Reading the Menaion with HTR – evaluating approaches to computer-assisted manuscript studies
Reading the Menaion with HTR – evaluating approaches to computer-assisted manuscript studies

Summary/Abstract: This paper evaluates the use of handwritten text recognition (HTR) for computer-assisted textual, manuscript, and linguistic analysis of Church Slavonic material, focusing on the office Menaion tradition. Using two fifteenth-century East Slavic manuscripts of the November Menaion, the study compares the performance of existing broad HTR models for historical Cyrillic handwriting with newly trained, specialized models developed in Transkribus. Model accuracy is assessed using character error rate (CER) and qualitative error analysis, with particular attention to abbreviations, diacritics, superscripts, and word division. The results show that while large, generic models provide strong baseline performance, specialized models trained on a smaller but homogeneous dataset can achieve comparable or improved accuracy for material from a specific period and region. The article further analyses how HTRgenerated transcriptions can support quantitative linguistic analysis, testing hypotheses concerning orthographic features in texts thought to have been written in Novgorod in the 15th century. The study highlights the methodological potential of HTR for historical Slavistics, textual criticism, and philology, while addressing current limitations and best-practice.

• Details
• Contents