Google's future voice-enabled universal translation system “moonshots”

Google’s new  voice-enabled universal translation system “moonshots” .Spiegel reports that Google’s Translate team has enormous ambitions to eventually eliminate language barriers with the development of a voice-enabled universal translation system. Although such a system is certainly a long way off, Google has already started working on a voice-enabled translator smartphone app that can translate roughly 24 different languages so far.

Franz Josef Och, who leads Google’s Translate department, admits to Spiegel that this voice-enabled translator is fairly slow and clunky right now but he points to improvements in Google’s text-based Translation service over the years and projects that the app will grow by similar leaps and bounds. The biggest barriers remain nailing down the more subtle nuances of language such as syntax and ambiguity.

What makes Google’s approach so interesting is that the company doesn’t actually employ a linguist anywhere on its team. Instead, the Translate app is simply designed to get better with experience. Since Google’s specialty is collecting, sorting an analyzing data, the Translate algorithms aren’t designed to understand grammatical rules.

The algorithms search through the clutter, gather data and learn along the way. For instance, when the program is tasked with translating a sentence from French into German, say, it searches for matching phrases that already exist and computes how to best compose a new sentence based on this information.

This often works well, but it is still far from perfect. Syntax, intonation and ambiguity remain a significant problem for automated translation programs. Often, the translations they produce are comprehensible, though barely, but are also a nightmare for linguists. The machine simply has no sense of aesthetics.

So does Google have to teach the program creativity to be truly successful? No, says Och, noting that it's really a question of "more effective learning." In other words, the computer needs to master the art of imitation.

This is why Google Translate, despite its vast improvements over the years, still delivers sentences that are understandable but that require some syntactical detective work to really sound like proper English. But given that Googlers used Translate more than 200 million times last year alone, it seems that the company will have all information it needs to keep perfecting the service.


Labels: , ,