ABSTRACT
The use of situational knowledge for automatic speech recognition in the ATC simulation domain is discussed. A cognitive model of the ATC controller is proposed that supports the recognition process with an estimation about the probability of clearances in the actual situation. The most probable clearances are written to a syntax file which is then handed to the speech recogniser. Experiments designed to analyse the power of this approach show that the use of dynamic syntax reduces the recognition error rate by about 50 percent, compared to the use of a static, non context-sensitive syntax.
