260 • CHAPTER 9 Knowledge
While active research on connectionism continues in many laboratories, some
researchers point out that there are limits to what connectionist networks can explain.
They are especially critical of the idea of the back propagation error-correcting mecha-
nism, because it isn’t clear exactly how it works, and it has been diffi cult to fi nd a simi-
lar mechanism in the brain (see O’Reilly, 1996, who proposes a possible mechanism,
and the commentary following Rogers & McClelland, 2008). Whatever the fi nal ver-
dict on connectionism, this approach has stimulated a great deal of research, some of
which has added to our understanding of both normal cognition and how brain damage
affects cognition.
Now that we have described how concepts are represented by a computer model
based on the operation of the brain, we will describe some recent discoveries about
how concepts are represented in the brain. This research involves recording from single
neurons in the monkey, studying the effects of brain damage in humans, and measuring
brain activity by brain scanning in humans.Categories and the Brain
How are different categories (or concepts) represented in the brain? Research on this
question has focused largely on visual objects like the ones we have been discussing in
this chapter—things like plants, animals, vehicles, and trees.SPECIFIC OR DISTRIBUTED ACTIVITY?
One possible answer to the question of how objects are represented is that different
categories of objects are represented by activity in specifi c areas of the brain. Two
examples of areas for specifi c categories are the fusiform face area (FFA) that responds
strongly to faces and the parahippocampal place area (PPA) that responds to houses,
rooms, and places (see page 32). Supporting the connection between the fusiform area
and faces is the condition prospoagnosia, the inability to recognize faces, which occurs
in people who have suffered damage to the temporal lobe.
But saying that one group of things, such as faces, is represented by activity in one
area of the brain and another group, like houses or rooms, is represented in another
area doesn’t go far enough. For one thing, we know that brain representations are usu-
ally distributed, so that even if a particular stimulus causes a large amount of activity in
one area, it also causes activity in many other areas as well.
One reason for this distributed representation is that objects consist of many
different properties. Consider, for example, the cat in Figure 9.1. It has perceptual
properties such as color, texture, and form. It has motor properties, which would
include how cats move when they are walking, running, and catching mice. It has
behavioral properties, which include catching mice, sleeping during the day, and
other aspects of cat behavior. Cats also can evoke affective responses, such as a
particular person’s emotional response to cats. Thus, the representation of the cat
consists of a distributed representation that would include activity in sensory areas
(for what the cat looks like when stationary and moving), motor areas (for how it
moves), higher level areas (that represent knowledge about the cat’s behavior and
other qualities), and emotional areas (for the emotional response elicited by the cat)
(Barbey & Barsalou, 2009).CATEGORY INFORMATION IN SINGLE NEURONS
The representation of categories in the brain has been studied by recording from single
neurons. To illustrate this, we will describe an experiment by David Freedman and
coworkers (2001, 2003, 2008). Freedman trained monkeys to classify stimuli like the
ones in ● Figure 9.24, which consisted of mixtures of “cat” and “dog” stimuli. In thisCopyright 2011 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part.