| Vom Thema belegte Seiten: < [1 2] | Poll: When will machine translation be infallible? Initiator des Themas: ProZ.com Staff
|
|---|
Dusan Rabrenovic 
Slowenien
Local time: 18:27
Mitglied (2008)
Deutsch > Slowenisch
+ ...
| Give it 15 years | Aug 8, 2009 |
Emin Arı wrote:
However, the machine will not be machine, it will be a creature with even a soul. that is the dilemma.
Yeah.. it will demand a paycheck, 401k and paid leave..
I do believe at least a third of us will be out of business for good within the next 15 years. If it will be all because of machine translation I couldn't say, but the business is going downhill faster than a rocket-fueled lead boxcar.
| | | | Alexandra Krah
Deutschland
Local time: 18:27
Deutsch > Rumänisch
+ ...
| Really never? | Aug 9, 2009 |
Miroslav Jeftic wrote: Interesting to see that almost everybody chose never.  While I agree that the time when a machine will be able to translate a book or a poem isn't near, regarding the translation of instruction manuals, help files, contracts and agreements, etc, I wouldn't bet we aren't going to see that in our life time. [Edited at 2009-08-08 11:29 GMT]
There are indeed a lot of technical translations for which a well programmed machine would do quite well. Moreover, the research in this field is ongoing, because a machine will always be more profitable than a human being on the long term. I have a friend who works for such a project, sometimes he scares me. The only field which may take longer than 100 years is literature and poetry in particular. We may just be able to get to the rent, but I wouldn't advise my children to become translators. The end is not far away!
| | | | Rob Edwards
Local time: 17:27
Mitglied (2008)
Deutsch > Englisch
| I doubt Machine Translation will ever offer much more than it does at the moment. | Aug 9, 2009 |
The reason for this is simple. Context. Computers are, as the name suggests number crunching machines, so however "clever" the latest MT programs appear to be, they are still working using an algorithm, in other words, what is produced is produced mathematically using previous translations as a corpus for comparison, where potential translations are identified based on their statistical similarity to the source text and a balance of probabilities with regard to the correct output of the target t... See more The reason for this is simple. Context. Computers are, as the name suggests number crunching machines, so however "clever" the latest MT programs appear to be, they are still working using an algorithm, in other words, what is produced is produced mathematically using previous translations as a corpus for comparison, where potential translations are identified based on their statistical similarity to the source text and a balance of probabilities with regard to the correct output of the target text.
In other words, computers are not processing language here. What they are doing is not translation. It is state of the art number crunching. It is very useful, but as a tool (albeit a less useful one than CAT tools).
[Edited at 2009-08-09 12:54 GMT] ▲ Collapse
| | | | | Machine will recognize world through explicitation | Aug 13, 2009 |
In rough way: explicitation - idea of the technique of making explicit in the target text the information that is implicit in the source text. So-called "translation explicitation" (not mathematic or linguistic one!) could be formalized till definite level in terms of one-way semiotic process (for example, we can move from implicit to explicit information, but not vice versa in case of semiosis) from the point of view of making choice while translating. "Till definite level" - because of its nat... See more In rough way: explicitation - idea of the technique of making explicit in the target text the information that is implicit in the source text. So-called "translation explicitation" (not mathematic or linguistic one!) could be formalized till definite level in terms of one-way semiotic process (for example, we can move from implicit to explicit information, but not vice versa in case of semiosis) from the point of view of making choice while translating. "Till definite level" - because of its nature, the source semiotic process can't be ever reconstructed. If formalization of explicitation is possible, it is possible to teach machine to explicitate information through modelling semiotic process of the source text. If so, we'll have the first generation of translation machine with the elements of artificial intelligence (nobody in the world knows what is artificial intelligence). In theory, there are no critical obstacles for machine to "guess" what the author meant in his text using the element of artificial intelligence - translation explicitation. And then, machine will recognize itself... ▲ Collapse
| | |
|
|
|
heikeb
Mitglied (2003)
Englisch > Deutsch
+ ...
| Too general question | Aug 14, 2009 |
Will MT ever be even comparable with human translators in more creative areas (with poetry at the extreme end)? Never. But lots of poetry cannot even be translated adequately by a human translator. More often than not, marketing slogans cannot really be translated, but need to be "re-invented" in the localization process.
Will MT ever be infallible in certain areas such as computer manuals or other restricted areas that can be customized to a large extent? MT already has the the abi... See more Will MT ever be even comparable with human translators in more creative areas (with poetry at the extreme end)? Never. But lots of poetry cannot even be translated adequately by a human translator. More often than not, marketing slogans cannot really be translated, but need to be "re-invented" in the localization process.
