Pages in topic:   < [1 2 3 4] >
MT, "neural" stuff, and the future
Thread poster: Chase Faucheux
Philip Lees
Philip Lees  Identity Verified
Greece
Local time: 14:19
Greek to English
On chalk and chess Apr 20, 2019

Tom in London wrote:

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]


For chess, almost certainly not. For AlphaGo the situation is more complex. Although we don't know precisely how it makes decisions, presumably the program experiences some kind of positive feedback when its game position improves and negative feedback when the opposite happens. That is how it learns to improve. To what extent these events correspond to human emotions is impossible to say, but the program is not static and immutable; it changes continually in response to its experience. It should keep on getting better indefinitely to the point when only other computer programs can challenge it.

Similar arguments could apply to machine translation, though the goal is not as clearly specified as it is in a game like Go.


Andrew Morris
 
Phil Hand
Phil Hand  Identity Verified
China
Local time: 20:19
Chinese to English
A minimal understanding of understanding Apr 20, 2019

I have an answer to this that I think will work. Basically, understanding should be regarded as the ability to take something linguistic, and do the appropriate non-linguistic thing; or transfer something non-linguistic into language correctly.

Simplest version: A smart speaker "understands" me if it turns off when I say "Off". If it cannot understand when I say, "Switch yourself off, please," or "Shut down now," then its grasp of English is very poor.

Or a computer sys
... See more
I have an answer to this that I think will work. Basically, understanding should be regarded as the ability to take something linguistic, and do the appropriate non-linguistic thing; or transfer something non-linguistic into language correctly.

Simplest version: A smart speaker "understands" me if it turns off when I say "Off". If it cannot understand when I say, "Switch yourself off, please," or "Shut down now," then its grasp of English is very poor.

Or a computer system can be said to "understand" colour if I can show it a colour, and it consistently and reliably gives me the right English (or whatever language) word for that colour.

This avoids the problem that has bedeviled computer language and computer translation research, which is that it is utterly constrained within the language system, with no way to check understanding.

It's also a good way to think about understanding because it's not binary. It's not helpful to ask: Does the computer understand or not? It's better to ask the human question: How well does the computer understand this text?

My prediction is that as computers build up a much greater ability to do things in the world (e.g. drive cars, serve food, answer questions), they will quickly develop much more facility with language, because the connection between language and reality, previously inaccessible to them, will become accessible.

But I'm curious to see if a computer ever really learns to use a human language. I am suspicious that a computer's infinitely greater acuity in certain areas (calculations, perceptual precision) will mean it just never "gets" why we talk the way we do. For example, how will a computer understand our 11 colour vocabulary? When it can quickly and easily represent colour much more precisely using a simple colour code, it will just never develop a feel for what "blue" means. It will never know why we talk about quantities the way we do, because it's easy for a computer just to say the correct, precise quantity, rather than rounding. So maybe they'll never get it. And translation will remain the one thing that computers can't really do.

[/speculative!]
Collapse


Annamaria Amik
 
Chase Faucheux
Chase Faucheux  Identity Verified
United States
Local time: 07:19
German to English
TOPIC STARTER
But are brains really just complex machines, and does it matter? Apr 20, 2019

This seems to be the underlying assumption in all of this, but it's not a settled question in neuroscience, and certainly not in philosophy. But I'll admit, that won't matter to translation agencies if computers are ever able to mimic human language passably enough (and they're toast too at that point).

Here's one major flaw I see in all this, however. Yes, a computer can probably develop the "it just feels right" approach to using certain turns of phrase, collocations, etc. Those t
... See more
This seems to be the underlying assumption in all of this, but it's not a settled question in neuroscience, and certainly not in philosophy. But I'll admit, that won't matter to translation agencies if computers are ever able to mimic human language passably enough (and they're toast too at that point).

Here's one major flaw I see in all this, however. Yes, a computer can probably develop the "it just feels right" approach to using certain turns of phrase, collocations, etc. Those things do have a certain statistical likelihood or not. The problem here is that these computers are trained on bilingual corpora ... whereas an actual human translator bases his or her knowledge on a monolingual "corpus". I don't know what the right choice is, or what "feels right" based on whether other translators have used it the same way, but based on whether people actually sound that way when speaking or writing English to be heard/read by other English speakers.

This is the big problem with services like Linguee, and the reason why, if I want to check whether a certain phrase or collocation actually exists, seeing Linguee pop up in the Google search results is a major red flag. The same thing goes for whenever I see sites with .de, .at, or even .nl among the top search results. I've worked with enough translated material to know a lot of it really doesn't sound like something a native speaker would ever produce. So knowing whether to use "big", "large", or "major" in a certain spot is based on my understanding of English usage and my native "feel" for the language, NOT based on whether other translators have used it before.

