Abstract

Simulating Time-Course Phenomena in Perceptual Similarity via Incremental Encoding Andrew Lovett ([email protected]) Dedre Gentner ([email protected]) Qualitative Reasoning Group, Northwestern University 2133 Sheridan Road, Evanston, IL 60201 USA Department of Psychology, Northwestern University 2029 Sheridan Road, Evanston, IL 60201 USA Kenneth Forbus ([email protected]) Qualitative Reasoning Group, Northwestern University 2133 Sheridan Road, Evanston, IL 60201 USA of comparison. We use SME to simulate both studies, providing evidence that supports the time course of encoding interpretation. Abstract If people are required to respond very quickly in a same- different task, their judgments of sameness are heavily reliant on attribute matches, despite the fact that when given ample time, the judgments seem to rely chiefly on relational matches (Goldstone & Medin, 1994). One interpretation of this temporal pattern is that attribute matches enter into the comparison process before relational matches. However, an alternate explanation, suggested by findings of Sloutsky & Yarlas (submitted) is that attributes are encoded before relations. In this case, if the comparison process begins before the encoding is completed, early matches will involve attributes but not relations. We show via a simulation that SME can model the Goldstone & Medin results, as well as the Sloutsky & Yarlas (submitted). Figure 1. Scenes from Goldstone & Medin (1994) 2. Time-Course Effects in Comparison 1. Introduction There is considerable evidence that the processes that govern analogical mapping may also apply to similarity comparisons (Markman & Gentner, 1996; Gentner & Markman, 1997). For example, Markman and Gentner (1996) found that when rating the similarity of two images, subjects attended more to differences connected to the common structure of the two images (alignable differences) than to differences unrelated to the common structure. These findings suggest that when asked to find a difference, participants first carried out a structural alignment between the images. Results such as this suggest that the same cognitive process may underlie both analogy and similarity. Consistent with this, the Structure-Mapping Engine (SME) (Falkenhainer, Forbus & Gentner 1989), a computational model of analogy, has successfully modeled perceptual similarity results (Kuehne, Gentner, & Forbus, 2000; Loewenstein & Gentner, 2005). A critical issue in modeling the psychological processes of analogy and similarity is simulating the time course of processing. In an important study, Goldstone and Medin (1994) found that participants in a similarity task showed relatively greater sensitivity to attribute matches early in processing, and to relational matches later. This suggests that in perceptual similarity computations, attribute matches are made before relational matches. We begin by reviewing this study and then discuss results from Sloutsky and Yarlas (submitted) that suggest that the lag between attributes and relations arises from the time course of encoding rather than In Experiment 1 of Goldstone and Medin’s (henceforth GM e.g., the top butterfly in one scene could match the bottom butterfly in the other. There were three deadlines, which varied within-subject: short (1 s), medium (1.84 s), and long (2.68 s). The dependent measure was the error rate on different trials: the