Photo via Flickr
Of all the artificial intelligence nuts to crack, language translation is one of the hardest. Not only are there some 6,000 spoken tongues around the world, you have to account for slang, accents, near-unrecognizable brogues, countless regional dialects, and increasingly, hybrid languages like Spanglish popping up in the melting pots of the globe.
For those very same reasons, a tool that can translate language in real-time would have an enormous payoff for society—if someone can crack it. So far, no one has really nailed the technology, despite a few close calls and decades of trying. But we're definitely getting closer.
Today, the Japanese telecom giant NTT Docomo unveiled its new augmented reality glasses that can translate text on menus and signs in near real-time. It uses character recognition technology to convert the foreign dialect to your native tongue and display the translated text on the device's screen. The goal is to have the gadget ready for action in time for the 2020 Olympics in Tokyo, so tourists can more easily navigate their way around the city.
While NTT Docomo perfects its gadget, US heavyweights like Google and Microsoft are also cramming to improve real-time translation. And unlike the menu-reading glasses, they’re tackling verbal conversations as well as text. Cutting down on the myriad awkward translation errors hinges on improving machine translation and speech/text recognition technologies, and that hinges on artificial intelligence.
To reach that level of accuracy, intelligent machines have to not just convert each word into a new language, but analyze entire phrases for context and infer their meaning before offering up a translation—the way a human interpreter does.
As it happens that’s something Google does quite well, and sure enough, this July an Android exec at Google divulged that the company is working on developing a real-time translation tool for speech. Android VP Hugo Barra said early prototypes are working "near-perfect" for certain languages, providing there's no noise pollution, the UK Times Online reported. Franz Och, Google’s head of translation services, predicted that speech-to-speech translation would work reasonably well in a few years’ time.
Meanwhile Microsoft is nipping at Google’s heels. Using “Deep Neural Networks” AI technology that mimics the human brain, researchers are developing context-dependent speech recognition, a Microsoft executive revealed last year.
On the startup side, there are more low-tech translation devices aplenty. The Word Lens app works similarly to Japan’s augmented glasses, only you have to hover your iPhone over the text to see the translated prose. ReadSpeaker, announced last week, combines voice-to-subtitles technology with subtitles-to-voice to finally get full circle translation geared toward business collaboration tools—so executives can Skype in four different languages.
Yet another device, SIGMO, just blew past its Indiegogo goal. It’s marketed to travelers: You speak into the device and the sentences that come out are translated into your language of choice (from a couple dozen options).
So progress is being made. But before the Babel Fish device of Hitchhiker's Guide to the Galaxy fame can really become reality, wearable tech—like Japan’s menu-reading augmented reality glasses, and of course Google Glass—will have to escape the niches of nerdom and enter the mainstream. The Google researchers working on real-time translation haven’t mentioned anything about incorporating the technology with Glass, but that's an obvious application once it launches.
If a product can offer accurate real-time language translation with an augmented device that someone would actually want to wear, that'll be a seriously world-changing thing. Forget love, technology is shaping up to be the universal language.