The Study And Realization Of Chinese Parsing With Semantic And Sentence Pattern Information

Parsing is one of the most fundamental research works in Natural Language Processing (NLP). It can reveal the inner structures of natural language. And high-accuracy parsing can benefit upper level applications.In this paper, we focus on probabilistic context-free grammar (PCFG) model and propose a method to incorporate Chinese semantic and sentence pattern information in PCFG to solve the problem that PCFG model lacks semantic information and the global restrictions among grammar rules.First, we try to incorporate semantic information into unlexical parsing method. On one hand, semantic information can alleviate the disambiguious in parsing and the F-score in standard Penn Chinese Tree Bank (CTB) increases 1.37% and achieves 81.63%2. On the other hand, inferring syntax and semantic at the same time can also bring some semantic information.Second, we try to incorporate sentence pattern information into PCFG model. The sentence pattern information is acquired from the co-occurrent information between trees and grammar rules. And we use this information in parsing for disambiguation which corrects some mistakes of tags which represent sentence pattern in Tsinghua Tree Bank. And the F-score acquired in Tsinghua Tree bank increases 0.17% and reaches 86.57%2...