Memory - Computational Explorations in Cognitive Neuroscience - page 314

cue, but nothing is provided to support the retrieval of
each individual word from the list. Based on everything
we have learned in this chapter, it seems clear that the
hippocampus is primarily responsible for encoding ar-
bitrary information such as words on a list. However,
the prefrontal cortex also plays a critical role in free re-
call by virtue of its ability to organize information dur-
ing encoding and recall, and implement a temporally
extended sequence of strategic steps necessary to re-
trieve the information from the hippocampus. We will
discuss more about this kind of strategic processing in
the prefrontal cortex in chapter 11.
Two of the key behavioral phenomena to be ex-
plained in a model of free recall are: serial position
effects , where words studied at the beginning and end-
ing of a list are recalled better than ones in the mid-
dle ( Burgess, 1995 presents a neural network model
of these effects). The standard explanation of such ef-
fects is that the last items are retained in active mem-
ory (i.e., the prefrontal cortex), while the first items
are encoded better in long-term memory (i.e., the hip-
pocampus); and chunking effects , where items that can
be grouped (“chunked”) together (e.g., along semantic
dimensions) are recalled better than ungroupable lists.
Data from frontally lesioned patients indicates that the
prefrontal cortex is responsible for strategic grouping in
free recall (Gershberg & Shimamura, 1995).

on encoding within a given area. Although we will not
elaborate these ideas at this point, we point to it as an
important area for future research.

9.7.5 Working Memory

As mentioned previously, working memory refers to the
active use of memory for cognitive processing. As we
saw, the prefrontal cortex appears to be specialized for a
kind of robust, rapidly updating active memory that fits
many people's description of what working memory is.
However, we also think that working memory can be
subserved in many cases by the hippocampal system,
which can encode conjunctions of information during
processing for later recall. For example, as one is read-
ing a series of paragraphs (such as these), there is sim-
ply too much information to keep all of it online in an
active state in the prefrontal cortex. Thus, the active
representations that emerge from sentence processing
(see chapter 10 for relevant models) can be encoded in
the hippocampus, and then accessed as necessary dur-
ing the processing of subsequent text. In addition to
the hippocampus, we also think that the posterior cortex
plays an important role in working memory, by facilitat-
ing the activation of appropriate representations accord-
ing to rich semantic links, for example. Thus, working
memory is a sufficiently general construct as to involve
virtually the entire cognitive apparatus. For more dis-
cussion of how the framework developed in this text
can be applied to understanding working memory, see
O'Reilly et al. (1999a).

9.7.4 Item Effects

In addition to the kind of task used to probe memory
(i.e., recognition, cued, and free recall), another impor-
tant aspect of memory that has been studied is the ef-
fects of different item properties. For example, are dis-
tinctive items recalled better than less distinctive items?
What about high versus low frequency items? In gen-
eral, our computational framework for understanding
memory provides a rich source of explanation for such
effects. Items can differ in the nature of their represen-
tations in each of the three main areas (posterior cor-
tex, hippocampal system, and frontal cortex), and in the
extent to which they depend on these different areas.
Thus, some item effects can be explained by the dif-
ferential involvement of different areas, whereas others
can be explained by the effects of item representations

9.8 Summary

Starting with a simple distinction between weight-
based and activation-based memory, based on the
properties of individual neurons, we have developed a
more complex overall picture of memory as a product
of interacting neural systems, each with their own dis-
tinctive properties. We focused on the three major com-
ponents of the cognitive memory system: the posterior
cortex ,the frontal cortex ,andthe hippocampal sys-
tem .
In the domain of weight-based memory, interfer-
ence from the rapid learning of arbitrary informa-