BGT5 – PRE-TRAINED T5 MODEL ON BULGARIAN DATA

Abstract

In recent years many resources were created in Natural LanguageProcessing (NLP) but most of them are available in other languages than Bulgarian.With the exception of few multilingual models, which support Bulgarian language, thereare not many others. In this work, we collect Bulgarian data from different sources andpre-train a T5 model on this data. We further evaluate its performance against othermultilingual models on different Natural language processing tasks.