Elamur Research⌕
☰
home / papers / 2014 / Automatic Identification of Arabic Language Varieties and Dialects in Social Media

Automatic Identification of Arabic Language Varieties and Dialects in Social Media

Paper · 2014 · Fatiha Sadat, Farzindar Kazemi and Atefeh Farzindar

Summary

Arabic social media text mixes Modern Standard Arabic (MSA) with regional dialects (AD). Most Arabic NLP tools assume MSA, so identifying the dialect is a necessary first step for translation, sentiment analysis and retrieval. Working within the ASMAT (Arabic Social Media Analysis Tools) effort, the authors built a hand-collected corpus from about eighteen Arabic blog and forum sites. Each sentence or paragraph is labelled with one of 18 countries, and the countries are also grouped into six regional groups. They then compared two probabilistic approaches, both using character uni-, bi- and tri-grams. The first is a character n-gram Markov language model. The second is a Naive Bayes classifier. Character features were chosen because dialects differ largely in their affixes, which short character sequences capture well. The best setup was Naive Bayes over character bi-grams, which reached about 98% overall accuracy across the 18 dialects. The six regional groups were also separated well. The result shows that cheap, transparent character statistics can handle fine-grained variety identification in noisy short texts.

Key points

  • The corpus was manually gathered from roughly 18 Arabic forums and blogs. Texts were segmented by hand into coherent sentences or paragraphs and stored as XML with sequence number, country, date and link.
  • There are 18 country-level labels and six regional groups: Egyptian, Levantine, Gulf, Iraqi, Maghrebi and Others (Sudan). The grouping follows Zaidan and Callison-Burch (2012).
  • Data size varies widely by country. Saudi Arabia has 9,882 sentences, while Palestine has only 316.
  • 100 randomly chosen sentences per dialect were held out for testing and excluded from training.
  • Naive Bayes with character bi-grams beat the uni-gram and tri-gram variants and reached about 98% overall accuracy. For context, the related work cites a prior MSA-vs-Egyptian system at 85.5%.

Interesting ideas

  • I1. Short character n-grams pick up variation in affixes without any lexicon, which makes them a robust signal for closely related varieties that share most of their vocabulary.
  • I2. The paper treats variety identification as the very first preprocessing stage, which decides how all later components should handle the text.
  • I3. Evaluation uses a fixed number of held-out test items per class, even though training sizes are very unbalanced, so every class is measured on equal footing.
  • I4. Labels are kept at two levels of granularity (country and regional group), so a model can be scored on both the fine and the coarse distinction.
  • I5. Each collected sentence carries provenance metadata (source link, date, origin), which makes later auditing and re-splitting of the data straightforward.

For Mova

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

  • I1 (tokens): A character bi-/tri-gram profile could flag spelling variety (British vs American) or informal register before tagging, and could guess the morphology of unknown words from their endings.
  • I2 (tokens): A lightweight register detector could send noisy social-media English to a normalization pass before the tokenizer and tagger. The decision would be logged so the explanation chain shows why normalization happened.
  • I3 (learn): Evaluation sets for genres or treebank sources could use equal-sized per-class slices, so small domains are not hidden by large ones in the overall score.
  • I4 (tree): When converting between UD dialects or treebank conventions, errors could be reported both per fine label and per coarse family. This would show whether mistakes stay within a family or cross families.
  • I5 (learn): Silver data agreed by witness parsers could store source and date with every sentence, so suspicious training items can be traced back and filtered.

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

🎓Université du Québec à MontréalMontreal · Canada2 works · 120 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 Fatiha Sadat, Farzindar Kazemi and Atefeh Farzindar. 2014. Automatic Identification of Arabic Language Varieties and Dialects in Social Media. Proceedings of the Second Workshop on Natural Language Processing for Social Media (SocialNLP).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.3115/v1/w14-5904

@inproceedings{sadat2014automatic,
  title = {{Automatic Identification of Arabic Language Varieties and Dialects in Social Media}},
  author = {Fatiha Sadat, Farzindar Kazemi and Atefeh Farzindar},
  booktitle = {Proceedings of the Second Workshop on Natural Language Processing for Social Media (SocialNLP)},
  year = {2014},
  doi = {10.3115/v1/w14-5904},
  url = {https://doi.org/10.3115/v1/w14-5904}
}

