Elamur Research⌕
☰
home / papers / 2019 / ADIDA: Automatic Dialect Identification for Arabic

ADIDA: Automatic Dialect Identification for Arabic

Paper · 2019 · Ossama Obeid, Mohammad Salameh, Houda Bouamor et al.

Summary

Written Arabic dialects are hard to tell apart automatically. They share one script and much vocabulary, and writers often mix dialect with Modern Standard Arabic (MSA). Most earlier dialect identification worked only at the level of regions or countries. This demo paper presents ADIDA, a public web tool built on the fine-grained classifier of Salameh et al. (2018). That classifier assigns any Arabic input, from a single word to a full sentence, a probability for each of 25 Arab cities plus MSA. It is a Multinomial Naive Bayes model trained on the parallel MADAR corpus, using TF-IDF-weighted word and character n-gram features together with language model scores. The authors' main contribution is the visualization layer, which spreads the 26-value output vector over a map of the Arab world, rendered as individual points or as a heat map. This makes it easier to spot regional similarities and confusions than reading a flat list of scores. On test sentences averaging 7 words, the model picks the exact city 67.9% of the time.

Key points

  • The classifier covers 26 classes: 25 cities from Rabat to Muscat, grouped into Maghreb, Nile Basin, Levant, Gulf and Yemen, plus MSA. It is trained with scikit-learn on the MADAR corpus.
  • City-level accuracy is 67.9% on sentences of about 7 words. An oracle study in the earlier paper suggests accuracy can exceed 90% with 16-word inputs.
  • The interface has two display modes:
    • The point map shows only cities whose probability is above 0.1%.
    • The heat map places a Gaussian on every city, so neighbouring cities blend into regional "presence."
    • MSA appears in neither map because it has no location. The sidebar lists the top five cities, MSA and a leftover "Other" mass.
  • In the paper's example, an Egyptian song verse gets Cairo confidence of 99.5% with 11 words, 92.5% with 4 words and 20.4% with 1 word. With one word, most of the probability mass is spread over unlisted cities.
  • The tool is built with Flask on the back-end and Vue.js, Leaflet, Mapbox and heatmap.js on the front-end. The data were released through the MADAR shared task (2019).

Interesting ideas

  • I1. Placing a classifier's flat probability vector onto a meaningful structure (here, geography) exposes patterns of confusion between related classes that a ranked list hides.
  • I2. Cutting the same input to shorter and shorter prefixes shows how confidence and probability spread depend on how much evidence the model has, which serves as a cheap sanity probe.
  • I3. The display always reports the top candidates together with an explicit remainder bucket ("Other"), and a class with no natural position (MSA) is shown separately rather than forced into the picture.
  • I4. The authors explicitly warn that the heat-map blending between cities is only a visual aid, not a claim about the places in between.
  • I5. The authors plan a feedback mode in which users rate outputs, turning a demo into a data-collection channel.

For Mova

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

  • I1 (tree): Head-attachment or label distributions from gparse could be drawn over the sentence itself, making near-miss attachments between neighbouring words visible while debugging.
  • I2 (learn): Truncated-input runs can act as a control: if confidence in a frame or parse barely drops when the key words are removed, the model is likely relying on spurious cues.
  • I3 (answer): Readers could report their top answers plus the unassigned confidence mass, and keep "no grounded answer" as a separate outcome instead of hiding it among the real candidates.
  • I4 (world): When displaying inferred states or entity relations, Mova should mark which links are interpolated for display and which are actually derived from the text, keeping every step explainable.
  • I5 (learn): A demo front-end for Mova parses or answers could gather human quality judgments that then serve as extra signal next to the silver data agreed on by witness parsers.

