Elamur Research⌕
☰
home / papers / 2026 / Tarab: A Multi-Dialect Corpus of Arabic Lyrics and Poetry

Tarab: A Multi-Dialect Corpus of Arabic Lyrics and Poetry

Paper · 2026 · Mo El-Haj

Summary

Arabic NLP corpora mostly come from news, encyclopaedic text, or social media. Creative language is largely missing: song lyrics and poetry, with their rhyme, repetition, dialect-heavy morphology and non-standard spelling. This paper presents the Tarab Corpus, which combines lyrics and poems in a single verse-level resource. It covers Classical Arabic, MSA and six regional dialect groups, spanning roughly fourteen centuries from the Pre-Islamic period to today.

The data comes from three sources: an open Kaggle poetry collection, the earlier Habibi lyrics corpus, and newly crawled lyric pages that were collected only where robots.txt allows it. Everything was mapped to one shared schema with minimal preprocessing, so that dialectal orthography is preserved. Each verse is tagged with variety, origin (a modern country or a historical era) and genre, and it stays linked to its parent work.

The author reports coverage statistics, lexical contrasts between genres, and baseline analyses for variety identification and genre separation. The result is described as the largest open Arabic creative-text corpus. It is useful for dialect modelling, authorship work, stylistics and diachronic study.

Key points

  • The corpus has 2,557,311 verses and 13,509,336 tokens, drawn from 89,166 works by 2,598 creators (2,060 singers and 538 poets) linked to 28 countries and several historical eras.
  • The genre split is 34,239 songs (1,170,028 verses) and 54,927 poems (1,387,283 verses). Compared with the Habibi corpus (527,896 verses), this is about 4.8 times larger.
  • Varieties covered are Classical Arabic, MSA, Egyptian, Gulf, Levantine, Iraqi, Sudanese and Maghrebi.
  • Schema fields include art_id, artist_id, artist_name, art_title, writer, composer, verse_order, verse_lyrics, origin, dialect and type. A corpus_version flag marks items inherited from Habibi.
  • Deduplication uses a composite key (art_id, artist_id, verse_order). The corpus is released on HuggingFace (drelhaj/Tarab) with a research-use statement and a takedown route.
  • Stated limitations:
    • Era-level dating for heritage poetry is coarse.
    • Dialect labels do not capture mixing within a single text.
    • The corpus has no prosodic or musical features.

Interesting ideas

  • I1. Musical style, artist origin and dialect are stored as independent fields rather than merged into one genre tag, so a Tunisian singer performing in a Gulf style stays representable.
  • I2. Preprocessing is deliberately light: spellings that carry regional signal are kept, and no stemming or stopword removal is applied.
  • I3. Each verse is a separate row, yet a stable work ID plus an order field lets the full text be rebuilt. This gives fine-grained analysis without losing document context.
  • I4. A version flag marks which items came from an earlier release, which allows controlled comparisons between the old and new parts of the data.
  • I5. Contributors are ranked by a score that divides each dimension (works, words, verses) by its maximum before summing, so no single axis dominates.

For Mova

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

  • I1 (tree): When converting UD dialects, record source treebank, register and annotation convention as separate attributes, so conversion errors can be traced to the right cause.
  • I2 (tokens): Keep the raw surface form next to the normalised lemma in crate en, so later stages can still use spelling variation as evidence.
  • I3 (world): Store sentence-level analyses with document IDs and positions, so the world builder can reassemble event order and track entities across sentences.
  • I4 (learn): Tag silver data with provenance and generation round, so evaluation can isolate whether gains come from newly added material.
  • I5 (learn): A similar max-normalised composite could rank candidate training texts or witness parsers across several metrics without one metric swamping the others.

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, topics and concepts

📚 Shelves 5The 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 worksLiterature and Literary Theory · 519 works

🧭 Topics 1Research subjects and languages the work is about.Semitic languages · 2,908 works

Cite

Cite as Mo El-Haj. 2026. Tarab: A Multi-Dialect Corpus of Arabic Lyrics and Poetry. Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.abjadnlp-1.5

@inproceedings{elhaj2026tarab,
  title = {{Tarab: A Multi-Dialect Corpus of Arabic Lyrics and Poetry}},
  author = {Mo El-Haj},
  booktitle = {Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script},
  year = {2026},
  doi = {10.18653/v1/2026.abjadnlp-1.5},
  url = {https://doi.org/10.18653/v1/2026.abjadnlp-1.5}
}

