How to custom train a language different than English when it comes to OCR?

Question

I am thinking that the process to train an OCR for any given language is that same as the English language with the input ground truth labels being dependent on the type of