--- title: "Approaches to Arabic Name Transliteration and Matching in the" description: "The paper describes how SAS Institute added Arabic personal-name handling to the DataFlux Quality… 92% accuracy. 2 references." authors: ["Brant N. Kay", "Brian C. Rineer"] year: 2012 venue: "Fourth Workshop on Computational Approaches to Arabic-Script-based Languages" source: "https://aclanthology.org/2012.amta-caas14.5/" licence: "CC BY-NC-SA 3.0 (ACL Anthology, before 2016)" licence_url: "https://creativecommons.org/licenses/by-nc-sa/3.0/" summary_licence: "Elamur Cosmic License" canonical: "https://research.elamur.ai/free/en/papers/kay-2012-approaches-arabic-name-transliteration-matching/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/kay-2012-approaches-arabic-name-transliteration-matching/) [home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2012](https://md.research.elamur.ai/free/en/papers/2012/page/2/) / Approaches to Arabic Name Transliteration and Matching in the DataFlux Quality Knowledge Base # Approaches to Arabic Name Transliteration and Matching in the DataFlux Quality Knowledge Base **Authors:** Brant N. Kay and Brian C. Rineer · **Year:** 2012 · **Venue:** [Fourth Workshop on Computational Approaches to Arabic-Script-based Languages](https://md.research.elamur.ai/free/en/venues/amta/) ([AMTA 2012](https://md.research.elamur.ai/free/en/venues/amta/2012/)) · **Publisher:** [Association for Machine Translation in the Americas](https://md.research.elamur.ai/free/en/publishers/association-for-machine-translation-in-the-americas/) · **Licence:** [CC BY-NC-SA 3.0 (ACL Anthology, before 2016)](https://creativecommons.org/licenses/by-nc-sa/3.0/) **Links:** [Publisher's page](https://aclanthology.org/2012.amta-caas14.5/) · [PDF](https://aclanthology.org/2012.amta-caas14.5.pdf) · [OpenAlex](https://openalex.org/W3166285349) **Citations:** 3 citations (OpenAlex, 2026-10-14) ## Summary The paper describes how SAS Institute added Arabic personal-name handling to the DataFlux Quality Knowledge Base (QKB). The QKB is a generic, deterministic data-quality component with no machine-learning infrastructure. The authors use minimal pairs to show that rules alone fail. Latin spellings of one Arabic name vary arbitrarily (Mohammed, Muhammet, …). Word-final letters behave inconsistently. Arabic omits short vowels, so vowel insertion cannot be predicted. Their solution is a hybrid. Transliteration in both directions first consults an Egyptian expert lexicon, then a large CJK Dictionary Institute lexicon, and only then falls back on Perl regex rules. For Arabic-to-Latin, these rules are modeled on Buckwalter. For matching, every name is converted to Latin script and phonetically reduced before comparison. Each component was tuned by having an Egyptian expert review a random sample of bank-customer names and fixing every error, repeating until internal quality thresholds were met. The final results were: - Arabic→Latin transliteration: 92% accuracy. - Latin→Arabic transliteration: 93.9% accuracy. - Cross-script matching: very low false-positive and false-negative rates. The work shows that a curated lexicon plus fallback rules is a practical approach when statistical models are unavailable. ## Key points - Arabic→Latin resources: about 55,000 Arabic-script variants from the CJK Dictionary Institute, plus about 10,000 Egyptian expert entries; the fallback rules follow Buckwalter, extended with ئ, ء and ؤ. - Latin→Arabic resources: 863,282 Latin-script variants from CJK, plus about 10,000 expert entries; the fallback is letter-by-letter rules with expert-written context rules (e.g., word-final Y/I becomes ى). - Transliteration testing used samples of 500 names each, drawn from about 9,000 Arabic-script and about 8,000 Latin-script bank records. Results were 92% after 12 iterations (Arabic→Latin) and 93.9% after 14 iterations (Latin→Arabic). - Matching was tested by cluster analysis on about 8,000 names drawn from a pool of about 17,000 (half in each script). After 6 iterations, 0.4% of matches were false positives and about 0.025% were missed matches. The authors admit they tolerated false positives generously. - The CJK lexicons were not yet available when matching was built, so matching currently relies more on rules; the authors plan to use the lexicons to reduce false positives. ## Interesting ideas - Lookups run in a fixed order: a small, curated regional-preference lexicon first, then a broad general lexicon, then generic rules, so local conventions override defaults without editing them. - Each development