Difference: GrmNGramForum (110 vs. 111)

Revision 1112016-07-07 - DanielRenshaw

Line: 1 to 1

OpenGrm NGram Forum

Line: 83 to 83
  Then use fstcompile to produce your filtering automaton. If you compose this with your LM FST, then use fstconnect (to prune out any paths that don't lead to a final state) then you'll have an LM restricted to that vocabulary. It will not, however, renormalize the model, it will simply give the original probabilities. If you need to renormalize, then I would suggest using the above method of printing the counts, stripping out what you don't want and recompiling. Hope that helps!

DanielRenshaw - 2016-07-07 - 04:37

Thanks Brian. That's useful.
Log In
This site is powered by the TWiki collaboration platform Powered by PerlCopyright © 2008-2019 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback