TJ Blog
Monday, August 13, 2007
Machine Translation
Making computers translate has been the dream and goal of computer scientists since the early days of computing. The idea of machine translation was born before word processing, spreadsheets, and electronic data transfer, and long before DTP and the Internet.
We have all seen horrible examples of machine translation (por ex., at http://accurapid.com/accurapid/goodbadfr.html#MT, which has prompted some experts to declare that "machines will never replace human translators." But let's not forget that the same experts had also predicted that computers would never play chess at the level of human grand masters.
The fact is that computer hardware and software is becoming more and more powerful, which has made the use of new techniques such as corpus-based translation possible. (Several articles of the Translation Journal describe how corpus-based machine translation works.) The successes achieved with these new techniques have given rise to predictions of the opposite extreme, namely that high-quality machine translation of any imaginable text is just around the corner.
Today millions of words are being translated by computers every day, and this number is bound to increase exponentially in the near future. Don't expect the works of Shakespeare to be translated into Quechua by a robotic superpoet tomorrow, but machine translation of many types of straightforward text between the major languages already meets the quality standards for practical applications.
Where does this leave the human translator? Those of us who lack specialization or skills to perform at a superior quality level will have to change careers or content themselves with doing pre- or post-editing of machine-translated texts. Only the true aces of the profession will survive and prosper in an era where the bulk of translation work will be performed by computers.
Do you agree with these statements? Where do you see our profession and yourself five or ten years from now?
We have all seen horrible examples of machine translation (por ex., at http://accurapid.com/accurapid/goodbadfr.html#MT, which has prompted some experts to declare that "machines will never replace human translators." But let's not forget that the same experts had also predicted that computers would never play chess at the level of human grand masters.
The fact is that computer hardware and software is becoming more and more powerful, which has made the use of new techniques such as corpus-based translation possible. (Several articles of the Translation Journal describe how corpus-based machine translation works.) The successes achieved with these new techniques have given rise to predictions of the opposite extreme, namely that high-quality machine translation of any imaginable text is just around the corner.
Today millions of words are being translated by computers every day, and this number is bound to increase exponentially in the near future. Don't expect the works of Shakespeare to be translated into Quechua by a robotic superpoet tomorrow, but machine translation of many types of straightforward text between the major languages already meets the quality standards for practical applications.
Where does this leave the human translator? Those of us who lack specialization or skills to perform at a superior quality level will have to change careers or content themselves with doing pre- or post-editing of machine-translated texts. Only the true aces of the profession will survive and prosper in an era where the bulk of translation work will be performed by computers.
Do you agree with these statements? Where do you see our profession and yourself five or ten years from now?
16 Comments:
I have been using both standalone and Web-based machine-translation (English & French into Spanish) for quite a while now.
The results depend greatly on the nature of the text. Texts of a declarative nature are best for machine-translation.
Between the extremes that you mention, I see machine-translation as another CAT tool, such as translation-memory and speech-recognition programs.
Given the infinite plasticity of language, it is not conceivable that human translation would be consigned to the dust heap of history.
I agree with your point that possession of superlative skills will determine who stays in our field in the next 5-10 years.
Day by day computers get more powerful and software becomes capable to dealing with amazing things. However languages have a lot of idiosyncrasies which I don;t think can be dealt with by a computer.
Sophie.
www.translate-to-chinese.com
I wonder what you mean by "aces", there.
I am not worried about my own place in the market. Reverting to the sort of unidiomatic translations that were done by pretenders with dictionaries, the ones that gave foreign (to the reader) manufacturers a bad name? It won't happen.
I think technology will shape the translation market but not replace the human translator, just as technology have done with so many other professions. Similar statements have been made about photography, using large photo databases an automatic image can be created in ways that are freaky to watch. I don't see any of the two disappearing, people will have to enhance skills faster just a everything is faster now, the entry barriers might become higher for some, and probably it will just make the market only available for highly skilled. But using the chest example even when computers can match humans, people wants to see humans competing. Even when a computer can produce a perfect translation, the new reader will look for subtle twists that the reader of today doesn't even perceive.
