Lexical Information Acquisition Using Accent Knowedge

在自然语言处理中，处理未登录词是不可避免的。我们的研究目的就是在词法分析中解决这一问题，特别是利用口音信息。我们的方法使得从文本中提取未登录词并利用语音信息获取其词性信息和词形变化成为可能。在从书面句子中提取未登录词的过程中，汉字和汉字扮演着重要的角色。虽然在语音交互中没有这样的信息，但我们可以使用口音信息。我们设计了一种利用重音信息获取未登录词的机制。我们的系统流程是：1.对带有重音信息的句子进行词法分析；2.估计未登录词的词性和词形变化。我们的方法在意义上是一种新颖的方法，它从语音上获取未登录词的信息，对于理解语言学习过程具有重要的认知心理学意义。为了实现规则的自动获取，我们研究了基于GMDH神经网络的形态分析方法。研究了HMM词法分析器及其从未标注语料中获取概率信息的技术，以及词性或词性之间的强度提取。作为相关研究，我们对音乐中的语音学进行了研究。

In natural language processing, to handle unknown words is not avoided. The aim of our research is to solve this problem in the morphological analysis, especially using accent information. Our method make it possible to estimate extract unknown words from texts and acquire its part of speech information and inflection using the phonetic information. In an extraction of unknown words from written sentences, kanjis and kanas play an important role. Though in phonetic interaction, there are not such information, we can use accent information. We designed a mechanism which aim is to acquire unknown words using accent information.The flow of our system is :1.analyze sentences with accent information morphologically2.estimate unknown words' part of speech and inflection.Our method below is a novel method in the meaning that it acquires unknown words' information in honetics and is significant in cognitive psychology which tries to understand the process of language learning.In this research, firstly, we built the morphological analyzer for kana sentences using regulations of accent information. To acquiring regulations automatically, we studied on the morphological analysis using the GMDH neural network. Also we studied on the HMM morphological analyzer and technique acquiring probability information from non-tagged corpus for it and the extraction of strength between parts of speech or words. As a relative research, we studied on phonetics in music.