Ka (digitalnom) izdanju rukopisa Gavrila Stefanovića Venclovića pomoću veštačke inteligencije

Abstract

The paper describes the process of creating and evaluating an HTR (Handwritten Text Recognition) model for the automatic text recognition of Serbian and Serbian Church Slavonic manuscripts by Gavrilo Stefanović Venclović using the Transkribus software platform, based on the principles of artificial intelligence, machine learning, and advanced neural networks. The training material for the model was taken from two manuscripts held in the Archives of the Serbian Academy of Sciences and Arts: Slova izabrana, written in the Serbian language and semi-uncial Cyrillic script; Razlagolnik, written in the Serbian Church Slavonic language and cursive Cyrillic script; and the Serbian Church Slavonic Menaion for October from the Archives of the Serbian Orthodox Diocese of Buda in Szentendre, written in semi-uncial Cyrillic script. The most important result of the work is the creation of the first version of the generic HTR model Vencl. 1.0 with exceptional performance – the percentage of incorrectly recognized characters is only 3%. The paper demonstrates how this model can be successfully used for the automatic transcription of other Serbian and Serbian Church Slavonic manuscripts by Venclović, which could finally make feasible not only a critical edition in the original script but also the creation of searchable digital editions, and subsequently, an electronic corpus of Venclović’s written legacy