Empirical Support for Empiricism

In the preceding section, I offered a theoretical argument for empiricism. If we consider what concepts are for, it is natural to suppose that they are built up from stored perceptual-motor states. But theoretical arguments require confirmation in the lab. Is there any empirical support for Hypothesis 3? Every chapter in this volume testifies to an affirmative answer. I will mention a few important finding here.

On the face of it, empiricism may seem incompatible with evidence from neuroscience. There are polymodal association areas of the brain that are thought to contain amodal symbols. These areas receive inputs from multiple sensory modalities, and they are especially prevalent in frontal cortex, which is associated with higher cognition. Do such findings refute concept empiricism? I don't think so. While these areas play a role in higher cognition, they do not function alone. Damasio (1989) has argued that polymodal association areas are "convergence zones" that serve to reactivate modality specific areas during cognitive tasks (see also Warrington & McCarthy 1983; Simmons & Barsalou, 2003). They store records about what sensory states were concurrently active during prior encounters with objects and events. Cognitive tasks require the reactivation of those sensory records. This is consistent with empiricism.

Barsalou and his colleagues have amassed a body of evidence that is consistent with convergence zone theory (see Simmons & Barsalou, 2003, for a review). Some of that evidence comes from neuroimaging. Simmons et al. (2003) conducted an fMRI study in which subjects were asked to confirm whether members of a category can have certain properties (e.g., "can leaves be rustling?"). Outside of the scanner, the categories and properties were rated for the extent to which they are experienced in each of five sense modality and the extent to which they are associated with motor response. For example, experiences of leaves might be rated as highly visual, somewhat auditory, and largely nonolfactory. These ratings were then correlated with brain activity in sensory specific areas. The category and property sensory ratings had an 0.7 correlation with activity in sensory areas, and explained almost 50% of the variance. These results show that property verification tasks cause spontaneous activation in modality-specific brain areas, and those activations can be predicted by the perceptual profile of the category. Other researchers have obtained similar results (see Martin & Chao, 2001). Chao, Haxby, and Martin (1999) found that the same modality-specific areas of the brain are active when subjects recognize pictures of a category and when they read the name. Chao and Martin (2000) found brain activation in areas associated with motor response when subjects viewed pictures of tools.

Barsalou and colleagues have also found behavioral evidence in support of empiricism. For example, Pecher et al. (2003) gave subjects a property verification task, in which they need to confirm whether or not a named feature is true of a named category. Subjects would answer "yes" to LEAVES-rustling and "no" to CARNATION-black. All of the features were colors, sounds, or other observable qualities. Pecher et al. wanted to know if these features were represented using an amodal code or a modality specific code. They reasoned that if subjects were reactivating sensory states to perform the task, there should be a temporal cost to shifting from a feature in one modality to a feature in another. This is just what they found. In performing conceptual tasks, subjects apparently reengaged perceptual representations. Category names also seem to engage motor responses. For example, Spivey et al. (2000) found that subjects tended to look up when they heard stories about tall buildings, and they tended to look down when they heard about canyons.

I think that the evidence for perceptual/motor grounding is very strong. Rather than utilizing purely amodal representations, we seem to call up perceptual features and action commands when we perform conceptual tasks. This suggests that concepts are collections of perceptual and behavioral features that are reactivated "off-line" in thought. But one might think these findings are limited to concrete concepts, i.e., concepts of objects and relations that we can readily observe. There has been comparatively little investigation of empiricist accounts of abstract concepts. Abstract concepts pose a significant challenge. If conceptual tasks reactivate perceptual states, then perceptual states must be reactivated when we use abstract concepts. That sounds bizarre. By definition, abstract concepts do not designate observable things. What perceptual features could possibly be involved?

0 0

Post a comment