1

I have this rule in a grammar :

VP->Verb NP NP   [0.05]
VP->Verb NP PP   [0.1]

I want to convert this PCFG(Probabilistic Context Free grammar ) to CNF(Chomsky Normal Form)

To do that I know we can split the rule into two non-terminals 

VP->@V_N NP
VP->@V_N PP
@V_N->Verb NP

Which probability to set for each rule?

Thanks

Nada Ghanem
  • 451
  • 6
  • 16
  • The conversion will depend on what you would want the CNF to do and also on the other rules. I assume that you want the resulting CNF probabilities to sum to one. If not, @vsevolod's answer resolves the problem. But he's taking a conditional approach to the probability assignment, i.e. `prob(VP->@V_N NP) = prob(@V_N) * prob(@V_N NP | @V_N)` so the sum of probabilities don't sum to 1 – alvas Dec 26 '14 at 06:50
  • Well ,I'll use it to fill CKY table . – Nada Ghanem Dec 26 '14 at 15:52

1 Answers1

3
VP->@V_N NP     [0.05]
VP->@V_N PP     [0.1]
@V_N->Verb NP   [1.0]

Because you need to get the same probability for the result by applying both the original and the converted set of rules.

Vsevolod Dyomkin
  • 9,343
  • 2
  • 31
  • 36
  • Thanks Vsevolod -shall we always set this sub prob to 1.0? ..I'm trying to solve CKY problem ,and I got across this case. – Nada Ghanem Dec 26 '14 at 15:53