Elamur Research⌕
☰
home / papers / 2007 / Finding Variants of Out-of-Vocabulary Words in Arabic

Finding Variants of Out-of-Vocabulary Words in Arabic

Paper · 2007 · Abdusalam F.A. Nwesri, S.M.M. Tahaghoghi and Falk Scholer

Summary

Foreign names and technical terms written in Arabic script often appear with many competing spellings. Transliterators choose different letters for sounds Arabic lacks, such as /g/. They also insert long vowels inconsistently, and they rarely write diacritics. A search for one spelling therefore misses documents that use another.

The authors compare nine ways of finding such out-of-vocabulary variants:

  • language-independent string measures: Edit Distance, gramCount, Dice, gramDist and LCS;
  • the earlier Asoundex-final;
  • two Arabic adaptations of their own: Soutex (from Soundex) and AEditex (from Editex);
  • NORM, a new normaliser that strips vowels and merges letters that sound alike.

They test on two sets. The first holds foreign words mined from a crawl of Arabic news sites, using a spellchecker plus manual checking. The second is a list of words transliterated by hand from English. Each word is used as a query, and results are scored with the PRR measure, which handles tied ranks.

Every method beats exact matching by a clear margin. NORM is best on the news data, while LCS is best on the manual set and is the most robust overall. NORM and Soutex can also be computed at indexing time, which makes them cheaper to run.

Key points

  • Crawled corpus: about 1 GB from twelve Arabic news sites, giving 18,873,073 tokens and 383,649 unique words.
  • The Microsoft Office 2003 Arabic spellchecker flagged 40,514 OOV words. After removing a leading preposition and the article "Al", 32,583 remained, and manual inspection found 2,039 foreign words among them.
  • Average PRR on the crawled data: NORM 0.660, LCS 0.619, AEditex 0.576, Edit Distance 0.572, Soutex 0.530, exact match 0.300.
  • Average PRR on the transliterated data: LCS 0.782, Edit Distance 0.700, AEditex 0.624, Soutex 0.590, NORM 0.536, gramDist 0.401, exact match 0.261.
  • All methods beat the exact-match baseline (p<0.0001). Asoundex-final scored poorly, which the authors attribute to its dependence on diacritics that ordinary text omits.

Interesting ideas

  • I1. Map every spelling to a reduced "skeleton" key (drop vowels, merge confusable letters) and store it in the index, so variants collapse cheaply without pairwise comparison at query time.
  • I2. Use an edit distance where swapping two letters from the same sound group costs less than swapping unrelated letters (the Editex/AEditex idea), giving a graded, interpretable similarity score.
  • I3. Treat every item in the gold clusters as a query against the whole pool, so the test does not favour phonetic or string methods by how the queries were chosen.
  • I4. Score ranked outputs that contain ties with PRR, which assumes the user picks randomly within a tied rank, instead of breaking ties arbitrarily.
  • I5. The authors note that their hand-made dataset was skewed by non-fluent transliterators, a reminder that how test data is produced can change which method appears best.

For Mova

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

  • I1 (tokens): such a key could serve as a fallback lookup for unknown proper nouns, so that spellings like Gaddafi/Qaddafi or colour/color land on one lemma or entity candidate.
  • I2 (world): this could rank Wikidata or WordNet label candidates for a mention with odd spelling, and record which substitutions justified the match.
  • I3 (learn): the same all-items-as-queries setup suits evaluating entity matching or coreference clustering, with unrelated words acting as natural negative controls.
  • I4 (learn): when several witness parsers or scorers give equal confidence to candidates, PRR-style scoring would keep the evaluation from rewarding lucky tie order.
  • I5 (learn): document how each silver or LLM-teacher set was generated and compare methods across sources, so artefacts of one source are not mistaken for real gains.

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

🎓RMIT UniversityMelbourne · Australia19 works · 704 citations

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksDecision Sciences · 249 worksManagement Science and Operations Research · 134 works

🧭 Topics 1Research subjects and languages the work is about.Semitic languages · 3,066 works

Cite

Cite as Abdusalam F.A. Nwesri, S.M.M. Tahaghoghi and Falk Scholer. 2007. Finding Variants of Out-of-Vocabulary Words in Arabic. Proceedings of the 2007 Workshop on Computational Approaches to Semitic Languages: Common Issues and Resources.

