[SRILM User List] How does -hmm option work?

Andreas Stolcke stolcke at speech.sri.com
Mon Mar 22 09:56:35 PDT 2010

If you want to do word-level tagging you should look at the disambig 
tool.   the ngram -hmm option is for a kind of language model that 
uses a hidden state, modeled by an HMM.  


In message <94c3f5461003212309t124c51bai2bf41c7827bad9cc at mail.gmail.com>you wro
> Hi:
> I have a question about using "-hmm" option in "ngram". In short,
> my task is to label a sequence of sentences, in which each sentence is
> to have one label. In the manual for "ngram", it says "each states
> emits one or more words". However when i try to label a sentence using
> -hmm option, instead of giving one label for each sentence/line, it is
> giving each word a label, which means in the hmm model, every states
> only emits one word. Where did I do wrong? Or "-hmm" option is not
> designed for this kind of purpose? Following is some information of
> the file and command i used.
> HMM file:
> INITIAL . lm1 0.5 lm2 0.5
> s1 lm1 s1 0.3 s2 0.6 FINAL 0.1
> s2 lm2 s1 0.8 s2 0.1 FINAL 0.1
> **(lm1 and lm2 are language model file from sentences of different types)
> The command i used is "ngram -hmm -lm hmmfile -ppl test -debug 2"
> **(Also, is there anyway to output decoded hmm sequence without using
> -debug option?)
> Thanks
> Qu ZH
> _______________________________________________
> SRILM-User site list
> SRILM-User at speech.sri.com
> http://www.speech.sri.com/mailman/listinfo/srilm-user

More information about the SRILM-User mailing list