Will MT ever be infallible in certain areas such as computer manuals or other restricted areas that can be customized to a large extent? MT already has the the ability. At this point, it might be a question of how long it takes to fully customize and train a system and whether the required time and effort are financially acceptable. The more this customization can be automatized (e.g. by taking advantage of large corpus analysis etc.), the more affordable a working - and maybe for restricted areas infallible - MT solution will be. Personally, I'm actually surprised that some companies that produce lots of rather similar manuals are not investigating MT options more actively (but maybe they are and we just don't know about it).
In some aspects, MT is already superior to human translators: consistency, spelling (which of course requires a dictionary free of spelling errors), and of course speed.
And when it comes to the probability of infallibility: Humans will never be infallible; computers at least have the potential of infallibility. So, which "system" has the better cards? ▲ Collapse
| | | | absciarretta
Vereinigte Staaten
Local time: 17:27
Mitglied (2008)
Englisch > Norwegisch
+ ...
For my language pair, never. There are too many words that sound the same but mean different things, too many idioms that are hard to translate etc. Plus it is a rather rare language.
| | | | Jeff Allen
Frankreich
Local time: 18:27
Mehrere Sprachen
+ ...
| you are referring only to Statistical MT | Aug 29, 2009 |
Rob Edwards wrote:
The reason for this is simple. Context. Computers are, as the name suggests number crunching machines, so however "clever" the latest MT programs appear to be, they are still working using an algorithm, in other words, what is produced is produced mathematically using previous translations as a corpus for comparison, where potential translations are identified based on their statistical similarity to the source text and a balance of probabilities with regard to the correct output of the target text.
In other words, computers are not processing language here. What they are doing is not translation. It is state of the art number crunching. It is very useful, but as a tool (albeit a less useful one than CAT tools).
Rob,
Your reply here is only partially correct. You are specifically referring to only one specific type of MT engines, a relatively new breed, which are called Statistical-based MT.
These are very different from Rule-based MT, which still comprise the majority of commercial MT products on the market today, and which are based on a combination of grammar rules and dictionary entries (both are linguistic elements) to generate the translation.
Only as of very recently (less than 2 months ago) have some of the commercial MT vendors started implementing hybrid Statistical and Rule-based systems.
And Example-based MT (basically what professional translators have referred to as TM for many years) has already been implemented in several commercial MT products over the past several years.
I voted for Other because the question, as many have stated, is not about MT becoming fallible. The real question is when it can start be used as one of the tools in the arsenal of a translator who wants to increase productivity, increase consistency, and achieve a higher throughput rate.
The answer to that question is that it is already being done by many who have learned to used MT successfully. I've posted a lot on this here on Proz (just look up the term "postediting"). And there are lots of new references to translation agencies investing in MT technologies and training their translators on new projects. One just cited recently that they are able to get the translator team up to speed, creating custom dictionaries on the customer-specific content, and achieve extremely high return on productivity within one month.
Jeff
| | | | Jeff Allen
Frankreich
Local time: 18:27
Mehrere Sprachen
+ ...
| companies are investing MT options | Aug 29, 2009 |
Heike Behl, Ph.D. wrote:
I'm actually surprised that some companies that produce lots of rather similar manuals are not investigating MT options more actively (but maybe they are and we just don't know about it).
They are. I have been getting requests about this from all kinds of companies over the years.
Whether or not it is practical to implement based on their workflow, their IT constraints, their content, their resource availability for dictionary customization, and many other factors, is what leads to them deciding what direction to take.
Jeff
| | |
|
|
|
Muriel Vasconcellos
Vereinigte Staaten
Local time: 09:27
Mitglied (2003)
Spanisch > Englisch
+ ...
| I worked in machine translation development for over 10 years | Sep 17, 2009 |
I know first-hand how hard it is to improve an MT system. More often than not, "improvements" degrade the development work that has already been done. There is no end in sight.
That said, I agree with the colleagues who have pointed out that NO translation is infallible. So how can we expect a computer program, which has been populated with rules and lexicons developed by people, to be any better that its developers?
| | | | | Vom Thema belegte Seiten: < [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 » Poll: When will machine translation be infallible? | TM-Town | Manage your TMs and Terms ... and boost your translation business
Are you ready for something fresh in the industry? TM-Town is a unique new site for you -- the freelance translator -- to store, manage and share translation memories (TMs) and glossaries...and potentially meet new clients on the basis of your prior work.
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 » |
|
| | | | X Sign in to your ProZ.com account... | | | | | |
Login to reply/comment 
