| Pages in topic: < [1 2 3 4] > | Poll: Will machine translation ever replace human translation entirely? Thread poster: ProZ.com Staff
|
|---|
Roberto Rey 
Colombia
Local time: 05:29
Member (2007)
Spanish to English
+ ...
-The world is flat
-Man will never go to the moon
-Little Albert (Einstein) will never amount to anything in his life
-The wall between Germany will never fall
-Rome will prevail for ever
-Man will never fly
-The car will never replace the horse
and a MILLION other examples
......are we also blind sighted ??
NEVER say NEVER
Then somebody will say: " I knew this would happen" "I saw it coming" " I predicted i... See more -The world is flat
-Man will never go to the moon
-Little Albert (Einstein) will never amount to anything in his life
-The wall between Germany will never fall
-Rome will prevail for ever
-Man will never fly
-The car will never replace the horse
and a MILLION other examples
......are we also blind sighted ??
NEVER say NEVER
Then somebody will say: " I knew this would happen" "I saw it coming" " I predicted it " ..and so forth.. ▲ Collapse
| | | | mediamatrix (X)
Local time: 06:29
Spanish to English
+ ...
| Misguided assumption | Oct 8, 2007 |
Kevin Jeffcoat wrote:
I do think that for technical translations or ones without need for tone or style that computer translation will come quite far.
Why do so many people - including people in this profession - so frequently go on record inferring - or even stating explicitly - that technical translations are devoid of 'tone' or 'style'?
Those attributes, which many like to think of as their mark of superiority when deployed in their own non-technical fields, have as much impact on the readability and intelligibility of technical writing as they have on literary prose - or even, dare I say it, on poetry.
I see no reason to assume that MT will satisfy the world of technical translation any sooner than it will satisfy those working with British humour, Molière or ... (you name it).
MediaMatrix
| | | | | Attempts to train translators prove the futility of machine translation | Oct 8, 2007 |
If it is not possible, even by providing extensive TMs and terminology databases, to always train human translators to produce the desired result, then how much more futile will it be to train mere robots...
From that perspective, the effort to train human beings to do something cheers me up!
Astrid
| | | | | More than "I don't think so"... | Oct 8, 2007 |
... I would say "I HOPE not!!!". For my job's sake!
| | |
|
|
|
| It's not about creativity, it's about processing power | Oct 8, 2007 |
Victor Dewsbery wrote:
I find this belief in the linguistic creativity of computers absolutely astounding. In view of the estimated total of more than 5,000 languages on the face of the earth, I find it hard to swallow the fact that over 120 intelligent linguists believe that computers will handle languages so thoroughly (at some point) that no human translator will ever have anything to do in any subject area and in any language pair in the whole world.
One way for MT to develop is not by attempting to re-create human lingusistic skills, but by sheer brute force.
As memory gets cheaper and processing gets faster, searching through previously translated material, tons and tons of it (or perhaps I should say tera-giga-mega-tera-again-googleplex-squared-megabytes of it!), becomes an increasing viable option.
Not in every field, maybe, but in many - IT itself, general commercial, some legal -OK, lets just say in the stuff I work in, I can see it happening. Maybe not in every language, but in the main ones (possibly using English as an intermediary).
That is why I believe MT will become, shall we say, potentially responsible for a large proportion of translation output, in my lifetime (I'm 42).
But probably not poetry.
However, once a suitable translation for "pull the other one" has been found, that'll be it. Done once, done forever.
| | | | heikeb
Member (2003)
English to German
+ ...
| Don't underestimate MT and computers | Oct 8, 2007 |
Astrid Elke Johnson wrote:
If it is not possible, even by providing extensive TMs and terminology databases, to always train human translators to produce the desired result, then how much more futile will it be to train mere robots...
From that perspective, the effort to train human beings to do something cheers me up!
Astrid
Computers are much more trainable than humans as they never forget and only have to be told once, their translation quality is never affected by working late hours or without breaks or lack of concentration. They will never leave out words or segments, make no typos, are 100% consistent and never careless, work incredibly fast and cheap (once the initial costs - which can be substantial - have been recovered)... In that respect alone, they surpass humans thousandfold!
