This paper is trying to explore the issue of whether analogical mapping may influence perception and cause re-interpretation of the target stimulus. An attempt is made to put this rather complicated issue at experimental test bed. Two experiments are described which demonstrate that the process of analogical mapping may reorganize the perceptual input and stimulate a re-representation of the input image thus allowing the participants to discover meanings that were not initially available to them. The experiments are not decisive on when this re-representation happens.

Analogical Mapping and Perception: Can Mapping Cause a Re-Representation of the Target Stimulus? Boicho Kokinov (bkokinov@nbu.bg) Svetoslav Bliznashki (valsotevs@gmail.com) Svetlin Kosev (skosev@cogs.nbu.bg) Penka Hristova (phristova@cogs.nbu.bg) Central and East European Center for Cognitive Science, Department of Cognitive Science and Psychology, New Bulgarian University, 21 Montevideo Street Sofia 1618, Bulgaria Abstract This paper is trying to explore the issue of whether analogical mapping may influence perception and cause re-interpretation of the target stimulus. An attempt is made to put this rather complicated issue at experimental test bed. Two experiments are described which demonstrate that the process of analogical mapping may reorganize the perceptual input and stimulate a re-representation of the input image thus allowing the participants to discover meanings that were not initially available to them. The experiments are not decisive on when this re-representation happens. Keywords: analogy; perception; representation; representation; psychological experimentation. re- Introduction The debate about the inseparability of perception and mapping in analogy-making was very hot in the 90s (Chalmers, French, & Hofstadter, 1992, Morrison, & Dietrich, 1995, Hoffman, 1995, Forbus, Gentner, Markman, Ferguson, 1998) but began to fade out later on even though it remained unresolved. This debate was probably too emotional and needed some break. However, it is an important debate that needs to be continued in a more constructive way. One problematic aspect of the debate as it was carried out in the 90s was that it stayed at a very abstract philosophical level and there were no proposals for how to try to answer the questions in an empirical way. This paper makes an attempt to reinitiate the debate, but to put it on experimental grounds. We fully realize that it is slippery to test empirically such subtle issues, but let us make some small steps in that direction. First of all, let us point out that since the time the debate was initiated a number of models were developed that addressed related issues. Thus Keane & Brayshaw (1988), Keane, Ledgeway, and Duff (1994) and Forbus, Ferguson, & Gentner (1994) have given up the “one step view” on the mapping process and introduced the so called “incremental view”. This approach allowed the mapping to dynamically adjust to the new coming information about the target. These approaches also allowed backtracking, i.e. old partial mappings can be destroyed and new mappings to be built on their place. However, none of these approaches ever discussed the possibility for re-representation of the target or the base themselves. In contrast, the CopyCat and TableTop models (Hofstadter, 1995, Mitchell, 1993, French, 1995) introduced the idea of re-representation of the target and the base, which are both perceived from the environment. Since the representation building is running in parallel with the mapping process in these models, there is constant re-representation of the target and the base (e.g. grouping and re-grouping of the elements, representing them as letters or numbers, etc.). The AMBR model introduced the idea of re-representation of the base while it is being retrieved from long term memory and mapped to the target (Kokinov & Petrov, 2000) and as result memory distortions were predicted which were later on experimentally confirmed. Finally, Yan, Forbus, and Gentner (2003) suggested a whole library of transformation functions that can allow for the re-representation of the target when it does not match well the base. Thus many models seem to have adopted the idea of re-representation to some degree and developed at least partial mechanisms for performing it (even though these mechanisms are often very different from each other). However, there are not many experimental studies that test whether such re- representations actually happen in human analogy-making. The closest experimental study we are aware of is performed by Medin, Goldstone and Gentner (1993). They demonstrated that comparing two objects may influence their encoding. They used ambiguous abstract figures that could be seen in two different ways. Each ambiguous target figure was presented in pair with another unambiguous figure and the participants were instructed to compare the two figures and to list their similarities and differences. The same target figure was paired with two different figures in a between group design. Since the participants produced different sets of attributes, depending on the figure the target was compared with, the authors concluded that they perceive and encode the target figures differently depending on the figure to be compared with. These results coupled together with the mounting evidence that alignment is a central aspect of similarity comparisons (Goldstone and Medin, 1993; Markman and Gentner, 1990) are backing up the hypothesis that analogical mapping may cause re- representation of the target stimulus. However, this study was not designed explicitly to test this hypothesis and that is why the task was not to make an analogy, but rather to list superficial similarities in most cases. Another problem with