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

@inproceedings{nwesri2007finding,
  title = {{Finding Variants of Out-of-Vocabulary Words in Arabic}},
  author = {Abdusalam F.A. Nwesri, S.M.M. Tahaghoghi and Falk Scholer},
  booktitle = {Proceedings of the 2007 Workshop on Computational Approaches to Semitic Languages: Common Issues and Resources},
  year = {2007},
  doi = {10.3115/1654576.1654586},
  url = {https://doi.org/10.3115/1654576.1654586}
}

References

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

  1. Ahmed Abdelali, Jim Cowie, and Hamdy S. Soliman. 2004. Arabic information retrieval perspectives. In Proceedings of the 11th Conference on Natural Language Processing, Journes d’Etude sur la Parole - Traitement Automatique des Langues Naturelles (JEP-TALN), Fez, Morocco. find
  2. Nasreen Abduljaleel and Leah S. Larkey. 2003. Statistical transliteration for English-Arabic cross-language information retrieval. In Proceedings of the International Conference on Information and Knowledge Management, pages 139–146, New Orleans, LA, USA. ACM Press. find
  3. Jamal B. S. Al-Qinal. 2002. Morphophonemics of loanwords in translation. Journal of King Saud University, 13:1–132. find
  4. Syed Uzair Aqeel, Steve Beitzel, Eric Jensen, David Grossman, and Ophir Frieder. 2006. On the development of name search techniques for Arabic. Journal of the American Society for Information Science and Technology, 57(6):728–739. ↗
  5. Christine L. Borgman and Susan L. Siegfried. 1992. Getty’s synoname and its cousins: A survey of applications of personal name-matching algorithms. Journal of the American Society for Information Science, 43(7):459–476. find
  6. Lee R. Dice. 1945. Measures of the amount of ecologic association between species. Ecology, 26(3):297–302, July. ↗
  7. T. Gadd. 1990. Phonix: the algorithm. Program, 24(4):363– 369. ↗
  8. Patrick A. V. Hall and Geoff R. Dowling. 1980. Approximate string matching. ACM Computing Surveys, 12(4):381–402. ↗
  9. Dan Melamed. 1995. Automatic evaluation and uniform filter cascades for inducing N-best translation lexicons. In David Yarovsky and Kenneth Church, editors, Proceedings of the Third Workshop on Very Large Corpora, pages 184–198, Somerset, New Jersey. Association for Computational Linguistics. ↗
  10. Abdusalam F Ahmad Nwesri, S. M. M. Tahaghoghi, and Falk Scholer. 2006. Capturing out-of-vocabulary words in Arabic text. In Proceedings of the 2006 Conference on Empirical Methods in Natural Language Processing (EMNLP 2006), pages 258–266, Sydney, Australia, 22–23 July. Association for Computational Linguistics.
  11. Ulrich Pfeifer, Thomas Poersch, and Norbert Fuhr. 1996. Retrieval effectiveness of proper name search methods. Inf. Process. Manage., 32(6):667–679. ↗
  12. Vijay Raghavan, Peter Bollmann, and Gwang S. Jung. 1989. A critical investigation of recall and precision as measures of retrieval system performance. ACM Trans. Inf. Syst., 7(3):205–229. ↗
  13. Bonnie Glover Stalls and Kevin Knight. 1998. Translating names and technical terms in Arabic text. In COLING/ACL Workshop on Computational Approaches to Semitic Languages, pages 34–41, Montreal, Quebc, Canada.
  14. Graham A Stephen. 1992. String search. Technical report, School of Electronic Engineering Science, University College of North Wales. find
  15. Esko Ukkonen. 1992. Approximate string-matching with q-grams and maximal matches. Theor. Comput. Sci., 92(1):191–211. ↗
  16. Robert A. Wagner and Michael J. Fischer. 1974. The string-tostring correction problem. J. ACM, 21(1):168–173. find
  17. Justin Zobel and Philip Dart. 1995. Finding approximate matches in large lexicons. Software - Practice and Experience, 25(3):331–345. ↗
  18. Justin Zobel and Philip Dart. 1996. Phonetic string matching: lessons from information retrieval. In The 19th annual international ACM SIGIR conference on Research and development in information retrieval, pages 166–172, New York, NY, USA. ACM Press. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/nwesri-2007-finding-variants-out-vocabulary-words/