Very interesting - but it's a tough challenge. I'd think the results couldn't be a lot better than what you get in text using Google Translate, and possibly worse given that people tend to speak more idiomatically than they write. You probably wouldn't want to have important conversations using it.
Links:
[1] http://technology.timesonline.co.uk/tol/news/tech_and_web/personal_tech/article7017831.ece
[2] http://rabble.ca/print/babble/humanities-science/google-working-real-time-voice-translation#comment-1110111
[3] http://rabble.ca/print/babble/humanities-science/google-working-real-time-voice-translation#comment-1110116
[4] http://rabble.ca/user
[5] http://rabble.ca/user/register
I would like to see how they translate idiomatic and colloquial phrases, vernacular dialects, accents and the cultural use of words in different contexts. Onward reductionism!
once they have the basics for the Big Languages I can't imagine that they'll have too much trouble getting it to smaller localized translations. I honestly think that this sort of capacity for accurate and worthwhile translations of numerous languages is only a decade or so away. Given that it will be expensive as anything to develop/study/catalogue/do all that stuff for every language you'll see Big Languages with lots of speakers becoming more normalized at the loss of smaller languages (and probably whole language families). So, you'll probably see Mandarin translator, English, Spanish and Arabic and French. Then smaller, more regional languages will follow, so probably Persian, Russian, ones like that. You'll probably also get a lot more standardization and centralization of 'literary' languages and some interesting slang will develop that these machines will chase to define. I can imagine there will be a very interesting rise amongst the intelligence community in pushing for the development of small language families based on if they will be needed. Once the software is available, it will only take time before linguists (the most employable of all academic pursuits!) will say 'hey, the DIA, CSIS, FSB are all running projects on computer translation - maybe my degree in Georgian, specializing in the ergative case, will actually make me money!' Plus, you know, diffusion of technology - once this stuff hits the Cloud City we call the internet individuals will pick it up and probably use it for all kinds of super neat stuff, find exploits, etc.
Needless to say, the coming years are going to see the greatest extinction of languages that this world has ever seen - and disparity technological advances will be making this all the more pronounced in the global south. Oh, and in countries that actively suppress local languages - I can't imagine that the control of information that is going to become possible (and a preferable norm) for governments is going to do much good for 'undesirable' 'obsolete' or 'backward' languages.