Semi-hidden Markov models for generation and analysis of sequences

In this work a new kind of stochastic model is presented, the semi-hidden Markov model (SHMM). The proposed model is related to the hidden Markov model (HMM), and it is called semi-hidden because generated sequences need less information than HMM sequences to infer the succession of states run by the source.The main feature of SHMM is that they work with statistical memory, i.e. the symbol’s emission probability distribution on the current state of the emitting source depends on a number of symbols already emitted in the previous state. The proposed model is useful for the generation and analysis of processes and symbolic sequences containing runs.