Duolingo. Luis von Ahn is developing software with a group at CMU that will help you learn a language and simultaneously translate the web. Its incredible.
Translation data is obtained by statistically combining translated responses from tons of people using the software. Why would tons of people use it? Ahn’s group is testing it now and it’s working. People spend a lot of money already on language learning software, so this provides an alternative. If all is designed well, people would use this instead, and contribute to a translated internet.
I just love how Ahn begins with the question of how can you create something great with a million people. He has created several architectures of mass collaboration. In this case the personal payoff is much more immediate for the users- they learn a language! For previous tools like captcha, recaptcha and image labeler, the gains were for digitizing books and indexing images for the last two, respectively- which would be harder to see immediate results for the users.
From a design and complexity standpoint, he is creating attractive local behaviors (the incentive to learn a language) from which globally a translated internet will emerge (using statistical natural language processing, perhaps Bayesian networks). I’m excited to see what sorts of tools and APIs are created using Duolingo.







