The library provides efficient implementation of the first-order,
linear-chain conditional random fields (CRFs) with position-wise
constraints imposed over label values.
It is strongly related to the simpler
http://hackage.haskell.org/package/crf-chain1
library where constraints are not taken into account and all
features which are not included in the CRF model are considered to have
probability of 0. Here, on the other hand, such features do not influence
the overall probability of the (sentence, labels) pair - they are
assigned the default potential of 0.
Efficient algorithm for determining marginal probabilities of individual
labels is provided.
The tagging is performed with respect to marginal probabilities.