Comment Re:Whats the problem? (Score 1) 535
I do research in computational linguistics, and I've done work at a major machine translation (MT) provider, and I can state with some certainty that we will NOT have perfected translation software in 10 years, if ever. MT is about as old as artificial intelligence itself, and it's still struggling. Researchers can get their work published if they improve BLEU scores (an MT evaluation metric) even by less than a point (on a 100-point scale). Statistical MT is reaching the limits of what it can do by simply adding more training data, and hand-built grammars never really took off as they've not yet proven to be feasible in terms of time and memory complexity. The dream of having instantaneous, accurate, speech-to-speech (or even text-to-text) translation is as far fetched as time travel and deep-space colonization.
It will, and does, help people understand documents written in other languages, but it will be a long time until computers can translate (or understand or generate language) as well as humans.