基于句法依存和CRFs的韵律短语识别.PDFVIP

文章编号：170 基于句法依存和CRFs 的韵律短语识别* 1,2 1 钱揖丽 ，张二萌 （1. 山西大学 计算机与信息技术学院，山西 太原 030006 ； 2. 山西大学 计算机智能与中文信息处理教育部重点实验室，山西 太原 030006 ） 摘要：正确划分句子的韵律结构对于提高合成语音的质量具有重要的意义。而特征的选择是韵律结构预测 的关键因素之一。在中文信息处理中，文本特征可以分为浅层文本特征与深层文本特征，浅层特征包括词、 词性、词长等，深层特征包括句法信息、语义信息等。该文在挖掘剖析句法结构、依存句法结构同韵律结 构之间关系的基础上，从文本中获取相关浅层和深层文本特征，并采用条件随机场模型实现韵律短语预测。 该文首先以浅层文本特征进行韵律短语识别，然后在此基础上加入句法依存深层文本特征进行模型构建。 实验结果证明，加入句法依存特征后，韵律短语预测精确度提高了 13.3%，召回率提高了14.7%，F 值提高 了14.1%。 关键词：韵律短语；句法依存；文本特征；条件随机场 中图分类号：TP391 文献标识码：A Identification of Prosodic Phrase Based on Syntax Dependency and CRFs QIAN Yili1,2 , ZHANG Ermeng1 (1. School of Computer Information Technology, Shanxi University, Taiyuan, Shanxi 030006, China; 2. Key Laboratory of Computational Intelligence and Chinese Information Processing of Ministry of Education, Shanxi University, Taiyuan, Shanxi 03006, China) Abstract: Correctly dividing the prosodic structure of sentences is of great significance to improve the quality of synthesized speech. And feature selection is one of the key factors of prosodic structure prediction. In Chinese information processing, text features can be divided into shallow text features and deep text features, shallow features including words, parts of speech, word length and so on, deep features including syntactic information, semantic information and so on. Based on the analysis of the relationship between syntactic dependency structure and prosodic structure, this paper obtains the relevant shallow and deep text features from the text, and uses conditional random field model to realize prosodic phrase prediction. In this paper, we first use the feature of shallow text to recognize the prosodic phrase, and then add the syntactic dependency deep text feature to construct the model. The experime