Bringing the old writings closer to us: Deep learning and symbolic methods in deciphering old Cyrillic Romanian documents

Bringing the old writings closer to us: Deep learning and symbolic methods in deciphering old Cyrillic Romanian documents

Blog Article

The paper addresses the problem of transliteration of scanned copies of old Romanian books written in the Cyrillic script into the Latin ORG OAT BRAN script.The motivation of this endeavor and attendees of such a technology are enumerated.Then, a number of peculiarities of these documents, which create difficulties for automatic processing, are exemplified.

The proposed technology is presented in the form of a pipeline of modules, each applying AI or symbolic methods.Then, the component parts are discussed individually, and solutions are presented.The research is presented as work in progress, which leaves space for further enhancements.

The data supporting training and evaluation of the modules is Match Holder rooted in the former DeLORo project.