| Pages in topic: < [1 2] | "Why computers still can't translate languages automatically"
|
|---|
| | heikeb 
Member (2003)
English to German
+ ...
Phil Hand wrote:
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.
Of course, I can't find the same patent again I used as reference. I just noticed that the usual boilerplate expressions were translated correctly and, as far as I could tell, the terminology. Sentence structure was OK as well, allowing for the fact that sentence structure in patents tends to follow its own kind of logic. I was just very surprised to see that kind of quality from an MT system, but as I mentioned before, patents are highly specialized texts.
It was most likely done with the official Patent Translate system of the EPO:
http://www.epo.org/searching/free/patent-translate.html
"What is Patent Translate?
Patent Translate is a machine translation service specifically "trained" to handle elaborate patent vocabulary and grammar. It is now available in Espacenet and the European publication server, as well as via Google Translate.
It takes a statistical approach, comparing the source document sentence by sentence to millions of patent documents previously translated by humans. The final translation profits from this "previous learning" by the translation engine.
How was Patent Translate optimised for patents?
In co-operation with the national patent offices in its member states, the EPO is providing Google with millions of official, human-translated patent documents. These are used to train the translation engine to handle technical subject-matter and the specific style and format used for patent documents.
What can I use it for?
The machine translation should give you the gist of any patent or patent-related document, and help you to determine whether it is relevant. You might decide on this basis whether you need to invest in a human translation of the document.
Please note that the engine cannot provide legally binding translations."
| | | | Michele Fauble
United States
Local time: 05:45
Member (2006)
Norwegian to English
+ ...
| Disambiguation | May 15, 2012 |
Heike Behl, Ph.D. wrote:
"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.
Replace "on marsupials" with "on skates" and the computer has a problem.
| | | | heikeb
Member (2003)
English to German
+ ...
Michele Fauble wrote: Heike Behl, Ph.D. wrote:
"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.
Replace "on marsupials" with "on skates" and the computer has a problem.
On what do you base your assumption?
One might consider the sentence "He gave a talk on skates ..." as ambiguous: The speaker is wearing skates, or his talk deals with skates. So, without context, either translation would be justifiable.
But the MT system could compare the order of adverbial supplements in the sentence with the most common/standard order and come to the conclusion that, if "on skates" as Manner adverb were to mean "wearing skates", it would be placed at a different position in the sentence and that it therefore, because of its direct proximity to "talk", would most likely be the supplement thereof.
There are so many different types of information available that can/could be coded into an MT system and/or gathered from bilingual corpora by statistical means. Therefore, you can never categorically say: MT would have a problem with this or that. There are so many different MT systems out there. How many are you really familiar with? How can you say what each of them is or is not capable of?
All other EG MT systems that I was familiar with while still working for that software company had problems with correct word order and many other issues that our system was really good at handling. Some MT systems were better at handling some of the things our system had problems with. But that was 10 or more years ago, and MT development has continued all those years. That MT has not made any generally obvious progress, I would attribute to mainly two reasons:
- Many of the older systems that have been around for 20, 30 or more years have reached their development limit. This was already visible over 10 years ago. The companies would add more languages, more dictionaries, more features such translation of websites, etc., but generally no real improvements of the actual translation quality.
- Many of the new(er) systems have changed to a more statistically-based approach, which is still pretty much in its infancy in comparison to a rule-based approach.
And for truly ambiguous sentences: How often do we human translators have to ask the clients what they actually meant to say?
| | | | Lingua 5B
Bosnia and Herzegovina
Local time: 14:45
Member (2009)
English to Croatian
+ ...
| Discordance between sciences | May 15, 2012 |
Obviously, for computer science language is a combination of TM units = finite, while in linguistics language is described as "infinite".
I realize there are some fields where it is more finite and limited; the bottom line, however, is that if we automate the language, it also means automating the thinking, and I don't think any society will profit from automated uniformed thinking.
| | |
|
|
|
Jeff Whittaker
United States
Local time: 08:45
Spanish to English
+ ...
| Neil Coffey
United Kingdom
Local time: 13:45
French to English
+ ...
| Computer models of language | May 16, 2012 |
Lingua 5B wrote:
Obviously, for computer science language is a combination of TM units = finite, while in linguistics language is described as "infinite".
Computer models of language can capture infinity fairly well (actually, better than many other linguistic phenomena if it comes to it...) and the "TM unit" is really just a preoccupation of CAT tools and *some* components of *some* MT systems.
