Constraint Grammar as a SAT problem

Paper i proceeding, 2015

We represent Constraint Grammar (CG) as a Boolean satisfiability (SAT) problem. Encoding CG in logic brings some new features to the grammars. The rules are interpreted in a more declarative way, which makes it possible to abstract away from details such as cautious context and ordering. A rule is allowed to affect its context words, which makes the number of the rules in a grammar potentially smaller. Ordering can be preserved or discarded; in the latter case, we solve eventual rule conflicts by finding a solution that discards the least number of rule applications. We test our implementation by parsing texts in the order of 10,000s–100,000s words, using grammars with hundreds of rules.