguistics. ↗
  19. Marwan Sayed, Sohaila Eltanbouly, May Bashendy, and Tamer Elsayed. 2025. Feature Engineering is not Dead: A Step Towards State of the Art for Arabic Automated Essay Scoring. In Proceedings of The Third Arabic Natural Language Processing Conference, pages 231–245, Suzhou, China. Association for Computational Linguistics. find
  20. Haneul Yoo, Jieun Han, So-Yeon Ahn, and Alice Oh. 2025. DREsS: Dataset for Rubric-based Essay Scoring on EFL Writing. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 13439–13454, Vienna, Austria. Association for Computational Linguistics. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/bashendy-2026-corruption-based-data-augmentation-arabic/