One more article as Machine Translation vs. Human Translation at Translationindia.com
Check out A Simple Model for Outlining Translation Technology which is a visual model based on the ideas of Alan Melby and Yehoshua Bar-Hillel. It's a great start to understanding the reaches and limits of machine translation and translation technology in general.
It is absolutely, thoroughly unrealistic to think that computers will ever, ever replace human translators. Computers may be able to do some basic "draft" translating, as they already do, but computers will likely never, ever be able to deal with the ambiguities and intercultural issues that frequently occur in natural human language. Human language is always changing, always evolving: computers won't be able to deal with that, either. No translator should be worried about losing his or her job.
That said, CAT tools are helpful *to human translators.* But CAT tools are often also overestimated in their usefulness.
I think it will happen, but gradually, over decades.
Most technologies develop in separate parallel processes - the Internet with the need for sharing information, familiarity with interactively accessing information off a screen (teletext), standardization in computer technology and the upsurge in electronic entertainment. Cars need roads, and the light bulb needs a power supply... you get the drift.
Machine translation will develop with networked translation pools, text quality checkers in the major languages, increased use of translatable boilerplate texts by international companies (the bread-and-butter jobs of most translators), lawyers, governments, and other organizations, increasing refinement and intelligence in text and context analysis technology including software that can access relevant text sources on the Web and use them as reference, taking in and applying cultural differences along the way. Not necessarily in that order.
I don't think there will be any noticeable sudden breakthrough, as this area is way too complex for any single development to make a noticeable difference. For the reasons that s.m. mentioned, I don't think good-quality translators need fear for their jobs for the next few decades; the network pooling and text standardization process alone will take decades, as it will rely on volume, and therefore wide acceptance. In the process, we'll probably see general language use standardizing to a more simple, translatable structure, as unidiomatic translated texts become more commonplace. Bland, but who talks like Shakespeare today.
One thing I think is as good as certain: there will come a point where the software has become so advanced that the probability of even the best translator making a mistake is higher than that of a machine (or the network it's hitched to) - although the mistake may arise for different reasons.
Machine translations will probably never reach perfection. But that also applies to human translators.
Btw: I don't use any kind of translation software - the hassle is more than it's worth.
Great article.
Language is so complex that even with human-made translations there are words or expressions that one have to ask for more context or explanation in order to have an accurate translation.
Machine translation could have this feature (ask for more explanation or disambiguation) to improve its results.
I do not fear computers taking my place in the future, in fact the better the computer the easier my job (I have translated En 2 Sp for 20 years) what scares me the most is new agencies offering lower and lower prices, translations made by non-native speakers which some times resembles the translations made currently by computers and with which I had to compete. The result is a User Guide User Guide in English with its translation to Spanish. Where you can find "carabiner clip" translated as "cierre tipo carabinero". You can try for eons but you will never find such a thing in Spanish. This translation just does not make any sense; the correct translation of carabiner clip is "mosquetón".
The problem that is likely to remain with Machine Translation is that the entire system is based on the source language, not on the target one.
A professional translator on the other hand will always focus on the target language, thus reducing the chances that the recipient of the translation will misunderstand the original content.
I've been working on a student project for two years and impressions of other students who translated something via online machine translator were so negative that the vast majority of them said they would never go for free machine translation solution when they need professional translations to be done
I'm still trying and see if pre-editing can help to solve some of the issues in MT.
@Ellen:
I tried that too, a few years ago; I don't know what your experiences will be, but I found that each step in the translation process was a source of error so the quality was poor, and the time saving was negligible.
Machine translation is definitely man-oriented, as it is nurtured by human knowledge only..machine translation might be very helpful but not entirely safe as it only conveys the conventional meaning of words and not their interactive purpose or spirit. In fact relying on machine translation alone, no matter how rich its vocabulary, could result to horrible mistakes and misunderstandings! That is why turning to professional human translators and agencies who represent them, is the safest way to choose if you want a high quality and reliable translation..professional translation services can guarantee specialized and effective translations since it is based on highly qualified professionals in the field who combine human knowledge with machine translation to convey the best possible results.
really nice blog and the information provided about Translation service companies
was nice and easy to understand
Post a Comment
<< Home