Also, if computers make an error, this error is logically explainable and as such can be much better fixed than the often illogical and completely inexplicable mistakes human translators make. Furthermore, each human translator has their own personal flaws, whereas all computers using the same program suffer from the same flaws. Ie. if you fix one flaw in the program, all computers running it will be "cured" at the same time. Correcting an error in one translator will have no effect whatsoever on all the other translators, even if the cause is exactly the same. On top of that, the corrected translator might make the very same mistake again in the future...
Kevin Jeffcoat wrote
I put no...I think that to really speak a language involves a lot more than just having enough information...the computer would have to have emotion and be able to sense tone and style in writing automatically...and this can vary depending on the writer himself or herself!
Computers don't have to have emotions but only need to be able to distinguish the implied emotions of certain words. And they have been able to do so for quite a while. Apart from the option to feed computers the lexicons in detailed hierarchical semantic fields (e.g. all nouns, verbs, adjectives etc belonging to a certain emotion), computers can anaylize corpora and group words according to their semantic fields without any knowledge of their semantic meaning as long as they have enough data. E.g. they can identify words for family members and relatives, food items or emotions as belonging into the same respective category just by analyzing the context (ie. which other words tend to be used in their respective contexts). Absolutely amazing!
There are already MT systems out there that don't look at the text sentence by sentence, but analyze the context of the entire document before and, based on that information, decide which of the numerous translations of a given term is correct.
I don't believe MT will replace translators entirely - particularly not in the more creative areas -, but there is no reason why they shouldn't take over specific areas and do a much better job much more cost-effectively than human translators will ever be able.
A couple of years ago, they were working on implementing on-the-fly MT in traffic reports (completely with a synthetic "human" voice which couldn't be distinguished from a real person's voice as it also included personal and regional idiosyncrasies) so that German drivers, for instance, could understand French traffic reports while driving in France. The driver would never know that it was MT and not a real person.
For 20 years already, Canada has been using the MT system METEO for weather reports:
METEO provides weather forecasts in French and English for the whole of Canada. Starting at 7,500 words a day in 1977, the system today translates close to 80,000 words a day or more than 30 million words a year. A complete computer-assisted translators' work centre, the current version of METEO incorporates two fully-automatic translation modules and two machine-assisted manual translation modules, as well as communications and administrative support functions. METEO now performs 91% of the workload of Environment Canada's translation team in Ville Saint-Laurent, Quebec.
In 1996, the U.S. National Weather Service selected METEO to ensure that all forecasts, watches, warnings and advisories issued for the Atlanta Olympic Games were available in French. METEO 96 translated more than 305,000 words in 16 days with better than 93% accuracy, a task which would have taken a human translator 7½ months. Output was edited by three bilingual Canadian meteorologists.
With a turn-around time of approximately four minutes per text, METEO and METEO 96 ensure that vital weather information is rapidly available to keep the general public weather-wise and weather-safe.
http://www.chandioux.com/press_meteo20_eng.html
And with MT systems becoming increasingly smarter, the range of uses where they would be preferred over human translators will undoubtedly grow accordingly.
[Edited at 2007-10-08 23:57]
| | | | | Depends on the context - and the readers | Oct 8, 2007 |
Charlie Bavington wrote:
However, once a suitable translation for "pull the other one" has been found, that'll be it. Done once, done forever.
Perhaps that is OK if the readers are other computers. Or if the text is only for information purposes (like the controlled language Canadian weather reports that Heike reports on).
But I will retain the freedom to translate "pull the other one" in different ways, depending on the context, the register and the needs of the moment.
I realise that for some purposes, a "one-translation-fits-all" approach may be good enough. But I don't work like that - not even on legal translations, where consistency is important, but where the actual message can often only be grasped by an alert brain aware of the undertones, so the specialist terminology has to be embedded into the sentence in different ways, otherwise the reader will get the wrong end of the stick.
| | | | heikeb
Member (2003)
English to German
+ ...
| Change in the way source texts are created | Oct 8, 2007 |
Victor Dewsbery wrote:
I realise that for some purposes, a "one-translation-fits-all" approach may be good enough. But I don't work like that - not even on legal translations, where consistency is important, but where the actual message can often only be grasped by an alert brain aware of the undertones, so the specialist terminology has to be embedded into the sentence in different ways, otherwise the reader will get the wrong end of the stick.
