Elamur Research⌕
☰
home / papers / 2012 / Using Arabic Transliteration to Improve Word Alignment from French- Arabic Parallel Corpora

Using Arabic Transliteration to Improve Word Alignment from French- Arabic Parallel Corpora

Summary

Linking names across French and Arabic is hard because one Arabic spelling can show up in Latin script in dozens of forms. Dialects, colonial-era French or English conventions and the lack of a shared standard all add variants. The authors address this inside a linguistically driven French–Arabic word aligner. They built an Arabic-to-Latin transliterator from finite-state automata and contextual rules about how Arabic personal names are structured. Its Latin-letter equivalences were collected bottom-up from real web usage, including the digit-based spellings common in SMS and social media.

The aligner links single words using a bilingual lexicon, named entities, cognates, word positions and grammatical tags. It links compound words through syntactic dependency relations. Transliteration lets Arabic names be compared to French words as cognates.

On 283 hand-aligned ARCADE II sentences, adding transliteration raised F-measure from 0.82 to 0.85. Precision and recall both went up. Feeding the aligner's output into Moses training data raised BLEU slightly. The work shows that a cheap, rule-based script bridge can help alignment for a language pair with few resources.

Key points

  • Graphematic equivalence tables were mined from multilingual web text. One Arabic letter maps to many Latin variants, and some letters map to digits, e.g. 7 for ḥāʾ and 3 for ʿayn.
  • The transliterator models the parts of an Arabic name with contextual rules, such as Kunya (Abu/Umm + name), Ism and Nasab (bin/bint).
  • After transliteration, two words count as cognates if their length ratio exceeds 0.8 and their shared-character ratio exceeds 0.5. A sentence-position check filters false matches such as Muhammad/mahmoud.
  • Compound words are aligned compositionally: Translation(A.B) = Translation(A).Translation(B), as in frais_inscription.
  • In the Moses experiment, the baseline was trained on 10,000 ARCADE II sentence pairs with a language model built from 10,800 Arabic sentences. Adding about 10,000 aligned word and compound pairs from 500 sentences moved BLEU from 12.50 to 12.82 on a 250-sentence test set.

Interesting ideas

  • I1. Build spelling-variant tables from observed usage rather than from a normative standard, keeping every attested form, because real text never follows one convention.
  • I2. Using two simple ratios with fixed thresholds (length similarity plus shared characters) is a transparent way to detect cognates, and a positional check removes many false positives.
  • I3. Decompose a personal name into culturally meaningful slots (kinship particle, given name, genealogy marker) instead of treating it as one opaque string.
  • I4. Align multiword units through their dependency structure and compose their translation from the parts' translations.
  • I5. Evaluate twice: intrinsically against a small hand reference, and extrinsically by measuring the effect on a downstream system.

For Mova

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

  • I1 (tokens): Mova could learn variant and nonstandard spellings (names, informal or SMS-style English) from corpus evidence and map them to a single lemma.
  • I2 (world): The same explainable scoring could merge entity mentions with slightly different spellings, keeping the reason for each merge.
  • I3 (world): Parsing names into structured parts could support kinship and identity facts, and help resolve "son of X" style references in tales.
  • I4 (back): In translation through the meta-world, Mova could render compounds by translating the head and modifier along the UD arc and recombining them, with a traceable derivation.
  • I5 (learn): Each new resource or repair rule could be checked both on a small gold set and by its effect on the answer stage, which shows whether local gains actually matter.

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 · Machine translation

Topics Semitic languages

Cite

Cite as Houda Saadane, Ouafa Benterki, Nasredine Semmar and Christian Fluhr. 2012. Using Arabic Transliteration to Improve Word Alignment from French- Arabic Parallel Corpora. Fourth Workshop on Computational Approaches to Arabic-Script-based Languages.

