--- title: "TAGARAB — paper by Maloney and Niv (1998)" description: "The paper tackles named entity recognition for Arabic newspaper text. 5 references, cited by 10 papers here." authors: ["John Maloney", "Michael Niv"] year: 1998 venue: "Computational Approaches to Semitic Languages" source: "https://aclanthology.org/W98-1002/" 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/maloney-1998-tagarab-fast-accurate-arabic-name/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/maloney-1998-tagarab-fast-accurate-arabic-name/) [home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [1998](https://md.research.elamur.ai/free/en/papers/1998/) / TAGARAB: A Fast, Accurate Arabic Name Recognizer Using High-Precision Morphological Analysis # TAGARAB: A Fast, Accurate Arabic Name Recognizer Using High-Precision Morphological Analysis **Authors:** John Maloney and Michael Niv · **Year:** 1998 · **Venue:** Computational Approaches to Semitic Languages · **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/W98-1002/) · [PDF](https://aclanthology.org/W98-1002.pdf) · [OpenAlex](https://openalex.org/W1992004647) **Citations:** 55 citations (OpenAlex, 2026-10-14) · [10](https://md.research.elamur.ai/free/en/papers/maloney-1998-tagarab-fast-accurate-arabic-name/cited-by/) in our library, of the 149,439 papers whose bibliographies we read ## Summary The paper tackles named entity recognition for Arabic newspaper text. Arabic script has no letter case, so there is no capitalization signal for where a name starts or stops. The authors built TAGARAB on top of SRA's NetOwl TurboTag pattern engine. Its front end is a morphological tokenizer that gives each token coarse grammatical features (part of speech, definite article, attached suffixes) and splits off clitic conjunctions and prepositions. A pattern-based Name Finder then marks persons, entities, locations, times and numerics with SGML tags. The analyzer is deliberately simple. It runs ordered regular expressions backed by small lists of frequent stems, takes the first match, and ignores context and ambiguity. Embedding it in the Flex-generated scanner kept the system very fast. On a blind set of Al-Hayat articles, the system scored about F 85. An ablation without morphological features dropped the score to about 75, with person names hit hardest. The study suggests that cheap, high-precision morphology matters a great deal for finding entity boundaries in caseless scripts. ## Key points - Pipeline: ISO-8859-6 text goes to the Morphological Tokenizer, which emits up to 32 feature bits per token (e.g. ARABIC, NOUN, PERFECT, IMPERFECT, CLOSED, CONJ, PREP, DEF_ART, SUFF). The Name Finder then produces SGML-tagged output. - Words stay whole, with features attached. The only splits are for attached *wa/fa* conjunctions and *la/ba/ka* prepositions. - The analyzer began as 31 sequential Perl regexes, which proved too slow. Moving it into the Flex specification gave about 46 MB/hour for the whole system and about 190 MB/hour for morphology alone on a Sun Ultra 1. - Besides closed-class items, the lexicon covers forms from three open classes: 1051 for nouns, 813 for verbs and 241 for adjectives. Broken plurals get their own entries, and both hamza and hamza-less spellings are listed. - Scores: training set overall F 90.9. Blind set recall 80.8 and F about 85.0. Without morphology the blind-set F falls to 75.3, and the person F drops from roughly 80.9 to 63.1. ## Interesting ideas - Morphology is used mainly as negative evidence: closed-class words, inflected verbs and suffixed forms are flagged as unlikely name parts, so an open-ended name pattern knows where to stop. - The lexicon favors precision on frequent words over full coverage. Only common items are listed, each with its most probable reading, because those are the words that most often constrain later stages. - The morphological features are packed into the tokenizer's bit-vector rather than computed by a separate module, which avoids costly calls between components and keeps throughput high. - An ablation that switches off one component (morphology) and reports the effect per entity type showed exactly which downstream patterns depended on it. - One wrong boundary spreads: a stray word absorbed into a name is later used to match supposed short variants of that name elsewhere in the article. ## For Mova How these ideas could serve Mova, our small open language model — its files, stages and tests: - **I1** (world): When entity mentions are grown from a head or title, tokens tagged as function words or finite verbs could serve as explicit, explainable stop conditions for the