Friday, August 5, 2011

Three Questions to Ask about Hidden Markov Models



Continuing through Chapter 9 of Foundations of Statistical Natural Language Processing (1999) by Christopher D. Manning and Hinrich Schütze.

Manning and Schütze describe three fundamental questions to ask about Hidden Markov Models (HMMs):
  1. Given a certain HMM (knowing its state transition probabilities, symbol emission probabilities, and initial state probabilities), how do we efficiently compute how likely a certain observation is?
  2. Given an observation sequence and a certain HMM, how do we choose a state sequence that best explains the observations?
  3. Given an observation sequence and a space of possible HMMs found by varying the model parameters (e.g., state transition probabilities, symbol emission probabilities, and initial state probabilities), how do we find the HMM that best explains the observed data?
Future posts will address how to approach these questions.

* * * * *

In other news, I've intermittently been playing with Python and the Natural Language Processing Toolkit.

No comments: