Sugata Mitra gave a great TED talk about minimally invasive education. I've long been a fan of his work with unsupervised learning and the hole-in-the-wall project, but what really captured my imagination was the project he described in which he left a computer running speech recognition with some children in Hyderabad who had some knowledge of English but were unable to speak it understandably. Apparently after several months with the system they were able to make the machine understand them reliably.
When I first heard about this I got the impression that the kids did not actually learn 'American English' but a 'Computer English'. I can imagine that over time 'Standard English' may come to be defined by the accuracy with which Google Speech can recognize your voice more than any particular geographic accent.
But back to education, I would love to explore more ways in which free speech recognition software could be used to facilitate English language learning in children who don't speak English at home. Low cost Android devices are probably the primary platform this should be developed for. I can also imagine a standalone slotmachine like device that could get set up in rural community centers as being effective too.
I find this concept especially fascinating as it is training children at a very young age in computer interaction. This may prove to be even more valueable to the students in their lives than human to human English communication.