Consider the predicament…
…of a young infant recently arrived in the world and trying to make sense of it. She has some resources at her disposal: sensory information about her environment, the ability to act on it, and in most cases a surrounding linguistic environment, family and culture that can help to teach her what she needs to know. Nevertheless the task is daunting. Suppose on one occasion that daddy gestures out the window and says “Look, a bunny!” To what is he referring? The field of green? The tall structures dotting the horizon? The brownish object streaking rapidly along the ground? Later in the evening mommy repeats the word, this time gesturing toward a white contour in a picture book—it is not moving, it is not brown, it is two-dimensional. At bedtime big brother says “Here’s your bunny,” this time handing her a soft pink fuzzy object. What on earth could they all be talking about!
And yet before she turns 10 she will know that the word “bunny” refers to a particular animal with long ears and a fluffy little tail, and what’s more, she will know that bunnies have blood and bones inside; that they can reproduce and grow and die; that they can feel pain and get hungry; that they are warm to the touch; that they live in holes in the ground; that some people believe it brings good luck to wear a bunny-foot on a chain. When she gets a new bunny-rabbit as a pet, she will be able to infer that all of these things are true, even though she has never before encountered this particular bunny; and when she brings her new pet to show-and-tell, she will be able to communicate all of these facts to her classmates simply by talking. And, this knowledge about bunny-rabbits constitutes a tiny fraction of the general factual world-knowledge she will have accumulated. Understanding the basis of these human abilities—to recognize, comprehend, and make inferences about objects and events in the world, and to comprehend and produce statements about them—is the goal of research in semantic memory.
In the Knowledge and Concepts Lab, we are developing a computational framework for understanding human semantic memory, and are bringing this framework to bear on aspects of conceptual development, adult conceptual abilities, and impaired semantic cognition in patients with degenerative semantic syndromes. In agreement with many others, we suggest that semantic knowledge arises from the interactive activation of modality-specific representations of perceptual attributes (such as shape, colour, patterns of movement, word sounds, etc.). A novel and key aspect of our theory is the proposal that these different modality-specific representations communicate with one another through a common set of distributed, amodal representations, that a) serve to capture conceptual similarity structure and thereby promote semantic generalization and induction, b) are sensitive to context and c) are coded in the anterior temporal lobes. The theory specifies that these intermediating representations are acquired through experience, and much of our work has focused on illustrating, through computer simulations, how domain-general mechanisms of acquisition can explain puzzling phenomena in the study of semantic cognition. Much of this work is described in a book available from MIT Press. Current research in the lab addresses empirical and computational puzzles arising from the basic theory. Links to descriptions of some current projects appear below.