Probabilistic Record Linkage for Arabic-Script Archival Material: A Fellegi–Sunter Framework with Patronymic-Chain Features

Le résumé fourni par la source

Author's original version (preprint). Submitted to Digital Scholarship in the Humanities (Oxford University Press), 25 August 2026, manuscript DSH-2026-1071. Cross-source record linkage in Arabic-script archival material poses challenges that mainstream genealogical and prosopographical infrastructure does not address: multi-token patronymic naming (ism, nasab, nisba) interacts with heterogeneous transliteration practice, so the same individual appears under many inconsistent surface forms across colonial-era and pre-modern sources. We present a reference implementation of the Fellegi-Sunter probabilistic framework with eight Arabic-aware comparison features, each assigned a log-likelihood-ratio weight from domain-tuned initial probabilities, with Beta(1,1)-smoothed Bayesian re-estimation from archivist review decisions. On a two-corpus synthetic test set the pipeline achieves 100% mention recall on narrative material and 87.5% cross-source merge recall with 100% precision on observed merges; per-fact provenance coverage, an invariant of the architecture, is verified at 100% over the consolidated graph.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.