Elamur Research⌕
☰
home / papers / 2014 / Foreign Words and the Automatic Processing of Arabic Social Media Text Written in Roman Script

Foreign Words and the Automatic Processing of Arabic Social Media Text Written in Roman Script

Paper · 2014 · Ramy Eskander, Mohamed Al-Badrashiny, Nizar Habash et al.

Summary

Arabic on social media is often dialectal, has no fixed spelling, and is frequently typed in Latin letters ("Arabizi"). Writers also mix in English and other foreign words. Before Arabizi can be converted to Arabic script for downstream tools, a system must decide which tokens are actually Arabic. The authors build the first end-to-end system for the full task defined by the LDC Arabizi corpus (Bies et al., 2014), which consists mostly of Egyptian SMS and chat. Every token gets one of five labels: Arabic, Foreign, Name, Punct or Sound. Emoticons are detected separately. Tokens labelled Arabic or Name are then passed to an existing Arabizi-to-Arabic transliterator (Al-Badrashiny et al., 2014), used as a black box. Foreign words are found with SVM and decision-tree classifiers that draw on lexical, character n-gram and surface features. Punctuation, sounds and emoticons are handled mainly with regular expressions. The authors also propose a new combined evaluation metric. On blind test data the system scores 83.8% overall accuracy against a 76.8% majority-tag baseline. This makes Arabizi usable for MT and sentiment analysis.

Key points

  • The data is the LDC Arabizi corpus, with about 113K training words and 32K test words. Darwish (2013), the closest prior work, used only about 5K words for training.
  • A token counts as Foreign only if it would keep the same spelling in its source language, which is almost always English. Borrowings that carry Arabic affixes or are respelled count as Arabic, and the corpus makes no distinction between borrowing and code switching.
  • The foreign-word classifiers use several feature groups:
    • word unigram scores from Arabic and English language models;
    • Arabic and foreign character n-grams;
    • word properties such as length, capitalization and speech effects;
    • a list of words already tagged as foreign.
  • The best Dev setup for foreign words used only the probabilistic features.
  • F-scores on the test set:
    • Arabic: 96.4
    • Punct: 94.0
    • Foreign: 87.6
    • Sound: 85.6
    • Name: 63.9 (the hardest label)
    • emoticon detection: 98.1
  • Overall accuracy rose from 65.7% to 82.5% on Dev and from 76.8% to 83.8% on Test. Error analysis showed that many apparent mistakes were annotation errors in the gold data: 100% for Punct, 89.6% for Sound, 49.1% for Foreign and 40.0% for Name.

Interesting ideas

  • I1. The overall metric is mixed: tokens that are never transliterated only need the right tag, while Arabic and Name tokens need both the right tag and the right transliteration. This lets one score a pipeline even where gold output does not exist for some token types.
  • I2. "Foreign" is defined by a concrete test: is the word unchanged from its source-language spelling? Words inflected with native morphology are treated as native, which turns a fuzzy borrowing-versus-switching question into a checkable rule.
  • I3. The classifier compares scores from separate per-language word and character n-gram models instead of using raw word identity. These features generalized best and suit noisy, out-of-vocabulary spellings.
  • I4. Errors are split into gold errors and system errors, and each into missed labels versus over-generated labels, with typos counted separately. This reveals how much remaining "error" is really annotation noise.

For Mova

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

  • I1 (learn): Score end-to-end runs so that tokens without a gold frame or world fact are judged only on their label, while fully annotated tokens must match label plus output.
  • I2 (tokens): Use a similar rule to decide when an English token gets a foreign/X treatment (unchanged non-English spelling) versus normal English morphology and lemmas (e.g. "café-ing").
  • I3 (tokens): Contrasting English and non-English character-model scores could help the tagger separate unknown names, foreign insertions and typos before lemmatization.
  • I4 (learn): Apply the same breakdown when auditing silver data from witness parsers, so that disagreements blamed on the silver labels are not used to train repair rules.

