This paper proposes a method for generating an adaptive knowledge base (AKB) involving two knowledge representations: rule
and case. Combining rules and cases makes it possible to solve problems accurately and quickly, and to acquire new cases from
problem-solving results. In general case-based problem-solving methods, the similarity metric must be defined for each problem
domain. In previous work using rules and cases, a threshold of negative case applications had to be adjusted. The proposed
AKB does not require manual adjustment of the threshold and the similarity metric.
This paper also proposes a Japanese-to-Braille translation system which uses the proposed AKB. Experimental results have showed
that the case acquisition and similarity weight adjustment can reduce errors, and that the threshold adjustment significantly
reduces segmentation errors.
Key words Knowledge base - Case-based reasoning - Threshold adjustment - Weighted similarity measure - Japanese-to-Braille translation
This work was presented in part at the 12th International Symposium on Artificial Life and Robotics, Oita, Japan, January
25–27, 2007