The Translator Google Made
People wonder how it is that Google succeeded so well in translation when it apparently failed at lots of other things.
There’s a little secret to that – the translator Google made is so successful because it directly benefits from having access to Google’s mega powerful computing centers – the ones it calls its search engine farms.
Whenever Google finds a way to exploit its overwhelming computing power, it wins.
It’s a widely publicized anecdote about the algorithm translator Google had licensed at one time; and it was popularized in the New York Times.
According to the story, one of Google’s founders tried translating something to the effect of “Google is a favorite search engine” through the algorithm. What he got out of that was, “Google green onion thing”.
That incensed the Google genius so much; he unleashed the company’s research department on the job.
Today, not only is the translator Google has come up with much better, it does dozens of languages across the world and handles a hundred million of translation requests every single day.
Traditionally, translation software that’s anywhere near as successful as Google’s has to be dedicated to one subject area. General-purpose translation software usually ends up saying something like “a green onion thing”.
As great a piece of success as this has been for Google, the company has shown how it is capable of making mistakes in the past – consider its failed social networks Buzz and Orkut, its collaboration engine Wave, and its book scanning project that’s been impossibly tangled up in court.
When Google succeeds though (the Google Phone initiative appears to be wildly successful right now), it can take over the world.
The thing is, Google has done this at a time no one else has been really taking the market for translation as all that important. Translation has forever been one of the Holy Grails of artificial intelligence.
The approach that scientists have always taken with machine translation has been to teach the computer the rules of the languages involved, and then fed in dictionaries.
For about 20 years now, scientists have found that computers that have access to millions of pieces of human translations are able to grasp things that the plain algorithms before never did.
Guess who has millions of snippets of human-generated translations? Google, that’s who.
Right now what the translator Google has created is capable of doing is it can take a message and give you a general idea of what it is about.
It’s not going to really create a finely nuanced translation that a human translator with begin to feel insecure about. What it does is, it gets the job done.
If you’re wondering where it gets all the human generated translation snippets in the millions from, Google does have access to the transcripts of everything the United Nations does, translated into dozens of languages. It also has everything that goes on in the European Parliament.
This is something that no one but Google could have done, given the amount of computing horsepower needed. It opened the Internet up to new audiences, which is a good thing for Google’s bottom line.