Study for Cognitive Model and Context Understanding System for Situation Dependency of Natural Language

本研究的目的是构建一个认知模型和一个自然语言理解系统，能够处理语言现象中的多义词和歧义。将认知语义学、认知心理学和自然语言处理三种方法相结合，对语言情景依赖进行消解。在认知语义学方法中，对动词、名词和形容词的多义性生成原理进行描述。对于多义词现象，明确了每一类词的功能作用。其次，通过设定理解对象和理解内容两个层面，在句子层面上解决语义歧义问题。认知心理学将词义歧义分为语言歧义和概念认知歧义两大类。后者是通过实验来确定的，实验表明这种歧义是在什么时候发生的。在自然语言处理过程中，构建了一个在线联想实验系统，并将其应用于一组基本名词的实验中，从中提取出一些有用的特征。近千个概念被系统化，并使用这些特征在测量的空间中表示。在测量空间上提取并表示事物概念(名词)、行为概念(动词)和表示其环境的情景概念。这些概念在空间上对词义的消解和自然语言理解系统的构建都是有用的。

The objective of this study is to construct a cognitive model and a natural language understanding system which can treat with polysemy and ambiguity included in linguistic phenomena. It can be obtained by integrating the three approaches, cognitive semantics, cognitive psychology and natural language processing to disambiguate the language situation dependency.In the cognitive semantics approach, a description method for the principle of generating polysemy in verbs, nouns and adjectives. Functional roles of each category are made clear for the polysemy phenomena. Secondly, setting two levels of understanding objects and contents, the polysemy problems at applying sense-makinq to utterances are solved at a sentence level.In the cognitive psychology, word meaning ambiguity is classified into two, linguistic one and concept cognitive one. The latter is identified by using experiments showing when such ambiguity happens. Next, a cognitive process model is constructed to disambiguate the cognitive ambiguity in the human language understanding process.In the natural language processing, an on-line association experiment system is constructed and by rising it, several useful features are extracted from the experiment applied to a fundamental set of nouns. Almost one thousand concepts are systematized and represented on a measured space using the features. Thing concepts(nouns), act ion concepts(verbs) and situation concepts which show their environment are extracted and represented on the measured space. Such concepts on the space are shown useful for word meaning disambiguation and for constructing natural language understanding system.