First get the following files from the web:

http://www.cs.toronto.edu/~hinton/csc321/matlab/hmmtrainenglish.m
http://www.cs.toronto.edu/~hinton/csc321/matlab/hmmtrainfrench.m
http://www.cs.toronto.edu/~hinton/csc321/matlab/hmmtestenglish.m
http://www.cs.toronto.edu/~hinton/csc321/matlab/hmmtestfrench.m
http://www.cs.toronto.edu/~hinton/csc321/matlab/hmmcost.m
http://www.cs.toronto.edu/~hinton/csc321/matlab/hmmgen2.m
http://www.cs.toronto.edu/~hinton/csc321/matlab/dhmm2.m
http://www.cs.toronto.edu/~hinton/csc321/matlab/rsum.m
http://www.cs.toronto.edu/~hinton/csc321/matlab/rdiv.m
http://www.cs.toronto.edu/~hinton/csc321/matlab/cdiv.m
http://www.cs.toronto.edu/~hinton/csc321/matlab/csum.m
http://www.cs.toronto.edu/~hinton/csc321/matlab/hinton.m
http://www.cs.toronto.edu/~hinton/csc321/matlab/blob.m

Train one HMM on the sequence of English letters that is in the script
hmmtrainenglish.m and then compute the log probability of both the
English and the French test sequences that are in the relevant scripts.

Then train another HMM on the sequence of French letters in the script
hmmtrainfrench.m and then compute the log probability of both the
English and the French test sequences that are in the relevant scripts.

Try training the HMM's for different numbers of epochs (i.e. cycles)
and also try using different numbers of hidden states. For each
experiment make a 2X2 table of the log probabilities of the two test
sequences under the two HMM's.

Write less than two pages showing that you understand why varying the
number of hidden states influences the 2X2 tables in the way it does.