HW 12: N-GRAM LANGUAGE MODELING
- Download the
Switchboard transcriptions
and replicate the plots of N-gram frequencies shown in
page 2 of lecture 32.
- Consider all possible N-grams that could have occurred assuming
any word in the unigram distribution could follow any other word
in this distribution. Add one to every count (a simple form of
smoothing). Compute the entropy of each distribution and compare
this to the total numbers of N-grams in the distribution.
Interpret these results.