Google's new AI-powered translation tool

[lips]

I don't see any reason why standardized texts like product descriptions, simple technical explanations, and standardized instructions cannot be produced (not translated) in multiple languages by a computer.  Maybe they are already.  I remember translating some financial product descriptions.  After at few I was thinking, "Surely a computer can write these in multiple languages without the need of translation."  For example, Ikea assembly instructions.  Surely a computer can write,"Attached handle A to door B using screws C" in many languages.  Maybe Ikea is already doing it, generating assembly instructions directly from engineering documents by a computer.

[imron]

IKEA are already way ahead of you. Their assembly instructions are picture only these days, no translators required.

[Angelina]

I remember translating some financial product descriptions.

 

 

 

Chinese water torture. 

[laurenth]

Off-topic: If you haven't already, go and read the short story "Profession" by Isaac Asimov.  You'll enjoy it.

 

 

I'd say entirely on-topic, on the contrary. Thanks for the recommendation imron, I enjoyed the story immensely.

[imron]

Off topic with respect to Google's translation tool, on topic with your comments :-)

[Angelina]

Excellent story! Thanks. 

 

If there are people afraid they are going to lose their jobs to AI, for example translators losing their jobs to Google Translate, they should read Profession. 

[Demonic_Duck]

I just translated a contract using Google translate as my primary tool. Was surprisingly good, but there were a few places where simply using the machine-translated version would have conveyed an incorrect meaning (as opposed to simply failing to convey the correct one). In this case it was a non-legally-binding contract, but I can imagine that for higher-stakes materials people might still shy away from using it for fear that that kind of thing would pass unnoticed.

[Lu]

I mostly translate literature (and that's never going to be replaced by any AI), but I can think of a few texts that I have translated recently that the client would probably have used computer translation for if 1) computer translation were really good and 2) they knew how and which program to use. So there is less work in the future, probably.

 

And now that I think of it, hey, I translate to Dutch. Take any errors that get into Chinese-English translation, that number squared is the number of mistakes that will end up in the Dutch translation, because Google doesn't actually offer Chinese to Dutch translation, it takes everything through English first. No need to worry for my livelihood just yet. And I can always go and rely on interpreting. The day a computer can handle a Sichuanese accent will be well after my retirement.

[li3wei1]

"because Google doesn't actually offer Chinese to Dutch translation"

but there's nothing to stop it from doing that. Once the principle is in place, all you have to do (as I understand it) is line up the Dutch and Chinese corpi and let a big fat computer look at if for a few minutes, let it loose on the public and have it learn from a few million interactions, and you're there. Maybe an extra server, a few hard drives, and a few keystrokes, and you've replaced how many humans?

Same with Sichuanese accents. Voice recognition is programed to learn, so given enough time and correction, it will learn whatever accent it works with. Yes, any given computer will take a while to learn the accent, but I bet there are plenty that already have. 

[Lu]

but there's nothing to stop it from doing that. Once the principle is in place, all you have to do (as I understand it) is line up the Dutch and Chinese corpi and let a big fat computer look at if for a few minutes, let it loose on the public and have it learn from a few million interactions, and you're there.

And yet, they haven't. Google Translated has existed for quite some years now and it doesn't offer this. I don't know what's stopping them either, but fact is that they haven't done this. Perhaps the corpi aren't quite big enough? I know that when translating highly specialised things I often find it easier to translate to English, because a lot of the vocabulary is much easier to find.

[Kenny同志]

Was surprisingly good, but there were a few places where simply using the machine-translated version would have conveyed an incorrect meaning (as opposed to simply failing to convey the correct one). 

Translation, particularly legal translation, is rarely free of mistakes. If Google Translate was able to translate a 60-page or even 20-page contract with only a few errors, it would equal, if not do better than most human translators in terms of accuracy. 

[Angelina]

Lu, translating literature is great    

[Lu]

Translation, particularly legal translation, is rarely free of mistakes. If Google Translate was able to translate a 60-page or even 20-page contract with only a few errors, it would equal, if not do better than most human translators in terms of accuracy.

Perhaps I have high expectations, but if I wanted a legal translation, I'd want an end product free of mistakes. For that, I'd generally need both a translator and someone to check the translator, but once they are both through with it I don't expect any errors left in the translation.

