基于语法树高度的汉语韵律短语预测.pdfVIP

ComputerEngineeringandApplications计算机工程与应用 2010，46(36) 139 @数据库、信号与信息处理◎ 基于语法树高度的汉语韵律短语预测 杨鸿武 ，王晓丽 ，陈 龙 ，裴 东’，郭威彤 ，蔡莲红 YANG Hong—wu ，WANG Xiao—li’，CHEN Long ，PEIDong’，GUO Wei—tong’，CAILian—hong 1．西北师范大学 物理与电子工程学院，兰州 730070 2．清华大学 深圳研究生院，广东 深圳 518005 3．清华大学 计算机系，北京 100084 1．CollegeofPh：ysics and ElectronicEngineering，NorthwestNormalUniversity，Lanzhou 730070，China 2．GraduateSchoolatShenzhen，Tsinghua University，Shenzhen，Guangdong 518005，China 3．DepartmentofComputerScienceandTechnology，TsinghuaUniversity，Bering100084，China YANG Hong—wn，WANG Xiao-li，CHEN Long。eta1．Predicting Chinese prosodic phrasewith height ofsyntax tree．Com— puterEngineeringandApplications，2010，46(36)：139-143． Abstract：Predicting preciseprosodic structure is one ofthe mostimportantaspectsfor improving thenaturalness ofsynthe— sized speech in textto speech synthesis．10000 sentences with part—of-speech tags are used for manually labeling prosodic word and prosodicphrase undera linguisticexpert’Sguidance．A new feature based on the heightofsyntax tree isproposed inthepaperfordescribingthedegreeofclosenessbetweentheadjacentlexiconwordsaccordingtothehierarchalsyntax structure．Analysis on the relationshipsbetween heightofsyntax tree and prosodicphrase boundary showsthatthe heightof syntax tree，part。of-speech and length oflexicon word are thethreemostimportantfeaturesforprosodic phrase boundary pre- diction．Thereforethese three featuresare employed to train aTBL based modelwith 400 training sentences．Experimentsdem— onstratethatthe approach achieves75．2％ ofprecision，77．1％ ofrecalland 76．1％ ofF-Score． Key words：prosodic structure prediction；syntax structure；prosodic phrase；heightof syntax tree；Transformation-based Er— ror-drivenLearning(TBL) 摘 要：在文语转换 系统中，从文本中预测出准确的韵律结构对于提 高合成语音的 自然度具有重要的作用。利用 10000句标注 了词性标记的文本语料，在语言学专家的指导下，人工标注了语料的韵律词和韵律短语。选择 了标注结果一致性最高的500句语 句，标注了语法层级结构，并利用语法树高度描述语法词之间连接的紧密程度。通过分析韵律短语边界与语法结构的关系，发现 韵律短语边界受语