Computational Analysis of Graphic Generation: Effects of surface and structural similarity

In studies of analogical reasoning, the distinction between surface and structural similarity has been repeatedly investigated. However, this distinction has not been investigated in a generative analogy where target representations are not provided in advance. This study uses computational methods to analyze how this distinction is involved in generative analogy. In the experiment, participants were given an example and asked to generate an original graphic by modifying the example. The procedure was repeated twice. The result indicated that the first graphics that the participants generated were similar in both surface and structural features to the presented example. However, the second graphics that they generated varied in the degree of surface and structural similarity. This result may indicate characteristics of generative analogy but is also compatible with the results of previous studies of analogical reasoning.