 

Of course, if machine translation gets really good, I could just make do with two proofreaders instead of a translator and a proofreader, which would be cheaper. Apparently this trend is already visible all through the translation industry (though I haven't yet encountered it myself). Some translators even refuse to work on machine translations, saying it will only serve to end the whole profession faster, but personally I think that's rather pointless.

[Demonic_Duck]

Translation, particularly legal translation, is rarely free of mistakes.

If anything, I'd expect legal translation to be both more formulaic (thus less mistakes occurring naturally) and higher stakes (thus more eyes being assigned to each project) than most other types of translations.

I don't typically work on legal translations, though (and this was no exception, due to the non-binding nature of the contract).

[realmayo]

Coincidentally, a long article in the guardian today https://www.theguardian.com/technology/2016/oct/11/crash-how-computers-are-setting-us-up-disaster about letting computers take over from humans: accidents/mistakes happen less often but when they do the humans either find it difficult to accept that the computer did actually come up with the wrong result, or they lack the skills to take over because those skills have atrophied through disuse (e.g. airline pilots).

 

Lu, when you (& others) talk about proofreaders, are you referring to people who are able to understand the two languages well enough to spot errors of meaning in the translation?  And does this kind of work currently pay less per hour than full-on translating?

[Kenny同志]

if I wanted a legal translation, I'd want an end product free of mistakes. For that, I'd generally need both a translator and someone to check the translator, but once they are both through with it I don't expect any errors left in the translation.

 

That would be the way to go. I imagine there would be more editing work for me and clients would still like to hire me to come up with a draft translation where creativity/rewriting is required but my gross income may suffer badly as editing is usually less paid for. 

 

Anyway, I will do whatever I can to continue on this career path. If I cannot translate + edit to make ends meet, I may get a new job (not sure what I can do except translating) and make translating a hobby.

[Lu]

Lu, when you (& others) talk about proofreaders, are you referring to people who are able to understand the two languages well enough to spot errors of meaning in the translation?  And does this kind of work currently pay less per hour than full-on translating?

Yes and yes. Although translating is typically paid per word/character and proofreading (to my knowledge) usually per hour. It makes sense, too, that proofreading would pay less: you have to look up things here and there of course, and correct things, but everything that is already correct don't have to think of any further, while the translator had to take the time to think about the meaning, make a good sentence and type it up. In my experience (but so far I've only proofread literary translations of my co-translator), proofreading takes time and effort, but less time and effort than translating.

 

Kenny, but didn't you have another job before this? Not sure what the right path for you would be. Does China have certified translators they use for official documents and such? That can sometimes pay well. Or you could consider learning another language, if that's not too much effort. I heard Dutch to Chinese pays crazy well (20+ eurocent per word), because so few people offer it. Or you could try getting into literary translation, which should at least ensure you ongoing work, although to my knowledge pay in China is pretty bad.

[Kenny同志]

Some translators even refuse to work on machine translations, saying it will only serve to end the whole profession faster, but personally I think that's rather pointless.

Google Translate seems to translate legal texts quite well but when it comes to texts that are much less formulaic or that follow traditional Chinese structures, the results it generates can be horrible--I tested a Chinese email I had translated from English with the program, and proofing this type of work would involve a whole lot of retranslating. So in some cases, it is wise to refuse the work.

[lips]

... translators even refuse to work on machine translations ...

I don't refuse to work on machine translations but I have to see the translated text first before I'd accept a proofreading/editing job. If it's a machine translation I'd just quote the translation rate. My experience is that at least 75% of a machine translation has to be corrected. The same applies to really bad human translations, which are common nowadays.

[Kenny同志]

Kenny, but didn't you have another job before this? Not sure what the right path for you would be. Does China have certified translators they use for official documents and such? That can sometimes pay well. Or you could consider learning another language, if that's not too much effort. I heard Dutch to Chinese pays crazy well (20+ eurocent per word), because so few people offer it. Or you could try getting into literary translation, which should at least ensure you ongoing work, although to my knowledge pay in China is pretty bad.

No, I had not had any other job for a significant period of time before I made translation a full-time job. 

 

Given the fast developments in artificial intelligence, I probably won't commit myself to learning a new language. Literary translation might be an option for me and I will see if I can get any opportunities. I know of a Beijing-based publisher who pays translators 0.25-0.30 RMB per word so it isn't  that bad, although this sounds too low compared with the rates I could charge an Australian or British client.