Elamur Research⌕
☰
home / papers / 2023 / Machine Translation of Omani Arabic Dialect from Social Media

Machine Translation of Omani Arabic Dialect from Social Media

Paper · 2023 · Khoula Al-Kharusi and Abdurahman AAlAbdulsalam

Summary

Omani Arabic (OA) dialects have almost no translation resources. The one public source with Omani content, the Muscat portion of MADAR, reads to a native speaker as close to MSA with some Bedouin traits, not the sedentary (Hadari) speech common in northern Oman. The authors built a public OA–English parallel corpus of roughly 2,900 posts collected from X (Twitter) and translated into English without professional translators. They used it to benchmark Google Translate, Microsoft Translation and Marian NMT. They manually classified the errors of Google and Marian using a linguistic taxonomy and found that choosing the wrong word sense was the most frequent failure. They then fine-tuned the open Helsinki-NLP Arabic–English Marian model on the Omani training split. BLEU rose by 9.88 points, which brought the open model close to the commercial engines. The work shows that a small, authentic dialect corpus plus transfer from a related standard-language model can close much of the gap for an under-documented variety.

Key points

  • Corpus split: 2,034 training, 436 validation and 437 test parallel posts (70/15/15). The dataset is released on GitHub (khoula-k/OmaniArabicTranslation).
  • Test results:
    • Google: BLEU 34.98, chrF 60.55
    • Microsoft: BLEU 34.26, chrF 61.28
    • Marian NMT: BLEU 24.22, chrF 49.02
    • Fine-tuned Marian: BLEU 34.11, chrF 59.81
  • The parent model was Helsinki-NLP/opus-mt-tc-big-ar-en, a 6+6-layer transformer trained on 126.6M OPUS Arabic–English pairs. It was fine-tuned with the Hugging Face Seq2Seq trainer and AdamW for five epochs, the point after which validation loss began to rise.
  • In the error taxonomy, wrong word-sense selection was the dominant error type for both evaluated systems.
  • OA vocabulary contains loans from Hindi, Persian, English and Portuguese (e.g., sekal 'bicycle', mez 'table').

Interesting ideas

  • I1. A native speaker checked whether the only existing benchmark really reflects the target variety and found it skewed toward the standard language, which shows that benchmark labels can mislead.
  • I2. Instead of reporting only a metric, the authors sorted system outputs into a linguistic error taxonomy, which revealed that sense choice, not grammar, was the main weakness.
  • I3. Fine-tuning a model built for a closely related standard variety on a few thousand in-domain examples recovered most of the gap, and the authors stress that closeness between parent and child matters.
  • I4. The authors compared the open model before and after adaptation against strong commercial systems used as reference ceilings.
  • I5. Reporting a character-level metric (chrF) next to word-level BLEU gives a second view of output quality for text with irregular spelling.

For Mova

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

  • I1 (learn): audit our evaluation sets and silver data for register mismatch, such as edited prose versus informal web English, before trusting scores on a variety.
  • I2 (back): tag generation and translation failures by linguistic category (word sense, agreement, order) so repair work targets the dominant class.
  • I3 (tree): adapt the gparse parser trained on standard UD English treebanks to informal or dialectal English with a small hand-checked in-domain set, rather than training from scratch.
  • I4 (learn): always report an unadapted baseline and an external reference beside the trained model, so improvements are read against fixed anchors.
  • I5 (back): score the text → graph → text compressor with both word- and character-level overlap to separate spelling-level drift from wording changes.

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

🎓Sultan Qaboos UniversityMuscat · Oman4 works · 12 citations

 Fields, topics and concepts

📚 Shelves 4The 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 works

🧭 Topics 2Research subjects and languages the work is about.Machine translation · 1,087 worksSemitic languages · 2,908 works

Cite

Cite as Khoula Al-Kharusi and Abdurahman AAlAbdulsalam. 2023. Machine Translation of Omani Arabic Dialect from Social Media. Proceedings of ArabicNLP 2023.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2023.arabicnlp-1.24

@inproceedings{alkharusi2023machine,
  title = {{Machine Translation of Omani Arabic Dialect from Social Media}},
  author = {Khoula Al-Kharusi and Abdurahman AAlAbdulsalam},
  booktitle = {Proceedings of ArabicNLP 2023},
  year = {2023},
  doi = {10.18653/v1/2023.arabicnlp-1.24},
  url = {https://doi.org/10.18653/v1/2023.arabicnlp-1.24}
}

