I remember re-reading the Asterix and Obelix comics and being incredibly impressed by the translation. It's not really something you think about as a child reading it, but the amount of language...
I remember re-reading the Asterix and Obelix comics and being incredibly impressed by the translation. It's not really something you think about as a child reading it, but the amount of language jokes in it is amazing - presumably almost all of them had to be entirely reworked from the original. I really can't see AI sticking the landing with things like that anytime soon.
It's not entirely the same as your example but it hits the same notes I think - translation is partially recreating the work, but also partially adapting the work, and in fiction specifically translators have to work hard to hit that balance.
Yeah this is my assessment for my own work as an audiobook narrator of fiction. AI isn’t taking my job away any time soon. I’m a character actor who inhabits each role in a bunch of ways. There’s...
Yeah this is my assessment for my own work as an audiobook narrator of fiction. AI isn’t taking my job away any time soon. I’m a character actor who inhabits each role in a bunch of ways. There’s so much context in each line…
Like in my own novels I’m narrating now, Alonso is a 50 year old scientist with a Cuban accent. He also loves to drink and sing opera. As the writer, I had him sing the opening line to Una Furtiva Lagrima while tired and drunk, but ultimately as a celebration.
There were so many layers to each syllable of sound. A Cuban-American singing in Italian. He is a victim of torture and many of his words echo the trauma. But he is also singing to his loved ones in a camp under the stars. It took many attempts to get all the elements right.
Sounds like you are one of the better audiobook narrators. I have heard some narrators that are much more mediocre. I would guess that AI could replace those mediocre narrators, but not the truly...
Sounds like you are one of the better audiobook narrators. I have heard some narrators that are much more mediocre. I would guess that AI could replace those mediocre narrators, but not the truly skilled/talented ones. As a consumer, I would be happy to rebuy a book for a better or different narrator.
Thanks. Yeah I am immensely picky myself when it comes to listening to other narrators, as you can imagine. But I have more sympathy for the bad ones than many of their listeners. Doing this well...
Thanks. Yeah I am immensely picky myself when it comes to listening to other narrators, as you can imagine. But I have more sympathy for the bad ones than many of their listeners. Doing this well generally takes a decade of preparation and each project is hundreds of hours of work, usually for a gamble or a fairly paltry fee.
Those awful narrators who get slammed so much online are almost all doing their best. Nobody can be a narrator and be lazy. Lazy people try and never get to completion.
To get back to the OP subject, I am thrilled at the tools AI will add to my workflow. I’m also pretty sure that most of us will prefer human made art forms, at least for another generation or two. But to be able to precisely cancel all my breaths and mouth noise and background sounds? Yes please. I could narrate a book while hiking in the forest.
I think that integrating AI tools into existing workflows is really the future of AI. AI tools are currently the most well developed for the programming industry, and we have already seen them...
I think that integrating AI tools into existing workflows is really the future of AI. AI tools are currently the most well developed for the programming industry, and we have already seen them become pretty integral to that industry. In 50 years, I don’t think we will find that AI has taken a bunch of jobs, but improved the workflow and quality of many different sectors. We don’t complain today about how nail guns, power drills, saws, and other technologies took the jobs of carpenters and other trades.
Both the professional translators I know say the same thing. Translating simple text and documents by machine has been essentially solved already - even before the fancy chatbots came along. But...
Both the professional translators I know say the same thing. Translating simple text and documents by machine has been essentially solved already - even before the fancy chatbots came along. But translating fiction is a huge step from what the ML systems can do right now. It requires so much contextual awareness and the kind of long-form consistency and deep cultural understanding that the machines are simply not capable of - so far. Perhaps they will be one day. But translating fiction seems like a fairly niche case, it's probably cheaper to just keep paying humans to do the job than pour a shedload of development and compute resources into automating it.
Just as an aside, the people I know who translate fiction for a living take many months, even years sometimes, to complete a work. Some authors write entire books faster than that! It's hard work and it's not that well paid either.
I've seen a czech fiction translator (an excellent one) say pretty much the same thing and the reasoning and examples he provided made perfect sense. However I'm still mildly skeptical because I...
I've seen a czech fiction translator (an excellent one) say pretty much the same thing and the reasoning and examples he provided made perfect sense.
However I'm still mildly skeptical because I see some evidence that LLMs could actually be considerably better at translating fiction than they are now if they were finetuned for it. The main reason being that while tasks like "translate x into y" provide results that are not nearly good enough for fiction for various reasons, LLMs seem to be surprisingly good at explaining even things that are very difficult for human translators to do.
I've seen GPT-4 correctly explain a difficult double entendre consisting of an idiom and a reference to a latin phrase by Terry Pratchett that even various human translators got partially wrong, and suggest a translation that was not as good as what the actual (imo world class) translator came up with, but it was, in my opinion, definitely acceptable.
I would deduce from this that LLMs have capabilities that are partially hidden by the way we use them nowadays, and that may be uncovered and become easier to use with time. Of course, this only solves some of the existing issues with translating fiction, not all of them, and it may take some years before enough development is done because translating fiction is famously underpaid, so there are many areas with a larger financial gain to focus on.
