Exploring higher order dependency parsers

Abstract

Parsing is one of the most important steps in understanding of natural languages. In this thesis, we focus on the dependency grammar formalism since, the core concepts of dependency grammar, namely the relational view of head and modifier asymmetry, has proven useful for diverse set of languages, especially accounting for the explanation of word order and relation between surface structure and meaning. Most of the recent efficient algorithms tor dependency parsing work by factoring the dependency trees. In most of these approaches, the parser loses much of the contextual information during the process of factorization. In this thesis we investigate how features (syntacto semantic) affect the higher order discriminative learning methods for dependency parsing.