Bouchaffra and Tan present a novel technique that merges syntactical and statistical information into a single probabilistic approach that relates visible observation sequences through their contribution to the same local structure built from an equivalence relation. This structural hidden Markov model (SHMM) approach decomposes the whole pattern into meaningful entities, and assigns them syntactical information. The approach involves probability evaluation, statistical decoding, local structure decoding, and parameter estimation, and supports capturing the whole pattern by revealing its local structures one at a time.

SHMM extends traditional hidden Markov models by incorporating the structural dimension within the statistical design. The authors illustrate this with an automotive application. However, more data is required in order to measure the global contribution of SHMM.