Running Automatic Transcription for the Most Challenging Sources: Tabular Parish Registers from Transylvania

Abstract The paper explores the challenges and adaptations required for applying Transkribus to the parish registers from Transylvania, used as main sources for the HPDT and the complexity introduced by linguistic, alphabetic, and structural diversity inherent to the sources. The denominational and ethnic mosaic of Transylvania has resulted in parish registers written in a variety of languages—Latin, Hungarian, German, and Romanian— and using diverse alphabets, including Latin, Cyrillic, and Kurrentschrift. In many instances, these multilingual registers coexist within records from the same village, further complicating the recognition process. The structure of these tables, which can vary significantly, adds another layer of complexity. The discussion is focused on adapting and developing the Transkribus platform to accurately recognize and convert the handwritten text into a format compatible with database structure. The adaptation process involves fine-tuning recognition models for each specific language, alphabet, and tabular configuration.