Elamur Research⌕
☰
home / papers / 2016 / Fast and Robust POS tagger for Arabic Tweets Using Agreement-based Bootstrapping

Fast and Robust POS tagger for Arabic Tweets Using Agreement-based Bootstrapping

Summary

Arabic POS taggers trained on newswire from the PATB perform poorly on Twitter text. Tweets are short and informal, often in dialect, and full of misspellings, joined words, stretched letters, transliterated English and tokens that exist only on Twitter. The authors collected tweets geolocated to the Arabian Peninsula without a language filter, so Romanised and mixed-script posts were kept, then removed the fully English ones. They built a gold standard of 390 tweets (5454 words) by hand-correcting tagger output, and added new tags for Twitter elements. They then evaluated AMIRA, MADA and Stanford on this data and grouped their mistakes, finding that most errors fall on unknown words.

Rather than writing a new tagger, they wrap the existing MSA taggers with two additions:

  • normalisation before tagging,
  • post-processing that draws on external knowledge.

They also apply agreement-based bootstrapping, where unlabelled tweets on which the taggers concur become extra training material. The result is significant gains over the well-known Arabic taggers while keeping tagging fast. This offers a low-cost route into a noisy, under-resourced domain.

Key points

  • Off-the-shelf precision on Arabic tweets is AMIRA 60.2%, MADA 65.8% and Stanford 49.0%, compared with roughly 96–97% for the same tools on newswire.
  • The gold set contains 390 tweets (5454 words) and introduces six Twitter-specific tags: replies are labelled REP, mentions MEN, hashtags HASH, links LINK, usernames USERN, and retweets RET.
  • All three taggers score 0.0% on Twitter-specific tokens and only 9.1–19.9% on non-Arabic tokens. On Arabic-script words they reach 54.0–73.8%.
  • Arabic-script words account for 73.5% (AMIRA), 68.1% (MADA) and 73.2% (Stanford) of the errors. Even well-formed MSA words fall to 71.8%, 79.3% and 62% respectively, which the authors attribute to noisy context, omitted function words and domain shift.
  • Error subcategories within Arabic words are MSA words, concatenation, repeated letters, named entities, spelling mistakes, slang, character deletion and transliteration.

Interesting ideas

  • I1. Break down errors on unknown words by token class (clean MSA, concatenated, letter-stretched, slang, transliterated, platform-specific), so each failure points to a specific fix instead of a single accuracy number.
  • I2. Treat tokens on which several independent taggers agree as automatically labelled training data, enlarging the corpus without manual annotation.
  • I3. Keep a strong existing tagger unchanged and fix the domain mismatch with a normalisation step before it and a knowledge-based correction step after it.
  • I4. Give platform elements (mentions, hashtags, links, retweet markers) their own tags rather than forcing them into ordinary word classes, where every tagger scored 0%.
  • I5. Measure clean, in-vocabulary words separately on the new domain, which separates damage caused by surrounding noise and domain shift from damage caused by the noisy words themselves.

For Mova

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

  • I1 (tokens): Build the same per-class error table for the en tagger on noisy English input, so we can tell whether misses come from normalisation gaps or from the model itself.
  • I2 (learn): This directly supports our silver data agreed by witness parsers, and the paper's approach is evidence that agreement filtering can produce useful in-domain training data for a noisy domain.
  • I3 (tokens): An explainable pre-pass could split joined words, collapse stretched letters and expand clipped forms before the en tagger runs, with each rewrite logged.
  • I4 (tokens): Add dedicated token types for such elements so they can be routed around the UD parser instead of corrupting the tree.
  • I5 (learn): Use a clean-word subset as a control in evaluations, so a parser regression can be traced to context effects rather than lexical noise.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Linguistics & languages

Topics Part-of-speech tagging Semitic languages

 Institutions

🎓University of ManchesterManchester · United Kingdom22 works · 42,781 citations

Cite

Cite as Fahad Albogamy and Allan Ramsay. 2016. Fast and Robust POS tagger for Arabic Tweets Using Agreement-based Bootstrapping. Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16).

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

@inproceedings{albogamy2016fast,
  title = {{Fast and Robust POS tagger for Arabic Tweets Using Agreement-based Bootstrapping}},
  author = {Fahad Albogamy and Allan Ramsay},
  booktitle = {Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16)},
  year = {2016},
  doi = {10.63317/4ds35xk9ijjc},
  url = {https://doi.org/10.63317/4ds35xk9ijjc}
}

