In the analysis of Ergodic HMMs, a high similarity was observed between HMM high output probability words and POSs. Therefore, an Ergodic HMM is able to obtain an SNG automatically.
Moreover, as we increase the number of states of an Ergodic HMM like to 4, 8, and 16 states, the word grouping become closer to the POSs. Therefore, we can obtain the grammar in more detail with increasing number of state.
By the way, the current POS is determined linguistically. However, by using an Ergodic HMM, we may be able to automatically obtain a new system of POS under the maximum likelihood.