For donkeys years automatic translation has been five to six years away. In many ways it has been a good example of a Techno-Ponzi scheme - appearing almost within the grasp if technologists, so allowing technologists to raise large sums of development cash to go the last mile.
But the last mile is still untravelled.
However, at this year's MWC, Google CEO Eric Schmidt demonstrated a prototype of an app which allows you to take a cellphone picture of a foreign menu and get an instant translation into English.
Already the Google Translate feature is the best source of machine translation around.
That's because Google has taken the approach of throwing huge computing capability at the problem, and Google has probably the most powerful computing resources on the planet with which to do this.
Google Translate is far from being good, but Google is finding that the more computer power you harness to the task, the better automatic translation gets.
So the menu thing will be a great application for Cloud Computing.
Obviously it's probably never going to be possible have a handheld device that can do automatic translation locally, but a handheld device that sends the German menu off to Google's enormous data processing facilities, which then send the English version back to your handheld is, apparently, going to become a possibility reasonably soon.
Good old Cloud Computing.
The problem for Google here is how do they pay for this ? Google translation is already used on many multiuser gaming platforms quite effectively but with no way of delivering adverts that can be seen I'm not sure they will be prepared to carry on with this philanthropy indefinitely.
Mike, It seems to me that one of the good things about Google is that it puts out a service first, then, if it takes off, tries to figure out a way to monetise it. To me this is smart thinking. After all the world went MS-DOS because you could copy the office version on a floppy and install it on your home PC for nothing. Then, wham bam, when everyone couldn't do without it, MS says everyone has to pay. That seems to me good business. The guys at the top of the telecoms network operators don't have the nous to follow that model. They seem to want it all worked out, and all under their control, from day 1, which is one reason why, in my opinion, they're always two steps behind the curve.
One of the great things about Google is they keep the rest of us on our toes. As a developer of machine translation solutions who's had a good run, I now find myself submitting every test passage to Google Translate to see how well or badly my approach does compared with theirs. Certainly keeps me from resting on my laurels.
"...The problem for Google here is how do they pay for this..."
Google Translate is not very expensive to provide to users on the fly per translation request. Processor-wise, most of the heavy lifting for Google Translate is performed ahead of time---pre-processing of the most probable translations is one of the keystones of Google Translate (which I am asked to reverse engineer from time to time by my Japanese clients).
The "expensive" partis Google's massive statisical-anaylsis approach that their researchers employ to Google's huge corpus/database of words. Even this expense is rather low for Google because it is being performed in a massively distributed manner utilising wasted CPU cycles like the SETI program (but more clever computationally, better design thoughtout and far better managed).
Taro
-- a surviver of world's most expensive state-funded scientific failure-Japan's Fifth Generation Computer Project (machine translation, AI, etc.)