/vt/ - Virtual Youtubers

>>1346660

i see a lot of people posting about computer translating japanese.
it's not going to happen.
i mean... i can't say it's never going to happen in the future, but i can assure you this: if we actually solve translation than having wasted time learning japanese is the least of our concern, because we most probably reached either artificial general intelligence or did such a big leap in AI that society will heavily change forever.

don't get me wrong we are making great progress in computational linguistic (natural language processing how you wanna call it) but these great progresses are still incredibly limited in scope. also all these progresses are in english and don't necessarily translate that well in other languages.

a lot of work is being done in machines understanding human language or producing it. translating is kind of a different thing.
for a program to be able to translate let's say a manga it would need to be able to watch the pictures and understand what those picture rapresent and how they are connected to the text. in practice for a computer to translate a manga it would need to understand not only how the material world works but also human culture.
if someone in the manga speaks about some birds for example since japanese has no plural you need to look at the pictures to undestand if they are talking about a single bird or more. so the program would need to look at the image recognize that some of those lines are a stilyzed depiction of the real world equivalent of a bird and that it needs to use the number of those birds to correctly disambiguate the translation.
this becomes even more insane when it needs to translate sentences about abstracts thoughs or human emotions and the manga makes use of visual metaphores to visualize these abstract thoughs. think about stuff like kaiji or baki where it's full of visual and verbal methaphores. now you would need your AI to understand that the manga isn't actually talking about a cobra and a moongose fighting despite the fact that both the text and the visuals are talking about a cobra and a moongoose fighting.

now i'm not saying this is not going to happen, it is technically possible. but this is a lot more complex than grabbing a huge japanese corpus running machine learning on it and we done. a lot of this actually goes outside of linguistic and it involves other fields of AI some of wich require the AI to be almost AGI.