三段論法における人間の思考過程のモデル化

抄録

It is well recognized that the present computer interfaces are not flexible due to the large gaps between humans and computers. The authors believe that having models of human thought processes on computers helps greatly lessen these gaps and contributes much to the realization of flexible interfaces. The present paper presents a new approach for modelling human thought processes in categorical syllogisms. A first characteristic of this new modelling is that it aims at modelling thought processes of an individual user instead of the average user. This is necessitated by the fact that behaviors of users greatly differ from one to another. A second characteristic is that it aims at real time modelling based on the interaction between a user and a computer. These two characteristics are vitally important especially when the resulting models are used in computer interfaces. The basic approach of this modelling is that the relationship between user's behavior and the candidates of behavior patterns in the form of rules is analyzed based on the interaction record. This analysis clarifies the behavior rules which the user seems to follow. This modelling, however, has the following difficulties. Firstly, there exist, in general, more than one behavior rules which match user's behavior. Secondly, often multiple behavior rules are simultaneously existing. Thirdly, data gathered from users are often contaminated by noise due to various reasons. The authors propose an algorithm which can locate the true cause of errors even in these difficult situations. We have selected categorical syllogisms as an example domain, and have developed an experimental system for modelling human thought processes. Interacting with a user by giving questions and receiving answers, this system models the behaviors of users. Because the system interacts with users, it is required that the number of problems be as small as possible. An efficient algorithm for selecting doubtful error causes is also proposed. Experimental results using 9 subjects demonstrate that the experimental system with the algorithm for locating the true cause of errors works well. This system also contribute to finding new behavior patterns of users in categorical syllogisms and to clarifying the existence of large differences of behavior patterns among users.