Pages in topic:   [1 2 3 4] >
MT, "neural" stuff, and the future
Thread poster: Chase Faucheux
Chase Faucheux
Chase Faucheux  Identity Verified
United States
Local time: 13:58
German to English
Apr 19, 2019

Hi guys,

I've spent the past few days reading through a lot of forum posts from the past two years about the threat to translators posed by newer forms of machine translation (from the pessimists and cynics), or the great opportunities this is supposed to open up for us (from the starry-eyed optimists). After reading through all this, and playing with some of the more touted systems myself (DeepL in particular), here's what I'm worried about:

I don't think there's any t
... See more
Hi guys,

I've spent the past few days reading through a lot of forum posts from the past two years about the threat to translators posed by newer forms of machine translation (from the pessimists and cynics), or the great opportunities this is supposed to open up for us (from the starry-eyed optimists). After reading through all this, and playing with some of the more touted systems myself (DeepL in particular), here's what I'm worried about:

I don't think there's any threat now or in the near (maybe even distant) future of machine translation ever actually overtaking human translation in terms of actually understanding the sense of a text. Any machine capable of doing that is capable of replacing the proofreaders, the post-editors, the PMs, and even the clients. I'm also skeptical about the rise of Skynet ... uh, AI ... because it rests on the premise that the human brain is itself just a complex biological computer, which is not even close to being a settled topic in the hard sciences, to say nothing of philosophy and metaphysics. Maybe it's true, but maybe all our projecting and prognosticating about the future is nonsense, too.

The problem I see for translators in the medium term is that these services like DeepL are able to mimic human translation well enough to be considered "adequate" by clients and some agencies. And really, the output they produce is at least as good as the output of human translators who don't bother with things like deeply researching topics and checking to see if people actually use certain collocations. In any event, we're already seeing a race to the bottom with rates offered for post-editing, "discounts" on fuzzy matches, etc. The tendency here will be to drive talented translators who care about quality out of the market, especially those who are younger and not as well-established. And that's what worries me -- the effect that these new developments are actually having on the market.

By way of example, I've been asked to do post-editing work for stuff done with a custom system trained with client-specific material in the tourism and marketing field. The problem, of course, is that the machine has been fed with mostly "good enough" sub-par translation nonsense. Stuff like checking to see whether a "Berg" in question is actually a "hill" or indeed a "mountain", or whether a certain "Stadt" has a population of 10,000 and is actually typically referred to as a "town" or something else in English; this is just not something that many translators even bother with. So these errors are repeated and then baked into the machine. I refused the job, even though it was claimed other post editors were able to go at a rate of more than 1500 words per hour (yeah right, I'm suuure they were diligently checking for accuracy and style). Even DeepL relies on the Linguee corpus, half of which is nonsense, especially the EU stuff. Yes, it cross-references other translations, but what it doesn't look at is how similar texts are written in English for English speakers.

This is getting long, so I'll end now, but any thoughts?
Collapse


Tom in London
CARL HARRIS
Michele Fauble
Daryo
Sandra & Kenneth Grossman
Tim Sheedy
 
Tom in London
Tom in London
United Kingdom
Local time: 18:58
Member (2008)
Italian to English
The point Apr 19, 2019

Chase Faucheux wrote:

I don't think there's any threat now or in the near (maybe even distant) future of machine translation ever actually overtaking human translation in terms of actually understanding the sense of a text.


That is the point.


Chase Faucheux
Kay-Viktor Stegemann
Matthias Brombach
Daryo
 
Elizabeth Tamblin
Elizabeth Tamblin  Identity Verified
United Kingdom
Local time: 18:58
French to English
Concerns Apr 19, 2019

An agency I work for just sent me a questionnaire "to collect some feedback about your experience and interest in Light-PEMT, and willingness to use this technology for later MT Post-Edition projects".

Sounds ominous to me, and if they go down this route, I won't be sticking around.

I've done this kind of thing before, and found it to be far harder and more time-consuming than translating from scratch. I still have to check everything, as I'm just not capable of overloo
... See more
An agency I work for just sent me a questionnaire "to collect some feedback about your experience and interest in Light-PEMT, and willingness to use this technology for later MT Post-Edition projects".

Sounds ominous to me, and if they go down this route, I won't be sticking around.

I've done this kind of thing before, and found it to be far harder and more time-consuming than translating from scratch. I still have to check everything, as I'm just not capable of overlooking areas I'm not sure of.

