Modeling Linguistic Features in Speech Recognition 1