Elamur Research⌕
☰
home / papers / 2014 / Domain and Dialect Adaptation for Machine Translation into Egyptian Arabic

Domain and Dialect Adaptation for Machine Translation into Egyptian Arabic

Paper · 2014 · Serena Jeblee, Weston Feely, Houda Bouamor et al.

Summary

The paper tackles translation from English into Egyptian Arabic, a dialect with almost no parallel data, no standard spelling, and poor support from Modern Standard Arabic (MSA) tools. The authors pivot through MSA. A core phrase-based system trained on millions of out-of-domain English–MSA sentence pairs produces MSA. Further phrase-based systems then act as automatic post-editors that rewrite that MSA output. Two designs are compared. The one-step design maps core MSA output straight to in-domain Egyptian. The two-step design first adapts the MSA to the target domain and then converts it to Egyptian. To train these adapters, they built a 100k-sentence English–MSA–Egyptian tri-parallel corpus. Its Egyptian side was generated by MSA-to-Egyptian rewrite rules, with a dialect language model choosing among candidate outputs. Both designs beat a direct English→Egyptian baseline. The two-step design scores best at 42.9 BLEU. This shows that dialect conversion can be handled like domain adaptation within one language. The authors caution that the Egyptian test data is synthetic rather than naturally written.

Key points

  • Core English→MSA system: 5M NIST 2012 sentence pairs for training, 1,356 MT08 sentences for tuning, and 1,313 MT09 sentences for testing. It uses a 5-gram SRILM language model trained on 260M words of Arabic Gigaword, with text tokenized by MADA v3.1 in the ATB scheme.
  • The Egyptian side of the tri-parallel corpus was generated by extending Mohamed et al. (2012), which uses 103 rules for nouns, verbs, pronouns and some lexical items. Instead of picking one candidate at random, the authors ranked candidates with a 5-gram Kneser-Ney language model trained on Egyptian Wikipedia and the Egyptian part of the AOC corpus.
  • The corpus split is 98,027 / 960 / 961 sentences for train / tune / test. Egyptian text is tokenized with MADA-ARZ v0.4 in the ATB scheme, with alif/ya normalization.
  • The baseline is a single Moses system trained only on the 100k corpus. The two-step adapter scores 2.8 BLEU above the one-step adapter, reaching 42.9 BLEU.
  • Fragmentation penalties fall with each stage: 0.204 for the core, 0.159 for one-step, and 0.189 followed by 0.139 for the two-step stages. The two-step system used a reordering window of 7, while the one-step system used 0.

Interesting ideas

  • I1. Train a translation model to map a system's own flawed output to clean references in the same language, so it works as a learned automatic post-editor and can stack on any core system.
  • I2. Treat a dialect shift as a variant of domain shift and chain the adaptations as separate stages: domain first, then dialect. Each stage then has a smaller, cleaner job.
  • I3. Generate missing training data with hand-written rewrite rules, then use a target-variety language model to choose among the many rule outputs instead of sampling one at random.
  • I4. Error analysis showed that word-level metrics on detokenized text unfairly zero out outputs whose clitics differ, and that valid synonyms get punished without synonym matching.
  • I5. They used the fragmentation penalty, alongside BLEU, to show which pipeline produces longer runs of correctly ordered words.

For Mova

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

  • I1 (learn): Repair rules could be induced from pairs of (parser output, gold tree) in the same way, learning systematic fixes for a weak parser instead of retraining it.
  • I2 (tree): UD dialect conversion between treebanks could be split into a domain or genre normalization pass followed by a convention-mapping pass, so each pass can be explained on its own.
  • I3 (learn): Rule-generated silver data (e.g. paraphrased or voice-shifted sentences) could be filtered by a fluency scorer before training.
  • I4 (answer): Graders for answers and generated text should compare at lemma or morpheme level and allow synonyms, so correct answers are not marked wrong for surface variation.
  • I5 (back): Generation from the meaning graph could report a word-order fragmentation measure next to overlap scores, giving an explainable signal of whether ordering is the problem.