I wonder if there is enough training material on the net in languages that are not French, German, Russian, Spanish, Chinese, Japanese or English. The subtlety and nuance required training....
I wonder if there is enough training material on the net in languages that are not French, German, Russian, Spanish, Chinese, Japanese or English.
The subtlety and nuance required training. English has the most content online.
It's not just fiction by extension but anything that wants to convey tone/meaning. If you're translating instructions yeah that's going to be easy, but if you're dealing with a discussion between...
It's not just fiction by extension but anything that wants to convey tone/meaning. If you're translating instructions yeah that's going to be easy, but if you're dealing with a discussion between two humans it's very very likely you'll get idioms and expressions that do not translate naturally/cleanly at all.
This is whatever/funny between friends and absolutely disastrous for things like business, law, and diplomacy.
I'm kicking myself because I remember reading (listening to? watching?) something where a translator was making a similar conclusion, but I can't find it either. The translator made a point where...
I read an interesting interview with a professional translator a while back — Unfortunately, I'm not able to find it anymore.
I'm kicking myself because I remember reading (listening to? watching?) something where a translator was making a similar conclusion, but I can't find it either. The translator made a point where there are domains where the cost of a machine mistranslating a document can be tremendous (e.g. contract law).
The author uses the lens of a video game localization to explore how a translator chooses to represent Japanese "profanity" in English. A process which has a surprising amount of depth, since there isn't a clear 1-to-1 for how swearing is done in English vs Japanese.
DeepL kicks Google Translate's ass at the languages it supports. But yeah I largely agree with @douchebag's expert -- translating fiction is an artform and we're not even close to having models...
DeepL kicks Google Translate's ass at the languages it supports.
But yeah I largely agree with @douchebag's expert -- translating fiction is an artform and we're not even close to having models that can even compete with a skilled human on the subtleties involved, much less outcompete them. The competition is stiffer with nonfiction, but I think many of this use is in contexts where a human translator would be unlikely to be hired in the first place. I wouldn't be hiring a translator to help write my emails in German; I'd just be writing worse emails. The only time I've hired a human translator has been when I needed a certified translation for documents I was submitting as part of a visa application, and AI is highly unlikely to be allowed in those contexts anytime soon -- I need someone to sue if they fuck it up.
Technical documentation is really the only area I can see AI actually replacing human translators, but even there I'm not sure, it'll depend a lot on the field whether the AI is sufficiently good at accurately translating technical nuances to replace a human.
A translation into English of an article in Chinese about translating from English to Chinese with the help of AI. From the article:
A translation into English of an article in Chinese about translating from English to Chinese with the help of AI. From the article:
I started using Google Translate as my “assistant” around 2021. I will start by letting it translate a first draft, and then I will proofread it, which can help me increase my efficiency by two to three times. But this method is only suitable for non-fiction articles exemplified by business or technical documents. This type of article has a simple sentence structure, straightforward language, and is relatively easy to follow standard patterns" - this is also in line with the characteristics of the positions that experts predict will be the first to be replaced by AI, that is, work that does not require high professional knowledge and creativity, and can be produced in a standard mode, such as the writing of template articles such as general reports and media announcements, as well as customer service, data analysis, and ordinary customized programming, etc.
When it comes to content that is complex or has subtle connotations, AI translation will still be incoherent or full of errors. In terms of translation quality alone, ChatGPT is not superior to Google Translate, and its translations are often very rough. However, it can bring about human-computer interactions and can — by following specific human instructions — adjust the translation, improve the quality, or meet certain special requirements for the language style of the translation.
I used to call my sister to discuss things when I was working as a translator. She studies Chinese linguistics. Sometimes I understand the meaning of a piece of English, but I can't think of an authentic Chinese expression. I would explain to her what I need to express, and she would give me some inspiration. Now ChatGPT can well assume the role of this linguist consultant and can be on call anytime and anywhere. But does this mean that with AI translation, humans can achieve near-costless “freedom of translation”? My answer is: "It's not that simple."
I remember re-reading the Asterix and Obelix comics and being incredibly impressed by the translation. It's not really something you think about as a child reading it, but the amount of language jokes in it is amazing - presumably almost all of them had to be entirely reworked from the original. I really can't see AI sticking the landing with things like that anytime soon.
It's not entirely the same as your example but it hits the same notes I think - translation is partially recreating the work, but also partially adapting the work, and in fiction specifically translators have to work hard to hit that balance.
Yeah this is my assessment for my own work as an audiobook narrator of fiction. AI isn’t taking my job away any time soon. I’m a character actor who inhabits each role in a bunch of ways. There’s so much context in each line…
Like in my own novels I’m narrating now, Alonso is a 50 year old scientist with a Cuban accent. He also loves to drink and sing opera. As the writer, I had him sing the opening line to Una Furtiva Lagrima while tired and drunk, but ultimately as a celebration.
There were so many layers to each syllable of sound. A Cuban-American singing in Italian. He is a victim of torture and many of his words echo the trauma. But he is also singing to his loved ones in a camp under the stars. It took many attempts to get all the elements right.
