汉语依存句法分析关键技术分析.docx

析联合模型中句法特征占据主导地位,导致词性特征无法贡献其消歧作用。对此,本文提出一种面向词性标注和依存句法分析联合模型的训练算法。算法分别不同的步长对词性特征和句法特征的权重进行更新。和传统的平均感知器和被动进取训练算法相比,分离被动进取训练算法可以很自然的增大词性特征的权重,从而更好的平衡联合模型中词性特征和句法特征的消歧作用。实验发现,我们的基于分离被动进取训练算法的联合模型在汉语和英语数据上都可以取得最好的词性和句法准确率。
4. 提出了基于准同步文法的多树库融合方法。汉语存在多个异构树库,而利用多个树库以提高依存句法分析准确率是一个非常有吸引力的课题。本文提出一种基于准同步文法的多树库融合方法,充分利用标注规范不同的多个单语树库,以提高句法分析准确率。我们设计了丰富的转换模式来刻画不同标注规范间的对应规律,然后基于这些转化模式形成准同步文法特征,从而增强基准依存句法分析模型。准同步文法特征用来指导句法模型做出更好的决策,并且可以很自然的融入到基于图的句法分析解码算法中。实验结果表明,我们的方法可以充分利用源树库的知识。从而提高句法模型在目标树库上的准确率。
总之,本文针对汉语特点,深入研究了基于标点的快速高阶依存句法分析方法、词性标注和依存句法分析的联合模型和多树库融合问题,大大提高了汉语依存句法分析处理实际文本数据的效率和准确率。本研究取得了一些初步的成果。我们期待这些研究成果可以进一步推动自然语言处理领域和其他上层应用如机器翻译、信息抽取的发展。
关键词: 依存句法分析;柱搜索;联合模型;分离被动进取训练算法;多树库融合
Abstract
Dependency parsing aims to analyze the syntactic structure of a given sentence, and converts the word sequence into a tree. A dependency consists of two words, and the one word modifies the other one. The label of the dependency represents the specific relation between the two words, such as subject, object, adverbial modifier, and so on. Among different syntax formulizations, the dependency grammar has gained more and more in- terest in munity due to its characteristics: (1) representation simplicity, (2) easy to annotate, (3) easy to make use of. The international conference putational Natural Language Learning (CoNLL) anized shared tasks on multilingual depen- dency parsing from 2006 to 2009, which largely promotes the research on dependency parsing. Meanwhile, dependency parsing has been more extensively applied to machine translation, question answering, text mining, information retrieval, and so on.
Research on dependency parsing has two important goals. One goal is to improve the parsing accuracy, while the other is to improve the parsing efficiency. Accurate parsing results can provide reliable syntactic structures for higher-level applications. Along with the quick growth of the web data, higher-level applications need to process a large amount of information in some limited time.