Elamur Research⌕
☰
home / papers / 2014 / YouDACC: the Youtube Dialectal Arabic Comment Corpus

YouDACC: the Youtube Dialectal Arabic Comment Corpus

Paper · 2014 · Ahmed Salama, Houda Bouamor, Behrang Mohit et al.

Summary

Arabic speakers usually write formal texts in Modern Standard Arabic (MSA) but use their spoken dialects online. Large labeled dialect corpora were scarce. The authors built YouDACC, a corpus of YouTube comments spanning several dialects and organized around the five groups Habash defines: Levantine (LV), Maghrebi (MG), Iraqi (IQ), Gulf (GU) and Egyptian (EG). They found videos through the YouTube Data API using region-specific keywords supplied by native speakers. They then collected comments, normalized the noisy spelling and labeled each comment automatically with a dialect. A comment kept its label only when the author's profile location matched the region of the search keyword. Two native speakers judged random samples and confirmed that this labeling is mostly reliable. The exception is Maghrebi, where many comments turned out to be MSA. The authors also trained per-dialect language-model classifiers on the corpus and got high recall. The work shows that platform metadata can stand in for costly manual annotation of dialects. YouTube's terms prevented releasing the data, so the authors described their pipeline in enough detail for others to rebuild it.

Key points

  • Unique normalized comments per dialect: GU 655,578; EG 201,528; IQ 118,675; MG 48,162; LV 61,651. Counting only comments exclusive to one dialect gives GU 322,765, EG 165,579, IQ 68,869, MG 32,215 and LV 41,389.
  • At most 999 comments per video were retrieved, which is an API limit. Only Arabic-script comments were kept, and Franco-Arabic (Latin-script) text was dropped.
  • MG sentences are the longest on average (17.19 words, 90.73 characters), while GU sentences are the shortest (7.48 words).
  • Quality check: 1,000 random sentences per dialect were judged by two native speakers. MG scored lowest because about 40% of its sample was MSA.
  • Classification used one language model per dialect, and a sentence went to the model that scored it highest. Word bigrams gave the best recall for GU (97.00), EG (96.40) and IQ (93.30). For MG, word unigrams scored highest (87.20). Letter 4-gram and final-letter 4-gram models came close, at roughly 85–94.

Interesting ideas

  • I1. Labels are assigned weakly and kept only where two independent signals agree (the keyword's region and the author's stated location), which removes many mislabeled items, such as comments from expats, at no annotation cost.
  • I2. Items that occur under more than one class are excluded from the per-class counts, so ambiguous material does not blur the class boundaries.
  • I3. The human audit offered judges an explicit "MSA or other language" option, which exposed a systematic contamination (the Maghrebi MSA share) that a plain right/wrong question would have hidden.
  • I4. Cheap character features such as initial or final letter n-grams nearly match word models for telling varieties apart, because they capture affixes.

For Mova

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

  • I1 (learn): Silver UD trees or frames could likewise be kept only when independent witnesses agree, for example a parser plus metadata or a lexical cue.
  • I2 (learn): When building silver training sets, sentences whose analyses differ across sources could be set aside rather than force-labeled.
  • I3 (learn): Evaluations of silver data should include an "out of scope / different register" answer so that systematic leakage shows up as its own category.
  • I4 (tokens): Prefix/suffix n-gram scores could help the tagger guess morphology for unknown words, or flag informal register before parsing.

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🎓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 Ahmed Salama, Houda Bouamor, Behrang Mohit and Kemal Oflazer. 2014. YouDACC: the Youtube Dialectal Arabic Comment Corpus. Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC'14).

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{salama2014youdacc,
  title = {{YouDACC: the Youtube Dialectal Arabic Comment Corpus}},
  author = {Ahmed Salama, Houda Bouamor, Behrang Mohit and Kemal Oflazer},
  booktitle = {Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC'14)},
  year = {2014},
  doi = {10.63317/35tyqjp42ypn},
  url = {https://doi.org/10.63317/35tyqjp42ypn}
}

References

14 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Cheng, Xu, Dale, Cameron, and Liu, Jiangchuan. (2007). Understanding The Characteristics Of Internet Short Video Sharing: YouTube As A Case Study. ↗
  2. Diab, Mona, Habash, Nizar, Rambow, Owen, Altantawy, Mohamed, and Benajiba, Yassine. (2010). COLABA: Arabic Dialect Annotation And Processing. In LREC Workshop on Semitic Language Processing, pages 66–74. find
  3. Habash, Nizar, Roth, Ryan, Rambow, Owen, Eskander, Ramy, and Tomeh, Nadi. (2013). Morphological Analysis and Disambiguation for Dialectal Arabic. In Proceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 426–432, Atlanta, Georgia, June.
  4. Habash, Nizar Y. (2010). Introduction to Arabic Natural Language Processing, volume 3. Morgan & Claypool Publishers. find
  5. Mukherjee, Subhabrata and Bhattacharyya, Pushpak. (2012). YouCat: Weakly Supervised Youtube Video Categorization System from Meta Data & User Comments using WordNet & Wikipedia. In Proceedings of COLING 2012, pages 1865–1882, Mumbai, India, December. find
  6. The COLING 2012 Organizing Committee. find
  7. Sajjad, Hassan, Darwish, Kareem, and Belinkov, Yonatan. (2013). Translating Dialectal Arabic to English. In Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 1–6, Sofia, Bulgaria. find
  8. Salloum, Wael and Habash, Nizar. (2013). Dialectal Arabic to English Machine Translation: Pivoting through Modern Standard Arabic. In Proceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 348–358, Atlanta, Georgia.
  9. Siersdorfer, Stefan, Chelaru, Sergiu, Nejdl, Wolfgang, and San Pedro, Jose. (2010). How Useful are Your Comments?: Analyzing and Predicting Youtube Comments and Comment Ratings. In Proceedings of WWW, pages 891–900, New York, NY, USA. find
  10. Wang, Zheshen, Zhao, Ming, Song, Yang, Kumar, Sanjiv, and Li, Baoxin. (2010). YoutubeCat: Learning To Categorize Wild Web Videos. In Computer Vision and Pattern Recognition (CVPR), 2010 IEEE Conference on, pages 879–886. IEEE. find
  11. Zaidan, Omar F. and Callison-Burch, Chris. (2011a). 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, Portland, Oregon, USA, June. Association for Computational Linguistics.
  12. Zaidan, Omar F. and Callison-Burch, Chris. (2011b). Crowdsourcing Translation: Professional Quality from Non-Professionals. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies, pages 1220–1229, Portland, Oregon, USA, June.
  13. Zbib, Rabih, Malchiodi, Erika, Devlin, Jacob, Stallard, David, Matsoukas, Spyros, Schwartz, Richard, Makhoul, John, Zaidan, Omar, and Callison-Burch, Chris. (2012). Machine Translation of Arabic Dialects. In Proceedings of the 2012 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 49–59, Montréal, Canada, June.
  14. Zhang, John R, Song, Yang, and Leung, Thomas. (2011). Improving Video Classification Via Youtube Video Co-Watch Data. In Proceedings of the 2011 ACM workshop on Social and behavioural networked media access, pages 21–26. ACM. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/salama-2014-youdacc-youtube-dialectal-arabic-comment/