References

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

  1. Rashid Al-Balushi. 2017. Omani arabic: More than a dialect. Macrolinguistics, 4:80–125. ↗
  2. Mohammed Al-Qaraghuli, Gheith Abandah, and Ashraf Suyyagh. 2021. Correcting arabic soft spelling mistakes using transformers. pages 146–151. IEEE. find
  3. Laith H. Baniata, Isaac K.E. Ampomah, and Seyoung Park. 2021. A transformer-based neural machine translation model for arabic dialects that utilizes subword units. Sensors, 21. ↗
  4. Houda Bouamor, Nizar Habash, and Kemal Oflazer. 2014. A multidialectal parallel corpus of arabic. Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC’14), pages 1240–1245.
  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. European Language Resources Association (ELRA). find
  6. Hani Elgabou and Dimitar Kazakov. 2017. Building dialectal Arabic corpora. In Proceedings of the Workshop Human-Informed Translation and Interpreting Technology, pages 52–57, Varna, Bulgaria. Association for Computational Linguistics, Shoumen, Bulgaria. find
  7. Ali Farghaly and Khaled Shaalan. 2009. Arabic natural language processing. ACM Transactions on Asian Language Information Processing, 8:1–22. ↗
  8. Dario Franceschini, Chiara Canton, Ivan Simonini, Armin Schweinfurth, Adelheid Glott, Sebastian Stüker, Thai-Son Nguyen, Felix Schneider, Thanh-Le Ha, Alex Waibel, Barry Haddow, Philip Williams, Rico Sennrich, Ondřej Bojar, Sangeet Sagar, Dominik Macháček, and Otakar Smrz. 2020. Removing european language barriers with innovative machine translation technology. find
  9. Vikrant Goyal, Sourav Kumar, and Dipti Misra Sharma. 2020. Efficient neural machine translation for low-resource languages via exploiting related languages. find
  10. Salima Harrat, Karima Meftouh, and Kamel Smaili. 2019. Machine translation for arabic dialects (survey). Information Processing and Management, 56:262–273. find
  11. Marcin Junczys-Dowmunt, Roman Grundkiewicz, Tomasz Dwojak, Hieu Hoang, Kenneth Heafield, Tom Neckermann, Frank Seide, Ulrich Germann, Alham Fikri Aji, Nikolay Bogoychev, André F. T. Martins, and Alexandra Birch. 2018. Marian: Fast neural machine translation in C++. In Proceedings of ACL 2018, System Demonstrations, pages 116–121, Melbourne, Australia. Association for Computational Linguistics.
  12. Guillaume Klein, Yoon Kim, Yuntian Deng, Jean Senellart, and Alexander Rush. 2017. OpenNMT: Open-source toolkit for neural machine translation. In Proceedings of ACL 2017, System Demonstrations, pages 67–72, Vancouver, Canada. Association for Computational Linguistics.
  13. Karima Meftouh, Salma Jamoussi, Mourad Abbas, Crstdla ‡ Algiers, and Algeria Kamel Smaili. 2015. Machine translation experiments on padic: A parallel arabic dialect corpus.
  14. Hala Al Nabhani. 2011. Language and identity in oman through the voice of local radio broadcasters. find
  15. Hassan Sajjad, Ahmed Abdelali, Nadir Durrani, and Fahim Dalvi. 2020. Arabench: Benchmarking dialectal arabic-english machine translation. pages 5094–5107. International Committee on Computational Linguistics. find
  16. Mohammad Salameh, Houda Bouamor, and Nizar Habash. 2018. Fine-grained Arabic dialect identification. In Proceedings of the 27th International Conference on Computational Linguistics, pages 1332–1344, Santa Fe, New Mexico, USA. Association for Computational Linguistics.
  17. Barret Zoph, Deniz Yuret, Jonathan May, and Kevin Knight. 2016. Transfer learning for low-resource neural machine translation. find
  18. Ângela Costa, Wang Ling, Tiago Luís, Rui Correia, and Luísa Coheur. 2015. A linguistically motivated taxonomy for machine translation error analysis. Machine Translation, 29:127–161. ↗
  19. Wael Salloum and Nizar Habash. 2011. Dialectal to standard arabic paraphrasing to improve arabic-english statistical machine translation. pages 10–21. Association for Computational Linguistics.
  20. Wael Salloum and Nizar Habash. 2012. Elissa: A dialectal to standard arabic machine translation system.
  21. Wael Salloum and Nizar Habash. 2013. Dialectal arabic to english machine translation: Pivoting through modern standard arabic. NAACL-HLT.
  22. Wael Salloum and Nizar Habash. 2014. Adam: Analyzer for dialectal arabic morphology. Journal of King Saud University - Computer and Information Sciences, 26:372–378. Special Issue on Arabic NLP. ↗
  23. Hassan Sawaf. 2010. Arabic dialect handling in hybrid machine translation. Proceedings of the Conference of the Association for Machine Translation in the Americas (AMTA).
  24. Genevieve A. Schmitt. 2020. Relevance of Arabic Dialects: A Brief Discussion, pages 1383–1398. Springer International Publishing, Cham. find
  25. Mohamed Ali Sghaier and Mounir Zrigui. 2020. Rule-based machine translation from tunisian dialect to modern standard arabic. Procedia Computer Science, 176:310–319. Knowledge-Based and Intelligent Information Engineering Systems: Proceedings of the 24th International Conference KES2020. find
  26. Toshiyuki Takezawa, Genichiro Kikui, Masahide Mizushima, and Eiichiro Sumita. 2007. Multilingual spoken language corpus development for communication research. find
  27. Jörg Tiedemann. 2020. The tatoeba translation challenge – realistic data sets for low resource and multilingual mt.
  28. Jörg Tiedemann and Santhosh Thottingal. 2020. Opus-mt – building open translation services for the world. find
  29. David Vilar, Jia Xu, Luis Fernando D’haro, and Hermann Ney. 2006. Error analysis of statistical machine translation output.
  30. Rabih Zbib, Erika Malchiodi, Jacob Devlin, David Stallard, Spyros Matsoukas, Richard Schwartz, John Makhoul, Omar F Zaidan, and Chris Callison-Burch. 2012. Machine translation of arabic dialects.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/kharusi-2023-machine-translation-omani-arabic-dialect/