Anonymous asked:
Do you think translation tools will ever be able to incorporate context for japanese
Yes. I would have been slightly more skeptical a few months ago, but at this point, I think it's only a matter of time. I've heard that GPT4 can translate things better than DeepL—which isn't all that high a bar, as I've complained in the past. But my main problem with modern machine translation is that it's trying to "cheat", by focusing on making the output convincing, rather than actually producing a more accurate translation. With recent advances, we now have AI that can identify and explain the logic behind a joke, and that's the same kind of understanding that a computer would require to translate a context-heavy language.
Assuming that AI actually did reach the point where it made human translation obsolete, you would only need two things. To start with, you would have to change the nature of text hooking programs to let the computer know what the context was in the first place. And secondly, you would need an open source AI rather than a commercial one, because inevitably some asshole ethicist is going to tamper with it and make it useless for our purposes.
Heh, hate to tell you, there's no "understanding" that joke there. All the current crop of "AI" are just statistical word machines - there is zero understanding of the content they're generating; they've simply been trained on a very large and strong corpus of data, and had the weights tuned heavily. All it knows is "this word follows that sequence of words", and that's not what humans do at all.
ReplyDeleteNot to say they can't do that with a combination of chatbot tech and more emphasis on actual understanding and learning of content, but I'd expect an excellent Japanese translation AI long before AI *actually* understands the logic of a joke.
I don’t care if it is a Chinese Room situation; the effect is the same. If a computer can correctly identify, by whatever means, that a joke must be about dogs because it revolves around dog stereotypes, then that same kind of “thought” can be used to parse one Japanese sentence with contextual knowledge from another. Nobody’s suggesting actual thinking and sentience from these things. Apart from that one researcher who claimed that his AI thought it was a person trapped in a computer, and tried to Google how to get itself out.
Delete