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

@inproceedings{saadane2012using,
  title = {{Using Arabic Transliteration to Improve Word Alignment from French- Arabic Parallel Corpora}},
  author = {Houda Saadane, Ouafa Benterki, Nasredine Semmar and Christian Fluhr},
  booktitle = {Fourth Workshop on Computational Approaches to Arabic-Script-based Languages},
  year = {2012},
  url = {https://aclanthology.org/2012.amta-caas14.6/}
}

References

16 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Abdulmalik Alsalman, Mansour Alghamdi, Khalid Alhuqayl and Salih Alsubay. 2007. Romanization System for Arabic Names. In Proceedings of the First International Symposium on Computer and Arabic Language ISCAL – 07, Riyadh, 214-227. find
  2. Bonnie Stalls and Kevin Knight. 1998. Translating Names and Technical Terms in Arabic Text. In Proceedings of the COLING/ACL Workshop on Computational Approches to Semitic Languages, Montreal, Québec. find
  3. Hasnaa Qunaiir. 2001. Romanizing Arabic names. Journal er-Riyadh, 12314. find
  4. Joseph Dichy. 2009. La polyglossie de l'arabe illustrée par deux corpus. M. Bozdemir et L.–J. Calvet (EDS), Politiques linguistiques en méditerranée, Paris: Honoré Champion, 82-102. find
  5. Kevin Knight and Jonathan Graehl. 1997. Machine transliteration. In Proceedings of the 34th ACL Conference, Madrid, Spain. our card
  6. Leah S. Larkey, Lisa Ballesteros and Margaret E. Connell. 2002. Improving Stemming for Arabic Information Retrieval: Light Stemming and Co-occurrence Analysis. In Proceedings of the 25th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval, Tampere, Finland. find
  7. Mansour Alghamdi. 2005. Algorithms for Romanizing Arabic names. Journal of King Saud University - Computer and Information Sciences. Riyadh, 17:01-27. find
  8. Mathieu Guidère. 2004. Le traitement de la parole et la détection des dialectes arabes. Langues stratégiques et défense nationale, Publications du CREC, Saint-Cyr, 53-75. Melissa. find
  9. B. Defradas, Rym Hamdi and François Pellegrino. 2004. De la caractérisation linguistique à l’identification automatique des dialectes arabes, In Proceedings of the MIDL 2004 Workshop, Paris, France. Nasreen AbdulJaleel and Leah. find
  10. S. Larkey. 2003. Statistical transliteration for English-Arabic Cross Language Information Retrieval. In Proceedings of the 12th ACM International Conference on Information and Knowledge Management, New Orleans, LA, USA. find
  11. Philipp Koehn, Hieu Hoang, Alexandra Birch, Chris C. Burch, Marcello Federico, Nicolas Bertoldi, Brooke Cowan, Wade Shen, Christine Moran, Richard Zens, Chris Dyer, Ondrej Bojar, Alexandra Constantin and Evan Herbst. 2007. Moses: Open source toolkit for statistical machine translation. In Proceedings of the 45th ACL Conference, demo session, Prague, Czech Republic. find
  12. Rada Mihalcea and Ted Pedersen. 2003. An evaluation exercise for word alignment. In Proceedings of the Workshop on Building and Using Parallel Texts: Data Driven Machine Translation and Beyond, Edmonton, Canada. our card
  13. Saleem Abuleil and Martha Evens. 2004. Named Entity Recognition and Classification for Text in Arabic. In Proceedings of the 13th International Conference on Intelligent & Adaptive Systems and Software Engineering, Nice, France. find
  14. Tao Tao, Su-Youn Yoon, Andrew Fister, Richard Sproat and ChengXiang Zhai. 2006. Unsupervised named entity transliteration using temporal and phonetic correlation. In Proceedings of the 2006 Conference on Empirical Methods in Natural Language Processing (EMNLP’06), Sydney, Australia. find
  15. Tarek Sherif and Grzegorz Kondrak. 2007. Bootstrapping a stochastic transducer for Arabic-English transliteration extraction. In Proceedings of the 45th ACL Conference, Prague, Czech Republic. June 2007 find
  16. Yaser Al-Onaizan and Kevin Knight. 2002. Translating named entities using monolingual and bilingual resources. In Proceedings of the 40th ACL Conference, Philadelphia, USA. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/saadane-2012-arabic-transliteration-improve-word-alignment/