Unlocking Burgenland Croatian: An Automatic Text Recognition Model on Transkribus

Burgenland Croatian — a minority South Slavic language born from 16th-century Habsburg migrations — now has its first ATR model on Transkribus, achieving a remarkable 2.48% CER and outperforming multilingual Slavic supermodels in 94% of test cases. This work demonstrates what targeted specialisation can achieve for low-resource minority languages that generic models consistently fail.