---
title: "AraT5-MSAizer: Translating Dialectal Arabic to MSA — Fares 2024"
description: "The paper describes AraT5-MSAizer, a system built for Task 2 of OSACT 2024: converting text in five… 21.79 BLEU. 23 references, cited by 2 papers here."
authors: ["Murhaf Fares"]
year: 2024
venue: "Proceedings of the 6th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT) with Shared Tasks on Arabic LLMs Hallucination and Dialect to MSA Machine Translation @ LREC-COLING 2024"
source: "https://aclanthology.org/2024.osact-1.16/"
licence: "CC BY-NC 4.0 (ELRA, LREC proceedings)"
licence_url: "https://creativecommons.org/licenses/by-nc/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/fares-2024-arat5-msaizer-translating-dialectal-arabic/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/fares-2024-arat5-msaizer-translating-dialectal-arabic/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2024](https://md.research.elamur.ai/free/en/papers/2024/page/4/) / AraT5-MSAizer: Translating Dialectal Arabic to MSA

# AraT5-MSAizer: Translating Dialectal Arabic to MSA

**Authors:** [Murhaf Fares](https://md.research.elamur.ai/free/en/authors/murhaf-fares/) · **Year:** 2024 · **Venue:** Proceedings of the 6th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT) with Shared Tasks on Arabic LLMs Hallucination and Dialect to MSA Machine Translation @ LREC-COLING 2024 · **Licence:** [CC BY-NC 4.0 (ELRA, LREC proceedings)](https://creativecommons.org/licenses/by-nc/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2024.osact-1.16/) · [PDF](https://aclanthology.org/2024.osact-1.16.pdf) · [OpenAlex](https://openalex.org/W7131787181)

**Citations:** 2 citations (OpenAlex, 2026-10-14) · 2 in our library, of the 140,884 papers whose bibliographies we read

## Summary
The paper describes AraT5-MSAizer, a system built for Task 2 of OSACT 2024: converting text in five regional Arabic dialects (Gulf, Egyptian, Levantine, Iraqi, Maghrebi) into Modern Standard Arabic. Diglossia matters here because most Arabic NLP resources target MSA, and dialectal input therefore poses problems for them. The author fully fine-tuned the encoder-decoder AraT5 v2. It was chosen partly because its pre-training included dialect-rich Twitter text.

All dialects were treated as a single source variety. The training data combined three gold parallel corpora (MADAR, PADIC, the North Levantine Corpus) with a large silver corpus. That silver corpus was produced by back-translating OPUS MSA sentences through a reverse MSA-to-dialect model, then cleaned with length filters and a fix for parenthesis mismatches. On the development set, adding the synthetic data clearly beat training on gold data alone. A model trained only on MADAR fell below a trivial copy-the-input baseline. The final system reached 21.79 BLEU on the official test set. This shows that a modest single-GPU setup with back-translation gives a competitive dialect normalizer.

## Key points
- **Task data:**
  - The development set has 1,001 pairs (200 per dialect).
  - The test set has 1,888 pairs, unevenly spread across dialects: Gulf 586, Levantine 568, Maghrebi 343, Egyptian 314, Iraqi 77.
  - Scoring used BLEU and Comet DA.
- **Gold corpora:**
  - MADAR: 88,200 pairs, with its 25 city dialects merged into one "DA" label.
  - North Levantine Corpus: about 120,600 Shami–MSA pairs, 90% of them used.
  - PADIC: 41,680 pairs.
- **Synthetic data:**
  - Back-translated OPUS-100 yielded 965,020 pairs, which were re-filtered to MSA lengths of 25–300 characters.
  - About 17,000 MSA→MSA identity pairs from MADAR were added.
  - The final training split had 700,386 pairs.
- **Training setup:**
  - One epoch, learning rate 2e-5, batch size 32.
  - Maximum input length 1024, maximum generation length 512.
  - Trained on a single RTX A6000.
- **Development BLEU:**
  - MADAR-only: 0.1140.
  - Gold: 0.2038.
  - Gold+synthetic: 0.2302.
  - Copy baseline: 0.1445.
  - The peak was 0.2325 at around 15k steps, but those weights were not restored.
- **Test result:** BLEU 0.2179, with Comet DA reported as 0.0016.

## Interesting ideas
- A "leave-as-is" baseline, which simply copies the source, exposed that the MADAR-only model was worse than doing nothing; comparing against a trivial identity output is a cheap sanity check for any transformation system.
- Training examples where source and target are already identical (MSA→MSA) teach the model when no change is needed, which discourages over-editing.
- A reverse-direction model was trained on gold data and run over plentiful monolingual text to manufacture silver source inputs, turning one-sided data into parallel training pairs.
- Silver pairs were filtered by length after noticing that synthetic quality drops as sentences get longer, which trades some coverage for cleaner supervision.
- The author flags that Levantine dominance in the gold data skews what the reverse model generates; when silver data comes from a model, its biases carry over into the silver corpus.