This again goes to the issue of quality, and why MT worries me. We're fast approaching the time when MT can mimic human speech reasonably well, and we're past the point where it produce things that sound better than the things lazy translators produce. That's enough for many agencies and clients, and that's all it takes to poison the whole field. I for one would rather wait tables again than spend most of my time post-editing machine-translated work, not matter how good.
Collapse


Elizabeth Tamblin
 
Tom in London
Tom in London
United Kingdom
Local time: 12:19
Member (2008)
Italian to English
Psycho Apr 20, 2019

The thing is: computers don't have a psychology. They do not have drives, desires, motivations, instincts, passions, convictions, principles, belief systems. The things that underlie all human intelligence. Artificial intelligence is DOA - dead on arrival.

[Edited at 2019-04-20 19:05 GMT]


Kaspars Melkis
Chase Faucheux
 
Chase Faucheux
Chase Faucheux  Identity Verified
United States
Local time: 07:19
German to English
TOPIC STARTER
People talk as if we've unlocked human understanding Apr 20, 2019

Tom in London wrote:

The thing is: computers don't have a psychology. They do not have drives, desires, motivations, instincts, passions, convictions, principles, belief systems. The things that underlie all human intelligence. Artificial intelligence is DOA - dead on arrival.

[Edited at 2019-04-20 19:05 GMT]


A lot of the more optimistic talk regarding AI and neural networks and stuff like that seems to rest on the assumption that we understand how the human mind works -- it's just a grey matter computer, innit? -- but from what I can tell, we don't.

This is why when someone says "DeepL learns just like the human brain does", it's basically complete nonsense.


Tom in London
MoSpa
 
Tom in London
Tom in London
United Kingdom
Local time: 12:19
Member (2008)
Italian to English
Experience Apr 20, 2019

Chase Faucheux wrote:

.....when someone says "DeepL learns just like the human brain does", it's basically complete nonsense.


"Many people used to believe that the 'seat' of the soul was somewhere in the brain. Since brains began to be opened up frequently, no one has seen 'the soul'. As a result of this and like revelations, many people do not now believe in the soul."

- R. D. Laing “The Politics of Experience”



[Edited at 2019-04-20 19:43 GMT]


Chase Faucheux
 
Elizabeth Tamblin
Elizabeth Tamblin  Identity Verified
United Kingdom
Local time: 12:19
French to English
By the way... Apr 20, 2019

Ian McEwan's new novel, Machines Like Me, just out

See more
Ian McEwan's new novel, Machines Like Me, just out

https://www.amazon.co.uk/Machines-Like-Me-Ian-McEwan/dp/1787331660/ref=sr_1_1?adgrpid=64644289627&hvadid=310588646609&hvdev=c&hvlocphy=9046425&hvnetw=g&hvpos=1t1&hvqmt=e&hvrand=12129381416910245872&hvtargid=aud-612972976242:kwd-544475489479&keywords=machines%20like%20me&qid=1555790008&s=gateway&sr=8-1

Haven't read it yet, but it's on my Kindle.
Collapse


Andrew Morris
 
Chase Faucheux
Chase Faucheux  Identity Verified
United States
Local time: 07:19
German to English
TOPIC STARTER
I don't quite get it Apr 20, 2019

Hey Tom, maybe I'm a bit dense, but I don't quite get where you're going with that quote. Would you mind explaining a bit?

 
Tom in London
Tom in London
United Kingdom
Local time: 12:19
Member (2008)
Italian to English
Irony Apr 20, 2019

Chase Faucheux wrote:

Hey Tom, maybe I'm a bit dense, but I don't quite get where you're going with that quote. Would you mind explaining a bit?


Even today, there seem to be lots of people who think the brain is a machine, or that it can be understood as though it were a machine. They operate on brains, trying to remove the bits that they think don't work. They feed pills to people, to try and shut down the bits they think aren't working properly. In every case, they think of the brain as a piece of physical, chemical, neurological machinery.

Laing (I don't have many heroes, but he's one) was being ironic about those people.

My use of the quote was a reference back to your earlier remark about "A lot of the more optimistic talk regarding AI and neural networks and stuff like that seems to rest on the assumption that we understand how the human mind works -- it's just a grey matter computer, innit? -- but from what I can tell, we don't."