span. - **I2** (tokens): A small curated list of high-frequency forms with a preferred reading could act as a fast first pass, with the heavier tagger handling only the remainder. - **I3** (tokens): The en crate could attach coarse morph flags during tokenization itself so that later stages read them for free. - **I4** (learn): Each stage could get a disable-this-feature run as a standard negative control, broken down by category, to show where a module actually earns its keep. - **I5** (frames): Coreference that links mentions by partial string match should check that the shared fragment is a content-bearing name part, otherwise a single error can contaminate a whole chain. *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 - [SRA International](https://md.research.elamur.ai/free/en/companies/sra-international/) — Company, Fairfax, United States · 3 works · 1,130 citations 📚 Shelves (5): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works · [👁️ Computer Vision and Pattern Recognition](https://md.research.elamur.ai/free/en/shelves/computer-science/computer-vision-and-pattern-recognition/) · 2,869 works 🧭 Topics (2): [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 2,908 works · [🧩 Morphology](https://md.research.elamur.ai/free/en/topics/morphology/) · 405 works ## Cite Cite as: John Maloney and Michael Niv. 1998. TAGARAB: A Fast, Accurate Arabic Name Recognizer Using High-Precision Morphological Analysis. Computational Approaches to Semitic Languages. The paper itself, not this page. BibTeX from the library's record of the work. ```bibtex @inproceedings{maloney1998tagarab, title = {{TAGARAB: A Fast, Accurate Arabic Name Recognizer Using High-Precision Morphological Analysis}}, author = {John Maloney and Michael Niv}, booktitle = {Computational Approaches to Semitic Languages}, year = {1998}, doi = {10.3115/1621753.1621756}, url = {https://doi.org/10.3115/1621753.1621756} } ``` - [← previous Dynamic Coreference-Based Summarization](https://md.research.elamur.ai/free/en/papers/baldwin-1998-dynamic-coreference-based-summarization/) - [next → Generalizing Case Frames Using a Thesaurus and the MDL Principle](https://md.research.elamur.ai/free/en/papers/li-1998-generalizing-case-frames-thesaurus-mdl/) - [↑ back to papers of 1998](https://md.research.elamur.ai/free/en/papers/1998/) ## References 5 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. Chinatsu Aone and John Maloney: "Reuse of a Proper Noun Recognition System in Commercial and Operational NLP Applications," in Proceedings of ACL'97 Workshop on From Research to Commercial Applications: Making NLP Technology Work in Practice, Madrid, Spain, 1997. [find](https://scholar.google.com/scholar?q=Chinatsu+Aone+and+John+Maloney%3A+%22Reuse+of+a+Proper+Noun+Recognition+System+in+Commercial+and+Operational+NLP+Applications%2C%22+in+Proceedings+of+ACL%2797+Workshop+on+From+Research+to+Commercial+Application) 2. Chinatsu Aone, Mary Ellen Okurowski, James Gorlinsky, and Bjornar Larsen: "A Scalable Summarization System Using Robust NLP," in Proceedings of ACL'97 Workshop on Intelligent Scalable Text Summarization, Madrid, Spain, 1997. [find](https://scholar.google.com/scholar?q=Chinatsu+Aone%2C+Mary+Ellen+Okurowski%2C+James+Gorlinsky%2C+and+Bjornar+Larsen%3A+%22A+Scalable+Summarization+System+Using+Robust+NLP%2C%22+in+Proceedings+of+ACL%2797+Workshop+on+Intelligent+Scalable+Text+Summarizati) 3. Jacob Landau. A Word Count of Modern Arabic Prose. American Council of Learned Societies, New York, 1959. [find](https://scholar.google.com/scholar?q=Jacob+Landau%2E+A+Word+Count+of+Modern+Arabic+Prose%2E+American+Council+of+Learned+Societies%2C+New+York%2C+1959%2E) 4. Moshe Levinger et al. "Learning Morpho-Lexieal Probabilities from an Untagged Corpus with an Application to Hebrew." Computational Linguistics, Vol. 2113, 1995. [find](https://scholar.google.com/scholar?q=Moshe+Levinger+et+al%2E+%22Learning+Morpho%2DLexieal+Probabilities+from+an+Untagged+Corpus+with+an+Application+to+Hebrew%2E%22+Computational+Linguistics%2C+Vol%2E+2113%2C+1995%2E) 5. MIS SPU INC 0 148 6 36 8 198 REC PRE F-M 96.6 9 7 . 3 97.0 69.5 7 5 . 6 72.4 78.6 9 0 . 2 84.0 73.1 9 0 . 5 80.9 64.0 6 2 . 1 63.1 71.6 79.4 75.3 202 64 148 666 1082 364 268 814 590 2040 Annemarie Schimmel, Islamic Names. Edinburgh University Press. Edinburgh, 1989. [find](https://scholar.google.com/scholar?q=MIS+SPU+INC+0+148+6+36+8+198+REC+PRE+F%2DM+96%2E6+9+7+%2E+3+97%2E0+69%2E5+7+5+%2E+6+72%2E4+78%2E6+9+0+%2E+2+84%2E0+73%2E1+9+0+%2E+5+80%2E9+64%2E0+6+2+%2E+1+63%2E1+71%2E6+79%2E4+75%2E3+202+64+148+666+1082+364+268+814+590+2040+Annemarie+Sch) *Updated 30 October 2026*