An Exploratory Application of Constraint Optimization in Mozart to Probabilistic Natural Language Processing

This paper describes an exploratory implementation in Mozart applying constraint optimization to basic subproblems of parsing and generation. Optimization is performed on the probability of a sentence using a dependency-style syntactic representation, which is computed using an adaptation of the English Penn Treebank as data. The same program solves both parsing and generation subproblems, providing the flexibility of a general architecture combined with practical efficiency. We show results on a sample sentence that is a classic in natural language processing.