References

25 references, 14 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Kareem Darwish. 2013. Arabizi detection and conversion to arabic. arXiv preprint arXiv:1306.6755. ↗
  2. Ahmed Abdelali, Hamdy Mubarak, Younes Samih, Sabit Hassan, and Kareem Darwish. 2020. Arabic dialect identification in the wild. arXiv preprint arXiv:2005.06557. ↗
  3. Mahmoud El-Haj. 2020. Habibi-a multi dialect multi national arabic song lyrics corpus. In Proceedings of the Twelfth Language Resources and Evaluation Conference, pages 1318–1326.
  4. Maged S Al-Shaibani, Zaid Alyafeai, and Irfan Ahmad. 2020. Meter classification of arabic poems using deep bidirectional recurrent neural networks. Pattern Recognition Letters, 136:1–7. ↗
  5. Mahmoud El-Haj and Rim Koulali. 2013. Kalimat a multipurpose arabic corpus. Culture, 2:1–359. find
  6. Ali Alhazmi, Rohana Mahmud, Norisma Idris, Mohamed Elhag Mohamed Abo, and Christopher Ifeanyi Eke. 2024. Code-mixing unveiled: Enhancing the hate speech detection in arabic dialect tweets using machine learning models. Plos one, 19(7):e0305657. ↗
  7. Mahmoud El-Haj, Paul Rayson, and Mariam Aboelezz. 2018. Arabic dialect identification in the context of bivalency and code-switching. In Proceedings of the 11th International Conference on Language Resources and Evaluation, Miyazaki, Japan., pages 3622–3627. European Language Resources Association.
  8. Wissam Antoun, Fady Baly, and Hazem Hajj. 2020. Arabert: Transformer-based model for arabic language understanding. arXiv preprint arXiv:2003.00104. ↗
  9. Mo El-Haj and Saad Ezzini. 2024. The multilingual corpus of world’s constitutions (mcwc). In 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, pages 57–66. find
  10. Mohamed Attia, Mohsen Rashwan, Ahmed Ragheb, Mohamed Al-Badrashiny, Husein Al-Basoumy, and Sherif Abdou. 2008. A compact arabic lexical semantics language resource based on the theory of semantic fields. In International Conference on Natural Language Processing, pages 65–76. Springer.
  11. Nizar Habash, Houda Bouamor, and Kemal Oflazer. 2014. A multidialectal parallel corpus of Arabic. Carnegie Mellon University.
  12. Yonatan Belinkov, Alexander Magidow, Maxim Romanov, Avi Shmidman, and Moshe Koppel. 2016. Shamela: A large-scale historical arabic corpus. arXiv preprint arXiv:1612.08989. ↗
  13. Nizar Y Habash. 2010. Introduction to Arabic natural language processing. Morgan & Claypool Publishers. ↗
  14. Piotr Bojanowski, Edouard Grave, Armand Joulin, and Tomas Mikolov. 2017. Enriching word vectors with subword information. Transactions of the association for computational linguistics, 5:135–146. ↗
  15. AM Mutawa and Ayshah Alrumaih. 2025. Determining the meter of classical arabic poetry using deep learning: a performance analysis. Frontiers in Artificial Intelligence, 8:1523336. ↗
  16. Houda Bouamor, Nizar Habash, Mohammad Salameh, Wajdi Zaghouani, Owen Rambow, Dana Abdulrahim, Ossama Obeid, Salam Khalifa, Fadhl Eryani, Alexander Erdmann, and 1 others. 2018. The madar arabic dialect corpus and lexicon. In Proceedings of the eleventh international conference on language resources and evaluation (LREC 2018). find
  17. Ossama Obeid, Nasser Zalmout, Salam Khalifa, Dima Taji, Mai Oudah, Bashar Alhafni, Go Inoue, Fadhl Eryani, Alexander Erdmann, and Nizar Habash. 2020. Camel tools: An open source python toolkit for arabic natural language processing. In Proceedings of the twelfth language resources and evaluation conference, pages 7022–7032.
  18. Mohamed Boudchiche, Azzeddine Mazroui, Mohamed Ould Abdallahi Ould Bebah, Abdelhak Lakhouaja, and Abderrahim Boudlal. 2017. Alkhalil morpho sys 2: A robust arabic morphosyntactic analyzer. Journal of King Saud University-Computer and Information Sciences, 29(2):141–146. find
  19. Abdenour Padillo-Saoud. 2019. Open Islamicate Texts Initiative (OpenITI), 2016 [Reseña]. Universidad Nacional de Educación a Distancia (España). find
  20. Robert Parker, David Graff, Ke Chen, Junbo Kong, and Kazuaki Maeda. 2011. Arabic gigaword fifth edition. Linguistic Data Consortium, Catalog Number LDC2011T11. ISBN 1-58563-595-2. ↗
  21. Abderrahim Boudlal, Abdelhak Lakhouaja, Azzeddine Mazroui, Abdelouafi Meziane, MOAO Bebah, and Mostafa Shoul. 2010. Alkhalil morpho sys1: A morphosyntactic analysis system for arabic texts. In International Arab conference on information technology, pages 1–6. Elsevier Science Inc New York, NY. find
  22. Faisal Qarah. 2024. Arapoembert: A pretrained language model for arabic poetry analysis. arXiv preprint arXiv:2403.12392. ↗
  23. Omar Zaidan and Chris Callison-Burch. 2011. The arabic online commentary dataset: an annotated dataset of informal arabic with high dialectal content. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies, pages 37–41.
  24. Omar F Zaidan and Chris Callison-Burch. 2014. Arabic dialect identification. Computational Linguistics, 40(1):171–202. ↗
  25. Imad Zeroual, Dirk Goldhahn, Thomas Eckart, and Abdelhak Lakhouaja. 2019. 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. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/haj-2026-tarab-multi-dialect-corpus-arabic/