| Страниц в теме: [1 2] > | "Why computers still can't translate languages automatically"
|
|---|
| | neilmac
Испания
Local time: 23:06
испанский => английский
+ ...
| Sighs of relief | May 14, 2012 |
Here's hoping they take at least 10 years to perfect it, by which time I hope to be able to stop worrying about losing my job to a machine and an infinite number of monkeys
| | | | John Fossey 
Канада
Local time: 17:06
Член ProZ.com c 2008
французский => английский
+ ...
| Long way to go yet | May 15, 2012 |
How do you teach a machine meaning? Semantics are at a deeper level than language - as any translator knows, you can read a sentence in the source language, understand the meaning, yet struggle to find a way to convey that meaning accurately in the target language. Yet the meaning is clear to you. Machine translators are not even close to working at that level. Calculating the probability of how likely a given string of target words represents a given string of source words misses the underl... See more How do you teach a machine meaning? Semantics are at a deeper level than language - as any translator knows, you can read a sentence in the source language, understand the meaning, yet struggle to find a way to convey that meaning accurately in the target language. Yet the meaning is clear to you. Machine translators are not even close to working at that level. Calculating the probability of how likely a given string of target words represents a given string of source words misses the underlying meaning altogether. If humans find semantics a challenge, how are computers going to manage it? I think we're a long way before seeing this happen, if ever. ▲ Collapse
| | | | Neil Coffey
Великобритания
Local time: 22:06
французский => английский
+ ...
| Probably won't happen until machines are genuinely conscious | May 15, 2012 |
"Proper" human-like machine translation probably won't happen until machines are genuinely conscious.
However, many biologists and computer scientists believe that future computers may well be genuinely conscious (whatever *exactly* that means-- when we get closer to that point, the exact test will probably prove non-trivial).
So... at some point, translation may well be just "one of the normal tasks" that machines perform, along with helping people across the road, dri... See more "Proper" human-like machine translation probably won't happen until machines are genuinely conscious.
However, many biologists and computer scientists believe that future computers may well be genuinely conscious (whatever *exactly* that means-- when we get closer to that point, the exact test will probably prove non-trivial).
So... at some point, translation may well be just "one of the normal tasks" that machines perform, along with helping people across the road, driving vehicles, diagnosing illnesses etc.
I suspect we're not going to be at that point for a few decades yet, but I also don't see much reason why we won't eventually get there. ▲ Collapse
| | | | Lingua 5B
Босния и Герцеговина
Local time: 23:06
Член ProZ.com c 2009
английский => хорватский
+ ...
I am so sorry software developers won't be able to create a revolutionary change in this field and achieve a gigantic profit. Some small advancements might be possible.
As for the above comments, it takes a driver to drive a car, it takes a doctor/specialist to diagnose an illness.
| | |
|
|
|
Phil Hand
Китай
Local time: 06:06
китайский => английский
| A touch of sang froid | May 15, 2012 |
I just didn't get a job producing the training materials that feed these systems. I'm sure they handed it to a cheaper non-native speaker, just like I've seen before. Dunno about other language pairs, but it's going to be a long long time before Chinese-English gets sorted, because they won't stump up the extra 3 cents to get proper training materials. Means a longer career for me in the long run...
| | | | Neil Coffey
Великобритания
Local time: 22:06
французский => английский
+ ...
| Don't underestimate long-term technological change | May 15, 2012 |
Lingua 5B wrote:
I am so sorry software developers won't be able to create a revolutionary change in this field and achieve a gigantic profit. Some small advancements might be possible.
It's a common phenomenon that we tend to:
- overestimate short-term technological change
- underestimate long-term technological change
Don't underestimate what the short-term increments are gradually turning into.
Lingua 5B wrote:
As for the above comments, it takes a driver to drive a car, it takes a doctor/specialist to diagnose an illness.
You may think that now, but US states are already changing their laws to allow driverless cars because actual working prototypes are being tested...
