Pages sur ce sujet:   < [1 2 3 4] >
MT, "neural" stuff, and the future
Auteur du fil: Chase Faucheux

Chase Faucheux  Identity Verified
États-Unis
Local time: 01:32
allemand vers anglais
AUTEUR DU FIL
Also... Apr 19

Also, how does it compare the quality of the text? I keep seeing that, but unless there is a person going through every single line, how would they know? Also, what is "quality"? Do they mean relatively trivial but prescriptively "bad" things like missing punctuation or typos, the kind of human errors even native speakers make? Or so they mean apparently sensible but actually deeply flawed issues like translating "ein hübsches kleines Haus" ("a hibsches Heisle" for the Swabians out there) as "a... See more
Also, how does it compare the quality of the text? I keep seeing that, but unless there is a person going through every single line, how would they know? Also, what is "quality"? Do they mean relatively trivial but prescriptively "bad" things like missing punctuation or typos, the kind of human errors even native speakers make? Or so they mean apparently sensible but actually deeply flawed issues like translating "ein hübsches kleines Haus" ("a hibsches Heisle" for the Swabians out there) as "a pretty small house", which DeepL did for me a few minutes ago, an error no native speaker would make?Collapse


 

Philip Lees  Identity Verified
Grèce
Local time: 09:32
Membre (2008)
grec vers anglais
Understanding Apr 20

Tom in London wrote:

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.


Is it though? It used to be thought that a computer program could never beat a human being at chess, because the computer couldn't "understand" the game. A number of computer programs have now passed the Turing test (at least according to some definitions). A computer has now beaten a world champion at Go.

I don't think the programmers would claim that their programs "understand" chess, or on-line chat, or Go, but that doesn't necessarily mean that a computer program cannot outperform a skilled human at these activities.

I think the "understanding" part is a red herring - at least until we can agree on an objective definition of what "understanding" means. I think that machine translation is already close to being good enough for many mundane and repetitive translation tasks and will reach a satisfactory level of competence at those tasks probably within the next decade.

Translating more subtle texts (for the purposes of the current discussion I mean those for which human translators might disagree amongst themselves about the best translation choices) is much further off, but I see no reason why it shouldn't be achievable eventually.

I'm not losing any sleep over it, though.


Sandra& Kenneth
 

Tom in London
Royaume-Uni
Local time: 07:32
Membre (2008)
italien vers anglais
THere is Apr 20

Chase Faucheux wrote:

..... unless there is a person going through every single line....


Apparently there is (see the link I posted earlier). Deepl employs translators to check things manually.


 

Tom in London
Royaume-Uni
Local time: 07:32
Membre (2008)
italien vers anglais
Start Apr 20

Philip Lees wrote:

..... until we can agree on an objective definition of what "understanding" means.


This might be a good place at which to begin:

https://www.iep.utm.edu/understa/


 

Chase Faucheux  Identity Verified
États-Unis
Local time: 01:32
allemand vers anglais
AUTEUR DU FIL
I suppose to some degree you're right Apr 20

I mean, I guess I can see how agencies will get away with selling clients the stuff spit out by machine translation and post-edited by "linguists" (I'm hesitant to consider this low-skilled, low-paid work part of the profession), at least enough to where it affects the market to the disadvantage of translators who care about accuracy and quality. If people can sell an inferior product for the same price, what's to stop them? Principle?

At the same time, I do not see how the translat
... See more
I mean, I guess I can see how agencies will get away with selling clients the stuff spit out by machine translation and post-edited by "linguists" (I'm hesitant to consider this low-skilled, low-paid work part of the profession), at least enough to where it affects the market to the disadvantage of translators who care about accuracy and quality. If people can sell an inferior product for the same price, what's to stop them? Principle?

At the same time, I do not see how the translation of more subtle texts could ever be achieved without real AI that not only understands the two languages in question, but is capable of real creativity and is able to understand the cultural context, understand emotional and aesthetic appeal, etc. Not even HAL 9000 would be capable of that. In the meantime, this obsession with MT will continue to depress rates and reduce overall quality. Because a reduction in quality is where this is headed, especially if the post-edited texts for slave wages are then used to "feed" the machine so it can "learn".
Collapse


 

Chase Faucheux  Identity Verified
États-Unis
Local time: 01:32
allemand vers anglais
AUTEUR DU FIL
Yes but to what extent? Apr 20

Tom in London wrote:

Chase Faucheux wrote:

..... unless there is a person going through every single line....


