句法分析与消解PCFG改进

摘要

概率型的句法分析模型具有一定的消岐能力，PCFG的最大特点上下文无关，对于开放领域的自然语言分析来说具有较长的鲁棒性。但上下文无关假设不可能很好地表示语言的特点，难t,X解决需要上下文信息才可以消解的句法分析岐义．而且它只考虑了词类信息而没有考虑词汇或语义信息，所以对语言的描述也是粗粒度的。本文结合知网定义句子成分结构语义关联度，对PCFG分析结果改进，句法岐义有效改善。%Probabilistic model of syntactic parsing has a capacity of disambiguation. The most important feature of PCFG is context-free and it has a long robustness to the parsing of natural language in the public domain. But the hypothesis of context- free may not represent language characteristics and hardly settle the syntactic parsing ambiguity which needs context information. And it considers parts of speech rather than vocabulary or semanteme, so its description to language is coarse- grained. The paper improves the PCFG parsing results with HowNet definition of semantic structure of sentence elements correlation, and improves syntactic ambiguity effectively.