I fully agree with you, but maybe in face of the more usable TM options that will probably be available in the future, the writing style of practical texts (ie. all texts written primarily for the purpose of conveying information) might be adjusted to work better with the "one-translation-fits-all" approach. Ie. knowing that a text will be translated by a machine will shape the way texts are written. Even with existing commercially available systems such as Babelfish, the output quality can be improved considerably when you are aware of typical MT problems and write your text accordingly. Or professional writers will be able to use/add meta information to avoid typical MT pitfalls or to correct disambiguation problems. If this approach can save a company millions of dollars, it's an approach worth considering.
| | |
|
|
|
| I was simplifying ! | Oct 8, 2007 |
Victor Dewsbery wrote: Charlie Bavington wrote:
However, once a suitable translation for "pull the other one" has been found, that'll be it. Done once, done forever. Perhaps that is OK if the readers are other computers.
Without wishing to sound at all patronising, I was just simplifying this contrasting approach, which no-one else had mentioned up to then.
All the "it'll never happen" folks (of which I used to be one) seem to be working on the assumption that MT necessarily has to somehow replicate human language production, recognising parts of speech, and grammar and what have you, plus somehow recognising the "mood" of a text - saracasm, disparaging, upbeat, whatever.
I'm saying (and Heike has provided more details) that I believe the approach will be more based on taking a vast quantity of data and analysing it to come up with a statistically most probable translation. This analysis could well, of course, include analysing the surrounding text. And metadata, if there is any. There won't be much artificially intelligence involved. Just sheer weight of data.
Imagine a single TM in your language pair with, I dunno, the Bible, all Microsoft documentation, all IBM, SAP, etc. documentation, all EU directives, ECJ ruings, and other flannel, all the stuff the UN churns out, international standards (accounting, technical, etc), all NGO/charity stuff, the major works of the major authors (Shakespeare, Moliere, Tolstoy...whatever), and so on and so on. There will come a time when you'll be able to fit that on a laptop, plus everything everyone on this site has ever done.
You won't get 100% matches for eveything, but I bet it would break the back of the job, at the very least
| | | | Kevin Kaland
United States
Local time: 03:29
Spanish to English
+ ...
I think you have made some good points about computers doing data crunching. They could certainly come up with many statistically probable translations, and I think that many could be correct.
But I don't think that they will ever totally replace human translators, or at least human language experts. We will always need, at the very least, proofreaders and editors who are language-sensitive, to verify the content. And because there will be so much output so fast, that means there wi... See more I think you have made some good points about computers doing data crunching. They could certainly come up with many statistically probable translations, and I think that many could be correct.
But I don't think that they will ever totally replace human translators, or at least human language experts. We will always need, at the very least, proofreaders and editors who are language-sensitive, to verify the content. And because there will be so much output so fast, that means there will need to be ever that many more translators!
But regarding that list of "it could never be done," I think that those were all purely technological tasks...e.g. flying or going to the moon. Only technical knowledge needed to be acquired to do that.
But translation is, you could call it, a human skill. I think that the technical and emotional "knowledge" required for translation (the latter which is quite fluctuating and dynamic), plus that there are other humans on the other end who need to understand it (thus increasing the complexity), ensures that translators will never die. They will never iron out all the bugs!
And how will they even crunch data for tinier languages, that don't have many translations done for them?
I agree a bit with both sides. But I believe that MT will NEVER _fully_ replace human translation ▲ Collapse
| | | | Darren Holliday (X)
Local time: 12:29
German to English
| Almost, but never quite. | Oct 9, 2007 |
Machine Translation will forever be a step behind the humans who design it, for we develop independently whereas technology develops directly as a result of our actions.
We could design the most perfect MT yet, when its design and test phases are complete, it would already require an update supporting the latest use of the word. By the time this update has been identified, programmed and uploaded it will already be time for another one.
If human error (design flaw, upd... See more Machine Translation will forever be a step behind the humans who design it, for we develop independently whereas technology develops directly as a result of our actions.
We could design the most perfect MT yet, when its design and test phases are complete, it would already require an update supporting the latest use of the word. By the time this update has been identified, programmed and uploaded it will already be time for another one.
If human error (design flaw, update error) occurs, the MT becomes even more fallible. If we program the MT to automatically update then it would have to be present with every human person the whole time (like those Star Trek universal translators). Not feasible, not practical and not economically viable.