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

🎓Carnegie Mellon UniversityPittsburgh · United States539 works · 179,533 citations🎓New York UniversityNew York · United States201 works · 100,180 citations🎓New York University Abu DhabiAbu Dhabi · United Arab Emirates122 works · 4,602 citations🎓Carnegie Mellon University QatarDoha · Qatar47 works · 2,773 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 1Research subjects and languages the work is about.Semitic languages · 2,908 works

Cite

Cite as Ossama Obeid, Mohammad Salameh, Houda Bouamor and Nizar Habash. 2019. ADIDA: Automatic Dialect Identification for Arabic. Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics (Demonstrations).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/n19-4002

@inproceedings{obeid2019adida,
  title = {{ADIDA: Automatic Dialect Identification for Arabic}},
  author = {Ossama Obeid, Mohammad Salameh, Houda Bouamor and Nizar Habash},
  booktitle = {Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics (Demonstrations)},
  year = {2019},
  doi = {10.18653/v1/n19-4002},
  url = {https://doi.org/10.18653/v1/n19-4002}
}

References

31 references, 17 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, Hassan Alhuzali, and Mohamed Elaraby. 2018. You tweet what you speak: A city-level dataset of Arabic dialects. In Proceedings of the Language Resources and Evaluation Conference (LREC), Miyazaki, Japan.
  2. Abdel-Rahman Abu-Melhim. 1991. Code-switching and linguistic accommodation in Arabic. In Proceedings of the Annual Symposium on Arabic Linguistics, volume 80, pages 231–250. ↗
  3. Mahmoud Al-Ayyoub, Aya Nuseir, Kholoud Alsmearat, Yaser Jararweh, and Brij Gupta. 2017. Deep learning for Arabic nlp: A survey. Journal of Computational Science. ↗
  4. Faisal Al-Shargi and Owen Rambow. 2015. Diwan: A dialectal word annotation tool for Arabic. In Proceedings of the Workshop for Arabic Natural Language Processing (WANLP), pages 49–58, Beijing, China.
  5. Reem Bassiouney. 2009. Arabic Sociolinguistics: Topics in Diglossia, Gender, Identity, and Politics. Georgetown University Press. find
  6. Yonatan Belinkov and James Glass. 2016. A Character-level Convolutional Neural Network for Distinguishing Similar Languages and Dialects. In Proceedings of the Workshop on NLP for Similar Languages, Varieties and Dialects (VarDial), pages 145–152, Osaka, Japan. ↗
  7. Fadi Biadsy, Julia Hirschberg, and Nizar Habash. 2009. Spoken Arabic Dialect Identification Using Phonotactic Modeling. In Proceedings of the Workshop on Computational Approaches to Semitic Languages (CASL), pages 53–61, Athens, Greece.
  8. Houda Bouamor, Nizar Habash, Mohammad Salameh, Wajdi Zaghouani, Owen Rambow, Dana Abdulrahim, Ossama Obeid, Salam Khalifa, Fadhl Eryani, Alexander Erdmann, and Kemal Oflazer. 2018. find
  9. The Abdulhadi Shoufan and Sumaya Al-Ameri. 2015. Natural language processing for dialectical Arabic: A survey. In Proceedings of the Workshop for Arabic Natural Language Processing (WANLP), page 36, Beijing, China.
  10. Christoph Tillmann, Saab Mansour, and Yaser Al-Onaizan. 2014. Improved Sentence-Level Arabic Dialect Classification. In Proceedings of the Workshop on Applying NLP Tools to Similar Languages, Varieties and Dialects, pages 110–119, Dublin, Ireland.
  11. Bert Vaux and Scott Golder. 2003. The Harvard Dialect Survey. Cambridge, MA: Harvard University Linguistics Department. find
  12. Janet CE Watson. 2007. The Phonology and Morphology of Arabic. Oxford University Press. find
  13. David Joseph Wrisley. 2016. Linguistic Landscapes of Beirut Project. http://llbeirut.org. ↗
  14. David Joseph Wrisley. 2017. Locating medieval French, or why we collect and visualize the geographic information of texts. Speculum, 92(S1):S145–S169. ↗
  15. David Joseph Wrisley. 2019. ”Aggregate map.” Visualizing Medieval Places. http://vmp.djwrisley.com/map/. ↗
  16. Omar Zaidan and Chris Callison-Burch. 2014. Arabic dialect identification. Computational Linguistics, 40(1):171–202. ↗