[Edited at 2019-04-20 20:11 GMT]


Chase Faucheux
 
Philip Lees
Philip Lees  Identity Verified
Greece
Local time: 14:19
Greek to English
Software versus hardware Apr 21, 2019

Tom in London wrote:

The thing is: computers don't have a psychology. They do not have drives, desires, motivations, instincts, passions, convictions, principles, belief systems. The things that underlie all human intelligence. Artificial intelligence is DOA - dead on arrival.


Tom, I think you're confusing hardware and software here. MS Word is recognisably the same program whether it's running on Windows or on a Mac. You won't get to understand how the program works by analysing the hardware of either type of computer. No, computers don't have the attributes you mention above, but some of them are quite easy to program into software.

The software running in our brains possesses all those attributes already. There are differing views about how that came to be. Some people believe it's possible to transfer the software from a human brain to a different hardware platform - the so-called "mind uploading". That's not directly relevant to this thread, but if you accept the possibility of mind uploading to a sufficiently sophisticated computer, then there seems no reason why an artificial intelligence could not be programmed into that computer that possesses all the attributes of an uploaded human mind (perhaps plus some of its own).

There is the view that human consciousness includes something ineffable that is not confined to the biological structure of the brain, but is connected to it in some way. Very well, that same ineffable something could switch its connection to somewhere else. Or, as others believe, it could be something that automatically comes into being when a thinking machine, biological or electronic, reaches a certain level of complexity.

All this stuff, in my view, goes in the "not proven" box, but I wouldn't dismiss artificial intelligence so quickly. Let me also remind you of Clarke's First Law - or indeed all of them.



[Edited repeatedly to correct the hyperlink, which ProZ broke]

[Edited at 2019-04-21 09:11 GMT]


 
Tom in London
Tom in London
United Kingdom
Local time: 12:19
Member (2008)
Italian to English
No... Apr 21, 2019

Philip Lees wrote:

The software running in our brains


There is no software running in our brains. Crazy idea. Terminates my participation in this discussion.

[Edited at 2019-04-21 09:21 GMT]


Kaspars Melkis
MoSpa
Chase Faucheux
John Fossey
 
Philip Lees
Philip Lees  Identity Verified
Greece
Local time: 14:19
Greek to English
Metaphor Apr 21, 2019

Tom in London wrote:

Philip Lees wrote:

The software running in our brains


There is no software running in our brains. Crazy idea. Terminates my participation in this discussion.

[Edited at 2019-04-21 09:21 GMT]


Er, it was a metaphor. Sorry if it was too difficult.


 
Dan Lucas
Dan Lucas  Identity Verified
United Kingdom
Local time: 12:19
Member (2014)
Japanese to English
Talent doesn't pay the bills Apr 22, 2019

Chase Faucheux wrote:
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.

I'm coming late to this, but if translators cannot find work that rewards quality (pays well) in a major language pair, then their talent is of a very limited kind. Perhaps only in the field of literature (?) is raw translation ability a dominant factor.

For most of us freelancers, to be successful you need to find and keep clients. That takes more than translation ability. It requires drive, perseverance, business savvy, the ability to market oneself, skill in human interactions and several other competencies. If you have these you will end up with a specialisation or two that work, a solid client list, and enough work to keep you very busy. If you don't, "talented" or not, you won't.

To my mind the core problem is not MT, but individuals. I see a huge and growing market, but with more rapid growth in automated translation than in traditional translation. There's still a place for humans, and will be for the foreseeable future. But translation is changing, like most professions, and you either change your tack to keep ahead of the breaking wave, or you get dragged down into the surf.

What is the wave? Today it's MT. Two decades ago it was CAT tools, the MT of its era. You'd think CAT as a theme was done and dusted, but a couple of years ago I was visiting Japan to meet other translators and clients (agencies). In some meetings translators were complaining about CAT, and how it was a racket to squeeze the translator. In my other meetings, my agencies were complaining that hardly anybody in my specialisation uses CAT. I was astonished to find that take-up rates for CAT were so low.

Fortunately I do use CAT tools and my business has just kept growing. I find it unfortunate that many other translators in my pair (still) don't use CAT and thus marginalise themselves, but from a personal viewpoint it's been great to see otherwise competent translators making themselves less competitive.

I am not a "talented translator". What I am is a specialised, competent, and reliable translator who is focused on keeping my clients happy. That simple recipe has delivered double-digit growth in my business, year after year, for half a decade. And if in future a form of MT comes along that I can use to maintain that situation, I'll use it, just as I use CAT tools now.

If work is the centre around which your life revolves, it is no doubt important that work is exciting, or fulfilling, or a generator of joy. I freely admit that I am not one of those people. For me, work is just a way to support myself and those dear to me, and to fund my interests outside work. I do find translation intermittently interesting and enjoyable, but I'm not passionate about it. (I realise that this is an unfashionable attitude in an era in which workers are supposed to declare their passion for every little aspect of their job, but there you go.)