Think about what computers were like in the 1960s and imagine suggesting then that in a few decades' time, the masses would be walking round with a device in their pocket that was orders of magnitude more powerful than their roomful of machinery-- and that they would do so for the mere purpose of listening to music?
[Edited at 2012-05-15 13:41 GMT]
| | | | Lingua 5B
Босния и Герцеговина
Local time: 23:06
Член ProZ.com c 2009
английский => хорватский
+ ...
| I don't deny the technological progress | May 15, 2012 |
Of course a significant progress in technology can be seen. However, each time you get a new innovation/device/application, it always takes a relevant guidance of a human and it can't produce much on its own (i.e. is not reliable or it's limited). For instance, you mentioned a driverless car, I am sure the number of performances that car will be able to do is very limited, for instance think of a tricky parking situation at a very busy place in a city (I'm talking about a real unsupervised situ... See more Of course a significant progress in technology can be seen. However, each time you get a new innovation/device/application, it always takes a relevant guidance of a human and it can't produce much on its own (i.e. is not reliable or it's limited). For instance, you mentioned a driverless car, I am sure the number of performances that car will be able to do is very limited, for instance think of a tricky parking situation at a very busy place in a city (I'm talking about a real unsupervised situation for the car, not a lab situation). For it to be programmed to drive in a straight line from A to B, it might be possible. But we need a driver who can think of many different aspects around him/her, what about safety?
Now using the same analogy for machine translation, yes it can provide simple sentences and common phrases. For more complicated string of thoughts, it's just anemic. Sorry if I disappointed the dreamers.
Machines can replace/help mechanical, repetitive actions, but not creative thinking.
[Edited at 2012-05-15 13:54 GMT] ▲ Collapse
| | | | John Fossey
Канада
Local time: 17:06
Член ProZ.com c 2008
французский => английский
+ ...
| Machines being genuinely conscious? | May 15, 2012 |
Neil Coffey wrote:
"Proper" human-like machine translation probably won't happen until machines are genuinely conscious.
However, many biologists and computer scientists believe that future computers may well be genuinely conscious (whatever *exactly* that means-- when we get closer to that point, the exact test will probably prove non-trivial)
When scientists don't even have much of a clue what consciousness is, how are they to transfer that knowledge to machines?
http://www.horizonresearch.org/main_nav_pages.php?cat_id=10
[Edited at 2012-05-15 14:01 GMT]
| | |
|
|
|
Neil Coffey
Великобритания
Local time: 22:06
французский => английский
+ ...
| My point of view | May 15, 2012 |
Lingua 5B wrote:
Now using the same analogy for machine translation, yes it can provide simple sentences and common phrases. For more complicated string of thoughts, it's just anemic. Sorry if I disappointed the dreamers.
Machines can replace/help mechanical, repetitive actions, but not creative thinking
This could well be true: the actual technology of *current* driverless cars is probably a bad example because it is indeed still in the realm of what we would call "artificial intelligence", and not actually mimicking the consciousness that humans have.
However, my take is that this may well eventually happen: the technology that we eventually use to build machinery may be similar enough in function to the human brain-- which at the end of the day is still just a big bunch of electrical connections-- that such machines are indeed 'conscious' in the same way as human beings and can indeed perform functions that we currently associate with human consciousness and creativity such as spontaneous intelligent decision-making, writing novels etc.
Now, it will still be true that the creation of such machines at some point will have required a human designer (even if they're designing "conscious machines" rather than designing specific algorithms). But we could also reach a point where that is no longer true and that the conscious machines are designing conscious machines (or self-replicating).
Now, in terms of what that means for translation and creative industries over the next few decades-- it could mean very little. We could be talking another century of development at least (though I repeat: as humans, we tend to underestimate long-term development). But I don't it is dangerous to stand about proudly asserting that "machines will never be able to do X because you need human creativity for that"... the tech may catch up with that sooner than you think.
| | | | Neil Coffey
Великобритания
Local time: 22:06
французский => английский
+ ...
| Consciousness | May 15, 2012 |
John Fossey wrote:
Neil Coffey wrote:
"Proper" human-like machine translation probably won't happen until machines are genuinely conscious.
