Elamur Research⌕
☰
home / papers / 2007 / Arabic diacritization in the context of statistical machine translation

Arabic diacritization in the context of statistical machine translation

Paper · 2007 · Mona Diab, Mahmoud Ghoneim and Nizar Habash

Summary

Arabic script usually omits diacritics, the optional marks for short vowels, gemination and nunation. This leaves words lexically and morphologically ambiguous. The authors ask whether restoring some or all of these marks helps phrase-based SMT from Arabic. Earlier preprocessing work mostly tried to shrink sparsity. Here the tokenization stays fixed and only selected word forms gain diacritics, so the vocabulary actually grows. The schemes were chosen by looking at where diacritics occur naturally in the Penn Arabic Treebank. They include none (NONE), passive-voice marking (PASS), C-M, shadda only (GEM), sukuun only (SUK) and full diacritization (FULL). Diacritics were assigned automatically with MADA. Each scheme was tested in two modes that separate the effect on word alignment from the effect on decoding. FULL clearly hurts translation. No partial scheme differs significantly from NONE. The useful finding is about what makes a scheme help: it needs to touch many tokens, and the tagger must assign that diacritic with a high F-score.

Key points

  • About 1.6% of tokens in ATB III v2 carry natural diacritics, and about 1.5% in the UN Arabic corpus. Among natural diacritics, nunation makes up 73.4%, shadda 20.8% and damma 3%. Damma mostly marks the passive.
  • Diacritics were assigned with MADA, which is built on BAMA. According to its reported results, MADA gets 14.9% of words wrong and 4.8% of diacritics wrong.
  • Translation was evaluated on MT03, MT04 and MT05 in two modes, ALIGNBASIC and ALIGNREMAP. For example, on MT03 in ALIGNBASIC, SUK scored 0.4482 BLEU and GEM scored 0.4354.
  • The schemes differed sharply in how many tokens they changed and how accurately:
    • PASS changed only 0.4% of tokens, with a MADA F-score of 62%. GEM changed 16.85% of tokens with an F-score of 95.6%. Both added 1.3% to the type count.
    • SUK changed 25.3% of tokens with an F-score of 96.6% and added 3.9% types.
  • C-M and FULL had the highest OOV rates, up to 2.44%, and also gave the worst BLEU. SUK beat GEM despite having an equal or higher OOV rate.

Interesting ideas

  • I1. Base partial marking on what writers mark on their own, adding only features that resolve real ambiguity rather than full annotation.
  • I2. Run the richer representation in two modes, so its effect on training-time alignment is measured apart from its effect on inference.
  • I3. Judge a restoration task by F-score on the marks it adds, not by token accuracy, since most tokens never change.
  • I4. A feature's downstream value depends on two things together: how many tokens it touches and how reliably it is predicted.
  • I5. Track the OOV rate between training and test data for each representation, as a cheap early warning that a richer encoding is fragmenting the vocabulary.

For Mova

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

  • I1 (tokens): Mova could mark only disambiguating features, such as passive participle versus past tense, instead of full feature bundles, and so avoid needless sparsity in later stages.
  • I2 (learn): Mova could test a new feature once in training only and once at inference too, to see where any gain or loss actually comes from.
  • I3 (learn): Mova's evaluation of rare tags or repair rules could report precision and recall on the affected items, so that trivial no-change cases do not inflate the scores.
  • I4 (frames): before passing an automatically inferred feature into frames, Mova could weigh its coverage against tagger reliability and drop features that are both rare and noisy.
  • I5 (tokens): Mova could log the lemma and form OOV rate for every tokenizer or morphology variant before running full parsing evaluations.

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, 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 Mona Diab, Mahmoud Ghoneim and Nizar Habash. 2007. Arabic diacritization in the context of statistical machine translation. Proceedings of Machine Translation Summit XI: Papers.