In effect, the agency will be paying a lot less for more work, and I'm sure they know it.
Collapse


Chase Faucheux
Tom in London
Andriy Yasharov
Mr.Q
Daryo
Sandra & Kenneth Grossman
Tim Sheedy
 
Chase Faucheux
Chase Faucheux  Identity Verified
United States
Local time: 13:58
German to English
TOPIC STARTER
I feel like I'm meeting a celebrity :) Apr 19, 2019

Tom, if you don't mind me saying so, I've read a LOT of what yo have had to say on these issues, and I found it very entertaining and sobering. I tend to agree with you on almost everything. I do not think anyone as well established as you seem to be has anything whatsoever to worry about. At the same time, those of us who are still finding our way in this field may be facing some challenges in the short to medium term because it may become harder to find work for clients who actually care about... See more
Tom, if you don't mind me saying so, I've read a LOT of what yo have had to say on these issues, and I found it very entertaining and sobering. I tend to agree with you on almost everything. I do not think anyone as well established as you seem to be has anything whatsoever to worry about. At the same time, those of us who are still finding our way in this field may be facing some challenges in the short to medium term because it may become harder to find work for clients who actually care about quality.

My translation philosophy is that a translation that reads like a translation is always in some sense, a failed translation (except in some literary areas where it may be desirable to retain some feel of the source language). I will spend hours going down rabbit holes of technical material to make sure I understand what's being said on both sides of the process. I will never, ever translate a compound or collocated adjective + noun that I've never used personally without checking to see if it's been used before in native-language materials (if Linguee shows up in the search results, that's a red flag).

This is the stuff that takes up most of my time as a translator. MT can save some time on the repetitive stuff, the simple phrases, etc. But is it going to go on Google Maps to figure out if ambiguous driving directions actually make sense? Is it going to try and deal with the fact that the way the two languages approach and talk about the same topic in STEM fields may be completely different (some German pedagogical materials in math are next to useless for an American audience, for example)? Of course it won't. The problem is that I suspect most translators won't either, especially if rates are going down. Only suckers like me will.
Collapse


 
Jean Dimitriadis
Jean Dimitriadis  Identity Verified
English to French
+ ...
A craftsman's approach to translation Apr 19, 2019

I think a craftsman approach such as the one you describe is still relevant today.

Just figure out which domain, document, client types etc., this is more likely to apply to.

Because you take more time than the average, just make sure you market your services at above (or at least) average rates.

Along with any relevant specializations, if you can offer premium quality, you should still be able to command premium rates.

[Edited at 2019-04-19 14:51 GM
... See more
I think a craftsman approach such as the one you describe is still relevant today.

Just figure out which domain, document, client types etc., this is more likely to apply to.

Because you take more time than the average, just make sure you market your services at above (or at least) average rates.

Along with any relevant specializations, if you can offer premium quality, you should still be able to command premium rates.

[Edited at 2019-04-19 14:51 GMT]
Collapse


 
Katalin Szilárd
Katalin Szilárd  Identity Verified
Hungary
Local time: 19:58
English to Hungarian
+ ...
Same words, same terms, same text styles: end clients will lose customers Apr 19, 2019

Chase Faucheux wrote:

Yes, it cross-references other translations, but what it doesn't look at is how similar texts are written in English for English speakers.



I totally agree with this statement. All translations will be the same, because same terms, same sentence parts, same style will be used. Sooner or later end clients will lose their customers. Because even if their products or services are different, the text style, terms, words, marketing texts, user manuals of their products/services will sound the same, since it's collected from the Big Database. It doesn't matter whether it's a phone or a new medical device or a scientific publication about a medical procedure. Customers will read them and they won't know which one to choose among similar products, only the brand names will be different but the text will be the same. So those end clients that will rely on MT sooner or later will lose customers and then they will lose money. A lot of money. Even if AI will be able to make brand new words, we humans won't understand it. I think I read somewhere that they had to stop an AI robot because it created a language humans didn't understand. lol

https://www.independent.co.uk/voices/facebook-shuts-down-robots-ai-artificial-intelligence-develop-own-language-common-a7871341.html

Once I was on holiday in a foreign country.
The resort had a library and a family member of mine was looking for something to read.
She found 2 Hungarian books. She began to read the first one, then the second one. When reading the second one she began to complain about that she was reading the same style, same terms, same proverbs, very similar conversations etc. but the plots were different. I checked the 2 books: the authors were different but the translator was the same. So 2 different authors, 2 different plots, but when you read both of them, you couldn't sometimes tell the difference because the styles were so similar. I think this is what is going to happen with MT and MTPE or NMT.

