In the Hitchhiker Guide to the Galaxy books, characters from across the universe can communicate using something called the Babel Fish, a small creature that burrows itself into your ear and translates any given language by sending signals directly into your brain.
Here on earth, we don't have a Babel Fish. But, thanks to a burgeoning form of artificial intelligence, honest-to-goodness translation technology is finally on its way. And it won't involve implanting a fish in your ear.
Later this year, Microsoft will release a beta version of Skype Translate, a near real-time translation tool. The app was unveiled last night at the Code Conference in Rancho Palos Verdes, where Microsoft vice president of Skype Gurdeep Pall used the tool to have a conversation with a German-speaking colleague named Diana. It's not the first time the company has showed off the technology that underpins this app--Microsoft Chief Research Officer Rick Rashid did a similar demo at an event in Tianjin, China in October 2012--but now, the software giant has wrapped this translation engine in a package that's ready for everyday internet users.
>Honest-to-goodness translation technology is finally on its way--and it won't involve implanting a fish in your ear.
According to an article published by Microsoft Research, the company has been working on this technology for over a decade. Real-time translation is a tough trick to pull off because it requires both accurate speech recognition and on-the-fly language translation, but the technology is finally becoming a reality thanks to advances in a field of computer science called "deep learning." This involves new kinds of "neural networks"--computer simulations of the connections between neurons in the brain--that can mimic the way the brain behaves, at least in some ways. Deep learning models are no where near as complex as the brain, which contains tens of billion of neurons, but the approach allows machines to "learn." They can improve their performance as they analyze more and more data.
The technique dates back to at least the early 1980s, but it stayed on the fringes of academia until the mid-2000s when powerful computers became cheap enough to start doing practical work. In 2009, Microsoft invited deep learning pioneer Geoff Hinton to visit its Redmond campus to work with its researchers on neural networks, and although Hinton has since joined Google, Microsoft has moved ahead with its own deep learning research, which became the foundation for both Skype Translate and its new digital assistant, Cortana.
Meanwhile, Facebook has hired another deep learning guru, Yann LeCun, and Netflix is experimenting with deep learning techniques in order to improve its recommendation algorithms.
Beau Cronin, a neuroscientist and machine learning expert at Salesforce.com, points out that just because something is "brain like," doesn't mean it's useful. "There are many ways a system can be like the brain, but only a fraction of these will prove important," he wrote in an article for O'Reilly Radar. Nor is deep learning the only way to power real-time translation. The startup Babelverse, for example, enables you to find a human translator available to do live translation via a mobile application. But what Microsoft has achieved is certainly impressive--at least on the surface.
Many think of Microsoft is a lumbering dinosaur, but projects like this--along with things like Kinect--show the benefits of being a large company that can spend several years, or even a couple of decades, developing new technology. Google may have taken the lead in turning blue-sky research into reality, but it's far too early to count Microsoft out.