References

21 references, 11 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Al-Sabbagh R. and Girju R. 2013. Yadac : Yet another dialectal arabic corpus. In N. C. C. Chair, K. Choukri, T. Declerck, M. U. Doan, B. Maegaard, J. Mariani, J. Odijk, and S. Piperidis, editors, Proceedings of the Eight International Conference on Language Resources and Evaluation (LREC'12), Istanbul, Turkey, May 2012.
  2. Almeman K. and Lee M. 2013. Automatic building of arabic multi dialect text corpora by bootstrapping dialect words. In Communications, Signal Processing, and their Applications (ICCSPA), 2013. find
  3. Baldwin T. and Lui M. 2010. Language identification: The long and the short of the matter. In Human Language Technologies: The 2010 Annual Conference of theNorth American Chapter of the Association for Computational Linguistics, HLT'10, pages 229{237, Stroudsburg, PA, USA, 2010. Association for Computational Linguistics. find
  4. Biadsy F., Hirschberg J., and Habash N. 2009. Spoken arabic dialect identification using phonotactic modeling. In Proceedings of the Workshop on Computational Approaches to Semitic Languages at the meeting of the European Association for Computational Linguistics (EACL), Athens, Greece.
  5. Cavnar W. B., Trenkle J. M. 1994. N-gram-based text categorization. 1994. In Proceedings of SDAIR-94, 3rd Annual Symposium on Document Analysis and Information Retrieval. Ann Arbor MI, 48113(2):161{175, 1994. find
  6. Dasigi P. and Diab M. 2011. Codact: Towards identifying orthographic variants in dialectal arabic. In Proceedings of the 5th International Joint Conference on Natural Language Processing (ICJNLP), Chiangmai, Thailand, 2011. ↗
  7. Dunning T. 1994. Statistical identification of languages. find
  8. Elfardy H. and Diab M. 2012. Simplified guidelines for the creation of large scale dialectal arabic annotations. In Proceedings of the 8th International Conference on Language Resources and Evaluation (LREC), Istanbul, Turkey.
  9. Elfardy H. and Diab M. 2013. Sentence-Level Dialect Identification in Arabic, In Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, ACL 2013, Sofia, Bulgaria. 2013.
  10. Elfardy H., Al-Badrashiny M., Elfardy M. and Diab M. 2013. Sentence Level Dialect Identification in Arabic. In Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 456–461, Sofia, Bulgaria, August 4-9 2013.
  11. Gotti F., Langlais P., and Farzindar A. 2013. Translating government agencies' tweet feeds: Specificities, problems and (a few) solutions. In Proceedings of the Workshop on Language Analysis in Social Media, Atlanta, Georgia, June 2013. Association for Computational Linguistics, Association for Computational Linguistics. find
  12. Habash N. and Rambow O. 2006. MAGEAD: A morphological analyzer and generator for the Arabic dialects. In Proceedings of the 21st International Conference on Computational Linguistics and 44th Annual Meeting of the Association for Computational Linguistics , pages 681–688, Sydney, Australia, July. Association for Computational Linguistics. ↗
  13. Habash N., Diab M., and Rabmow O. 2012. Conventional orthography for dialectal arabic. In Proceedings of the Language Resources and Evaluation Conference (LREC), Istanbul, Turkey, 2012.
  14. Joachims T. 1998. Text categorization with support vector machines: Learning with many relevant features. ↗
  15. Kikui G.-i. 1996. Identifying, the coding system and language, of on-line documents on the internet. In Proceedings of the 16th Conference on Computational Linguistics - Volume 2, COLING '96, pages 652{657, Stroudsburg, PA, USA, 1996. Association for Computational Linguistics. find
  16. Kirchhoff K. and Vergyri D. 2004. Cross-dialectal acoustic data sharing for arabic speech recognition. In Acoustics, Speech, and Signal Processing, 2004. Proceedings.(ICASSP'04). IEEE International Conference on, volume 1, pages I-765. find
  17. Peng F. and Schuurmans D. 2003. Combining naive bayes and n-gram language models for text classi_cation. In Advances in Information Retrieval, pages 335{350. find
  18. Salloum W. and Habash N. 2011. Dialectal to standard arabic paraphrasing to improve arabic-english statistical machine translation. In Proceedings of the First Workshop on Algorithms and Resources for Modelling of Dialects and Language Varieties , pages 10–21. Association for Computational Linguistics. ↗
  19. Suliman A. F. 2008. Automatic Identification of Arabic Dialects USING Hidden Markov Models. Doctoral Dissertation, University of Pittsburgh. 2008. find
  20. Zaidan O. F. and Callison-Burch C. 2011. The arabic online commentary dataset: an annotated dataset of informal arabic with high dialectal content. In Proceedings of ACL, pages 37–41, 2011.
  21. Zaidan O. F. and Callison-Burch C. 2012. Arabic dialect identi_cation. volume 1, Microsoft Research, 2012. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/sadat-2014-automatic-identification-arabic-language-varieties/