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

🎓Columbia UniversityNew York · United States241 works · 54,053 citations🎓New York University Abu DhabiAbu Dhabi · United Arab Emirates122 works · 4,602 citations🎓George Washington UniversityWashington · United States51 works · 8,148 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 Ramy Eskander, Mohamed Al-Badrashiny, Nizar Habash and Owen Rambow. 2014. Foreign Words and the Automatic Processing of Arabic Social Media Text Written in Roman Script. Proceedings of the First Workshop on Computational Approaches to Code Switching.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.3115/v1/w14-3901

@inproceedings{eskander2014foreign,
  title = {{Foreign Words and the Automatic Processing of Arabic Social Media Text Written in Roman Script}},
  author = {Ramy Eskander, Mohamed Al-Badrashiny, Nizar Habash and Owen Rambow},
  booktitle = {Proceedings of the First Workshop on Computational Approaches to Code Switching},
  year = {2014},
  doi = {10.3115/v1/w14-3901},
  url = {https://doi.org/10.3115/v1/w14-3901}
}

References

29 references, 17 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Mohamed Al-Badrashiny, Ramy Eskander, Nizar Habash, and Owen Rambow. 2014. Automatic Transliteration of Romanized Dialectal Arabic. In Proceedings of the Eighteenth Conference on Computational Natural Language Learning, pages 30– 38, Ann Arbor, Michigan, June. Association for Computational Linguistics.
  2. Ann Bies, Zhiyi Song, Mohamed Maamouri, Stephen Grimes, Haejoong Lee, Jonathan Wright, Stephanie Strassel, Nizar Habash, Ramy Eskander, and Owen Rabmow. 2014. Transliteration of Arabizi into Arabic Orthography: Developing a Parallel Annotated Arabizi-Arabic Script SMS/Chat Corpus. In Arabic Natural Language Processing Workshop, EMNLP, Doha, Qatar.
  3. Tim Buckwalter. 2004. Buckwalter Arabic Morphological Analyzer Version 2.0. LDC catalog number LDC2004L02, ISBN 1-58563-324-0. ↗
  4. Achraf Chalabi and Hany Gerges. 2012. Romanized Arabic Transliteration. In Proceedings of the Second Workshop on Advances in Text Input Methods (WTIM 2012).
  5. Eleanor Clark and Kenji Araki. 2011. Text normalization in social media: Progress, problems and applications for a pre-processing system of casual english. Procedia - Social and Behavioral Sciences, 27(0):2 – 11. Computational Linguistics and Related Fields. find
  6. Kareem Darwish, Walid Magdy, and Ahmed Mourad. 2012. Language Processing for Arabic Microblog Retrieval. In Proceedings of the 21st ACM International Conference on Information and Knowledge Management, CIKM ’12, pages 2427–2430, New York, NY, USA. ACM. find
  7. Kareem Darwish. 2013. Arabizi Detection and Conversion to Arabic. CoRR. ↗
  8. Kareem Darwish. 2014. Arabizi Detection and Conversion to Arabic . In Arabic Natural Language Processing Workshop, EMNLP, Doha, Qatar.
  9. Leon Derczynski, Alan Ritter, Sam Clark, and Kalina Bontcheva. 2013. Twitter part-of-speech tagging for all: Overcoming sparse and noisy data. In Proceedings of the International Conference Recent Advances in Natural Language Processing RANLP 2013, pages 198–206, Hissar, Bulgaria, September. INCOMA Ltd. Shoumen, BULGARIA.
  10. Mona Diab, Mohamed Al-Badrashiny, Maryam Aminian, Mohammed Attia, Pradeep Dasigi, Heba Elfardy, Ramy Eskander, Nizar Habash, Abdelati Hawwari, and Wael Salloum. 2014. Tharwa: A Large Scale Dialectal Arabic - Standard Arabic - English Lexicon. In Proceedings of the Language Resources and Evaluation Conference (LREC), Reykjavik, Iceland.
  11. Heba Elfardy, Mohamed Al-Badrashiny, and Mona Diab. 2013. Code Switch Point Detection in Arabic. In Proceedings of the 18th International Conference on Application of Natural Language to Information Systems (NLDB2013), MediaCity, UK, June. find
  12. Heba Elfardy, Mohamed Al-Badrashiny, and Mona Diab. 2014. AIDA: Identifying Code Switching in Informal Arabic Text. In Workshop on Computational Approaches to Linguistic Code Switching, EMNLP, Doha, Qatar, October.