I am not talented, but I am professional. And apparently that's good enough. So take heart, any younger freelancers reading this. There's hope for you yet.

Regards,
Dan


[Edited at 2019-04-22 13:02 GMT]


Maciek Drobka
Rachel Waddington
Mr.Q
Michele Fauble
Tansy Tazewell
Susanna Martoni
 
Chase Faucheux
Chase Faucheux  Identity Verified
United States
Local time: 07:19
German to English
TOPIC STARTER
So in other words, the good news is that quality loses... Apr 22, 2019

By "talent" what I mean is the ability to fully understand the source language in all its details and nuances and to render the meaning of the source text in the target language in a way that reads like it was written in that language to begin with. I can understand aiming for less than that if one's attitude is "hey, it's just a job, and I can get more jobs done in less time this way", but if that's the case, then it sure isn't quality.

I'm not saying "talented translators" can't f
... See more
By "talent" what I mean is the ability to fully understand the source language in all its details and nuances and to render the meaning of the source text in the target language in a way that reads like it was written in that language to begin with. I can understand aiming for less than that if one's attitude is "hey, it's just a job, and I can get more jobs done in less time this way", but if that's the case, then it sure isn't quality.

I'm not saying "talented translators" can't find work, but that the market in its current form seems to reward mediocrity over quality. If someone can post-edit god-awful machine-translated language at 2000 words/hr and be fine with what they submit, that's a nice paycheck. The agency is happy, the client's English isn't that good anyway, so it's a win all around, I guess. I've done enough proofreading and post-editing to see what passes for quality among some prolific translators, and it's depressing. But it does pay the bills, so if there's no incentive to not write crap like "This robust and beautiful product convinces with its two speed and power settings" or "With its location right on the ski piste, this hotel will have outdoor lovers' hearts beating faster", then those of us who can't tolerate that and actually try to think about how to render it in proper English idiom need to move aside.

All I'm trying to say is that the proliferation of jobs where the bulk of the work is done by either low-paid speed translators with an iffy grasp of how people write in the target language OR by machines with no actual grasp of anything, that makes it that much harder for quality and dedication to stand out. Not impossible, but harder. And if the machines are getting fed with more mediocre work of the kind one regularly finds on Linguee, then agencies should just start advertising "good enough" and be done with it.
Collapse


 
Kaspars Melkis
Kaspars Melkis  Identity Verified
United Kingdom
Local time: 12:19
English to Latvian
+ ...
With all due respect, Dan... Apr 22, 2019

This is not about the income of individual translators but about good translation practice and counteracting negative trends in the industry.

Overestimation of MT capabilities can be very dangerous because it sends the wrong message which creates false beliefs. The results can be very damaging to everyone involved. We as professionals have a duty to educatate the public about involved risks so that costly mistakes can be avoided.

Ultimately, our input can even set globa
... See more
This is not about the income of individual translators but about good translation practice and counteracting negative trends in the industry.

Overestimation of MT capabilities can be very dangerous because it sends the wrong message which creates false beliefs. The results can be very damaging to everyone involved. We as professionals have a duty to educatate the public about involved risks so that costly mistakes can be avoided.

Ultimately, our input can even set global standards and support legislative and regulative changes. For example, if a company decides to provide product information that is illegibile due to MT (which nowadays happens a lot), does the law considers it merely as insuficient quality that can be fixed by an update or a dishonest attempt to sell the product without IFU?

The tendency to overestimate IT capabilities has become epidemic in several industries. For example, Tesla cars with autopilots that drive off the road for no reason killing its occupants. Or recent Boeing 737 MAX fiasco where the management in order to avoid retraining pilonts and make more money in this way, decided to compensate the design flaws with poorly written software that cost the lives of > 300 people.

Yes, MT can be the tool, but the problem is that the general message by unscrupulous vendors clearly overestimate its capabilities. Unlike airplanes, generally bad translations do not result in deaths of hundreds of people but nevertheless selling MT solutions as equal (or almost equal) to professional translations is very dishonest that we should not tolerate.
Collapse


Chase Faucheux
MoSpa
Philip Lees
Luca Tutino
 
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 »
Wordfast Pro
Translation Memory Software for Any Platform

Exclusive discount for ProZ.com users! Save over 13% when purchasing Wordfast Pro through ProZ.com. Wordfast is the world's #1 provider of platform-independent Translation Memory software. Consistently ranked the most user-friendly and highest value

Buy now! »