特定人主题受限连续汉语人机对话系统的研究

Research on Speaker-dependent Topic-constrained Continuous Chinese Human-machine Dialogue system

本文描述了一个以火车售票和信息查询事项为对话主题的专人使用的连续汉语人机对话系统原型,着重讨论语言理解和应答文生成,这部分是连接语音识别和语音合成的纽带,起着承上启下的作用。语言理解和应答文生成采用上下文无关文法(CFG),结合黑板以及局部上下文相关(CSG),对连续汉语语音识别输出的结果进行理解,生成人机对话系统中的应答文,作为语音合成的输入。利用堆栈技术处理不完全信息和信息证实,使得该系统具有自动提问获取知识的能力和对关键信息进行确认的能力,从而使人机对话能连续进行。语义处理提取动词等关键信息,使该系统具有容错性。我们对时间的描述采用模糊数学方法处理,以适应现实对话中不精确的时间表示。

We describe a speaker-dependent topic-constrained continuous Chinese speech-understanding system which enables its user to obtain timetable and fare information through a spoken dialogue. A method is described which understands the results of speech recognition and generates the response to be used by the synthesiser using context-free grammer (CFG), context-senstive gramar (CSG) and blackboard. Stack technique is used to process insufficient information and information confirmation. Sematics processing obtains verb and other key information, so that this system has fault-tolerant ability, We describe the time using fuzzy mathematics method.