  17. Omar F 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 Conference of the Association for Computational Linguistics (ACL), pages 37–41.
  18. Marcos Zampieri, Shervin Malmasi, Nikola Ljubešić, Preslav Nakov, Ahmed Ali, Jörg Tiedemann, Yves Scherrer, and Noëmi Aepli. 2017. Findings of the VarDial Evaluation Campaign 2017. In Proceedings of the Workshop on NLP for Similar Languages, Varieties and Dialects (VarDial), pages 1–15, Valencia, Spain.
  19. Marcos Zampieri, Shervin Malmasi, Preslav Nakov, Ahmed Ali, Suwon Shon, James Glass, Yves Scherrer, Tanja Samardžić, Nikola Ljubešić, Jörg Tiedemann, Chris van der Lee, Stefan Grondelaers, Nelleke Oostdijk, Antal van den Bosch, Ritesh Kumar, Bornini Lahiri, and Mayank Jain. 2018. Language identification and morphosyntactic tagging: The second VarDial evaluation campaign. In Proceedings of the Fifth Workshop on NLP for Similar Languages, Varieties and Dialects (VarDial), Santa Fe, USA.
  20. 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. In Proceedings of the Conference of the North American Chapter of the Association for Computational Linguistics (NAACL), pages 49–59, Montréal, Canada.
  21. Fei Huang. 2015. Improved Arabic Dialect Classification with Social Media Data. In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 2118–2126, Lisbon, Portugal.
  22. Josh Katz and Wilson Andrews. 2013. How Yall, Youse and You Guys Talk. https://www.nytimes.com/interactive/ 2014/upshot/dialect-quiz-map.html. ↗
  23. Salam Khalifa, Nizar Habash, Dana Abdulrahim, and Sara Hassan. 2016. A Large Scale Corpus of Gulf Arabic. In Proceedings of the Language Resources and Evaluation Conference (LREC), Portorož, Slovenia.
  24. Shervin Malmasi, Marcos Zampieri, Nikola Ljubešić, Preslav Nakov, Ahmed Ali, and Jörg Tiedemann. 2016. Discriminating between Similar Languages and Arabic Dialect Identification: A Report on the Third DSL Shared Task. In Proceedings of the Workshop on NLP for Similar Languages, Varieties and Dialects (VarDial), pages 1–14, Osaka, Japan.
  25. Elise Michon, Minh Quang Pham, Josep Crego, and Jean Senellart. 2018. Neural network architectures for Arabic dialect identification. In Proceedings of the Fifth Workshop on NLP for Similar Languages, Varieties and Dialects (VarDial 2018), pages 128– 136.
  26. Ossama Obeid, Salam Khalifa, Nizar Habash, Houda Bouamor, Wajdi Zaghouani, and Kemal Oflazer. 2018. MADARi: A Web Interface for Joint Arabic Morphological Annotation and Spelling Correction. In Proceedings of the Language Resources and Evaluation Conference (LREC), Miyazaki, Japan.
  27. Fabian Pedregosa, Gaël Varoquaux, Alexandre Gramfort, Vincent Michel, Bertrand Thirion, Olivier Grisel, Mathieu Blondel, Peter Prettenhofer, Ron Weiss, Vincent Dubourg, Jake Vanderplas, Alexandre Passos, David Cournapeau, Matthieu Brucher, Matthieu Perrot, and Édouard Duchesnay. 2011. Scikit-learn: Machine learning in python. Journal of Machine Learning Research, 12:2825–2830. ↗
  28. Fatiha Sadat, Farnazeh Kazemi, and Atefeh Farzindar. 2014. Automatic Identification of Arabic Dialects in Social Media. In Proceedings of the Workshop on Natural Language Processing for Social Media (SocialNLP), pages 22–27, Dublin, Ireland. find
  29. Hassan Sajjad, Kareem Darwish, and Yonatan Belinkov. 2013. Translating dialectal Arabic to English. In Proceedings of the Conference of the Association for Computational Linguistics (ACL), pages 1–6, Sofia, Bulgaria. find
  30. Mohammad Salameh, Houda Bouamor, and Nizar Habash. 2018. Fine-grained arabic dialect identification. In Proceedings of the International Conference on Computational Linguistics (COLING), pages 1332–1344, Santa Fe, New Mexico, USA.
  31. Yves Scherrer and Philipp Stoeckle. 2016. A quantitative approach to Swiss German dialectometric analyses and comparisons of linguistic levels. Dialectologia et Geolinguistica, 24(1):92–125. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/obeid-2019-adida-automatic-dialect-identification-arabic/