Computer Translator Ready for Testing in Iraq

cgibby98 wrote to mention a Wired News story about a battle-zone translation technology that may allow near real-time conversations between English and Arabic speakers. From the article: "Funded by Darpa, the system would allow troops to communicate in Arabic through a laptop computer equipped with voice recognition and translation software. Troops could speak in English and have their words instantly translated into Iraqi Arabic, 'spoken' by a computerized man's voice. The program also translates Arabic into English. Will it replace the need for an interpreter when you're having some sort of high-level conversation? Absolutely not. But it is absolutely to the point where it could be useful in some carefully chosen situations."

Comments

This was already possible

[matt me]

I managed something similar a year or so back, in an attempt to create a 'babelfish'. Of course the input/output had to be specified, and it had a very limited range of languages - certainly no universal translator but it did use all free software (as that's all I have).
0) Input recording of English languagge
1) Voice recognition software (Sphinx) pipes output to
2) Script using online translator to convert between language
3) Festival stumbles out an imhuman gramatically-wrong rendition of the input.

It wasn't exactly in realtime, I just fed it recordings, for which it would then output an audio file in the other language. The worst step was the voice recognition, which didn't work great even when given the output of the voice syntethisier.

Sphinx http://cmusphinx.sourceforge.net/html/cmusphinx.ph p

Bad substitute for Arabic _training_ for _humans_

[dpbsmith]

"No one in the military would make life or death decisions based on a machine translation." That's pure CYA for the first time this device gets someone killed. If a life or death decision needs to be made and the only thing you have at hand is a machine translation, what are you going to use?

I don't know how representative of the state of the art they are, but I've been massively underwhelmed at Babelfish's ability to understand foreign-language text and by ViaVoice's ability to understand speech. I can't imagine the effect of layering machine translation errors on top of machine voice interpretation errors.

According to href="http://www.theatlantic.com/doc/200512/iraq-a rmy">James T. Fallows, "The U.S. military does everything in Iraq worse and slower than it could if it solved its language problems. It is unbelievable that American fighting ranks have so little help. Soon after Pearl Harbor the U.S. military launched major Japanese-language training institutes at universities and was screening draftees to find the most promising students. America has made no comparable effort to teach Arabic. Nearly three years after the invasion of Iraq the typical company of 150 or so U.S. soldiers gets by with one or two Arabic-speakers. T. X. Hammes says that U.S. forces and trainers in Iraq should have about 22,000 interpreters, but they have nowhere near that many. "

Instead of doing the obvious thing--give soldiers training in Arabic and offer big bonuses for Arabic-speaking recruits--the U.S. does nothing for a couple of years and then tries to throw a cheap technical fix at the problem.

If we must throw gadgets at the problem, why not a satellite phone linked to a big building full of human Arabic/English simultaneous translators?