Representing Properties Locally

Theories of knowledge such as feature lists, semantic networks, and localist neural nets typically use a single global symbol to represent a property that occurs in multiple concepts. Thus, a global symbol represents mane across HORSE, PONY, and LION. Alternatively, perceptual theories of knowledge, as well as distributed representational systems, assume that properties take different local forms in different concepts. Thus, different local forms of mane exist for HORSE, PONY, and LION, each capturing the specific form that mane takes in its respective concept. Three experiments used the property verification task to assess whether properties are represented globally or locally (e.g., Does a PONY have mane?). If a single global form represents a property, then verifying it in any concept should increase its accessibility and speed its verification later in any other concept. Verifying mane for PONY should benefit as much from having verified mane for LION earlier as from verifying mane for HORSE. If properties are represented locally, however, verifying a property should only benefit from verifying a similar form earlier. Verifying mane for PONY should only benefit from verifying mane for HORSE, not from verifying mane for LION. Findings from three experiments strongly supported local property representation and ruled out the interpretation that object similarity was responsible (e.g., the greater overall similarity between HORSE and PONY than between LION and PONY). The findings further suggest that property representation and verification are complicated phenomena, grounded in sensory-motor simulations.