Ibero-Romance Handwriting in Arabic Script (IRHAS): A First Step to Manuscript Text Recognition in Spanish Ajami (Aljamiado)

The Ibero-Romance Handwriting in Arabic Script (IRHAS) dataset provides ground-truth data for Automatic Text Recognition (ATR) of Spanish Ajami or Aljamiado (Spanish written in Arabic script) manuscripts. Comprising 22 extracts from 19 Mudejar and Morisco codices from the late 15^th to the early 17^th centuries (663 pages, 10,825 lines, 550,299 characters in total), the corpus was curated via eScriptorium and controlled using Choco-Mufin. Stored and documented on Zenodo under an open-access licence, IRHAS enables the fine-tuning of generic ATR models, achieving an average Character Accuracy Rate of 92.54%, which will facilitate the production of scholarly editions and, eventually, text mining and analysis.