Theory Of Hidden Markov Models And Its Applications

A Hidden Markov process is a double stochastic process:an urderlying process which is hidden from observation and an observable proccess which is determined by the underlying process.In the last tenyears, most of the work performed on HMM has been done on first order systems (i.e, first-order HMM:HMM1) All these studies assumed that the state-transition probability at timet+1 depends on thestate of the Markov chain at timet, and the probability of the observations'output at timet depends only on the stat of the Markov chain at time tIn this paper,we improve HMMl's conditions of the Markov assumption on state transition and observations'output. we first describe the structure of modified Hidden Markov Models(Second-order HMM:HMM2) on condition that observation noise is not independent of the Markov chain.Second, we research the Baum-Welch algorithm of modified HMM and derive the update parametric estimation equations for modified HMM based on traditional HMM.Finally,we present application of modified HMM in computational lingustics.