抄録
Smartphone, Ultrabook and Tablet PC put close the distance between human and computer device. But there are many users who can't use that Personal Digital Assistant efficiently. They need to improve its operability. Intelligent agent system has been attracting attention to solve such a problem. In order to develop intelligent agent system, it is necessary to improve the recognition rate of a text conversation for these users. In this paper, we propose a mathematical model for estimating the sentence patterns from the word order of conversation text (in Japanese). We use Hidden Markov Model (HMM) to affiliated labeling for estimating the sentence patterns from the word order of Japanese conversation text. A sentence is composed of a sequence of attributes and basic components as subject, verb, and adjective. Thus we can infer stochastically word order of basic components from pattern of postpositional in the text. We are trying to affiliate labeling Japanese conversation text using HMM. In this study, Two models, the model containing both the sentence which does not contain a subject, and the sentence containing a subject, and the model of only the sentence which does not contain a subject, were prepared, and comparative experiments were conducted.
