How can I calculate the startProbabilitise in hidden markov model for POS tagging? is that means: (# repeats of one tag)/(# repeats of all tags)? or (number of sentences beginning with one tag)/(all sentences)? if not, what?
Asked
Active
Viewed 115 times
1 Answers
0
Considering a second order HMM, Maximum Likelihood Estimate gives:
P(SomeTag | <BOS>,<BOS>) = count(<BOS>,<BOS>,SomeTag) / count(<BOS>,<BOS>)
It corresponds to your second proposal:
(number of sentences beginning with one tag)/(all sentences)

Pierre
- 1,204
- 8
- 15