Elamur Research⌕
☰
home / papers / 2016 / YAMAMA: Yet Another Multi-Dialect Arabic Morphological Analyzer

YAMAMA: Yet Another Multi-Dialect Arabic Morphological Analyzer

Paper · 2016 · Salam Khalifa, Nasser Zalmout and Nizar Habash

Summary

Arabic morphological processing is hard: words carry many inflections, the script is orthographically ambiguous (about a dozen full analyses per word on average), and dialects such as Egyptian (EGY) sit alongside Modern Standard Arabic (MSA). Existing tools force a choice. MADAMIRA gives rich output (diacritics, lemma, POS, gloss, all inflectional features) but runs slowly. FARASA is fast but only segments words, and only for MSA.

YAMAMA keeps MADAMIRA's analyzers and output format but replaces its context-based disambiguation with a simple context-free model inspired by FARASA. Known words get their most frequent analysis from the training data, looked up in a precomputed dictionary. Unknown words are analyzed and ranked by lemma and tag unigram probabilities. All decisions are made once per word type rather than per token.

Trained and tested on PATB and the ARZ Treebank, YAMAMA is roughly five times faster than MADAMIRA on large text. Its accuracy is slightly lower for MSA, mainly in diacritization, and comparable or better for EGY. Machine translation quality barely changes. This makes it a practical middle point between speed and richness.

Key points

  • Data and resources.
    • Training and test data: PATB parts 1–3 (MSA) and the ARZ Treebank (EGY), using the splits from Diab et al. (2013).
    • Out-of-vocabulary analysis uses the SAMA database for MSA and CALIMA ARZ for EGY.
  • Accuracy and speed on the MSA test set.
    • YAMAMA reaches ALL-features accuracy of 78.8 versus MADAMIRA's 86.0.
    • The gap comes mostly from diacritization (81.0 vs 88.0).
    • YAMAMA runs in 15.4 s against 57.7 s.
  • EGY results. YAMAMA slightly beats MADAMIRA on ALL (79.3 vs 78.4) and wins on most features except POS.
  • FARASA tokenization scheme.
    • On MSA ATB, YAMAMA and MADAMIRA both score 98.7, while FARASA scores 89.6.
    • On EGY, YAMAMA scores 94.4 against FARASA's 73.3.
    • On WikiNews, all three systems are close.
  • Large-corpus timing and MT.
    • On 7.5M words of Gigaword, MADAMIRA took 2,305 s, YAMAMA 398 s and FARASA 99 s.
    • In Moses SMT on the UN Corpus, YAMAMA scored 39.49 BLEU, close to MADAMIRA's 39.52 and above FARASA's 37.73.

Interesting ideas

  • I1. Deciding analyses once per word type and then expanding back to tokens exploits a low type-to-token ratio on big corpora (3.7% at 8M words), which gives large speedups at no accuracy cost for that step.
  • I2. A two-tier design: a lookup table of the most frequent analysis for known words, and for unknown ones, analyzer candidates ranked by the product of lemma and tag unigram probabilities, with the winner added back into the table.
  • I3. Keep the rich analyzer and full output format, and swap only the disambiguation component, so downstream consumers notice no difference.
  • I4. To compare systems with different conventions fairly, map every system's output and the gold data into one shared scheme, and report a normalized variant alongside it.
  • I5. In extrinsic MT evaluation, rerun after dropping OOV words to separate one system's transliteration artifact from genuine segmentation quality.

For Mova

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

  • I1 (tokens): The en crate could cache lemma/morphology decisions per surface form so that each form is resolved only once and reused across a corpus.
  • I2 (tokens): Mova's lemmatizer could use this as a fast, explainable baseline, where each choice traces to either "most frequent in treebank" or "lemma×tag score".
  • I3 (tokens): Mova could offer fast and accurate tagger modes behind one identical feature interface, so later stages run unchanged.
  • I4 (learn): When scoring witness parsers that emit different UD dialects, first convert all of them and the gold trees to a common scheme, and report both strict and normalized scores.
  • I5 (learn): Evaluations could include such ablation controls so that a single quirky failure mode does not dominate comparisons between components.

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

🎓New York UniversityNew York · United States201 works · 100,180 citations🎓New York University Abu DhabiAbu Dhabi · United Arab Emirates122 works · 4,602 citations

 Fields, topics and concepts

📚 Shelves 6The 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 worksSocial Sciences · 1,772 worksEducation · 203 works

🧭 Topics 2Research subjects and languages the work is about.Semitic languages · 2,908 worksMorphology · 405 works

Cite

Cite as Salam Khalifa, Nasser Zalmout and Nizar Habash. 2016. YAMAMA: Yet Another Multi-Dialect Arabic Morphological Analyzer. Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: System Demonstrations.

