Netgain Challenge

Universal language translation will be a fundamental tech breakthrough. Imagine conversing with anyone in the world. You speak English, your Russian friend hears and replies in Russian, and you hear English, all facilitated by software on your mobile device. Innovations are near real-time, continuous translation, but prototypes (i.e. Skype Translator) are limited. There are 6,500 spoken languages, and 2,000 have less than 1,000 speakers. This challenge proposes a universal translator for the 3,000 most popular languages. The bottleneck for LVCSR (large vocabulary, continuous speech recognition) and MT (machine translation) is collecting large amounts of high-quality data to train statistical models. We propose an open source Wikipedia-like platform to crowdsource data. A DIY toolkit can collect clean speech samples and source even obscure languages from speakers and scientists. This challenge connects humanity while preserving a precious and endangered resource: language diversity.