Generative Modeling of Coordination by Factoring Parallelism and Selectional Preferences

We present a unified generative model of coordination that considers parallelism of conjuncts and selectional preferences. Parallelism of conjuncts, which frequently characterizes coordinate structures, is modeled as a synchronized generation process in the generative parser. Selectional preferences learned from a large web corpus provide an important clue for resolving the ambiguities of coordinate structures. Our experiments of Japanese dependency parsing indicate the effectiveness of our approach, particularly in the domains of newspapers and patents.