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

@inproceedings{khalifa2016yamama,
  title = {{YAMAMA: Yet Another Multi-Dialect Arabic Morphological Analyzer}},
  author = {Salam Khalifa, Nasser Zalmout and Nizar Habash},
  booktitle = {Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: System Demonstrations},
  year = {2016},
  url = {https://aclanthology.org/C16-2047/}
}

References

19 references, 10 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Ahmed Abdelali, Kareem Darwish, Nadir Durrani, and Hamdy Mubarak. 2016. Farasa: A fast and furious segmenter for Arabic. In Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Demonstrations, pages 11–16, San Diego, California.
  2. Satanjeev Banerjee and Alon Lavie. 2005. METEOR: An Automatic Metric for MT Evaluation with Improved Correlation with Human Judgments. In Proceedings of the ACL 2005 Workshop on Intrinsic and Extrinsic Evaulation Measures for MT and/or Summarization.
  3. Kareem Darwish and Hamdy Mubarak. 2016. Farasa: A new fast and accurate Arabic word segmenter. In Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC 2016).
  4. Mona Diab, Nizar Habash, Owen Rambow, and Ryan Roth. 2013. LDC Arabic treebanks and associated corpora: Data divisions manual. arXiv preprint arXiv:1309.5652. ↗
  5. Andreas Eisele and Yu Chen. 2010. Multiun: A multilingual corpus from united nation documents. In Proceedings of the Language Resources and Evaluation Conference (LREC), pages 2868–2872.
  6. David Graff, Mohamed Maamouri, Basma Bouziri, Sondos Krouna, Seth Kulick, and Tim Buckwalter. 2009. Standard Arabic Morphological Analyzer (SAMA) Version 3.1. Linguistic Data Consortium LDC2009E73. find
  7. 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. ↗
  8. Nizar Habash, Owen Rambow, and Ryan Roth. 2009. MADA+TOKAN: A toolkit for Arabic tokenization, diacritization, morphological disambiguation, POS tagging, stemming and lemmatization. In Khalid Choukri and Bente Maegaard, editors, Proceedings of the Second International Conference on Arabic Language Resources and Tools. The MEDAR Consortium, April. find
  9. 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.
  10. Nizar Y Habash. 2010. Introduction to Arabic natural language processing, volume 3. Morgan & Claypool Publishers. find
  11. Kenneth Heafield, Ivan Pouzyrevsky, Jonathan H. Clark, and Philipp Koehn. 2013. Scalable modified Kneser-Ney language model estimation. In Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 690–696, Sofia, Bulgaria, August.
  12. 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
  13. Philipp Koehn. 2004. Statistical significance tests for machine translation evaluation. In Proceedings of EMNLP 2004, pages 388–395, Barcelona, Spain, July. Association for Computational Linguistics.
  14. Mohamed Maamouri, Ann Bies, Tim Buckwalter, and Wigdan Mekki. 2004. The Penn Arabic Treebank: Building a Large-Scale Annotated Arabic Corpus. In NEMLAR Conference on Arabic Language Resources and Tools, pages 102–109, Cairo, Egypt. find
  15. Mohamed Maamouri, Ann Bies, Seth Kulick, Michael Ciul, Nizar Habash, and Ramy Eskander. 2014. Developing an Egyptian Arabic Treebank: Impact of Dialectal Morphology on Annotation and Tool Development. In Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC-2014). European Language Resources Association (ELRA). find
  16. Franz Josef Och and Hermann Ney. 2003. A Systematic Comparison of Various Statistical Alignment Models. Computational Linguistics, 29(1):19–52. ↗
  17. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. BLEU: a Method for Automatic Evaluation of Machine Translation. In Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics, pages 311–318, Philadelphia, PA.
  18. Robert Parker, David Graff, Ke Chen, Junbo Kong, and Kazuaki Maeda. 2009. Arabic Gigaword Fourth Edition. LDC catalog number No. LDC2009T30, ISBN 1-58563-532-4. ↗
  19. Arfath Pasha, Mohamed Al-Badrashiny, Ahmed El Kholy, Ramy Eskander, Mona Diab, Nizar Habash, Manoj Pooleery, Owen Rambow, and Ryan Roth. 2014. MADAMIRA: A Fast, Comprehensive Tool for Morphological Analysis and Disambiguation of Arabic. In In Proceedings of LREC, Reykjavik, Iceland.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/khalifa-2016-yamama-yet-another-multi-dialect/