I made a career switch from translation (JP > EN) into tech in 2022, and since ChatGPT was released very shortly after I made that decision I feel I got lucky with the timing. Many, many jobs are now "machine translation post-editing", which means you spend almost as much time manually checking and re-translating, but get paid significantly less.
Machine translation certainly has improved a lot, but still has the usual AI issues- namely that the AI confabulates details in places where a human translator would realize that something is unclear and refer to context, other parts of the text or external research for clarification. A common example in Japanese to English is that Japanese often omits pronouns, while the English version of the sentence requires "he"/"she"/"me"/"you" to be grammatical. A human translator can check this, but AI will often autofill whatever is the most common in its dataset, which could be completely wrong in context. So I think the risk with machine translation is that it can seem reliable and trustworthy 90+% of the time, but if you have no ability to read the original text and compare, you're rather quickly going to get caught out by something that looks like a professional, clean, correct translation, but is a total misrepresentation of what was actually said.
Yeah, the "explaining the difference between 'butt dial' and 'booty call'" problem.
Another thing I notice is that, if a translator is not familiar with how people in a certain culture interact, and what words they use in different situations, they can end up doing things like translating something as a polite, gentle explanation when it's clearly a mocking, condescending one. The literal meaning of the words is there, but they haven't picked up on the tone. Even English does this- "I beg your pardon", "With all due respect" and "As per my last email" are frequently (but not exclusively) hostile even though the literal words are formal/polite. This is also hard for AI to deal with.
Yeah but only because it's probably the only foreign language left with large fanbase of popular media, and thus a good number of critics for bad translation.
I'm surprised it's only a third of translators. The translation capabilities of ChatGPT are shockingly good in my opinion. And it's probably the first machine translation where you explain the context to it in order to get the best translation.
There's still translators needed for legal stuff like contracts, official government documents, etc. Not sure how much of the profession they make up but it sounded like they were the only ones not dwindling in numbers before ChatGPT came along.
I once worked on a contract translated with ChatGPT from German to Polish and it was the most infuriating experience. Text for a layperson seemed to be ordinary legal speak, but in reality it made no sense(or even less than ordinarily). I had to explain meaning of every legal term used and why it made no sense in given context.
I agree, since the pogrom on October 7th in Israel, I've been able to get translations from all the gory telegram channels from the respective sides of the conflict, and the superimposed foreign words on both the captions as well as images and videos. let alone the comments.
a lot of it turns out to be slang, and I've been able to ask ChatGPT for nuanced context on how its used, immediately.
I've never been able to do that in history. Duolingo would never do that, formal education wouldn't either. A dictionary wouldn't. Google translate couldn't.
can search more content and conversations using the topical slang as the keyword.
Yep. Transformers were originally developed for language translation, not chatbots or image generators. The fact that they proved useful in so many other contexts was serendipitous.
I assume they’re trying to say “let more”, which would be a pretty depressing read as it’s a very “ai maximalist” “gosh I hope everyone loses their job and the world magically adapts and we all live in a work-free paradise”, which I think has about as much likelihood as me winning the lottery tomorrow…
Or not a native English speaker. “AI has led to more people lose their job and still no visible benefit to society” doesn’t sound like AI maximalism to me
Machine translation certainly has improved a lot, but still has the usual AI issues- namely that the AI confabulates details in places where a human translator would realize that something is unclear and refer to context, other parts of the text or external research for clarification. A common example in Japanese to English is that Japanese often omits pronouns, while the English version of the sentence requires "he"/"she"/"me"/"you" to be grammatical. A human translator can check this, but AI will often autofill whatever is the most common in its dataset, which could be completely wrong in context. So I think the risk with machine translation is that it can seem reliable and trustworthy 90+% of the time, but if you have no ability to read the original text and compare, you're rather quickly going to get caught out by something that looks like a professional, clean, correct translation, but is a total misrepresentation of what was actually said.
reply