I'm curious as to why you characterise the whole of computational linguistics in this way or why you see the treatment of infinity as the major restriction on computer models of language.
[Edited at 2012-05-16 00:40 GMT]
| | | | | Out of sight, out of mind! | May 16, 2012 |
This story is more than 30 years old.
The English expression, "Out of sight, out of mind" was fed into a computer for being translated into Russian.
The resulting Russian sentence was fed into another computer for translation back into English. The final result was, "invisible idiot".
Regards,
N. Raghavan
| | | | Phil Hand
China
Local time: 20:45
Chinese to English
You were doing so well, and then this:
"But the MT system could compare the order of adverbial supplements in the sentence with the most common/standard order and come to the conclusion that, if "on skates" as Manner adverb were to mean "wearing skates", it would be placed at a different position in the sentence and that it therefore, because of its direct proximity to "talk", would most likely be the supplement thereof."
No, those positions are fully ambiguous.
<... See more You were doing so well, and then this:
"But the MT system could compare the order of adverbial supplements in the sentence with the most common/standard order and come to the conclusion that, if "on skates" as Manner adverb were to mean "wearing skates", it would be placed at a different position in the sentence and that it therefore, because of its direct proximity to "talk", would most likely be the supplement thereof."
No, those positions are fully ambiguous.
He gave a talk on the bridge on skates.
He gave a talk on skates on the bridge.
Absent context, there is no way to distinguish between at least three different meanings of the on- clauses in those sentences.
And then this:
"And for truly ambiguous sentences: How often do we human translators have to ask the clients what they actually meant to say?"
Almost never. Because almost every sentence is ambiguous, but humans can get through whole books without ever failing to understand the mean. This is why syntax-based MT failed, remember? Fruit flies like a banana and all that?
I dunno, I just find that in every case I've see where MT is supposed to "work", it's actually doing something different to the claim. Here - "MT can translate patents". No, the EPO says on its website that MT translated patents cannot be used for legal purposes.
The MT translations can be used for browsing and finding relevant patents. But I wonder if a decent search of terms in the relevant subject wouldn't provide just as much accuracy in browsing. ▲ Collapse
| | |
|
|
|
Jeff Whittaker
United States
Local time: 08:45
Spanish to English
+ ...
| Even simple things can require a lot of thought | May 16, 2012 |
What about the opening of the first Harry Potter book:
"Mr. and Mrs. Dursley, of number four, Privet Drive, were proud to say that they were perfectly normal, thank you very much."
"It sort of preempts the idea that someone would say differently. It's also often meant to end the line of conversation. An example would be: "Those bags look heavy. Would you like some help?" "I am perfectly capable of carrying my own bags, thank you very much." The implication is that the person s... See more What about the opening of the first Harry Potter book:
"Mr. and Mrs. Dursley, of number four, Privet Drive, were proud to say that they were perfectly normal, thank you very much."
"It sort of preempts the idea that someone would say differently. It's also often meant to end the line of conversation. An example would be: "Those bags look heavy. Would you like some help?" "I am perfectly capable of carrying my own bags, thank you very much." The implication is that the person saying "thank you very much" is annoyed and/or affronted by the idea that someone would think that he/she looked like he/she needed help. In the case of Mr. and Mrs. Dursley, it means that they're preempting the idea that someone might question whether or not they were normal."
-----------
"[The translator] said she still encountered problems right from the start.
She said: "When I was doing the translation, the very first sentence took me ages because it said 'The Dursleys were perfectly normal, thank you very much'."
She said that the expression 'thank you very much' was not intended to mean what it normally would.
"I worked and worked at that sentence and then I said 'I'm still at the first sentence, help'. I went on, because I could have been stuck for ages," she said.
"After a while I worked out a technique to get around that." ▲ Collapse
| | | | | My AutoSuggest dictionary sails around with homonyms | May 16, 2012 |
As an early experiment with SDL Studio, I created an AutoSuggest dictionary from an amalgamation of my TMs.
It is quite useful, but it suggests 'boats' at odd intervals, when there is nothing remotely nautical in the context.
It has picked up on the Danish word 'både', which is occurs frequently and means something like 'both' (as in both apples and oranges).
This word is not consistently translated in one particular way, as it often sounds better in Engli... See more As an early experiment with SDL Studio, I created an AutoSuggest dictionary from an amalgamation of my TMs.
It is quite useful, but it suggests 'boats' at odd intervals, when there is nothing remotely nautical in the context.
It has picked up on the Danish word 'både', which is occurs frequently and means something like 'both' (as in both apples and oranges).
