The Early Islamic Conquests Database–Balādhurī (EICD-B): large language model extraction of conquest data from Futūḥ al-buldān

Abstract

Abstract This article introduces the Early Islamic Conquests Database–Balādhurī (EICD-B), a structured dataset extracted from al-Balādhurī’s ninth-century chronicle Futūḥ al-buldān using large language models (LLMs). Building on the OpenITI/KITAB digital corpus, a four-stage workflow transforms narrative historical material into machine-readable conquest records: pre-processing, schema-guided extraction, normalization, and validation. Each record contains the Hijrī year, toponym, commander name, caliph name, and mode of conquest (Capitulation/ṣulḥan, Subjugation/ʿanwatan, or Reconsolidation), with a textual evidence snippet enabling source verification. The validated dataset comprises 226 conquest events dated primarily to 11–97 AH (632–716 CE), with three events (99 and 221 AH) retained as temporal outliers, all successfully geocoded. Results indicate 52.2% Capitulation, 42.5% Subjugation, and 5.3% Reconsolidation. ʿUmar’s caliphate dominates with 99 conquests (43.8%), while fitna periods show a marked decline in conquest activity. EICD-B provides the first machine-readable, evidence-linked dataset of early Islamic conquest events extracted from a classical Arabic chronicle, enabling quantitative testing of historiographical hypotheses such as Noth’s schemata thesis. The study demonstrates a replicable LLM-based workflow for schema-guided extraction from historically inflected Arabic, while emphasizing the methodological importance of retaining textual evidence spans for verification. The transparent methodology and reproducible dataset enable new computational approaches to early Islamic history.

The authors' abstract, as published at the source. Digital Scholarship in the Humanities, 2026 · DOI ↗