“The language model describes in a statistical fashion the likelihood of specific word sequences occurring at a given time. In other words, what is the probability of the word "America" following the words "United States of"? These statistics are determined by analyzing a large collection of written and spoken text. Because of their statistical nature, they are flexible - any word sequence has some, albeit small, probability of occurring - but lack the rigor required for high recognition accuracy.”
This I have seen firsthand. Longhorn seems to have some difficulties understanding me when I say “Get this program on the web.” However, if I were to say “This program is a simulation of a spider's web,” it has a much easier time. The human mind processes speech to thought using such contextual clues as well.
Previous Topic