## For Mova

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

- **I1** (learn): Add a no-op control to every evaluation, such as an unchanged input tree or a default-frame guess, so that a trained component must beat it before being trusted.
- **I2** (tree): When training UD dialect-conversion rules or models, include trees already in the target convention so the converter learns to leave them alone.
- **I3** (learn): Text produced from meaning graphs by the generation stage could serve as synthetic input whose gold structure is known, adding silver training pairs for the parser.
- **I4** (learn): Filter or down-weight silver parses agreed by witness parsers by sentence length, since agreement on long sentences is less reliable.
- **I5** (learn): Track genre and source balance of the treebanks behind any teacher model, and audit the silver data it produces for the same skew.

*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](https://md.research.elamur.ai/free/en/shelves/linguistics/) · [Machine translation](https://md.research.elamur.ai/free/en/shelves/machine-translation/)

Topics: [🐪 Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/)

## Cite

Cite as: Murhaf Fares. 2024. AraT5-MSAizer: Translating Dialectal Arabic to MSA. Proceedings of the 6th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT) with Shared Tasks on Arabic LLMs Hallucination and Dialect to MSA Machine Translation @ LREC-COLING 2024.

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

```bibtex
@inproceedings{fares2024arat5,
  title = {{AraT5-MSAizer: Translating Dialectal Arabic to MSA}},
  author = {Murhaf Fares},
  booktitle = {Proceedings of the 6th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT) with Shared Tasks on Arabic LLMs Hallucination and Dialect to MSA Machine Translation @ LREC-COLING 2024},
  year = {2024},
  doi = {10.63317/26nxp6cqn8q2},
  url = {https://doi.org/10.63317/26nxp6cqn8q2}
}
```

- [← previous Analyzing Politeness in Arabic Tweets: A Preliminary Study](https://md.research.elamur.ai/free/en/papers/khalifa-2024-analyzing-politeness-arabic-tweets-preliminary/)
- [next → ASOS at OSACT6 Shared Task: Investigation of Data Augmentation in Arabic Dialect-MSA Translation](https://md.research.elamur.ai/free/en/papers/nacar-2024-asos-osact6-shared-task-investigation/)
- [↑ back to papers of 2024](https://md.research.elamur.ai/free/en/papers/2024/page/4/)

