Higher-Level Cognition - Computational Explorations in Cognitive Neuroscience - page 400

such mapping systems. The lower-level mappings can
subserve more reflexive and well-learned mappings,
while the frontal cortex is important for more novel
and temporally-extended, complex mappings. This is
clearly very consistent with our activation-based pro-
cessing framework. Another important example is the
framework of Goldman-Rakic (1987), which is based
on the idea of actively maintained representations in
frontal cortex. Although her research has focused on
relatively simple maintenance tasks that do not place
significant demands on either dynamic control mech-
anisms or representational complexity (e.g., more ab-
stract goallike representations), her basic ideas can be
extended to cover these issues, as we have done within
our framework.
Finally, there are a range of other theoretical perspec-
tives that we do not have the space to cover, but which
also share aspects in common with our framework, and
offer a variety of different ideas as well (e.g., Baddeley,
1986; Grafman, 1989; Petrides, 1996).

Novel,
sustained
or weak

Novel,
transient

Controlled

Automatic

Familiar,
sustained
or weak

Routine,
transient

Strong/Transient
(PFC Independent)

Weak/Sustained
(PFC Dependent)

Bias

Figure 11.16: Ways in which the hippocampus and
frontal cortex contribute to the automatic versus controlled-
processing distinction. Bias is provided by the frontal cortex,
and can be used to perform sustained processing, can facilitate
the processing of weakly-learned (i.e., relatively infrequent)
tasks, and can serve to coordinate processing across different
areas. Binding is provided by the hippocampus, and can be
used to rapidly learn and store the information necessary to
perform novel tasks or processing. Controlled processing can
involve either or both of these contributions, whereas auto-
matic processing can be performed independent of them.

11.6 Interacting Specialized Systems and

Cognitive Control

Although we have focused our discussion to this point
on the role of the frontal cortex in higher-level cognition
and controlled processing, it is clear that these phenom-
ena depend on the interaction of multiple brain areas. In
this section we elaborate some specific ideas regarding
the nature of these interactions as they contribute to con-
trolled (versus automatic) processing. These ideas are
based on the cognitive architecture comprised of three
interacting specialized systems: the posterior cortex,
the hippocampus, and the frontal cortex as sketched in
chapter 7 and explored in the context of memory func-
tion in chapter 9.
We focus on the interactions between frontal cortex
and hippocampus here, because a relatively simple story
can be told. This should not be taken to exclude other
interactions. The goal is to understand the nature of the
continuum between controlled and automatic process-
ing in terms of the differential involvement of the frontal
cortex and hippocampus (figure 11.16). As we've seen,
the frontal cortical contribution can be characterized in
terms of top-down biasing . The hippocampal contribu-

tion can be characterized in terms of a binding function
for rapidly encoding and binding together information
(Cohen & O'Reilly, 1996). This binding can be use-
ful in the context of controlled processing for encoding
the different components of a task as given by verbal
instructions, or encoding intermediate states of process-
ing for later retrieval. Thus, we hypothesize that the
extent of controlled processing in a task is defined by
the extent to which the following conditions exist:

Sustained, weakly learned (i.e., relatively infre-
quent), or coordinated processing is required (engag-
ing the frontal cortex biasing function).

Novel information must be rapidly stored and ac-
cessed (engaging the hippocampus binding function).

The extent of automatic processing is defined by the rel-
ative absence of these factors.
We can apply this framework to the case of text
comprehension to obtain a fuller picture of the kinds
of interactions that are possible.

As one is reading,