I wrote the below opinion of mine (last 2 sentences of this post) in a previous thread about a half year ago, but somebody reported it and it was hidden because "I hurt those translators' feelings who do MT and MTPE".
If somebody feels that I hurt her/his feelings, just ask me and I'll delete this part.

Furthermore I think the translator who is doing MTPE may lose the ability to translate (to think logically and to understand what the text is about) or actually think as a human. Because she/he is doing such a monotone robotic job, that sooner or later the translator will say OK, this was accepted before, then I'll accept it, too.
This is the first step to give up thinking.


[Edited at 2019-04-19 15:41 GMT]


Recep Kurt
Tanja Oresnik
Alison Jenner
 
Chase Faucheux
Chase Faucheux  Identity Verified
United States
Local time: 13:58
German to English
TOPIC STARTER
Good points all around Apr 19, 2019

Jean D. – I agree; however, I'm approaching this issue as someone who is relatively new to the business end of translation. I know the quality I can deliver, I know where I have specialized knowledge (and how to do proper research when such is lacking), and I know the feedback I receive from clients, from agencies, and from other translators. But the agencies and clients I apply to work for don't know any of that to start out. If I'm given a translation test/evaluation, then I can usually get ... See more
Jean D. – I agree; however, I'm approaching this issue as someone who is relatively new to the business end of translation. I know the quality I can deliver, I know where I have specialized knowledge (and how to do proper research when such is lacking), and I know the feedback I receive from clients, from agencies, and from other translators. But the agencies and clients I apply to work for don't know any of that to start out. If I'm given a translation test/evaluation, then I can usually get my foot in the door. But just delivering quotes, when I won't work for peanuts, is a frustrating task. Of course, maybe the agencies that pay peanuts aren't worth working for in the long run, as that necessarily means there are serious ethical issues at play.

Katalin Sz. – Apart from a personal interest in all things Magyar, I wish I were working to or from a language like Hungarian sometimes. As much of a blessing as it is to be a native speaker of English, as so much work is available into English, it also means there's a lot of substandard work out there, either from lazy native speakers or from non-native speakers who overestimate their English skills (or from native speakers who have been living in the SL country for too long). These are the kind of people who will use the word "cheeky" in a text that's supposed to be American English.

DeepL and other projects seem to leave languages like Hungarian alone, but they tend to be fed by huge quantities of work into and out of English. This means that such common gems as "This product convinces with its ..." or "the dreamy alleys of [TOWN]" get reproduced ad nauseam. But since these inane constructions are so, so common in English translation, how is a client ever supposed to know (or care) how dumb they sound?
Collapse


 
Katalin Szilárd
Katalin Szilárd  Identity Verified
Hungary
Local time: 19:58
English to Hungarian
+ ...
Inititals Apr 19, 2019

Chase Faucheux wrote:

Katalin Sz. –


Chase, can I ask you how do you know that my name's initial is Katalin Sz. and not Katalin S.?

[Edited at 2019-04-19 16:26 GMT]


 
Chase Faucheux
Chase Faucheux  Identity Verified
United States
Local time: 13:58
German to English
TOPIC STARTER
What infuriates me most Apr 19, 2019

I think what infuriates me most about all this stuff is that in other industries where automation has replaced human jobs, it has often been because the product produced using automation did a better job than humans at completing some repetitive task, to the point where there was a recognizable increase in quality. All I'm seeing here is a recognizable increase in mediocrity that makes the field less attractive to people who care about quality.

Also, re: the ethical issue at stake -
... See more
I think what infuriates me most about all this stuff is that in other industries where automation has replaced human jobs, it has often been because the product produced using automation did a better job than humans at completing some repetitive task, to the point where there was a recognizable increase in quality. All I'm seeing here is a recognizable increase in mediocrity that makes the field less attractive to people who care about quality.

Also, re: the ethical issue at stake - These MT services use massive databases of expropriated human-produced content, certainly without the consent of the translators whose work is being used to put those translators out of business. And let's be real, that is the ultimate goal of the people behind these projects - to produce a system that can do automatic, quality translation without the need of humans at all. Is there any legal case that could ever be made that the whole edifice is built upon expropriated creative content, or is all the content so anonymized and passed through so many owners at this point that it doesn't matter?
Collapse


 
Chase Faucheux
Chase Faucheux  Identity Verified
United States
Local time: 13:58
German to English
TOPIC STARTER
Because it's a different letter? Apr 19, 2019

Chase, can I ask you how do you know that my name's initial is Katalin Sz. and not Katalin S.?icon_wink.gif


Is that correct? I've been fascinated with Hungarian for some time (not easy to find actual speakers in Louisiana though), so I've done some amateur studying up on the language, and if I recall, are the digraphs (sz, cs, zs, etc.) treated as separate letters?


