Learning Constraint Grammars
Constraint Grammar style part-of-speech disambiguation rules can be learned automatically from disambiguated parallel corpora.
In statistical approach Constraint Grammar style rules are learned by calculating n-gram probabilities of word and part-of-speech tag groups. Current work on implementing such a system is at nuboro's Github repository, and it is based on the paper Inducing Constraint Grammars.
A subfield of Machine Learning, called Inductive Logic Programming, has been used to learn Constraint Grammar style disambiguation rules. See for example the branch mil-pos-tagger