Dual decomposition for parsing with non-projective head automata

Abstract

This paper introduces algorithms for non-projective parsing based on dual decomposition. We focus on parsing algorithms for non-projective head automata, a generalization of head-automata models to non-projective structures. The dual decomposition algorithms are simple and efficient, relying on standard dynamic programming and minimum spanning tree algorithms. They provably solve an LP relaxation of the non-projective parsing problem. Empirically the LP relaxation is very often tight: for many languages, exact solutions are achieved on over 98% of test sentences. The accuracy of our models is higher than previous work on a broad range of datasets.