But if the question is: "Will MT be used more than human translation in the future?" then the answer is a very likely yes as companies seek to meet their targets and push actual quality or preciseness further down their priorities. I have already seen this to be the case, especially within outsourced callcentres.
But, the human translation will live on eternally. ▲ Collapse
| | | | | Hurt feelings! ;-) | Oct 9, 2007 |
Heike Behl, Ph.D. wrote:
In 1996, the U.S. National Weather Service selected METEO to ensure that all forecasts, watches, warnings and advisories issued for the Atlanta Olympic Games were available in French. METEO 96 translated more than 305,000 words in 16 days with better than 93% accuracy, a task which would have taken a human translator 7½ months. Output was edited by three bilingual Canadian meteorologists.
Ok. But honestly, I don't think that was a great achievement. Even back in 1996, with a translation memory any translator can translate sentences in the spirit of "Mostly sunny in the morning...then partly sunny with a 20 percent chance of showers and thunderstorms in the afternoon. Highs in the upper 80s. West winds 5 to 10 mph.".
Also, let's look at the statement about the 305.000 words in 16 days. Ok. How long would this take to the average translator with a translation memory? My estimate is that, with a translation memory fed along the project, an average translator could translate a minimum of 5000 words a day of these weather reports. This means some 25.000 words per week working only 5 days a week, or 3 months of translation time with ONE person. And this person would produce 100% final text.
Now, let's replace the three bilingual Canadian meteorogists with 3 translators with a common translation memory, the time comes down to 1 month. Yes, ok, 16 days compared to 1 month but.... how many people worked in maintaining the system in operation? (Please note that I am not mentioning the people who created the system, which could have easily been a team of 20 people). Easily 2 people just to keep the system up and running. Replace those 2 people with 2 translators, and you reduce the translation time to... 2,44 weeks?
Yes, you had a system you could reuse and improve to enter other areas. But we are in 2007 and it does not look like this system or other automatic translation systems have made any big advances in these 11 years, has it?
Yes, I know, I know. I am defending the idea of 5 efficient workers with metal files competing against a CNC machine. I was just being mean! 
| | |
|
|
|
| "Video killed the radio star"... | Oct 9, 2007 |
...Trados and similar CAT tools killed the high costs of retranslating repetitive material and thus killed the big urge to reduce costs by developing automatic translation systems.
Nevertheless, companies with tons of repetitive/similar items like Microsoft with their Knowledge Base (95% of which is automatic translated) do need and use automatic translation, and these cases will certainly be a success.
| | | |
Tomás Cano Binder wrote:
Yes, you had a system you could reuse and improve to enter other areas. But we are in 2007 and it does not look like this system or other automatic translation systems have made any big advances in these 11 years, has it? :o How do you know?
I fully agree with Heike when she says:
I don't believe MT will replace translators entirely - particularly not in the more creative areas -, but there is no reason why they shouldn't take over specific areas and do a much better job much more cost-effectively than human translators will ever be able.
You know Tomás, the fact that you haven't heard of or seen a performant MT system lately doesn't mean there does not exist one.
I cannot quote any customer or brand name here of course, but I am sure many translators will have the same experience as I because the customer in question is a very big one with millions of words translated per year. I did what they call "post-edition" for them, which is basically about editing machine translations. But they are not using babelfish, my dear, they are using their own system which is already very performant (translations are in the technical field). Actually, in about 50% of the cases, translation is just perfect. Say, in 30-40% you have to change something and in 10-20% the translation is totally messed up and you have to rewrite it.
But such systems will surely improve in the future, even if they will not replace creative or innovative translation.
| | | | Henry Hinds
United States
Local time: 04:29
English to Spanish
+ ...
In memoriam
Never say never...
I'm sorry, but I will say "never". Language is human, and when a machine can replace a human in all functions, then ask me if I will say never. It cannot happen.
No two humans speak the same language; no machine can ever speak our language. It is an absolute.
| | | | | Pages in topic: < [1 2 3 4] > | To report site rules violations or get help, contact a site moderator: You can also contact site staff by submitting a support request » Poll: Will machine translation ever replace human translation entirely? | 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 » |
| | 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 
