An Approach Based on Integrated Learning to Generating Stories from Stories

Abstract

Unlike the Explanation-Based Learning (EBL) methods, the Similarity-Based Learning (SBL) methods cannot be applied in a complex domain. However, the SBL methods can learn truly new things, whereas the EBL methods cannot. The integration of EBL and SBL can help overcome their shortcomings, retaining the advantages. We propose an approach based on a combination of these two techniques to generating (new) stories from given stories. One important point in our approach is that new representations of the examples are created which reveal paradigmatic similarities in the input stories. Another point is that a notion of substitutability is provided which allows microplots to be shifted through different stories. We apply these ideas to the domain of folktales and present an example from a computer program.