They've been talking about this for 20 years, but now real time translation is "good enough":
- Most Americans likely have difficulty understanding the broadcasts of Al Jazeera, the Arab news network, but several government agencies now can watch it while simultaneously receiving an English translation of the programming.
Virage Inc., a San Mateo, Calif., a maker of Internet video technologies, has recently supplied several unnamed United States intelligence agencies with a system that will provide real-time voice recognition and English translation of foreign-language news broadcasts.
The system, which was financed last year by the Pentagon's Defense Advanced Research Projects Agency, or DARPA, can run on any fast personal computer, generating scrolling text displays of both Arabic and translated English text.
....Under ideal conditions, the BBN recognition software can be 95 percent accurate in recognizing spoken English, said Bradley Horowitz, a Virage founder and its chief technology officer.
However, in practice such a system would be most useful in spotting individual words rather than creating reliable transcripts, he said.
"This is what I call 'good enough' accuracy," said Mr. Horowitz. [NY Times]
That this is just coming together is funny to me because 20 years ago I worked as the US correspondent for a Japanese pop music show: I conducted interviews with various stars including Sting, Stevie Wonder, Herbie Hancock, George Duke, Devo, Stray Cats, Spandau Ballet, Soft Cell, Lionel Richie, and on and on, and sent them on to Japan. There they translated the interviews, then FAKED A SIMULTANEOUS COMPUTER TRANSLATION, making the translation of my voice and the stars' voices sound robotic and artificial even though they were really just the voices of human translators, because they thought it was cool and high-tech. I love the Japanese, but they can be really goofy.
Article comments
1 - Rob
As a translator, I'd hate to see a machine that can replace a human, but this machine could definitely speed up translation. The problem is that most machine programs cannot translate words in context, and they have problems with idiomatic speech. Still I'd love to work with something like this for rough translation, then humans can fine tune it.