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 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 2Research subjects and languages the work is about.Machine translation · 1,087 worksSemitic languages · 2,908 works

Cite

Cite as Serena Jeblee, Weston Feely, Houda Bouamor, Alon Lavie, Nizar Habash and Kemal Oflazer. 2014. Domain and Dialect Adaptation for Machine Translation into Egyptian Arabic. Proceedings of the EMNLP 2014 Workshop on Arabic Natural Language Processing (ANLP).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.3115/v1/w14-3627

@inproceedings{jeblee2014domain,
  title = {{Domain and Dialect Adaptation for Machine Translation into Egyptian Arabic}},
  author = {Serena Jeblee, Weston Feely, Houda Bouamor, Alon Lavie, Nizar Habash and Kemal Oflazer},
  booktitle = {Proceedings of the EMNLP 2014 Workshop on Arabic Natural Language Processing (ANLP)},
  year = {2014},
  doi = {10.3115/v1/w14-3627},
  url = {https://doi.org/10.3115/v1/w14-3627}
}

References

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

  1. Hitham Abo Bakr, Khaled Shaalan, and Ibrahim Ziedan. 2008. A Hybrid Approach for Converting Written Egyptian Colloquial Dialect into Diacritized Arabic. In Proceedings of the 6th International Conference on Informatics and Systems (IN-FOS2008). Cairo University. find
  2. Abdel-Rahman Abu-Melhim. 1991. Code-switching and Linguistic Accommodation in Arabic. In Perspectives on Arabic Linguistics III: Papers from the Third Annual Symposium on Arabic Linguistics, volume 80, pages 231–250. John Benjamins Publishing. ↗
  3. Kamla Al-Mannai, Hassan Sajjad, Alaa Khader, Fahad Al Obaidli, Preslav Nakov, and Stephan Vogel. 2014. Unsupervised Word Segmentation Improves Dialectal Arabic to English Machine Translation. In Proceedings of EMNLP Workshop on Arabic Natural Language Processing, Doha, Qatar. find
  4. Houda Bouamor, Nizar Habash, and Kemal Oflazer. 2014. A Multidialectal Parallel Corpus of Arabic. In Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC’14), pages 1240–1245, Reykjavik, Iceland.
  5. Rahma Boujelbane, Mariem Ellouze Khemekhem, and Lamia Hadrich Belguith. 2013. Mapping Rules for Building a Tunisian Dialect Lexicon and Generating Corpora. In Proceedings of the Sixth International Joint Conference on Natural Language Processing, pages 419–428, Nagoya, Japan. find
  6. Jonathan Clark, Chris Dyer, Alon Lavie, and Noah Smith. 2011. Better Hypothesis Testing for Statistical Machine Translation: Controlling for Optimizer Instability. In Proceedings of the Association for Computational Linguistics (ACL), Portland, Oregon. find
  7. Ryan Cotterell and Chris Callison-Burch. 2014. A Multi-Dialect, Multi-Genre Corpus of Informal Written Arabic. In Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC’14), pages 241–245, Reykjavik, Iceland.
  8. Michael Denkowski and Alon Lavie. 2011. Meteor 1.3: Automatic metric for reliable optimization and evaluation of machine translation systems. In Proceedings of the EMNLP 2011 Workshop on Statistical Machine Translation, pages 1–4, Edinburgh, Scotland.
  9. Marcello Federico, Nicola Bertoldi, and Mauro Cettolo. 2008. IRSTLM: an open source toolkit for handling large scale language models. In INTER-SPEECH, pages 1618–1621. find
  10. Nizar Habash and Owen Rambow. 2005. Arabic Tokenization, Part-of-speech Tagging and Morphological Disambiguation in One Fell Swoop. In Proceedings of the Association for Computational Linguistics, Ann Arbor, Michigan.
  11. Nizar Habash and Owen Rambow. 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. ↗
  12. Nizar Habash, Ramy Eskander, and Abdelati Hawwari. 2012. 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.
  13. Nizar Habash, Ryan Roth, Owen Rambow, Ramy Eskander, and Nadi Tomeh. 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.
  14. Nizar Habash. 2010. Introduction to Arabic Natural Language Processing, volume 3. Morgan & Claypool Publishers. find
  15. Ahmed Hamdi, Nuria Gala, and Alexis Nasr. 2014. Automatically Building a Tunisian Lexicon for Deverbal Nouns. In Proceedings of the First Workshop on Applying NLP Tools to Similar Languages, Varieties and Dialects, pages 95–102, Dublin, Ireland. find
  16. Kenneth Heafield, Ivan Pouzyrevsky, Jonathan H. Clark, and Philipp Koehn. 2013. Scalable Modfied Kneser-Ney Language Model Estimation. In In Proceedings of the Association for Computational Linguistics, Sofia, Bulgaria. find
  17. Kenneth Heafield. 2011. KenLM: Faster and Smaller Language Model Queries. In Proceedings of the EMNLP 2011 Sixth Workshop on Statistical Machine Translation, pages 187–197, Edinburgh, Scotland, United Kingdom, July.