round draws a fresh random sample for expert review, turns every observed error into a lexicon entry, and stops when the per-round gains flatten out. - A lossy key (stripping vowels so Khaled and Kholoud both become KHLD) merges distinct names; mapping every known variant to one canonical form avoids such collisions. - Clustering quality is reported separately for false merges and missed merges, together with an honest note on how lenient the judging was. - Minimal pairs (Muhammet/Muhammed vs. Falahat) serve as a crisp argument that no context rule can cover a phenomenon, which justifies using a lexicon instead. ## For Mova How these ideas could serve Mova, our small open language model — its files, stages and tests: - **I1** (tokens): Lemma and morphology lookup could use the same precedence (hand exceptions → large lexicon → rules) and record which layer produced each analysis, which keeps every lemma explainable. - **I2** (learn): Self-induced repair rules could follow this loop, with teacher-tier review of sampled outputs and errors converted into exceptions, plus an iteration-by-iteration accuracy curve to decide when to stop. - **I3** (world): Entity identity in the text world should use canonical IDs linked to known variant lists rather than aggressive string normalization, so different entities are not silently merged. - **I4** (frames): Coreference chains could be evaluated with separate over-merge and under-merge rates, since a single score can hide a merge-happy resolver. - **I5** (learn): Minimal-pair cases could be included as negative controls, to check whether a proposed rule actually generalizes or just memorizes. *Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.* 📚 Shelves (6): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 9,086 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 8,889 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 9,296 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 9,206 works · [🎯 Decision Sciences](https://md.research.elamur.ai/free/en/shelves/decision-sciences/) · 85 works · [🎯 Management Science and Operations Research](https://md.research.elamur.ai/free/en/shelves/decision-sciences/management-science-and-operations-research/) · 36 works 🧭 Topics (2): [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 1,484 works · [🕸️ Knowledge graphs & ontologies](https://md.research.elamur.ai/free/en/topics/knowledge-graphs/) · 103 works ## Cite Cite as: Brant N. Kay and Brian C. Rineer. 2012. Approaches to Arabic Name Transliteration and Matching in the DataFlux Quality Knowledge Base. 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. ```bibtex @inproceedings{kay2012approaches, title = {{Approaches to Arabic Name Transliteration and Matching in the DataFlux Quality Knowledge Base}}, author = {Brant N. Kay and Brian C. Rineer}, booktitle = {Fourth Workshop on Computational Approaches to Arabic-Script-based Languages}, year = {2012}, url = {https://aclanthology.org/2012.amta-caas14.5/} } ``` - [← previous Towards a Semantic Annotation of English Television News - Building and Evaluating a Constraint Grammar FrameNet](https://md.research.elamur.ai/free/en/papers/bick-2012-semantic-annotation-english-television-news/) - [next → ARNE - A tool for Namend Entity Recognition from Arabic Text](https://md.research.elamur.ai/free/en/papers/shihadeh-2012-arne-tool-namend-entity-recognition/) - [↑ back to papers of 2012](https://md.research.elamur.ai/free/en/papers/2012/page/2/) ## References 2 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. Jack Halpern. 2007. The Challenges and Pitfalls of Arabic Romanization and Arabization. In Proceedings of the Second Workshop on Computational Approaches to Arabic Script-based Languages. Palo Alta, CA. [find](https://scholar.google.com/scholar?q=Jack+Halpern%2E+2007%2E+The+Challenges+and+Pitfalls+of+Arabic+Romanization+and+Arabization%2E+In+Proceedings+of+the+Second+Workshop+on+Computational+Approaches+to+Arabic+Script%2Dbased+Languages%2E+Palo+Alta%2C+C) 2. U. Hermjakob, K. Knight, and H. Daumé III. 2008. Name Translation in Statistical Machine Translation - Learning when to Transliterate. In Proceedings of the Annual Meeting of the Association of Computational Linguistics (ACL), pages 389–397, Columbus, Ohio, June. [find](https://scholar.google.com/scholar?q=U%2E+Hermjakob%2C+K%2E+Knight%2C+and+H%2E+Daum%C3%A9+III%2E+2008%2E+Name+Translation+in+Statistical+Machine+Translation+%2D+Learning+when+to+Transliterate%2E+In+Proceedings+of+the+Annual+Meeting+of+the+Association+of+Comput) *Updated 24 October 2026*