Algorithms of POS Tagging HMM (generative) Maximum Entropy Markov Model (discriminative) Conditional Random Field (discriminative)
Three Fundamental Problems of HMM 1. Statistical Inference of the State Sequence Find P(S|O) by Viterbi Algorithm 2. Observation Determination Find P(O|S) by Forward Algorithm, Backward Algorithm 3. Model Estimation Given S,O estimate model parameters (Transition and Emission probabilities) by Baum-Welch Algorithm (based on Expectation Maximization)
HMM Examples HMM Classic Example NLP Example (Urn Problem) (POS Tag Example)
URN Problem ^ R G B $ - Observation Seq U3 U3 U3 U1 U1 U1 U0 U2 U2 U2 UF U3 U3 U3 Trellis or Stak Graph . . . . . .
POS Tag Problem J J J ^ People Jump High $ - Observation Seq N N N ^ V V V $ J J J Trellis or Stak Graph . . . . . .