## References

23 references, 10 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Muhammad Abdul-Mageed, AbdelRahim Elmadany, Chiyu Zhang, El Moatez Billah Nagoudi, Houda Bouamor, and Nizar Habash. 2023. NADI 2023: The fourth nuanced Arabic dialect identification shared task. In Proceedings of ArabicNLP 2023, pages 600–613, Singapore (Hybrid). Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Muhammad+Abdul%2DMageed%2C+AbdelRahim+Elmadany%2C+Chiyu+Zhang%2C+El+Moatez+Billah+Nagoudi%2C+Houda+Bouamor%2C+and+Nizar+Habash%2E+2023%2E+NADI+2023%3A+The+fourth+nuanced+Arabic+dialect+identification+shared+task%2E+In+Pr)
2. Emily Bender. 2019. The# benderrule: On naming the languages we study and why it matters. The Gradient, 14. [find](https://scholar.google.com/scholar?q=Emily+Bender%2E+2019%2E+The%23+benderrule%3A+On+naming+the+languages+we+study+and+why+it+matters%2E+The+Gradient%2C+14%2E)
3. A. Bergman and Mona Diab. 2022. Towards responsible natural language annotation for the varieties of Arabic. In Findings of the Association for Computational Linguistics: ACL 2022, pages 364–371, Dublin, Ireland. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=A%2E+Bergman+and+Mona+Diab%2E+2022%2E+Towards+responsible+natural+language+annotation+for+the+varieties+of+Arabic%2E+In+Findings+of+the+Association+for+Computational+Linguistics%3A+ACL+2022%2C+pages+364%E2%80%93371%2C+Dubl)
4. Houda Bouamor, Nizar Habash, and Kemal Oflazer. 2014. A multidialectal parallel corpus of Arabic. In Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC’14), pages 1240–1245, Reykjavik, Iceland. European Language Resources Association (ELRA). [our card](https://md.research.elamur.ai/free/en/papers/bouamor-2014-multidialectal-parallel-corpus-arabic/)
5. Houda Bouamor, Nizar Habash, Mohammad Salameh, Wajdi Zaghouani, Owen Rambow, Dana Abdulrahim, Ossama Obeid, Salam Khalifa, Fadhl Eryani, Alexander Erdmann, and Kemal Oflazer. 2018. The MADAR Arabic dialect corpus and lexicon. In Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), Miyazaki, Japan. European Language Resources Association (ELRA). [find](https://scholar.google.com/scholar?q=Houda+Bouamor%2C+Nizar+Habash%2C+Mohammad+Salameh%2C+Wajdi+Zaghouani%2C+Owen+Rambow%2C+Dana+Abdulrahim%2C+Ossama+Obeid%2C+Salam+Khalifa%2C+Fadhl+Eryani%2C+Alexander+Erdmann%2C+and+Kemal+Oflazer%2E+2018%2E+The+MADAR+Arabic+di)
6. Charles A. Ferguson. 1959. Diglossia. WORD, 15(2):325–340. [find](https://scholar.google.com/scholar?q=Charles+A%2E+Ferguson%2E+1959%2E+Diglossia%2E+WORD%2C+15%282%29%3A325%E2%80%93340%2E)
7. Markus Freitag and Yaser Al-Onaizan. 2017. Beam search strategies for neural machine translation. In Proceedings of the First Workshop on Neural Machine Translation, pages 56– 60, Vancouver. Association for Computational Linguistics. [our card](https://md.research.elamur.ai/free/en/papers/freitag-2017-beam-search-strategies-neural-machine/)
8. Nizar Y Habash. 2022. Introduction to Arabic natural language processing. Springer Nature. [find](https://scholar.google.com/scholar?q=Nizar+Y+Habash%2E+2022%2E+Introduction+to+Arabic+natural+language+processing%2E+Springer+Nature%2E)
9. Vu Cong Duy Hoang, Philipp Koehn, Gholamreza Haffari, and Trevor Cohn. 2018. Iterative back-translation for neural machine translation. In Proceedings of the 2nd Workshop on Neural Machine Translation and Generation, pages 18–24, Melbourne, Australia. Association for Computational Linguistics. [our card](https://md.research.elamur.ai/free/en/papers/hoang-2018-iterative-back-translation-neural-machine/)
10. Abdullah Khered, Ingy Abdelhalim, Nadine Abdelhalim, Ahmed Soliman, and Riza Batista-Navarro. 2023. UniManc at NADI 2023 shared task: A comparison of various t5-based models for translating Arabic dialectical text to Modern Standard Arabic. In Proceedings of ArabicNLP 2023, pages 658–664, Singapore (Hybrid). Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Abdullah+Khered%2C+Ingy+Abdelhalim%2C+Nadine+Abdelhalim%2C+Ahmed+Soliman%2C+and+Riza+Batista%2DNavarro%2E+2023%2E+UniManc+at+NADI+2023+shared+task%3A+A+comparison+of+various+t5%2Dbased+models+for+translating+Arabic+dia)
11. Mateusz Krubiński, Hashem Sellat, Shadi Saleh, Adam Pospíšil, Petr Zemánek, and Pavel Pecina. 2023. Multi-parallel corpus of North Levantine Arabic. In Proceedings of ArabicNLP 2023, pages 411–417, Singapore (Hybrid). Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Mateusz+Krubi%C5%84ski%2C+Hashem+Sellat%2C+Shadi+Saleh%2C+Adam+Posp%C3%AD%C5%A1il%2C+Petr+Zem%C3%A1nek%2C+and+Pavel+Pecina%2E+2023%2E+Multi%2Dparallel+corpus+of+North+Levantine+Arabic%2E+In+Proceedings+of+ArabicNLP+2023%2C+pages+411%E2%80%93417%2C+Si)