References

24 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Abdul-Mageed, M., Kübler, S., and Diab, M. (2012). SAMAR: A system for subjectivity and sentiment analysis of Arabic social media. In Proceedings of the 3rd Workshop in Computational Approaches to Subjectivity and Sentiment Analysis, pages 19–28. Association for Computational Linguistics. find
  2. Al-Sabbagh, R. and Girju, R. (2012). A supervised POS tagger for written Arabic social networking corpora. In Proceedings of KONVENS, pages 39–52. find
  3. Albogamy, F. and Ramsay, A. (2015a). Towards POS tagging for Arabic tweets. In Proceedings of ACL Workshop on Noisy User-generated Text. find
  4. Albogamy, F. and Ramsay, A. (2015b). POS tagging for Arabic tweets. In Proceedings of Recent Advances in Natural Language Processing, pages 1–8. find
  5. Barbosa, L. and Feng, J. (2010). Robust sentiment detection on Twitter from biased and noisy data. In Proceedings of the 23rd International Conference on Computational Linguistics: Posters, pages 36–44. Association for Computational Linguistics. our card
  6. Brill, E. (1995). Transformation-based error-driven learning and natural language processing: A case study in part-of-speech tagging. Computational linguistics, 21(4):543–565. our card
  7. Clark, S., Curran, J. R., and Osborne, M. (2003). Bootstrapping POS taggers using unlabelled data. In Proceedings of the Seventh Conference on Natural Language Learning at HLT-NAACL 2003-Volume 4, pages 49–55. Association for Computational Linguistics. find
  8. Cucerzan, S. and Yarowsky, D. (2002). Bootstrapping a multilingual part-of-speech tagger in one person-day. In Proceedings of the 6th Conference on Natural Language Learning-Volume 20, pages 1–7. Association for Computational Linguistics. find
  9. Derczynski, L., Ritter, A., Clark, S., and Bontcheva, K. (2013). Twitter part-of-speech tagging for all: Overcoming sparse and noisy data. In RANLP, pages 198–206. our card
  10. Diab, M. (2009). Second generation AMIRA tools for Arabic processing: Fast and robust tokenization, POS tagging, and base phrase chunking. In 2nd International Conference on Arabic Language Resources and Tools. find
  11. El-Fishawy, N., Hamouda, A., Attiya, G. M., and Atef, M. (2014). Arabic summarization in Twitter social network. Ain Shams Engineering Journal, 5(2):411–420. find
  12. Elsahar, H. and El-Beltagy, S. R. (2014). A fully automated approach for Arabic slang lexicon extraction from microblogs. In Proceedings of the 15th International Conference on Computational Linguistics and Intelligent Text Processing - Volume 8403, CICLing 2014. find
  13. Foster, J., Çetinoglu, Ö., Wagner, J., Le Roux, J., Hogan, S., Nivre, J., Hogan, D., and Van Genabith, J. (2011). # hardtoparse: POS tagging and parsing the twitterverse. In AAAI 2011 Workshop on Analyzing Microtext, pages 20–25. find
  14. Gadde, P., Subramaniam, L., and Faruquie, T. A. (2011). Adapting a WSJ trained part-of-speech tagger to noisy text: preliminary results. In Proceedings of the 2011 Joint Workshop on Multilingual OCR and Analytics for Noisy Unstructured Text Data, page 5. ACM. find
  15. Gimpel, K., Schneider, N., O’Connor, B., Das, D., Mills, D., Eisenstein, J., Heilman, M., Yogatama, D., Flanigan, J., and Smith, N. A. (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, pages 42–47. Association for Computational Linguistics. our card
  16. Habash, N., Rambow, O., and Roth, R. (2009). Mada+ tokan: A toolkit for Arabic tokenization, diacritization, morphological disambiguation, POS tagging, stemming and lemmatization. find
  17. Han, B. and Baldwin, T. (2011). Lexical normalisation of short text messages: Makn sens a# twitter. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies-Volume 1, pages 368–378. Association for Computational Linguistics. our card
  18. Java, A., Song, X., Finin, T., and Tseng, B. (2007). Why we Twitter: understanding microblogging usage and communities. In Proceedings of the 9th WebKDD and 1st SNA-KDD 2007 workshop on Web mining and social network analysis, pages 56–65. ACM. find
  19. Mourad, A. and Darwish, K. (2013). Subjectivity and sentiment analysis of modern standard Arabic and Arabic microblogs. In Proceedings of the 4th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis, pages 55–64. find
  20. Mustafa, S. H. (2012). Word stemming for Arabic information retrieval: The case for simple light stemming. Abhath Al-Yarmouk: Science & Engineering Series, 21(1):2012. find
  21. Refaee, E. and Rieser, V. (2014). An Arabic Twitter corpus for subjectivity and sentiment analysis. In Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC14), Reykjavik, Iceland, may. European Language Resources Association (ELRA). find
  22. Ritter, A., Clark, S., Etzioni, O., et al. (2011). Named entity recognition in tweets: an experimental study. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, pages 1524–1534. Association for Computational Linguistics. our card
  23. Toutanova, K., Klein, D., Manning, C. D., and Singer, Y. (2003). Feature-rich part-of-speech tagging with a cyclic dependency network. In Proceedings of the 2003 Conference of the North American Chapter of the Association for Computational Linguistics on Human Language Technology-Volume 1, pages 173–180. Association for Computational Linguistics. our card
  24. Zavrel, J. and Daelemans, W. (2000). Bootstrapping a tagged corpus through combination of existing heterogeneous taggers. In Proceedings of the Second International Conference on Language Resources and evaluation LREC-2000, pages 17–20. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/albogamy-2016-fast-robust-pos-tagger-arabic/