Personally, I don't think there will ever be a software program powerful enough to accurately interpret and render differences in tone, pronunciation and accentuation proper to spoken language. Not to mention dialect and homonyms, which abound in some languages. A Belgian company once claimed to have achieved just that, but they were later exposed as frauds who used company funds to buy their own stock in order to (artificially) fortify the company's position on Wall Street. (Google "Lernaut & Hauspie" for more info).
Food for thought, this quiz comparing machine translation to the work of Nobel Prize winning author William Faulkner:
http://reverent.org/sounds_like_faulkner.html