 
Tom in London
Tom in London
United Kingdom
Local time: 18:58
Member (2008)
Italian to English
A question Apr 19, 2019

Chase Faucheux wrote:

.....These MT services use massive databases of expropriated human-produced content....


Where do they get that content?


 
Chase Faucheux
Chase Faucheux  Identity Verified
United States
Local time: 13:58
German to English
TOPIC STARTER
From us Apr 19, 2019

From us, or at least, from those of us who actually deliberately help it by "correcting" its "mistakes". But isn't the vast majority of the data just mined from the Internet? Isn't that how Linguee works?

 
John Fossey
John Fossey  Identity Verified
Canada
Local time: 14:58
Member (2008)
French to English
+ ...
Human input Apr 19, 2019

Chase Faucheux wrote:

MT can save some time on the repetitive stuff, the simple phrases, etc. But is it going to go on Google Maps to figure out if ambiguous driving directions actually make sense?


Interesting analogy. Anyone who has both Google Maps and Waze on their phones knows that directions by Waze are more reliable than Google Maps (even though both are owned by Google), because Waze depends on real-time input from human users! Is there a lesson there that also applies to translation?

MT depends on bilingual corpora, but as can be seen with Linguee that corpora base is becoming increasingly corrupted because it includes more and more poor quality machine translations as input to the MT's algorithms.

[Edited at 2019-04-19 17:24 GMT]


Chase Faucheux
 
Tom in London
Tom in London
United Kingdom
Local time: 18:58
Member (2008)
Italian to English
PLease explain Apr 19, 2019

Chase Faucheux wrote:

From us, or at least, from those of us who actually deliberately help it by "correcting" its "mistakes". But isn't the vast majority of the data just mined from the Internet? Isn't that how Linguee works?


How do we "correct" its mistakes? And how can translations can be "mined" ? How does anyone know that a text is in fact a translation? I would like to know EXACTLY, in DETAIL, how Deepl/GT/etc works

So far I've found this:

https://slator.com/technology/deepl-got-machine-translation-plans-make-money/

Quote "Linguee is fed by web crawling, where the quality of the translated sentence pairs is evaluated....."

But how does the crawler find translations (source text and target text)?

For instance,. are my examples of my work, on Proz (see profile) being mined and used by Linguee?

[Edited at 2019-04-19 17:29 GMT]


 
Chase Faucheux
Chase Faucheux  Identity Verified
United States
Local time: 13:58
German to English
TOPIC STARTER
Don't these services allow you to make suggestions? Apr 19, 2019

Don't these services allow you to make suggestions? I never do, because I see no point in helping to make DeepL "better". I actually do not know how they know it's a translation, but I know that Linguee works by showing translated equivalents side by side. Of course, the machine doesn't actually "know" anything.

Nevertheless, I find Linguee itself to be only partially reliable, as most of the time, the source text is not English. The only real way to know how something should be exp
... See more
Don't these services allow you to make suggestions? I never do, because I see no point in helping to make DeepL "better". I actually do not know how they know it's a translation, but I know that Linguee works by showing translated equivalents side by side. Of course, the machine doesn't actually "know" anything.

Nevertheless, I find Linguee itself to be only partially reliable, as most of the time, the source text is not English. The only real way to know how something should be expressed in English is to be familiar with how native speakers express similar concepts in monolingual material. Since this is something a LOT of translators apparently don't know ("see? There are 1000+ results on Linguee for 'the product convinces with its features'!", I can't expect it's anything close to what these dumb machaines do (I say "dumb" not as a pejorative, but as a statement of fact).

As for Google maps – I just mean I look at the satellite view of the actual roads in question and look at the street view of the actual site if available. Of course I don't trust their directions
Collapse


 
Pages in topic:   [1 2 3 4] >


There is no moderator assigned specifically to this forum.
To report site rules violations or get help, please contact site staff »


MT, "neural" stuff, and the future







Anycount & Translation Office 3000
Translation Office 3000

Translation Office 3000 is an advanced accounting tool for freelance translators and small agencies. TO3000 easily and seamlessly integrates with the business life of professional freelance translators.

More info »
Protemos translation business management system
Create your account in minutes, and start working! 3-month trial for agencies, and free for freelancers!

The system lets you keep client/vendor database, with contacts and rates, manage projects and assign jobs to vendors, issue invoices, track payments, store and manage project files, generate business reports on turnover profit per client/manager etc.

More info »