However, many biologists and computer scientists believe that future computers may well be genuinely conscious (whatever *exactly* that means-- when we get closer to that point, the exact test will probably prove non-trivial)
When scientists don't even have much of a clue what consciousness is, how are they to transfer that knowledge to machines?
To some extent. it's probably one of those things that we'll "recognise when it happens".
But for building the technology this isn't necessarily a problem because it's not really "transferring knowledge" but more coming up with an overall *type* of machine that can mimic a brain more closely. (Mimicking current the single example of a biologically evolved brain that we have may also not be necessary, but we're starting with that because it's what we've got to go on at the moment...)
| | | | Lingua 5B
Босния и Герцеговина
Local time: 23:06
Член ProZ.com c 2009
английский => хорватский
+ ...
| Fancy and imagination | May 15, 2012 |
How are we going to teach a machine to use fancy and imagination as these processes are greatly involved in production of language? Remember, among people with medically healthy brains, not all have equally developed fancy and imagination.
I don't wanna be hypocritical, if it wasn't for technology we couldn't be here typing this right now, however, it's still just an aid, not an independent tool in its own.
| | | | heikeb
английский => немецкий
+ ...
| Patent translation | May 15, 2012 |
Well, I recently came across a patent that had been machine-translated from English into German (or the other way round) by an MT system customized for this purpose. And the translation was very good. I didn't realize it had been translated automatically. Maybe a closer scrutiny would have revealed some problems, but they were not blatanly obvious.
- There is a large corpus of bilingual patents around to harvest terminology
- There are lots of standard expressions and formulations... See more Well, I recently came across a patent that had been machine-translated from English into German (or the other way round) by an MT system customized for this purpose. And the translation was very good. I didn't realize it had been translated automatically. Maybe a closer scrutiny would have revealed some problems, but they were not blatanly obvious.
- There is a large corpus of bilingual patents around to harvest terminology
- There are lots of standard expressions and formulations
It is more a question of time that needs to be invested into customizing a decent MT system to start getting really good results (maybe not perfect, but very much acceptable plus much faster and cheaper than human translation) vs. the initial high costs.
Will MT ever be good enough for more creative texts? Maybe, but not within the next years/decades. And I'm not even sure that anybody is truely trying to achieve that. Will it be ever good enough for (semantically and syntactically) relatively well structured and predictable environments? With the ncessary "training", it already is.
There is a lot of what one might call "world knowledge" that can be implemented into MT, including semantic information. A machine can differentiate the differences between the different prepositional phrases and can re-order the sentence into the correct target word order:
"He gave a talk on marsupials on Monday on the bridge over the River." (Not a great sentence, but I can't think of anything better...)
A good machine "knows" that "on Monday" is an advervial phrase of time, that "on marsupials" belongs together with "talk", that "over the river" belongs together with "bridge" and that the entire prepositional phrase is an adverbial supplement of place. The system can then produce the correct translation and correct target word order: "Er hielt am Montag auf der Brücke über den Fluss eine Rede über Beuteltiere." In addition to the correct word order, also the preposition "on" has been translated correctly as "an", "auf" and "über". How do I know that MT systems are able to do that? Because I used to work for an MT company and taught the computer to do just that. And this is just a fairly simple example.
MT systems are able to translate highly complex sentences (including idiomatic expressions and terminology) 100% correct, where no human translator could do it any better. Is MT able to do it with every single sentence (without customization for a specific environment)? No, not yet.
What most people get to see are the often embarrassing results of something like Google Translate or other freely available online MT engines. And then, more often than not, to proof MT's shortcomings, the original translation result is sent through the MT system a second time for a back translation. If one would do this with a not too great human tanslator/translation, the results wouldn't be any better.