This word is not consistently translated in one particular way, as it often sounds better in English to rephrase the sentence and use expressions like
as well, not only... but also... too...
Sometimes I simply leave out the 'både' as superfluous in English, or I may actually translate it as 'both'.
But on the few occasions when the subject is nautical, the Danish for a boat is 'en båd', and the plural is ... både.
This is consistently translated as boats, and that is enough to fool the AutoSuggest compiler.
Naturally, AutoSuggest is only a small and simple tool compared with real machine translation, but the problem illustrates what computers are up against.
They have to deal with the apparent inconsistencies that are no problem for humans, but cannot be reduced to statistics and formulae. ▲ Collapse
| | | | | get ready, it's just round the corner | May 17, 2012 |
Translation will probably be one of the last tasks that computers "crack", but take a look at this: http://www.youtube.com/watch?v=WFR3lOm_xhE. We're nearly there...
It must be remembered that computer performance is rated against the best human performance, not the average or even the weakest. Google Translate already produces better translations than 95% of the human population could m... See more Translation will probably be one of the last tasks that computers "crack", but take a look at this: http://www.youtube.com/watch?v=WFR3lOm_xhE. We're nearly there...
It must be remembered that computer performance is rated against the best human performance, not the average or even the weakest. Google Translate already produces better translations than 95% of the human population could manage (including me when it comes to Finnish > French for example). My prediction is that human translators will cease to exist within about 15 years (but so will most other professions), and will have switched to editing computer translations. As for conscious machines... my neighbor says he's "conscious", but why should I believe him any more than I would a computer that (who) affirms the same? Ray Kurzweil makes a very convincing case for the development of superhuman machine intelligence around 2035. Once that happens, no-one can imagine where things will lead ▲ Collapse
| | | | Jeff Whittaker
United States
Local time: 08:45
Spanish to English
+ ...
| How Watson Know What It Knows | May 17, 2012 |
Just a simple search algorithm:
http://npinopunintended.wordpress.com/2011/02/15/how-does-watson-know-what-watson-knows/
david young wrote: Translation will probably be one of the last tasks that computers "crack", but take a look at this: http://www.youtube.com/watch?v=WFR3lOm_xhE. We're nearly there... It must be remembered that computer performance is rated against the best human performance, not the average or even the weakest. Google Translate already produces better translations than 95% of the human population could manage (including me when it comes to Finnish > French for example). My prediction is that human translators will cease to exist within about 15 years (but so will most other professions), and will have switched to editing computer translations. As for conscious machines... my neighbor says he's "conscious", but why should I believe him any more than I would a computer that (who) affirms the same? Ray Kurzweil makes a very convincing case for the development of superhuman machine intelligence around 2035. Once that happens, no-one can imagine where things will lead
| | |
|
|
|
| Patenttranslate and human translations | May 18, 2012 |
Heike Behl, Ph.D. wrote: Phil Hand wrote:
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.
Of course, I can't find the same patent again I used as reference. I just noticed that the usual boilerplate expressions were translated correctly and, as far as I could tell, the terminology. Sentence structure was OK as well, allowing for the fact that sentence structure in patents tends to follow its own kind of logic. I was just very surprised to see that kind of quality from an MT system, but as I mentioned before, patents are highly specialized texts. It was most likely done with the official Patent Translate system of the EPO: http://www.epo.org/searching/free/patent-translate.html
The notice that "the translation is machine-generated" can be a bit misleading because in some cases the patent claims were translated by human translators.
You can check this by scrolling down to the claims before clicking on the patenttranslate button.
| | | | | Intelligent? | May 18, 2012 |
Machine translators are about as intelligent as intelligent computers that one can read about.
On mine I have to press the Start button to switch it off...
| | | | | Computers are teachers or supervisors or the like, not translators | May 18, 2012 |
In the long run, they will teach us how to prepare properly the input material for processing. Then we will be able to perform the basic act of translation: "I say it in plain language, first, you do home assignment" (human-to-human interface)
[Edited at 2012-05-18 09:17 GMT]
| | | | | Pages in topic: < [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 Studio 2022 Freelance | The leading translation software used by over 270,000 translators.
Designed with your feedback in mind, Trados Studio 2022 delivers an unrivalled, powerful desktop
and cloud solution, empowering you to work in the most efficient and cost-effective way.
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! » |
|
| | | | X Sign in to your ProZ.com account... | | | | | |
Login to reply/comment 
Translation news
