Elamur Research⌕
☰
home / papers / 2006 / Combination of Arabic Preprocessing Schemes for Statistical Machine Translation

Combination of Arabic Preprocessing Schemes for Statistical Machine Translation

Paper · 2006 · Fatiha Sadat and Nizar Habash

Summary

Phrase-based statistical MT accepts almost any input representation, as long as training and testing use the same one. This leaves many options for preparing Arabic, a morphologically rich language with clitics, orthographic variation, and templatic inflection. The authors compare eleven word-level preprocessing schemes, ranging from simple tokenization (ST) through orthographic normalization, decliticization of increasing depth, and lemma- and English-like variants. All schemes are produced by a single technique: BAMA proposes analyses, MADA (trained on Penn Arabic Treebank part 1) picks one in context, and Aragen regenerates a normalized surface form once the chosen features have been removed. They then test two ways of combining schemes. One merges phrase tables from all schemes in the decoder while a "privileged" scheme preprocesses the input (DRC). The other rescores the pooled outputs (ROC). Combination beats every single scheme, reaching 38.67 BLEU, and bootstrap tests confirm the gains are significant. The work shows that no single segmentation is best and that pooling several views of the same text helps.

Key points

  • Eleven schemes are compared: ST, ON, D1, D2, D3, WA, TB, MR, L1, L2, EN. All of them except the baseline rely on BAMA + MADA + Aragen, and orthographic normalization is applied throughout.
  • MADA sometimes produces lexeme ties among analyses with the same features. These are resolved arbitrarily but consistently by taking the first analysis in a sorted list.
  • In decoder-level combination, the best privileged scheme was TB (38.24 BLEU, 1-best). Rescoring the pooled outputs of all 11 schemes gave 38.67 BLEU. Pooling only the 4 best (D2, TB, D1, WA) gave a lower 37.73.
  • In bootstrap resampling, DRC had the top BLEU in 100% of samples. With DRC excluded, ROC won 97.7% of samples, D2 2.2%, and TB 0.1%.
  • The best scheme depends on the amount of training data, consistent with earlier findings that morphological preprocessing helps most on smaller corpora.

Interesting ideas

  • I1. After splitting off a clitic, the word is regenerated with a morphological generator rather than simply cut. This keeps the remaining stem in its canonical form and avoids extra spurious variants that increase sparsity and ambiguity.
  • I2. The paper separates the "scheme", meaning the target representation, from the "technique", meaning the tool that produces it. This lets representations and tools be compared independently.
  • I3. The input is processed in one privileged representation, while the model consults knowledge tables built from every alternative representation. This exploits the complementary strengths of the different views.
  • I4. Bootstrap resampling is used to report how often each system wins, which is more informative than reporting a single score difference.
  • I5. Ambiguous analyses are resolved arbitrarily but deterministically, so the same word always maps to the same representation and training and test data stay consistent.

For Mova

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

  • I1 (tokens): When the crate en splits contractions or possessives, it could regenerate the canonical host form instead of keeping a truncated string, so lemmas stay unified.
  • I2 (learn): Evaluation logs could record representation choice and producing tool as separate axes, making it clear which one caused a gain or a regression.
  • I3 (tree): The parser could read one primary tokenization but draw features from alternative segmentations, such as split versus unsplit multiword tokens, to stay robust when the tokenizer errs.
  • I4 (learn): Competing parsers or repair rules could be compared by their win share over resampled test sets, giving an explainable confidence figure for each claimed improvement.
  • I5 (world): When WordNet or other lexicon sense ties cannot be resolved, a fixed deterministic tie-break keeps entity and concept IDs stable across runs and keeps explanations reproducible.

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🏛️National Research Council CanadaOttawa · Canada59 works · 21,677 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 Fatiha Sadat and Nizar Habash. 2006. Combination of Arabic Preprocessing Schemes for Statistical Machine Translation. Proceedings of the 21st International Conference on Computational Linguistics and 44th Annual Meeting of the Association for Computational Linguistics.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.3115/1220175.1220176

@inproceedings{sadat2006combination,
  title = {{Combination of Arabic Preprocessing Schemes for Statistical Machine Translation}},
  author = {Fatiha Sadat and Nizar Habash},
  booktitle = {Proceedings of the 21st International Conference on Computational Linguistics and 44th Annual Meeting of the Association for Computational Linguistics},
  year = {2006},
  doi = {10.3115/1220175.1220176},
  url = {https://doi.org/10.3115/1220175.1220176}
}

