基于语用的自然语言处理研究与应用初探  

Pragmatic Information Based NLP Research and Application

作　　者：李蕾[1] 周延泉[1] 钟义信[1] 

机构地区：[1]北京邮电大学智能科学技术研究中心,北京100876

出　　处：《智能系统学报》2006年第2期1-6,共6页CAAI Transactions on Intelligent Systems

基　　金：国家自然科学基金资助项目(60575034);国家"863"资助项目(2004AA117010;2005AA117010).

摘　　要：首先分析了语用信息的必要性和重要性,认为只有融入语用研究的自然语言处理技术才能显示“以人为本"和智能化的特色,只有语用、语义和语法信息的研究都成熟了,才能使计算机真正获得自然语言所表达的信息,达到与人类交流对话的水平.接着介绍了语用学的产生、发展和运用状况,剖析了存在的主要问题,提出了基于语用的自然语言处理.然后结合典型应用背景——奥运多语言信息服务示范终端“CityGuide"语音识别后文本的检错纠错需求,探索并尝试了一种基于语用信息的自然语言处理检错纠错方法,并通过真实语料的测试来检验效果.结果表明,当前算法可以使中文语音识别正确率提高29%.Pragmatic information is looked on as the next focus for natural language processing （NLP） research. The necessity and importance of pragmatic information are analyzed firstly. It is pointed out that NLP could be charaterized as humanity and intelligence only after pragmatic information are integrated into it. And only when syntactic, semantic and pragmatic information are all fully studied could computers understand the information expressed in human natural language. Thus computers could really communicate with human. Then details of pragmatics research are introduced, including its origin, growing history and applications. Problems are also analyzed for its current status. As a result, pragmatic information based NLP is put forward. Then a grope research of this, i.e. the sentence error detection and correction in the application domain of ＂CityGuide＂ Speech Recognition （SR） interface is reported. The ＂CityGuide＂ is a demo terminal for the National 863 project of ＂Olympics Oriented Multilingual Information Service＂. A method containing pragmatic information analysis is studied and tested using realistic corpus. Results show that the precision of Chinese SR can be improved by 29%.

关 键 词：自然语言处理 语用信息 语音识别检错纠错 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]