Train a first-order (i.e., the probability of a tag depends only on the previous tag) HMM part-of-speech tagger. Find the MAP estimate of the parameters of the model using add-1 smoothing.