>>106558511
>>106558650
As an avid ex enjoyer I can say that MTL is already used and is an issue as it is not good. That said the general meaning is usually understood such that you can proofread and correct it with minimal interactions with the source material. So yes such software already exists. The issues are that:
OCR is not solved for CJK languages, esp. Japanese as there is less training data and most researchers in the field are chinks not japs
(In fact you are more likely to have some tranime faggot make an Japanese OCR than an actual Japanese)
Japanese is not as gendered as English, so often the translation software will need to make a guess for pronouns. This should not be an issue with current AI models though, which are especially good at translation. You may need to tell it which characters are male or female at the start of the prompt though.
You still need to re-typeset everything at least for now.
I read xianxia using OCR to look up words I don't know and it works well but is somewhat taxing.