Higher-Level Cognition - Computational Explorations in Cognitive Neuroscience - page 401

the limited capacity of the frontal active memory sys-
tem will necessitate the offloading of information to
the hippocampal system. Thus, the comprehension of
prior paragraphs is encoded only within the hippocam-
pus, and must be recalled as necessary during later pro-
cessing (e.g., when encountering a reference like, “this
would be impossible, given Ms. Smith's condition,”
which refers to previously introduced information that
may not have remained active in the frontal cortex).
This later reference can be used to trigger recall of the
previous information from the hippocampus, perhaps
with the addition of some strategic activation of other
relevant information that has persisted in the frontal cor-
tex (e.g., the fact that Ms. Smith lives in Kansas). A suc-
cessful recall of this information will result in the acti-
vation of appropriate representations within the both the
frontal and posterior cortex, which combined with the
current text results in comprehension (e.g., Ms. Smith
was hit by a tornado, and can't come into work for an
important meeting).
Although we can visualize these kinds of interac-
tions in terms of the hippocampal and frontal models
we have explored, models that incorporate both of these
components, together with a distributed posterior corti-
cal system, are just beginning to be developed. Such
models are very complex, and represent the frontier
of current computational capacity. Therefore, it is too
early to say how well our account of controlled pro-
cessing works in practice, but as a conceptual model,
it clearly captures important aspects of the controlled-
versus-automatic processing distinction.

We first explored this framework by modeling several
patterns of human performance on the Stroop task, the
critical condition of which involves a conflict between
a more dominant word-reading pathway and a weaker
color-naming pathway when subjects are asked to name
the color of a color word written in another color (e.g.,
“red” written in blue ink). The Stroop model is based on
the principle of top-down biasing from the prefrontal
cortex, which can support the weaker color-naming
pathway and help it compete against the stronger word-
reading pathway. In the absence of this top-down sup-
port (e.g., in frontal patients and in schizophrenics, or
in the lesioned model), the word-reading pathway dom-
inates, resulting in slower and more error-prone perfor-
mance in the conflict condition.
Building on the basic principles of top-down biasing
from the Stroop task, we explored a dynamic catego-
rization task based on the Wisconsin Card Sorting
Te s t , which is widely known to be sensitive to frontal
function. The model demonstrates that the frontal cor-
tex can more rapidly switch categorization rules by us-
ing activation-based processing, whereas the posterior
cortex depends on less flexible weight-based process-
ing. Thus, frontal lesions result in perseverations when
the system resorts to the weight-based processing, and
the model can account for two different kinds of perse-
verations that result from lesions in two different frontal
areas. The model uses the same dynamic control
mechanism based on the neuromodulator dopamine as
was explored in chapter 9 for controlling updating and
maintenance in active memory. This mechanism results
in trial-and-error search in activation-space, which is
what produces more flexible behavior.
Having explored two important manifestations of the
activation- versus weight-based framework, we then
reviewed a range of phenomena from monkey elec-
trophysiology, the behavior of frontally damaged pa-
tients, and functional neuroimaging, which can all be
accounted for within the activation-based processing
framework. These phenomena include inhibition , per-
severation , fluency/flexibility , executive control ,and
monitoring/evaluation . We concluded with a discus-
sion of the ways in which other brain areas, including
the hippocampus , can interact with the frontal cortex
in higher-level cognitive phenomena.

11.7 Summary

We have presented a speculative but hopefully intrigu-
ing framework for approaching the issue of higher-level
cognition. This framework is centered around the no-
tion of activation-based processing as distinct from
weight-based processing , with the idea that the frontal
cortex is specialized for activation-based processing.
Two key issues for activation-based processing are the
nature of the control mechanisms for updating and
maintenance, and the nature of the activation-based
representations , which need to exist prior to their use
and be flexibly recombined to solve novel tasks.