The punctuation mark recognition is not accurate

[LindaEr]

If the language type is ['ch_sim', 'en'], it will be output according to English punctuation marks. But in fact, what I want is Chinese punctuation marks, but there are letters like ABCD in the text. It needs to be recognized as English. Is there any solution? For example, is there any priority ordering of language types?
Because I only use the python class library, I don’t understand model training and have no experience with it.