Elamur Research⌕
☰
home / papers / 2026 / ARCADE: A City-Scale Corpus for Fine-Grained Arabic Dialect Tagging

ARCADE: A City-Scale Corpus for Fine-Grained Arabic Dialect Tagging

Paper · 2026 · Omer Nacar, Serry Sibaee, Adel Ammar et al.

Summary

Most Arabic dialect identification resources label speech only by country or broad region, so they say little about variation inside a country. ARCADE addresses this with a speech corpus labelled at the city level. The authors built a pipeline that locates public Arabic radio streams by searching online station directories, including World Radio Map and Radio Garden. It checks that each stream works and is of usable quality, then records 30-second clips at random times of day. Eleven trained native-speaker annotators gave each clip one to three judgements covering:

  • emotion,
  • speech type,
  • a keep/skip decision,
  • dialect category (MSA, dialect, mixed, not applicable),
  • their confidence.

The confidence label helps flag guests who are not local to the station's city. The final corpus has 3,790 clips (about 31.6 hours) and 6,907 annotations from 58 cities in 19 countries. The paper reports label distributions and audio-quality statistics, but it runs no model benchmarks. The resource matters because it is the first manually verified, city-granular Arabic speech set, and it enables multi-task and geolinguistic research.

Key points

  • The corpus has 3,790 unique 30-second clips (about 31.6 h) and 6,907 annotations, covering 58 cities in 19 countries, with at least 10 kept recordings per city.
  • In the comparison table, ADI17 and MGB-5 rely on weak channel-based labels, and MGB-2 and QASR are lightly supervised. ARCADE is the only city-level resource there, and its labels are manual with multiple annotations per clip.
  • The 30-second clip length follows prior work reporting above 81% dialect-ID accuracy at that length. It also makes a single speaker and a uniform style more likely.
  • Neutral emotion dominates and dialectal speech is the largest category. The data are unevenly spread across cities, and audio quality (SNR) differs by region: Moroccan stations score high, while some Sudanese and Levantine stations score low.
  • The dataset is released on Hugging Face as riotu-lab/ARCADE-full under CC BY 4.0, which the authors describe as intended for non-commercial research use. Benchmarking is left to future work.

Interesting ideas

  • I1. The city label comes cheaply from source metadata (the station's location), and human validity and confidence flags remove clips where the speaker does not match that provenance, which is a useful way to turn noisy origin labels into trusted ones.
  • I2. The authors check their manual keep/skip decisions against objective measures (SNR, silence ratio, dynamic range, spectral centroid) and find systematic differences, which is a cheap sanity test of human filtering.
  • I3. The paper warns that recording quality varies by region and can masquerade as a dialect signal, which shows how a channel artifact can become a shortcut correlated with the label.
  • I4. Alongside the two pure dialect classes (MSA and dialect), annotators can choose an explicit mixed category and a not-applicable option, so ambiguous items are recorded honestly instead of forced into one class.
  • I5. Each clip carries several independent labels (emotion, speech type, dialect, confidence), so one annotation pass supports several learning tasks and lets one attribute be separated from another.

For Mova

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

  • I1 (learn): Silver sentences labelled by their source (genre, corpus, dialect) could carry a witness-disagreement flag, so items whose content contradicts the source label are filtered out instead of trusted.
  • I2 (learn): Accepted and rejected silver parses could be compared on independent statistics such as sentence length, OOV rate, and parser margin, to confirm the agreement filter is not discarding data arbitrarily.
  • I3 (learn): Negative controls could test whether a model learns source formatting (punctuation style, line breaks) instead of syntax, by shuffling or normalising those artifacts and checking that accuracy holds.
  • I4 (tokens): A register tag on sentences (standard / nonstandard / mixed / n.a.) could tell the tagger and lemmatizer when to switch on informal-spelling handling.
  • I5 (frames): Gold examples for speech acts could also record speaker stance and annotator confidence, so the prag crate can be trained and evaluated on several pragmatic layers at once.

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

🎓Prince Sultan UniversityRiyadh · Saudi Arabia17 works · 120 citations🎓Tuwaiq AcademyRiyadh · Saudi Arabia11 works · 7 citations🎓Shaqra UniversityRiyadh · Saudi Arabia2 works · 14 citations

Also on the byline: Institut Supérieur des Études Technologiques en Communications de Tunis.