How often have translators complained about the really bad quality of some translation they've come across and how many times do they automatically assume it has had to be done by a machine - when certain types of errors clearly illustrate that it must have been done by a human translator! This just shows that most people are not fully aware of the actual strengths and weaknesses of MT systems. I'd prefer an MT translation any time over a poor human translation - OK, maybe not by Google Translate for English German, but there are other systems around:
- MT is consistent
- spelling, agreement, gender etc. are usually correct (at least with engines that incorporate some kind of syntax/grammar analysis)
- good MT systems are able to handle the most common idiomatic expressions
- good MT systems are generaly able to handle different grammatical structures (e.g. gerund clauses > relative clauses in English > German)
- if you know the way MT systems "think", you can actually figure out what it was supposed to say. Good luck with that when facing the rubbish some so-called translators of the human kind produce.
- even more important: If you know how MT systems think, you can write source texts in a way that reduces the likelyhood of MT errors.
@Lingau 5B: For which practical texts do you need fancy and imagination? Not for those kind of texts where MT would really mean huge savings in time and money such as manuals, patents, reports, etc. ▲ Collapse
| | |
|
|
|
Phil Hand
Китай
Local time: 06:06
китайский => английский
| If you could show us an example... | May 15, 2012 |
Heike, that would be cool.
I've heard of this English-German system for patents, and heard that it's good. Don't know much about it. The problem is that when I search for good examples, I don't really find them.
Dunno how relevant this website is: http://library.dialog.com/bluesheets/html/bl0324.html#RC
Here's what it gives:
"The avail... See more Heike, that would be cool.
I've heard of this English-German system for patents, and heard that it's good. Don't know much about it. The problem is that when I search for good examples, I don't really find them.
Dunno how relevant this website is: http://library.dialog.com/bluesheets/html/bl0324.html#RC
Here's what it gives:
"The available invention concerns a mechanism for programming remote control devices of the controls of electrical household appliances, in particular in the form of timers or by means of keys programmable clocks or also energy savings devices, furthermore of thermostats and Alarmvorrichtungen.Die mechanism is characterized by the fact that she consists configuring and diagnostic device (1) essentially of a programming -, which exhibit an electronic circuit as interface to programming means (2) one in the device mentioned (1) equipment (3) which..."
Not terrible, but not right, either. Can't actually work out what claim (1) and (2) mean. ▲ Collapse
| | | | Christine Andersen
Дания
Local time: 23:06
Член ProZ.com c 2003
датский => английский
+ ...
| Let the computers have the boring jobs... | May 15, 2012 |
I used to love Data when my son was into Star Trek...
who so charmingly fixed a myriad of problems, but struggled desperately to understand humour!
I think it is going to be like that for many years to come. Some kind of Super CAT will take over the routine stuff and won't even make typos...
Even then, there are quite a lot of problems stil to solve.
But as long as they let us do the fun bits, I'm quite happy to let the computers do the rest.
| | | | Neil Coffey
Великобритания
Local time: 22:06
французский => английский
+ ...
| It's not about "teaching" a machine | May 15, 2012 |
Lingua 5B wrote:
How are we going to teach a machine to use fancy and imagination as these processes are greatly involved in production of language?
We're not necessarily going to "teach" a machine to have such properties, just as nothing "taught" our brains to have those properties.
But, just like our brains, a device which turned out to have those properties could nonetheless be built in principle.
| | | | | Страниц в теме: [1 2] > | To report site rules violations or get help, contact a site moderator: You can also contact site staff by submitting a support request » "Why computers still can't translate languages automatically" | Trados Business Manager Lite | Create customer quotes and invoices from within Trados Studio
Trados Business Manager Lite helps to simplify and speed up some of the daily tasks, such as invoicing and reporting, associated with running your freelance translation business.
More info » |
| | CafeTran Espresso | You've never met a CAT tool this clever!
Translate faster & easier, using a sophisticated CAT tool built by a translator / developer.
Accept jobs from clients who use Trados, MemoQ, Wordfast & major CAT tools.
Download and start using CafeTran Espresso -- for free
Buy now! » |
|
| | | | X Sign in to your ProZ.com account... | | | | | |
Login to reply/comment 
Translation news