I don’t see AI replacing that work anytime soon.
Sounds like you are one of the better audiobook narrators. I have heard some narrators that are much more mediocre. I would guess that AI could replace those mediocre narrators, but not the truly skilled/talented ones. As a consumer, I would be happy to rebuy a book for a better or different narrator.
Thanks. Yeah I am immensely picky myself when it comes to listening to other narrators, as you can imagine. But I have more sympathy for the bad ones than many of their listeners. Doing this well generally takes a decade of preparation and each project is hundreds of hours of work, usually for a gamble or a fairly paltry fee.
Those awful narrators who get slammed so much online are almost all doing their best. Nobody can be a narrator and be lazy. Lazy people try and never get to completion.
To get back to the OP subject, I am thrilled at the tools AI will add to my workflow. I’m also pretty sure that most of us will prefer human made art forms, at least for another generation or two. But to be able to precisely cancel all my breaths and mouth noise and background sounds? Yes please. I could narrate a book while hiking in the forest.
I think that integrating AI tools into existing workflows is really the future of AI. AI tools are currently the most well developed for the programming industry, and we have already seen them become pretty integral to that industry. In 50 years, I don’t think we will find that AI has taken a bunch of jobs, but improved the workflow and quality of many different sectors. We don’t complain today about how nail guns, power drills, saws, and other technologies took the jobs of carpenters and other trades.
Both the professional translators I know say the same thing. Translating simple text and documents by machine has been essentially solved already - even before the fancy chatbots came along. But translating fiction is a huge step from what the ML systems can do right now. It requires so much contextual awareness and the kind of long-form consistency and deep cultural understanding that the machines are simply not capable of - so far. Perhaps they will be one day. But translating fiction seems like a fairly niche case, it's probably cheaper to just keep paying humans to do the job than pour a shedload of development and compute resources into automating it.
Just as an aside, the people I know who translate fiction for a living take many months, even years sometimes, to complete a work. Some authors write entire books faster than that! It's hard work and it's not that well paid either.
I've seen a czech fiction translator (an excellent one) say pretty much the same thing and the reasoning and examples he provided made perfect sense.
However I'm still mildly skeptical because I see some evidence that LLMs could actually be considerably better at translating fiction than they are now if they were finetuned for it. The main reason being that while tasks like "translate x into y" provide results that are not nearly good enough for fiction for various reasons, LLMs seem to be surprisingly good at explaining even things that are very difficult for human translators to do.
I've seen GPT-4 correctly explain a difficult double entendre consisting of an idiom and a reference to a latin phrase by Terry Pratchett that even various human translators got partially wrong, and suggest a translation that was not as good as what the actual (imo world class) translator came up with, but it was, in my opinion, definitely acceptable.
I would deduce from this that LLMs have capabilities that are partially hidden by the way we use them nowadays, and that may be uncovered and become easier to use with time. Of course, this only solves some of the existing issues with translating fiction, not all of them, and it may take some years before enough development is done because translating fiction is famously underpaid, so there are many areas with a larger financial gain to focus on.
I wonder if there is enough training material on the net in languages that are not French, German, Russian, Spanish, Chinese, Japanese or English.
The subtlety and nuance required training. English has the most content online.
It's not just fiction by extension but anything that wants to convey tone/meaning. If you're translating instructions yeah that's going to be easy, but if you're dealing with a discussion between two humans it's very very likely you'll get idioms and expressions that do not translate naturally/cleanly at all.
This is whatever/funny between friends and absolutely disastrous for things like business, law, and diplomacy.
I'm kicking myself because I remember reading (listening to? watching?) something where a translator was making a similar conclusion, but I can't find it either. The translator made a point where there are domains where the cost of a machine mistranslating a document can be tremendous (e.g. contract law).
Separately, the video 999 - How to Make Good Use of Bad Language by Aquason was on an eye opener on how much of an art a good translation/localization is.
The author uses the lens of a video game localization to explore how a translator chooses to represent Japanese "profanity" in English. A process which has a surprising amount of depth, since there isn't a clear 1-to-1 for how swearing is done in English vs Japanese.
DeepL kicks Google Translate's ass at the languages it supports.
But yeah I largely agree with @douchebag's expert -- translating fiction is an artform and we're not even close to having models that can even compete with a skilled human on the subtleties involved, much less outcompete them. The competition is stiffer with nonfiction, but I think many of this use is in contexts where a human translator would be unlikely to be hired in the first place. I wouldn't be hiring a translator to help write my emails in German; I'd just be writing worse emails. The only time I've hired a human translator has been when I needed a certified translation for documents I was submitting as part of a visa application, and AI is highly unlikely to be allowed in those contexts anytime soon -- I need someone to sue if they fuck it up.
Technical documentation is really the only area I can see AI actually replacing human translators, but even there I'm not sure, it'll depend a lot on the field whether the AI is sufficiently good at accurately translating technical nuances to replace a human.
A translation into English of an article in Chinese about translating from English to Chinese with the help of AI. From the article: