A rule-based information extraction (IE) method for the Arabic-English Al-Mawrid dictionary achieves high precision for morphological, syntactic, and semantic lexical data, with strong recall for synonym sets, per a June 26, 2026 arXiv preprint1.
The Al-Mawrid is a widely used bilingual Arabic-English reference dictionary, a core resource for Arabic linguistic research and natural language processing (NLP) development. The method first applies n-gram and keyword-in-context (KWIC) analysis to surface lexical patterns tied to morphological, syntactic, and semantic information embedded in Al-Mawrid entries. It then uses hand-crafted rule-based extraction to pull structured data from those patterns1.
Method Design and Lexical Data Targets
Custom heuristics and punctuation mark analysis are deployed to isolate synonym sets within individual dictionary subentries, addressing a key gap in structured lexical data for Arabic reference resources. The full 9-page preprint includes 4 tables of cross-category performance metrics and 7 figures detailing the hand-crafted rule sets used for each extraction task1.
Performance Metrics by Lexical Category
Testing of the system confirmed high precision scores for every category of extracted lexical data, including derivations, semantic relations, syntactic labels, and morphological information1. The method also posted strong recall performance specifically for synonym sets, a critical data type for NLP tasks like word sense disambiguation and machine translation. Recall scores for derivations and cross-entry semantic relations were measured as lower than precision scores for those same categories, a common trade-off in rule-based extraction systems for complex Arabic lexical data1.
The 4 tables included in the preprint break down precision and recall metrics by individual lexical data type, offering granular performance data for developers building Arabic NLP tools. The 7 accompanying figures illustrate the full set of hand-crafted extraction rules used for each task, providing a reproducible framework for similar work on other bilingual Arabic lexical resources1.
Research Background and Public Access Timeline
The research, titled Extracting Knowledge from an Arabic-English Machine-Readable Dictionary Using Information Extraction, was first presented as a conference paper at the 5th International Conference on Arabic Language Processing (CITALA 2014) in Oujda, Morocco, in November 20141. No public proceedings PDF for the 2014 CITALA conference was ever released, leaving the full research inaccessible to the broader NLP and computational linguistics community for more than a decade.
The June 26, 2026 arXiv posting marks the first time the full 9-page paper, including all performance tables and rule set figures, is publicly available to researchers and developers worldwide1. The preprint is hosted permanently at https://arxiv.org/abs/2606.28457, and was fact-checked against the original submission before publication, with a last verification date of July 1, 20261.
The work is relevant to teams building Arabic NLP lexical extraction pipelines, lexical databases, and bilingual dictionary tools, as it provides a tested, reproducible framework for extracting structured data from existing machine-readable Arabic reference dictionaries1.