Apparently there is (see the link I posted earlier). Deepl employs translators to check things manually.


Are you saying that there is someone, someone who has not yet jumped off of a bridge, who actually goes through and checks the quality of all the stuff that ends up on Linguee and used by DeepL?


 

Tom in London
Royaume-Uni
Local time: 07:32
Membre (2008)
italien vers anglais
Yes Apr 20

Chase Faucheux wrote:

Are you saying that there is someone, someone who has not yet jumped off of a bridge, who actually goes through and checks the quality of all the stuff that ends up on Linguee and used by DeepL?


I'm not saying it; Deepl is.

"DeepL employs 20 full-time staff, about half of whom are engineers. The rest are product managers and quality editors that liaise with its more than 500 freelancers (mostly translators) on quality checking and monitoring."

See the link I posted earlier.


 

Philip Lees  Identity Verified
Grèce
Local time: 09:32
Membre (2008)
grec vers anglais
Kinds of understanding Apr 20

Tom in London wrote:

Philip Lees wrote:

..... until we can agree on an objective definition of what "understanding" means.


This might be a good place at which to begin:

https://www.iep.utm.edu/understa/


That looks like an interesting article. Thanks for the link. I've bookmarked it to read later.

However, from a quick glance I see that in section 1 it specifically excludes the kind of understanding we're discussing in this thread: "what we can call linguistic understanding—namely, the kind of understanding that is of particular interest to philosophers of language in connection with our competence with words and their meanings".


 

Chase Faucheux  Identity Verified
États-Unis
Local time: 01:32
allemand vers anglais
AUTEUR DU FIL
Ugh Apr 20

Tom in London wrote:

Chase Faucheux wrote:

Are you saying that there is someone, someone who has not yet jumped off of a bridge, who actually goes through and checks the quality of all the stuff that ends up on Linguee and used by DeepL?


I'm not saying it; Deepl is.

"DeepL employs 20 full-time staff, about half of whom are engineers. The rest are product managers and quality editors that liaise with its more than 500 freelancers (mostly translators) on quality checking and monitoring."

See the link I posted earlier.



Jeez, and I though I hated "proofreading" jobs. Why would any translator be interested in doing that? I somehow doubt they're being paid rates that make meticulous quality control worthwhile.


 

Philip Lees  Identity Verified
Grèce
Local time: 09:32
Membre (2008)
grec vers anglais
They don't understand us and we don't understand them Apr 20

Chase Faucheux wrote:
... I do not see how the translation of more subtle texts could ever be achieved without real AI that not only understands the two languages in question, but is capable of real creativity and is able to understand the cultural context, understand emotional and aesthetic appeal, etc. Not even HAL 9000 would be capable of that.


An interesting sidebar is that the AlphaGo program, which trounced a world champion a couple of years ago, taught itself to play the game. Its programmers don't understand how AlphaGo decides on the best move.

So AlphaGo, if sentient, might argue that human beings could never play Go at its own level because they didn't "understand" the game to the degree that AlphaGo does.


 

Tom in London
Royaume-Uni
Local time: 07:32
Membre (2008)
italien vers anglais
Whet Apr 20

Philip Lees wrote:

... I see that in section 1 it specifically excludes the kind of understanding we're discussing in this thread: "what we can call linguistic understanding—namely, the kind of understanding that is of particular interest to philosophers of language in connection with our competence with words and their meanings".


Well...if my first link whets your appetite there's always Wittgenstein and surrounding territory. e.g. here:

https://brewminate.com/linguistic-understanding-and-the-philosophy-of-language/


However it may be, we are certainly a long way from anything that AI or MT will ever be able to grasp.

[Edited at 2019-04-20 06:25 GMT]


 

Chase Faucheux  Identity Verified
États-Unis
Local time: 01:32
allemand vers anglais
AUTEUR DU FIL
Maybe, but Apr 20

Philip Lees wrote:

Chase Faucheux wrote:
... I do not see how the translation of more subtle texts could ever be achieved without real AI that not only understands the two languages in question, but is capable of real creativity and is able to understand the cultural context, understand emotional and aesthetic appeal, etc. Not even HAL 9000 would be capable of that.


