Home Languages Articles Links Downloads About Contact

Languages

bulletEnglish
bulletSpanish
bulletFrench
bulletGerman
bulletLatin
bulletChinese
bulletJapanese

Other Topics

bulletTranslation prices
bulletMachine translation
bulletHarry Potter
bulletTranslation quotes


Free website Translation Service

Search


Advertisement

Machine Translation Reality Check





Recent news articles make real machine translation seem as though it's just around the corner, and a recent discussion on the Honsyaku list (a Japanese translators group online) led to strong opinions and heated debate about the future of computers in translation.

So let's take a step back and see where we are, then we'll look at where we're likely to be in five, ten, and fifteen years. First, at present there is no machine translation system that can even begin to approach the quality a mediocre human translator can easily produce. Hype from MT companies notwithstanding, the simple reality is that the technology does not do what its proponents claim. You can test this yourself by trying to translate this page with Google, Yahoo!, or other services and see what the results are like. Word salad is the commonly used phrase.

But machine translation is getting better. The word salad is now useful enough so that, at least in some cases, you can get the gist of a document in a matter of seconds. If all you need to know is roughly what a document is about, then machine translation can play a role in your work. And with carefully prepared source texts on certain subjects using highly customized (read expensive) MT systems, the results can be better than word salad, those considerable post-editing by human translators is still essential to produce readable copy acceptable to the general public.

Where will we be in five years? That depends who you ask. Futurists use the "five year" benchmark frequently, and have since the 1950s been claiming that human-level MT is only "five years away." The most accurate futurist these days, Ray Kurzweill, thinks good MT is about 12 years away. Most translators think that MT is impossible, or at least sufficiently difficult that they won't see it within their professional lifetimes.

"Prediction is difficult," said Neils Bohr, "especially about the future." So what breakthroughs in cognitive science, neural computing, artificial intelligence, and computational linguistics might be around the corner are impossible to say, and whether or not they will result in human-level MT is also impossible to say. But improvement is inevitable, and a pattern in artificial intelligence is emerging that may help us see what is going to come.

For many years, AI researchers were trying to solve the "hard" problems of getting a computer to play chess well or prove mathematical theorems, assuming that the "easy" problems of getting a robot to maneuver on its own through the average home's living room were too simple to bother with. It turns out they had it all wrong.

Playing chess is easy. A very fast, dedicated computer, IBM's Deep Blue, is now the best chess player on the planet. Computers routinely contribute to proving mathematical theorems, solving intractable equations, and predicting or modeling the behavior of hurricanes, El Niño, or solar flare activity. But the average robot still cannot navigate the typical living room, at least not without bumping into everything along the way, and essentially working at random to solve a problem that a cockroach can handle easily.

In other words, AI researchers are no longer working from the top down, but instead are working from the bottom up. The apparently easy problems turn out to be very hard, and language seems to be the hardest of them all. So MT is likely to make its appearance at the bottom of the industry, first handling such simple and predictable material like diplomas, immigration documents, college transcripts, and such. It will then, as improvements are made and breakthroughs come, be able to handle simple business letters and other correspondence. But it will get to the level of dealing with research articles, political discourse, and literature only after many years, possibly decades of improvement. Put simply, translators' jobs are not immediately at risk.

There is, however, another approach, one that is currently entirely theoretical but may prove in about 25 to 30 years to be practicable. If a conscious artificial intelligence that can communicate in human language can be created (and this is a big assumption), then all that remains is to teach it to translate (assuming it wants to). But that approach is pure science fiction at this point, and although machine consciousness remains a hot topic of debate amongst theorists, there is nothing even on the drawing boards in research laboratories at present.

So sit back and relax when you work with your computer. It is not about to put you out of a job. Instead, machine-assisted translation and other tools will likely change how you work significantly in the next five to ten years. Moreover, changes in economic, political, and social situations around the world are far more likely in the coming years to affect a translators' career than something as far off as true MT.

Below is the foregoing paragraph as translated by Google's Language Tools:

[French version] Se reposer ainsi en arrière et détendre quand vous travaillez avec votre ordinateur. Il n'est pas sur le point de vous mettre hors d'un travail. Au lieu de cela, la traduction automatisée et d'autres outils changeront probablement comment vous travaillez de manière significative en cinq à dix années à venir. D'ailleurs, les changements des situations économiques, politiques, et sociales autour du monde sont bien en prochaines années pour affecter une carrière des traducteurs que quelque chose aussi loin au loin que la TA vraie

[Spanish version] Sentarte tan detrás y relajar cuando trabajas con tu computadora. No está a punto de ponerte de un trabajo. En lugar, la traducción automatizada y otras herramientas cambiarán probablemente cómo trabajas perceptiblemente en los cinco a diez años próximos. Por otra parte, los cambios en situaciones económicas, políticas, y sociales alrededor del mundo son más probables lejano en los años que vienen afectar una carrera de los traductores que algo tan lejos apagado como TA verdadera.

And below we have the same from Altavista's Babelfish:

[French version] Reposez-vous ainsi en arrière et détendez quand vous travaillez avec votre ordinateur. Il n'est pas sur le point de vous mettre hors d'un travail. Au lieu de cela, la traduction automatisée et d'autres outils changeront probablement comment vous travaillez de manière significative en cinq à dix années à venir. D'ailleurs, les changements des situations économiques, politiques, et sociales autour du monde sont bien en prochaines années pour affecter une carrière des traducteurs que quelque chose aussi loin au loin que la TA vraie.

[Spanish version] Siéntese tan detrás y relaje cuando usted trabaja con su computadora. No está a punto de ponerle de un trabajo. En lugar, la traducción automatizada y otras herramientas cambiarán probablemente cómo usted trabaja perceptiblemente en los cinco a diez años próximos. Por otra parte, los cambios en situaciones económicas, políticas, y sociales alrededor del mundo son más probables lejano en los años que vienen afectar una carrera de los traductores que algo tan lejos apagado como TA verdadera.

You can see the problems and limits of this technology, and you can try other experiments of your own with Google, Babelfish, or whichever other system you prefer.


Back to Machine Translation. top