基于错误驱动的汉语词性标注研究终极版.doc

目 录 摘 要 I ABSTRACT II 第1章 绪论 1 1.1论文背景 1 1.2论文意义 1 1.2.1国外主要语料库简介 2 1.2.2国内主要语料库简介 3 1.2.3词性标注方法研究现状 4 1)基于规则的方法 4 2)基于统计的方法 5 3)规则与统计相结合的方法 7 1.2.4论文各章节安排 8 第2章 基于极大熵的词性标注 9 2.1最大熵原理 9 2.2最大熵建模 9 2.2.1样本特征描述 11 2.2.2数据训练 11 2.3特征选择 12 2.3.1问题的提出 12 2.3.2特征选择实验 13 第3章 汉语兼类词和词性标注错误研究 16 3.1汉语的词性兼类问题 16 3.2兼类词标注中的错误分析和使用方法 18 3.2.2训练模型进行兼类词标注中的错误输出分析 19 3.2.3兼类词的概率特征函数 25 第4章 实验方案和分析 27 4.1实验设计 27 4.1.1实验语料描述 27 4.1.2实验工具 28 4.1.3实验方案 30 1.概率特征模型 30 2.外部知识模型 31 3.子分类模型 32 4.2实验结果及分析 33 4.3下一步的工作 34 总结与展望 35 5.1论文工作总结 35 5.2展望 36 参考文献 37 致 谢 39 附 录 40 1.主程序源代码： 40 2.相关的结果截图 48 ABSTRACT In the recent years, with the rapid development and enlargement of the Chinese Corpus and annotation technologies, a large scale of language block based at nationality language and different types of tagging feature musters appeared. The researches of the deep-processing methods and relevant algorithms are in need for the advancement of Nature Language Processing. Just like the other language, the first step toapproach Chinese corpus knowledge is part-of-speech tagging.Annotation systems which can run on the computers supports the computational linguistics which have attracted wide concerns from therelated fields such as Artificial Intellegence. There are several annotating solutions which mostly base statistical algorithm and rules which was writted manually. Such as the Maxent Entropy model and Hidden Markov ModelRule, which integrated different rules-templates can provide tagging tools for Natual Laguage.But the tagging results are not good enough to apply to the deep level annotation. According to the statiscal examples which are collected from multiwo:rds annotation error results in system, this essay will introduce three parts of appending models for Part-of-Speech task based at Maxent Entropy model. A new error-based method composed of events with feature probability which was calculated in advanced was held out to choose features templates for multi-word. KEYWORD: error-dr