  13. Ramy Eskander, Nizar Habash, Owen Rambow, and Nadi Tomeh. 2013. Processing Spontaneous Orthography. In Proceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT), Atlanta, GA. find
  14. Kevin Gimpel, Nathan Schneider, Brendan O’Connor, Dipanjan Das, Daniel Mills, Jacob Eisenstein, Michael Heilman, Dani Yogatama, Jeffrey Flanigan, and Noah A. Smith. 2011. Part-of-speech tagging for twitter: Annotation, features, and experiments. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies: Short Papers - Volume 2, HLT ’11, pages 42–47, Stroudsburg, PA, USA. Association for Computational Linguistics. ↗
  15. Stephan Gouws, Donald Metzler, Congxing Cai, and Eduard Hovy. 2011. Contextual bearing on linguistic variation in social media. In Proceedings of the Workshop on Languages in Social Media, LSM ’11, pages 20–29, Stroudsburg, PA, USA. Association for Computational Linguistics. find
  16. Nizar Habash, Abdelhadi Soudi, and Tim Buckwalter. 2007. On Arabic Transliteration. In A. van den Bosch and A. Soudi, editors, Arabic Computational Morphology: Knowledge-based and Empirical Methods. Springer. ↗
  17. Nizar Habash, Mona Diab, and Owen Rabmow. 2012a. Conventional Orthography for Dialectal Arabic. In Proceedings of the Language Resources and Evaluation Conference (LREC), Istanbul.
  18. Nizar Habash, Ramy Eskander, and Abdelati Hawwari. 2012b. A Morphological Analyzer for Egyptian Arabic. In Proceedings of the Twelfth Meeting of the Special Interest Group on Computational Morphology and Phonology, pages 1–9, Montréal, Canada.
  19. Nizar Habash. 2009. REMOOV: A tool for online handling of out-of-vocabulary words in machine translation. In Khalid Choukri and Bente Maegaard, editors, Proceedings of the Second International Conference on Arabic Language Resources and Tools. The MEDAR Consortium, April. find
  20. Mark Hall, Eibe Frank, Geoffrey Holmes, Bernhard Pfahringer, Peter Reutemann, and Ian H. Witten. 2009. The WEKA data mining software: an update. SIGKDD Explorations, 11(1):10–18. ↗
  21. LDC. 2014a. BOLT Phase 2 SMS and Chat Arabic DevTest Data – Source Annotation, Transliteration and Translation. LDC catalog number LDC2014E28. find
  22. LDC. 2014b. BOLT Phase 2 SMS and Chat Arabic Training Data – Source Annotation, Transliteration and Translation R1. LDC catalog number LDC2014E48. find
  23. LDC. 2014c. BOLT Program: Romanized Arabic (Arabizi) to Arabic Transliteration and Normalization Guidelines. Version 3. Linguistic Data Consortium. find
  24. Marco Lui, Jey Han Lau, and Timothy Baldwin. 2014. Automatic detection and language identification of multilingual documents. In Proceedings of LREC. ↗
  25. Arfath Pasha, Mohamed Al-Badrashiny, Mona Diab, Ahmed El Kholy, Ramy Eskander, Nizar Habash, Manoj Pooleery, Owen Rambow, and Ryan M. Roth. 2014. MADAMIRA: A Fast, Comprehensive Tool for Morphological Analysis and Disambiguation of Arabic. In Proceedings of the Language Resources and Evaluation Conference (LREC), Reykjavik, Iceland.
  26. Alan Ritter, Sam Clark, Mausam, and Oren Etzioni. 2011. Named entity recognition in tweets: An experimental study. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, EMNLP ’11, pages 1524–1534, Stroudsburg, PA, USA. Association for Computational Linguistics.
  27. Clare Voss, Stephen Tratz, Jamal Laoudi, and Douglas Briesch. 2014. Finding Romanized Arabic Dialect in Code-Mixed Tweets. In Nicoletta Calzolari (Conference Chair), Khalid Choukri, Thierry Declerck, Hrafn Loftsson, Bente Maegaard, Joseph Mariani, Asuncion Moreno, Jan Odijk, and Stelios Piperidis, editors, Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC’14), Reykjavik, Iceland, may. European Language Resources Association (ELRA).
  28. 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 ACL, pages 37–41.
  29. Ines Zribi, Rahma Boujelbane, Abir Masmoudi, Mariem Ellouze, Lamia Belguith, and Nizar Habash. 2014. A Conventional Orthography for Tunisian Arabic. In Proceedings of the Language Resources and Evaluation Conference (LREC), Reykjavik, Iceland.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/eskander-2014-foreign-words-automatic-processing-arabic/