 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 worksSignal Processing · 452 works

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

Cite

Cite as Omer Nacar, Serry Sibaee, Adel Ammar, Yasser Alhabashi, Nadia Samer Sibai, Yara Farouk Ahmed, Ahmed Saud Alqusaiyer, Sulieman Mahmoud AlMahmoud, Abdulrhman Mamdoh Mukhaniq, Lubaba Raed, Sulaiman Mohammed Alatwah, Waad Nasser Alqahtani, Yousif Abdulmajeed Alnasser, Mohamed Aziz Khadraoui and Wadii Boulila. 2026. ARCADE: A City-Scale Corpus for Fine-Grained Arabic Dialect Tagging. arXiv:2601.02209.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2601.02209 · arXiv 2601.02209

@misc{nacar2026arcade,
  title = {{ARCADE: A City-Scale Corpus for Fine-Grained Arabic Dialect Tagging}},
  author = {Omer Nacar, Serry Sibaee, Adel Ammar, Yasser Alhabashi, Nadia Samer Sibai, Yara Farouk Ahmed, Ahmed Saud Alqusaiyer, Sulieman Mahmoud AlMahmoud, Abdulrhman Mamdoh Mukhaniq, Lubaba Raed, Sulaiman Mohammed Alatwah, Waad Nasser Alqahtani, Yousif Abdulmajeed Alnasser, Mohamed Aziz Khadraoui and Wadii Boulila},
  year = {2026},
  doi = {10.48550/arxiv.2601.02209},
  eprint = {2601.02209},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2601.02209}
}

