An Argument for the Correctness of M.E.
Maximum entropy is a very
exible method of statistical modeling which turns
on the notion of futures", histories", and features". Futures are dened as the
possible outputs of the model. Like in the decision tree method, we are seeking to
choose from the set of 29 dierent named entity tags described in section 2.3. A
maximum entropy solution to this, or any other similar problem allows the computation
of p(fjh) for any f from the space of possible futures, F, for every h from
the space of possible histories, H. As with decision trees, a history" in maximum
entropy is all of the conditioning data which enables you to assign probabilities to
the space of futures. In the named entity problem, we could reformulate this in
terms of nding the probability of f associated with the token at index t in the
test corpus as: