-
Consider the problem of constructing the "best" sentence from
the output of a DUR recognizer having a four word vocabulary -
cat,fat,sat,that
-
Let them be referred mathematically as w(1) = cat and so on
-
The probability,P(in,n), for different words occurring
at different time instants are given
-
Also, we have information about the probabilities for bigram
statistics,Q(i/j) -- the probability that word i is spoken at
time n+1 , given that word j was spoken at time n
-
One may define a correctness measure C, that quantifies the
possibility that this sentence was spoken
-
The optimality problem is then to find that particular sentence
which maximizes the measure C