Learning Hidden Markov Models with Distributed State Representations for Domain Adaptation