Because it's a statistical model of the words used in similar problem domains.