Voice Recognition

by Stephen J. Andriole

It's been about three decades since the first promises about voice recognition were made by researchers at the Defense Advanced Research Projects Agency (DARPA -- the same Defense Department agency that created the underlying technology for the Internet). The idea was simple enough: develop hardware and software that could understand what people said -- as well as what they meant by the words they used -- and then, through "natural language understanding," respond coherently in the same (or even another) language, as appropriate. Full voice input and output was -- and remains -- the goal, which includes "semantic understanding," or the ability to understand the meaning and context of language, not just the structure of sentences ("syntactic understanding"). Impressive progress was made early on, but real-time, knowledge-based language understanding proved difficult; so hard, in fact, that the most impressive progress was made on the syntactic side (with limited vocabularies with less-than-perfect speech input recognition devices).