Research On Key Issues Of Dependency Parsing

Dependency parsing is an approach to syntactic analysis inspired by dependen-cy grammar. In recent years, several domains of Natural Language Processing havebenefted from dependency representations, such as synonym generation, relation ex-traction, and, machine translation. A primary reason for using dependency structuresinstead of more informative constituent structures is that they are usually easier to beunderstood and is more amenable to annotators who have good knowledge of the tar-get domain but lack of deep linguistic knowledge while still containing much usefulinformation needed in application.This paper presents generalized probabilistic models for high-order projective de-pendency parsing and an algorithmic framework for learning these statistical modelsinvolving dependency trees. Partition functions and marginals for high-order depen-dency trees can be computed efciently, by adapting our algorithms which extend theinside-outside algorithm to higher-order cases. Meanwhile, We present and imple-ment a fourth-order projective dependency parsing algorithm that efectively utilizesboth “grand-sibling”style and “tri-sibling”style interactions of third-order and“grand-tri-sibling”style interactions of forth-order factored parts for performance en-hancement. This algorithm requires O(n~5) time and O(n~4) space. We implement andevaluate the parser on two languages—English and Chinese, both achieving state-of-the-art accuracy.