Improving Dialogue Smoothing with A-priori State Pruning