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

@inproceedings{diab2007arabic,
  title = {{Arabic diacritization in the context of statistical machine translation}},
  author = {Mona Diab, Mahmoud Ghoneim and Nizar Habash},
  booktitle = {Proceedings of Machine Translation Summit XI: Papers},
  year = {2007},
  url = {https://aclanthology.org/2007.mtsummit-papers.20/}
}

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. S. Ananthakrishnan, S. Narayanan, and S. Bangalore. 2005. Automatic diacritization of arabic transcripts for asr. In Proceedings of ICON, Kanpur, India. find
  2. Tim Buckwalter. 2002. Buckwalter Arabic Morphological Analyzer Version 1.0. Linguistic Data Consortium, University of Pennsylvania, 2002. LDC Catalog No.: LDC2002L49. ↗
  3. Chris Callison-Burch, Miles Osborne, and Philipp Koehn. 2006. Re-evaluating the Role of BLEU in Machine Translation Research. In Proceedings of the 11th conference of the European Chapter of the Association for Computational Linguistics (EACL’06), pages 249–256, Trento, Italy.
  4. Sharon Goldwater and David McClosky. 2005. Improving Statistical MT Through Morphological Analysis. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, pages 676–683, Vancouver, Canada.
  5. Nizar Habash and Owen Rambow. 2005. Arabic Tokenization, Part-of-Speech Tagging and Morphological Disambiguation in One Fell Swoop. In Proceedings of the 43rd Annual Meeting of the Association for Computational Linguistics (ACL’05), pages 573–580, Ann Arbor, Michigan, June. Association for Computational Linguistics.
  6. Nizar Habash and Owen Rambow. 2007. Arabic Diacritization through Full Morphological Tagging. In Proceedings of the 8th Meeting of the North American Chapter of the Association for Computational Linguistics/Human Language Technologies Conference (HLT-NAACL07).
  7. Nizar Habash and Fatiha Sadat. 2006. Arabic Preprocessing Schemes for Statistical Machine Translation. In Proceedings of the 7th Meeting of the North American Chapter of the Association for Computational Linguistics/Human Language Technologies Conference (HLT-NAACL06), pages 49–52, New York, NY.
  8. Katrin Kirchhoff and Dimitra Vergyri. 2005. Cross-dialectal data sharing for acoustic modeling in arabic speech recognition. Speech Communication, (46):37– 51. ↗
  9. Philipp Koehn. 2004. Pharaoh: a Beam Search Decoder for Phrase-based Statistical Machine Translation Models. In Proceedings of the Association for Machine Translation in the Americas, pages 115–124. find
  10. Young-Suk Lee. 2004. Morphological Analysis for Statistical Machine Translation. In Proceedings of the 5th Meeting of the North American Chapter of the Association for Computational Linguistics/Human Language Technologies Conference (HLT-NAACL04), pages 57– 60, Boston, MA. ↗
  11. 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
  12. Sonja Nießen and Hermann Ney. 2004. Statistical Machine Translation with Scarce Resources using Morpho-syntactic Information. Computational Linguistics, 30(2). ↗
  13. Franz Josef Och and Hermann Ney. 2003. A Systematic Comparison of Various Statistical Alignment Models. Computational Linguistics, 29(1):19–52. ↗
  14. Franz Josef Och. 2003. Minimum Error Rate Training for Statistical Machine Translation. In Proceedings of the 41st Annual Conference of the Association for Computational Linguistics, pages 160–167, Sapporo, Japan, July. find
  15. 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.
  16. Maja Popović and Hermann Ney. 2004. Towards the Use of Word Stems and Suffixes for Statistical Machine Translation. In Proceedings of the 4th International Conference on Language Resources and Evaluation (LREC), pages 1585–1588, Lisbon, Portugal, May. find
  17. Andreas Stolcke. 2002. SRILM - an Extensible Language Modeling Toolkit. In Proceedings of the International Conference on Spoken Language Processing (ICSLP), pages 901–904. find
  18. Dimitra Vergyri and Katrin Kirchhoff. 2004. Automatic Diacritization of Arabic for Acoustic Modeling in Speech Recognition. In Ali Farghaly and Karine Megerdoomian, editors, COLING 2004 Workshop on Computational Approaches to Arabic Script-based Languages, pages 66–73, Geneva, Switzerland. ↗
  19. Imed Zitouni, Jeffrey S. Sorensen, and Ruhi Sarikaya. 2006. Maximum Entropy Based Restoration of Arabic Diacritics. In Proceedings of the 21st International Conference on Computational Linguistics and 44th Annual Meeting of the Association for Computational Linguistics, pages 577–584, Sydney, Australia.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/diab-2007-arabic-diacritization-context-statistical-machine/