Perceptual simulation in conceptual tasks

concepts directly, we have found that three mechanisms are central (1) event simulations that frame focal concepts (e.g., Fillmore, 1985; Langacker, 1986), (2) perceptual symbols from introspection, (3) propositional construal. Our conjecture is that any abstract concept can be represented directly using these three mechanisms. 2. Empirical Research The primary goal of our empirical work has been to test the assumption that perceptual simulation underlies conceptual processing. Whereas standard views assume that people process concepts using amodal symbol systems such as feature lists, frames, and semantic networks, we explore the possibility that people process concepts by simulating their referents perceptually. Although these simulations may at times be conscious, they may often proceed unconsciously (1.2). Note that our empirical efforts thus far do not constitute a full test of our theory. Indeed, most aspects of the theory remain untested. Instead, our research has only explored the theory's core assumption, namely, that perceptual simulation lies at the heart of conceptual processing. To explore this issue, we have subjects perform standard conceptual tasks, such as feature listing and property verification. In the critical conditions, subjects do not see pictorial stimuli, nor are they asked to perform perceptual processing, such as imagery. Instead, subjects only receive linguistic stimuli and are asked in as neutral a manner as possible to perform conceptual processing. Of interest is whether subjects perform perceptual simulation spontaneously. 2.1. Instructional equivalence and perceptual work Barsalou, Solomon, and Wu 6 Across several lines of experimentation, we have sought two general forms of evidence for perceptual simulation: instructional equivalence and perceptual work. To examine instructional equivalence, each experiment includes neutral and imagery subjects. As just described, neutral subjects receive standard task instructions with nothing said about images. For feature listing, subjects are asked to list the properties typically true of a concept. For property verification, subjects are asked to verify whether a concept has a property. In contrast, imagery subjects are asked to perform these conceptual tasks using images. For feature listing, subjects are asked to construct images for the referents of concepts and then describe these images. For property verification, subjects are asked to construct an image of a concept and then verify the truth or falsity of a property by attempting to find it on the image. To assess instructional equivalence, we compare detailed performance profiles of the neutral and imagery subjects. We can assume that imagery subjects use images as instructed, given that much previous work indicates that subjects adopt images when asked to do so (for reviews, see Finke, 1989, and Kosslyn, 1980). As discussed shortly, we also have independent verification from perceptual work that imagery subjects use images. Thus, the key issue is how neutral subjects compare to imagery subjects: If neutral subjects use images during perceptual simulation, then their detailed performance profiles should be essentially the same as those for imagery subjects. If both groups perform the task similarly, it is likely that neutral subjects use images, assuming that imagery subjects use them. On the other hand, if neutral subjects use amodal representations such as feature lists, frames, or semantic nets, then the detailed performance profiles of the two groups should differ considerably.1 Turning to perceptual work, we manipulate a variable in each experiment that is known to affect perceptual processing, such as the visibility, size, or position of features. In both perception and imagery, it is well known that these variables affect performance (Finke, 1989; Kosslyn, 1980; Shepard & Cooper, 1982). If neutral subjects in our conceptual tasks perform perceptual simulation, then we should see effects of these variables on their performance. The more perceptual work a subject has to do, the more difficult processing should be. In contrast, the amodal view predicts no such effects because perceptual simulation is irrelevant. Perceptual work provides an independent check on whether imagery subjects Barsalou, Solomon, and Wu 7 use images as instructed. If they do, then perceptual work variables should affect their performance. To the extent that perceptual work variables affect neutral and imagery subjects similarly, instructional equivalence is established