The HMMDecoder function finds the state sequence with the highest probability, given the learned model and observed sequences.
If a set of phrases or words was trained using HMMSupervised or HMMUnsupervised function, you can use HMMDecoder to decode the parts of speech in the set.