An interesting sidebar is that the AlphaGo program, which trounced a world champion a couple of years ago, taught itself to play the game. Its programmers don't understand how AlphaGo decides on the best move.

So AlphaGo, if sentient, might argue that human beings could never play Go at its own level because they didn't "understand" the game to the degree that AlphaGo does.


The thing is, though, that both chess and go are games with a finite number of moves that either player could make, even if those numbers are astronomically high (especially for go). And even then, the machine just needs to be able to calculate more possible outcomes than a human does. Well, machines are better at calculation than we are. And in any event, DeepL and its like are not AI, as far as I know. Just very sophisticated regurgitators of work done by humans.


Recep Kurt
 

Chase Faucheux  Identity Verified
États-Unis
Local time: 01:32
allemand vers anglais
AUTEUR DU FIL
Back to the topic at hand Apr 20

I'm more concerned in the short to medium term with the popularity of post-editing, "discounts" on fuzzy matches, and all kinds of other new techie stuff done at the expense of the translator, yet which produces mediocre results that are themselves based on mediocre translations. If it drives people who translate stuff like "It will make ski-fans' hearts beat faster" and "this product convinces with its beautiful and comfortable design" out of the market, then so be it. But I just wish it wouldn... See more
I'm more concerned in the short to medium term with the popularity of post-editing, "discounts" on fuzzy matches, and all kinds of other new techie stuff done at the expense of the translator, yet which produces mediocre results that are themselves based on mediocre translations. If it drives people who translate stuff like "It will make ski-fans' hearts beat faster" and "this product convinces with its beautiful and comfortable design" out of the market, then so be it. But I just wish it wouldn't be affecting work for the rest of us.Collapse


 

Philip Lees  Identity Verified
Grèce
Local time: 09:32
Membre (2008)
grec vers anglais
Not just calculation Apr 20

Chase Faucheux wrote:

The thing is, though, that both chess and go are games with a finite number of moves that either player could make, even if those numbers are astronomically high (especially for go). And even then, the machine just needs to be able to calculate more possible outcomes than a human does. Well, machines are better at calculation than we are.


The Deep Blue chess program worked through sheer number-crunching and computer power. AlphaGo was quite different, though. It used strategies it had developed for itself by playing the game many many times. This is similar to the way humans learn.

When we have a computer program that can learn a human language in the same way, reaching a level of competence through repeated practice, then it may be hard to insist that the program doesn't "understand".

As for humans playing chess, I remember watching a chess tournament on TV many years ago (thrilling, I know) where the players were providing comments about their thinking at various points in the game. One thing that struck me was that they were generally unable to explain why they picked one particular move from the small set of choices dictated by the tactical situation. Usually the best they could come up with was something along the lines of "R to e8 looks good. I'll go with that."

I suspect that if I were asked to explain why I picked one particular word or expression in preference to another while translating a literary text, I might sometimes have trouble coming up with a better answer than just, "It looks good," or "It feels better," or similar flavours of fudge.


 

Tom in London
Royaume-Uni
Local time: 07:32
Membre (2008)
italien vers anglais
Purpose/intention Apr 20

Philip Lees wrote:

.... they were generally unable to explain why....


That's it! Does a chess-playing computer know WHY it is playing chess? Does it feel gratified and stimulated by a well-planned move? Is it disappointed when things go wrong?

[Edited at 2019-04-20 07:38 GMT]


 
Pages sur ce sujet:   < [1 2 3 4] >


Aucun modérateur n'est spécifiquement affecté à ce forum
Pour rapporter des violations des règles du site ou pour obtenir de l'aide, veuillez contacter l'équipe du site »


MT, "neural" stuff, and the future

Advanced search







PerfectIt consistency checker
Faster Checking, Greater Accuracy

PerfectIt helps deliver error-free documents. It improves consistency, ensures quality and helps to enforce style guides. It’s a powerful tool for pro users, and comes with the assurance of a 30-day money back guarantee.

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 »



Forums
  • All of ProZ.com
  • Recherche par terme
  • Travaux
  • Forums
  • Multiple search