Abstract
While optical character recognition for printed Arabic and other right-to-left scripts has been principally solved, production-level deployment continues to pose critical challenges. These include demands of scale and precision, such as achieving near-perfect word accuracy, while navigating constraints like limited human resources and continually advancing technologies. Drawing on a use case centred on premodern named entities, we explore these challenges and identify best practices from a data engineering perspective. Rather than focusing on software development, the discussion addresses the practical realities of applications that remain perpetually evolving. Within this context, best practices designate kraken as the preferred OCR engine and eScriptorium as the primary tool for manual data manipulation.
In: Working Papers in Corpus Linguistics and Digital Technologies: Analyses and Methodology, 2026 (12), p.27-37.
Special Issue: AI in linguistics: opportunities and challenges. Editors: Markus Kunzmann (University of Vienna); Nicole Palliwoda (Christian-Albrechts-University of Kiel); Markus Pluschkovits (University of Vienna); Manuel Raaf (Ludwig-Maximilians-Universität Munich); Ines Röhrer (Ludwig-Maximilians-Universität Munich)
https://ebook.ek.szte.hu/index.php/btk-magyarnyelviirodalmi-intezet/catalog/book/440/chapter/740