References

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

  1. S. Bangalore, G. Bordel, and G. Riccardi. 2001. Computing Consensus Translation from Multiple Machine Translation Systems. In Proc. of IEEE Automatic Speech Recognition and Understanding Workshop, Italy. find
  2. T. Buckwalter. 2002. Buckwalter Arabic Morphological Analyzer Version 1.0. Linguistic Data Consortium, University of Pennsylvania. Catalog: LDC2002L49. find
  3. C. Callison-Burch, M. Osborne, and P. Koehn. 2006. Re-evaluating the Role of Bleu in Machine Translation Research. In Proc. of the European Chapter of the Association for Computational Linguistics (EACL), Trento, Italy.
  4. M. Diab, K. Hacioglu, and D. Jurafsky. 2004. Automatic Tagging of Arabic Text: From Raw Text to Base Phrase Chunks. In Proc. of the North American Chapter of the Association for Computational Linguistics (NAACL), Boston, MA.
  5. R. Frederking and S. Nirenburg. 2005. Three Heads are Better Than One. In Proc. of Applied Natural Language Processing, Stuttgart, Germany. find
  6. S. Goldwater and D. McClosky. 2005. Improving Statistical MT through Morphological Analysis. In Proc. of Empirical Methods in Natural Language Processing (EMNLP), Vancouver, Canada.
  7. N. Habash and O. Rambow. 2005. Tokenization, Morphological Analysis, and Part-of-Speech Tagging for Arabic in One Fell Swoop. In Proc. of Association for Computational Linguistics (ACL), Ann Arbor, Michigan. find
  8. N. Habash and F. Sadat. 2006. Arabic Preprocessing Schemes for Statistical Machine Translation. In Proc. of NAACL, Brooklyn, New York.
  9. N. Habash. 2004. Large Scale Lexeme-based Arabic Morphological Generation. In Proc. of Traitement Automatique du Langage Naturel (TALN). Fez, Morocco. find
  10. S. Jayaraman and A. Lavie. 2005. Multi-Engine Machine Translation Guided by Explicit Word Matching. In Proc. of the Association of Computational Linguistics (ACL), Ann Arbor, MI. find
  11. P. Koehn. 2004a. Pharaoh: a Beam Search Decoder for Phrase-based Statistical Machine Translation Models. In Proc. of the Association for Machine Translation in the Americas (AMTA). find
  12. P. Koehn. 2004b. Statistical Significance Tests for Machine Translation Evaluation. In Proc. of the EMNLP, Barcelona, Spain.
  13. Y. Lee. 2004. Morphological Analysis for Statistical Machine Translation. In Proc. of NAACL, Boston, MA. ↗
  14. Y. Lee. 2005. IBM Statistical Machine Translation for Spoken Languages. In Proc. of International Workshop on Spoken Language Translation (IWSLT). find
  15. M. Maamouri, A. Bies, and T. Buckwalter. 2004. The Penn Arabic Treebank: Building a Large-scale Annotated Arabic Corpus. In Proc. of NEMLAR Conference on Arabic Language Resources and Tools, Cairo, Egypt. find
  16. E. Matusov, N. Ueffing, H. Ney 2006. Computing Consensus Translation from Multiple Machine Translation Systems Using Enhanced Hypotheses Alignment. In Proc. of EACL, Trento, Italy. find
  17. S. Nießen and H. Ney. 2004. Statistical Machine Translation with Scarce Resources Using Morphosyntactic Information. Computational Linguistics, 30(2). find
  18. T. Nomoto. 2004. Multi-Engine Machine Translation with Voted Language Model. In Proc. of ACL, Barcelona, Spain. find
  19. F. Och. 2003. Minimum Error Rate Training in Statistical Machine Translation. In Proc. of the ACL, Sapporo, Japan. find
  20. F. Och. 2005. Google System Description for the 2005 Nist MT Evaluation. In MT Eval Workshop (unpublished talk). find
  21. K. Papineni, S. Roukos, T. Ward, and W. Zhu. 2001. Bleu: a Method for Automatic Evaluation of Machine Translation. Technical Report RC22176(W0109-022), IBM Research Division, Yorktown Heights, NY.
  22. M. Paul, T. Doi, Y. Hwang, K. Imamura, H. Okuma, and E. Sumita. 2005. Nobody is Perfect: ATR’s Hybrid Approach to Spoken Language Translation. In Proc. of IWSLT. find
  23. M. Popović and H. Ney. 2004. Towards the Use of Word Stems and Suffixes for Statistical Machine Translation. In Proc. of Language Resources and Evaluation (LREC), Lisbon, Portugal. find
  24. F. Sadat, H. Johnson, A. Agbago, G. Foster, R. Kuhn, J. Martin, and A. Tikuisis. 2005. Portage: A Phrase-based Machine Translation System. In Proceedings of the ACL Workshop on Building and Using Parallel Texts, Ann Arbor, Michigan. find
  25. A. Stolcke. 2002. Srilm - An Extensible Language Modeling Toolkit. In Proc. of International Conference on Spoken Language Processing. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/sadat-2006-combination-arabic-preprocessing-schemes-statistical/