If I generate a language model with SRILM's ngram-count and then use ngram -unk -ppl text -lm model to get log probabilities and perplexity values, are the perplexities normalized for sentence length?
Add a comment
|
1 Answer
Hope it is not too late.
Both ppl and ppl1 are normalized according http://www.speech.sri.com/projects/srilm/manpages/srilm-faq.7.html , which is called entropy rate.