  18. Pierre Isabelle, Cyril Goutte, and Michel Simard. 2007. Domain Adaptation of MT Systems through Automatic Post-Editing. In Proceedings of MT Summit XI, pages 255–261, Copenhagen, Denmark. find
  19. Philipp Koehn, Hieu Hoang, Alexandra Birch, Christopher Callison-Burch, Marcello Federico, Nicola Bertoldi, Brooke Cowan, Wade Shen, Christine Moran, Richard Zens, Christopher Dyer, Ondrej Bojar, Alexandra Constantin, and Evan Herbst. 2007. Moses: Open Source Toolkit for Statistical Machine Translation. In Proceedings of the 45th Annual Meeting of the Association for Computational Linguistics Companion Volume Proceedings of the Demo and Poster Sessions, pages 177–180, Prague, Czech Republic. find
  20. Emad Mohamed, Behrang Mohit, and Kemal Oflazer. 2012. Transforming Standard Arabic to Colloquial Arabic. In Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 176–180, Jeju Island, Korea.
  21. NIST Multimodal Information Group. 2010a. NIST 2008 Open Machine Translation (OpenMT) Evaluation LDC2010T21. Web Download. find
  22. NIST Multimodal Information Group. 2010b. NIST 2009 Open Machine Translation (OpenMT) Evaluation LDC2010T23. Web Download. find
  23. Franz Josef Och and Hermann Ney. 2003. A Systematic Comparison of Various Statistical Alignment Models. In Computational Linguistics, pages 19– 51. ↗
  24. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. BLEU: a Method for Automatic Evaluation of Machine Translation. In Proceedings of the Association for Computational Linguistics, Philadelphia, Pennsylvania.
  25. Robert Parker, David Graff, Ke Chen, Junbo Kong, and Kazuaki Maeda. 2011. Arabic Gigaword Fifth Edition LDC2011T11. Web Download. find
  26. Arfath Pasha, Mohamed Al-Badrashiny, Mona Diab, Ahmed El Kholy, Ramy Eskander, Nizar Habash, Manoj Pooleery, Owen Rambow, and Ryan Roth. 2014. MADAMIRA: A Fast, Comprehensive Tool for Morphological Analysis and Disambiguation of Arabic. In Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC’14), pages 1094–1101, Reykjavik, Iceland.
  27. Hassan Sajjad, Kareem Darwish, and Yonatan Belinkov. 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
  28. Ahmed Salama, Houda Bouamor, Behrang Mohit, and Kemal Oflazer. 2014. YouDACC: the Youtube Dialectal Arabic Comment Corpus. In Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC’14), pages 1246– 1251, Reykjavik, Iceland.
  29. Wael Salloum and Nizar Habash. 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.
  30. Wael Salloum, Heba Elfardy, Linda Alamir-Salloum, Nizar Habash, and Mona Diab. 2014. Sentence Level Dialect Identification for Machine Translation System Selection. In Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 772– 778, Baltimore, Maryland. find
  31. Hassan Sawaf. 2010. Arabic Dialect Handling in Hybrid Machine Translation. In Proceedings of the Ninth Conference of the Association for Machine Translation in the Americas (AMTA 10), Denver, Colorado.
  32. Mohammed Q Shatnawi, Muneer Bani Yassein, and Reem Mahafza. 2012. A Framework for Retrieving Arabic Documents Based on Queries Written in Arabic Slang Language. Journal of Information Science, 38(4):350–365. ↗
  33. Michel Simard, Cyril Goutte, and Pierre Isabelle. 2007. Statistical Phrase-Based Post-Editing. In Proceedings of NAACL-HLT-2007 Human Language Technology: the Conference of the North American Chapter of the Association for Computational Linguistics, pages 508–515, Rochester, NY.
  34. Matthew Snover, Bonnie Dorr, Richard Schwartz, Linnea Micciulla, and John Makhoul. 2006. A Study of Translation Edit Rate with Targeted Human Annotation. In Proceedings of Association for Machine Translation in the Americas, Cambridge, Massachusetts.
  35. Andreas Stolcke. 2002. SRILM - An Extensible Language Modeling Toolkit. In Proceedings of the International Conference on Spoken Language Processing, vol. 2, pages 901–904, Denver, CO, USA. find
  36. 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 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies, pages 37–41, Portland, Oregon, USA.
  37. Rabih Zbib, Erika Malchiodi, Jacob Devlin, David Stallard, Spyros Matsoukas, Richard Schwartz, John Makhoul, Omar F. Zaidan, and Chris Callison-Burch. 2012. Machine Translation for Arabic Dialects. In Proceedings of North American Chapter of the Association for Computational Linguistics, Montreal, Canada. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/jeblee-2014-domain-dialect-adaptation-machine-translation/