Finally,
with an increasing number of HMM-based classifiers for large
vocabulary cursive handwriting recognition, it is anticipated that
the discriminative training techniques reported in this paper may
be of general use to improve the performance of those systems