Would it be possible to create an AI that translates Japanese games accurately?

If you were to feed an AI untranslated Japanese media and the translated version of the same media, it could compare and contrast and learn how to translate it properly without doing literal 1 to 1 translations that machine translation does.

I've been playing a lot of untranslated rpgmaker hgames using machine translation, and for example whenever characters are orgasming they say "I'm going to go, I'm going to go", which I'm guessing the machine translator is trying to translate the word "iku".
But if you were to translate it with the AI, it could just learn through the material it's been fed that in this context, human translators translated "iku" to "I'm cumming".

Would it be possible to create this sort of AI?

Attached: cat stare.jpg (980x652, 37.08K)

Google translate already does that, don't let the nippon mysticism fools you.

Wrong

Is this really true?

Sooner or later. I'd subscribe to that.

deepl?

Impossible.
I know nip and have a CS degree and I tell you, you fucking can't.

We will forever be stuck with translations that are at best 90-95% correct but every now and then it'll completely shit the bed and make no sense.
AI can't learn Japanese (just like you).

>Impossible.
That's what they said about the horseless carriage.

It's possible but the tech probably needs another decade to be close to decent.

Extremely difficult. East asian languages are incredibly unpractical and often rely in context and nuances that an AI will not catch. The chinese famously often have trouble understanding each other, imagine an AI trying that.

You should have a biological neural network that you can just train to do exactly this

You're describing the way machine translation worked 20 years ago.

Attached: Sigmund_Freud,_by_Max_Halberstadt_(cropped).jpg (1200x1632, 270.31K)

I already answered you on Any Forums

Not entirely. You're right that east Asian languages are heavily context-dependent, which makes translation to a more precise language like English difficult, but things like domain identification (what's the topic of the sentence or paragraph?) and full-work translation can get around issues like that.

AIは日本語が出来ない

DeepL is downright magical but I don't speak Nip so I can't verify it.
That said at this point a semi-intelligible machine translation sounds preferrable to whatever the fuck those schizo "localisers" are doing.

Horseless carriage is a matter of physical efficiency, not a matter of "Here's a language quirk that not even Japanese know 90% of the time" "Here's a pun about a historical figure most Japanese haven't even heard the name of".

deepL does a really good job, but it still fucks some things up every now and then

Now that's something that's truly difficult even for robust translation NNs capable of domain identification. Japanese written language puns are pretty impenetrable without a full-on world history database explorer like Watson

You've been here for over a decade and you never bothered learning Japanese?
What were you doing all this time...

Attached: 1659267986420.jpg (1600x1066, 589.41K)

You don't even need to know nip to see it fucks up constantly, skips parts of the text and messes up pronouns.

>talking with bros on voice chat in JP
>I read a lot of ln/vn
>use a word that is extremely common in those formats while talking
>friends ask what the fuck I just said
>type it out
>they literally can't read the fucking word either
>"use normal words dude"
the Japanese themselves don't fucking know their own language half the time and it makes me wonder if they straight up just mind read half the things they hear and see rather than understanding the literal words being said or read.

Machine translation will never be good go learn Japanese you lazy fuck