Via UWR.org
How close are we to breaking down global language barriers? Microsoft researchers demonstrate that we’re closer than anyone could’ve imagined.
For the last 60 years, computer scientists have been working tirelessly to build systems that can genuinely understand what a person says when they talk. The most primitive of techniques involved simple pattern matching, which was flawed by the fact that various people said the same things differently, and even one particular individual could enunciate or speak the same phrase in a multitude of ways.
We saw modest improvements over the years, but nothing truly significant until two years ago, when Microsoft engineers working with the University of Toronto implemented a technique called Deep Neural Networks that more dramatically mimicked the speech recognition and translation process the human brain goes through.