Prediction and Verification of Hypotheses in a Speech Recognition System