【作者】 刘彤；

【导师】 黄德根；

【作者基本信息】 大连理工大学 ， 计算机科学与技术， 2018， 硕士

【摘要】 介词短语在汉语中使用的频率很高,其复杂多变的结构造成了识别的困难,而其识别准确率的高低也会影响到一系列句法分析任务的结果。在自然语言处理的研究中,提高介词短语的识别效果可以降低句法分析的复杂度,提高文本分类的分类效果,并且对机器翻译的性能也有很大的提升。本文对于介词短语的语法特点进行分析研究,结合近年来研究现状和难点,提出了一种基于细粒度短语的多模型融合介词短语识别方法,针对嵌套并列等结构复杂的介词短语进行了改进。不仅能够提高嵌套型复杂介词短语的识别精度,也从整体上对介词短语的识别效果进行提升。首先,利用细粒度短语识别模型识别出语料中的短语信息并进行融合,精简句子结构,降低介词短语内部复杂性;其次,用CRF模型识别具有嵌套结构的内层介词短语,即若嵌套存在则识别内层介词短语,否则直接识别该介词短语;最后,将初始语料中识别出来的内层介词短语进行分词融合并修改其特征信息,重新训练外层介词短语识别模型进行识别。在内、外层介词短语识别后,都利用双重错误校正系统对识别的介词短语进行校正。细粒度短语融合的方法,能够在描述语句信息的同时简化结构,使介词短语跨度缩小。分层嵌套多模型识别方法,将同一层次的介词短语同时进行识别,并使用不同的模型识别不同层介词短语,更适合存在嵌套、并列的介词短语。双重错误校正系统利用了规则的方法,将统计与规则相结合,对实验效果进一步提升。在2000年《人民日报》新闻语料中进行五倍交叉实验,结果显示,本文方法识别的介词短语正确率、召回率、F值分别为94.33%,94.28%,94.30%,比基于简单名词短语的介词短语识别方法(baseline)分别提高了1.31、1.33、1.32个百分点,有效提高了介词短语识别的性能。更多还原

【Abstract】 The use of prepositional phrase is very frequent in Chinese,it’s complex and changeable structure caused the difficulty of recognition.The accuracy of its recognition will affect the result of a series of parsing tasks.In Natural Language Processing research,improving the recognition effect of prepositional phrases can reduce the complexity of syntactic analysis,improve the classification effect of text classification,and improve the performance of Machine Translation greatly.In this paper,by analyzing the grammatical features of prepositional phrases and referring to the research status and difficulties in recent years,a multi model fusion prepositional phrase recognition method based on fine-grained phrases is proposed,which is mainly improved for complex prepositional phrases such as nested juxtaposition.It not only identifies parallel prepositional phrases,but also improves the recognition accuracy of embedded prepositional phrases.First,a fine-grained phrase recognition model is used to identify and merge the phrases in the corpus in order to reduce internal complexity of prepositional phrases;Then,the CRF model is used to identify the inner layer of the nested prepositions phrases,i.e.if the preposition phrases is nested,recognize the inner layer,otherwise,recognize the whole preposition phrase;Finally,merge the recognized inner prepositional phrases in the corpus and modify the feature information in order to train a new model for outer prepositional phrase recognition.In addition,after the recognition of both inner and outer prepositional phrases,a double error correction system is used to correct the recognized phrases.The method of fusing fine-grained phrase to simplify sentence structure while keeping sentence information,and shorten the span of prepositional phrase.The hierarchical nested multi model prepositional phrase recognition method identifies the same level prepositional phrases at the same time,and uses different models to identify different layer’s prepositional phrases,which is more suitable for the existence of nested and parallel prepositional phrases.The double error correction system makes use of the rule method to combine statistics and rules to further improve the experimental results.Five-fold experiments are conducted on the corpus of People’s Daily of 2000 including 7028 prepositional phrases,and the results achieve 94.33% in precision,94.28% in recall,and 94.30% in F-measure,which are improved by 1.31%,1.33%,1.32% respectively than the simple noun phrases based prepositional phrase identification method(baseline).更多还原