12. Pierre Lison, Jörg Tiedemann, and Milen Kouylekov. 2018. OpenSubtitles2018: Statistical rescoring of sentence alignments in large, noisy parallel corpora. In Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), Miyazaki, Japan. European Language Resources Association (ELRA). [our card](https://md.research.elamur.ai/free/en/papers/lison-2018-opensubtitles2018-statistical-rescoring-sentence-alignments/)
13. Karima Meftouh, Salima Harrat, Salma Jamoussi, Mourad Abbas, and Kamel Smaili. 2015. Machine translation experiments on PADIC: A parallel Arabic DIalect corpus. In Proceedings of the 29th Pacific Asia Conference on Language, Information and Computation, pages 26– 34, Shanghai, China. [find](https://scholar.google.com/scholar?q=Karima+Meftouh%2C+Salima+Harrat%2C+Salma+Jamoussi%2C+Mourad+Abbas%2C+and+Kamel+Smaili%2E+2015%2E+Machine+translation+experiments+on+PADIC%3A+A+parallel+Arabic+DIalect+corpus%2E+In+Proceedings+of+the+29th+Pacific+Asia)
14. Karima Meftouh, Salima Harrat, and Kamel Smaïli. 2018. PADIC: extension and new experiments. In 7th International Conference on Advanced Technologies ICAT. [find](https://scholar.google.com/scholar?q=Karima+Meftouh%2C+Salima+Harrat%2C+and+Kamel+Sma%C3%AFli%2E+2018%2E+PADIC%3A+extension+and+new+experiments%2E+In+7th+International+Conference+on+Advanced+Technologies+ICAT%2E)
15. El Moatez Billah Nagoudi, AbdelRahim Elmadany, and Muhammad Abdul-Mageed. 2022. AraT5: Text-to-text transformers for Arabic language generation. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 628–647, Dublin, Ireland. Association for Computational Linguistics. [our card](https://md.research.elamur.ai/free/en/papers/nagoudi-2022-arat5-text-text-transformers-arabic/)
16. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. BLEU: a method for automatic evaluation of machine translation. In Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics, pages 311–318, Philadelphia, Pennsylvania, USA. Association for Computational Linguistics. [↗](https://doi.org/10.3115/1073083.1073135) [our card](https://md.research.elamur.ai/free/en/papers/papineni-2002-bleu-method-automatic-evaluation-machine/)
17. Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. 2020. Exploring the limits of transfer learning with a unified text-to-text transformer. The Journal of Machine Learning Research, 21(1):5485–5551. [our card](https://md.research.elamur.ai/free/en/papers/raffel-2019-exploring-limits-transfer-learning-unified/)
18. Ricardo Rei, José G. C. de Souza, Duarte Alves, Chrysoula Zerva, Ana C Farinha, Taisiya Glushkova, Alon Lavie, Luisa Coheur, and André F. T. Martins. 2022. COMET-22: Unbabel-IST 2022 submission for the metrics shared task. In Proceedings of the Seventh Conference on Machine Translation (WMT), pages 578–585, Abu Dhabi, United Arab Emirates (Hybrid). Association for Computational Linguistics. [our card](https://md.research.elamur.ai/free/en/papers/rei-2022-comet-22-unbabel-ist-2022/)
19. Yves Scherrer, Aleksandra Miletić, and Olli Kuparinen. 2023. The Helsinki-NLP submissions at NADI 2023 shared task: Walking the baseline. In Proceedings of ArabicNLP 2023, pages 670– 677, Singapore (Hybrid). Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Yves+Scherrer%2C+Aleksandra+Mileti%C4%87%2C+and+Olli+Kuparinen%2E+2023%2E+The+Helsinki%2DNLP+submissions+at+NADI+2023+shared+task%3A+Walking+the+baseline%2E+In+Proceedings+of+ArabicNLP+2023%2C+pages+670%E2%80%93+677%2C+Singapore+%28H)
20. Rico Sennrich, Barry Haddow, and Alexandra Birch. 2016. Improving neural machine translation models with monolingual data. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 86–96, Berlin, Germany. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Rico+Sennrich%2C+Barry+Haddow%2C+and+Alexandra+Birch%2E+2016%2E+Improving+neural+machine+translation+models+with+monolingual+data%2E+In+Proceedings+of+the+54th+Annual+Meeting+of+the+Association+for+Computationa)
21. Jörg Tiedemann. 2012. Parallel data, tools and interfaces in OPUS. In Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC’12), pages 2214–2218, Istanbul, Turkey. European Language Resources Association (ELRA). [find](https://scholar.google.com/scholar?q=J%C3%B6rg+Tiedemann%2E+2012%2E+Parallel+data%2C+tools+and+interfaces+in+OPUS%2E+In+Proceedings+of+the+Eighth+International+Conference+on+Language+Resources+and+Evaluation+%28LREC%E2%80%9912%29%2C+pages+2214%E2%80%932218%2C+Istanbul%2C+Turk)
22. Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, Rémi Louf, Morgan Funtowicz, et al. 2020. 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. [our card](https://md.research.elamur.ai/free/en/papers/wolf-2020-transformers-state-art-natural-language/)
23. Biao Zhang, Philip Williams, Ivan Titov, and Rico Sennrich. 2020. Improving massively multilingual neural machine translation and zero-shot translation. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 1628–1639, Online. Association for Computational Linguistics. [our card](https://md.research.elamur.ai/free/en/papers/zhang-2020-improving-massively-multilingual-neural-machine/)