Higher-Level Cognition - Computational Explorations in Cognitive Neuroscience - page 395

the model would require a more complex network ca-
pable of representing 16 different stimuli, whereas our
simplified model can only represent 4. One of the most
interesting phenomena that could be addressed by such
a model is the finding that a second reversal does not
produce the same patterns of frontal deficits as observed
with the first reversal. We suspect that this may emerge
as a result of the posterior system establishing a more
equal balance among the representations involved, but
this may not entirely account for the effect.
Another aspect of the model worth noting is the way
in which object and location information has been rep-
resented in the hidden layer. Although we know that
object (“what”) and location (“where”) information are
represented in separate pathways in the brain (chap-
ter 8), we have used combined what/where represen-
tations in the hidden layer, so that each hidden unit
represents an object feature in a particular location.
This avoids having to deal with the what-where bind-
ing problem that is likely resolved by more complex se-
quential attention mechanisms in the real system (again
see chapter 8 for models of this). The basic principles
of the frontal involvement in this task are not likely to
be affected by these implementational factors.

work, which we then discuss in relation to other major
theoretical frameworks for understanding frontal func-
tion. The following functions have generally been at-
tributed to the frontal cortex:

Activation-based working memory This attribution
is based largely on monkey electrophysiology
data in simple delayed response tasks that require
information to be maintained during the delay (e.g.,
Fuster, 1989; Goldman-Rakic, 1987; Miller et al.,
1996; see chapter 9 for a more detailed discussion of
these data).
Inhibition The frontal cortex appears to be important
for inhibiting prepotent responses, (e.g., word read-
ing in the Stroop task) or “reflexive” or “instinctual”
behaviors.
Flexibility Frontal patients tend to lack flexibility, for
example by perseverating in producing old responses
even after the task has changed (e.g., in the reversal
conditions of the dynamic categorization task).
Fluency Frontal patients have difficulty generating a
variety of responses to a given stimulus, for example
in coming up with novel uses for a familiar object.
Executive control Frontal cortex is important for goal-
directed behavior like planning, coordinating, and the
like.
Monitoring/evaluation To control behavior, an execu-
tive also needs to monitor and evaluate the status of
ongoing behavior — some frontal areas seem to be
specialized for these functions.

11.5 General Role of Frontal Cortex in

Higher-Level Cognition

We have now explored two primary ways in which
frontal cortex contributes to higher-level cognitive func-
tion. In the Stroop model, robust active maintenance
provides top-down activation to facilitate weaker pro-
cessing in the posterior cortex, and in the dynamic cat-
egorization model frontal cortex contributes to flexible
behavior by virtue of dynamic activation-based process-
ing control mechanisms.

After discussing the data in support of these attribu-
tions, we will see how each of them can be understood
within the common framework of an activation-based
processing model of frontal function. Because chap-
ter 9 covers the working memory data, we focus on the
other themes here.

11.5.1 Functions Commonly Attributed to Frontal

Cortex

Inhibition

In this section, we more broadly review several func-
tions, not necessarily mutually exclusive, that have been
attributed to the frontal cortex based on cognitive neu-
roscience data. We show that these functions can be un-
derstood within the activation-based processing frame-

Failures of inhibition are sometimes the most salient ef-
fect of frontal damage. A classic example is the case
of Phineas Gage, who had a metal spike penetrate his