References

22 references, 13 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Serry Sibaee, Omer Nacar, Yasser Al-Habashi, Adel Ammar, and Wadii Boulila. SHAMI-MT: A Syrian Arabic Dialect to Modern Standard Arabic Bidirectional Machine Translation System, 2025. ↗
  2. Jihad Zahir. IADD: An Integrated Arabic Dialect Identification Dataset. Data in Brief, 40:107777, 2021. ↗
  3. Omer Nacar, Serry Taiseer Sibaee, Samar Ahmed, Safa Ben Atitallah, Adel Ammar, Yasser Alhabashi, Abdulrahman S Al-Batati, Arwa Alsehibani, Nour Qandos, Omar Elshehy, et al. Towards inclusive arabic llms: A culturally aligned benchmark in arabic large language model evaluation. In Proceedings of the First Workshop on Language Models for Low-Resource Languages, pages 387–401, 2025.
  4. Muhammad Abdul-Mageed, Amr Keleg, AbdelRahim Elmadany, Chiyu Zhang, Injy Hamed, Walid Magdy, Houda Bouamor, and Nizar Habash. Nadi 2024: The fifth nuanced arabic dialect identification shared task. In Proceedings of the Arabic Natural Language Processing Conference, 2024. arXiv preprint arXiv:2407.04910. ↗
  5. Bashar Talafha, Hawau Olamide Toyin, Peter Sullivan, AbdelRahim A Elmadany, Abdurrahman Juma, Amirbek Djanibekov, Chiyu Zhang, Hamad Alshehhi, Hanan Aldarmaki, Mustafa Jarrar, et al. Nadi 2025: The first multidialectal arabic speech processing shared task. In Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks, pages 720–733, 2025.
  6. Suwon Shon, Ahmed Ali, Younes Samih, Hamdy Mubarak, and James Glass. ADI17: A Fine-Grained Arabic Dialect Identification Dataset. In Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 8244–8248, 2020. find
  7. Karima Meftouh, Salima Harrat, Salma Jamoussi, Mourad Abbas, and Kamel Smaili. Machine translation experiments on PADIC: A parallel Arabic DIalect corpus. In Hai Zhao, editor, Proceedings of the 29th Pacific Asia Conference on Language, Information and Computation, pages 26–34, Shanghai, China, October 2015.
  8. Amal Nayouf, Tymaa Hammouda, Mustafa Jarrar, Fadi Zaraket, and Mohamad-Bassam Kurdy. Nâbra: Syrian Arabic dialects with morphological annotations. In Hassan Sawaf, Samhaa El-Beltagy, Wajdi Zaghouani, Walid Magdy, Ahmed Abdelali, Nadi Tomeh, Ibrahim Abu Farha, Nizar Habash, Salam Khalifa, Amr Keleg, Hatem Haddad, Imed Zitouni, Khalil Mrini, and Rawan Almatham, editors, Proceedings of ArabicNLP 2023, pages 12–23, Singapore (Hybrid), December 2023. Association for Computational Linguistics. ↗
  9. Salima Medhaffar, Fethi Bougares, Yannick Estève, and Lamia Hadrich-Belguith. Sentiment analysis of Tunisian dialects: Linguistic ressources and experiments. In Nizar Habash, Mona Diab, Kareem Darwish, Wassim El-Hajj, Hend Al-Khalifa, Houda Bouamor, Nadi Tomeh, Mahmoud El-Haj, and Wajdi Zaghouani, editors, Proceedings of the Third Arabic Natural Language Processing Workshop, pages 55–61, Valencia, Spain, April 2017. Association for Computational Linguistics.
  10. Omer Nacar, Anis Koubaa, Serry Sibaee, Yasser Al-Habashi, Adel Ammar, and Wadii Boulila. GATE: General Arabic text embedding for enhanced semantic textual similarity with matryoshka representation learning and hybrid loss training. arXiv preprint arXiv:2505.24581, 2025. ↗
  11. Alexandra Canavan, George Zipperlen, and David Graff. CALLHOME egyptian arabic speech. Linguistic Data Consortium, 1997. LDC97S45. find
  12. Peter Bell, Mark JF Gales, Thomas Hain, Jonathan Kilgour, Pierre Lanchantin, Xunying Liu, Andrew McParland, Steve Renals, Oscar Saz, Mirjam Wester, et al. The MGB challenge: Evaluating multi-genre broadcast media recognition. In 2015 IEEE Workshop on Automatic Speech Recognition and Understanding (ASRU), pages 687–693. IEEE, 2015. find
  13. Ahmed Ali, Peter Bell, James Glass, Yacine Messaoui, Hamdy Mubarak, Steve Renals, and Yifan Zhang. The MGB-2 challenge: Arabic multi-dialect broadcast media recognition. In 2016 IEEE Spoken Language Technology Workshop (SLT), pages 279–284, 2016.
  14. Ahmed Ali, Stephan Vogel, and Steve Renals. Speech recognition challenge in the wild: Arabic MGB-3. In 2017 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU), pages 316–322, 2017.
  15. Ahmed Ali, Suwon Shon, Younes Samih, Hamdy Mubarak, Ahmed Abdelali, James Glass, Steve Renals, and Khalid Choukri. The mgb-5 challenge: Recognition and dialect identification of dialectal arabic speech. In 2019 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU), pages 1026–1033. IEEE, 2019. find
  16. Hamdy Mubarak, Amir Hussein, Shammur Absar Chowdhury, and Ahmed Ali. QASR: QCRI aljazeera speech resource a large scale annotated Arabic speech corpus. In Chengqing Zong, Fei Xia, Wenjie Li, and Roberto Navigli, editors, Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 2274–2285, Online, August 2021. Association for Computational Linguistics. ↗
  17. Fadi Biadsy, Julia Hirschberg, and Nizar Habash. Spoken arabic dialect identification using phonotactic modeling. In Proceedings of the EACL 2009 Workshop on Computational Approaches to Semitic Languages, Semitic ’09, page 53–61, USA, 2009. Association for Computational Linguistics.
  18. Nour Qandos, Serry Sibaee, Samar Ahmad, Omer Nacar, Adel Ammar, Wadii Boulila, and Yasser Alhabashi. Anplers at iqraeval shared task: Adapting whisper-large-v3 as speech-to-phoneme for qur’anic recitation mispronunciation detection. In Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks, pages 464–468, 2025. — ANPLers at IqraEval Shared task: Adapting Whisper-large-v3 as Speech-to-Phoneme for Qur'anic Recitation Mispronunciation Detection
  19. Kees Versteegh. Arabic language. Edinburgh University Press, 2014. find
  20. Raya Kalaldeh. Acoustic analysis of modern standard arabic vowels by jordanian speakers. International Journal of Arabic-English Studies, 18(1), 2018. ↗
  21. Emad Abdul-Latif. Bassiouney, reem (2009) arabic sociolinguistics. edinburgh university press: Edinburgh; pp. 311+ i-xiv. International Journal of Arabic Linguistics, 2(2):154–158, 2016. find
  22. Alvan R Feinstein and Domenic V Cicchetti. High agreement but low kappa: I. the problems of two paradoxes. Journal of clinical epidemiology, 43(6):543–549, 1990. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/nacar-2026-arcade-city-scale-corpus-fine/