When Working Memory Is Unlimited

When Working Memory Is Unlimited
A limited capacity working memory is a central concept in cognitive
psychology. Since Miller (1956) and Atkinson and ShiVrin (1968), most
discussions of human cognitive architecture have incorporated a limited
capacity short-term or working memory. It is appropriate that they should
do so because working memory is limited when dealing with new
information. Nevertheless, capacity limitations only apply when dealing
with new, not old information. When dealing with previously learned
material, the only discernible limit on working memory is the amount that
has been learned and stored in long-term memory. Massive, seemingly
unlimited amounts of information can be processed by working memory
providing they have previously been incorporated in schemas. A schema
may contain a large amount of information but will be processed in working
memory as a single element.
The tension between a very limited working memory when dealing
with new information and an unlimited working memory when dealing with
224 John Swellerlearned material can be seen as far back as Miller’s (1956) paper. Miller’s
concept of ‘‘chunking,’’ which today can be incorporated in the more
sophisticated conception of schema construction, altered the amount of
information that short-term memory could hold. By chunking together
elements of information, the amount of information held by short-term
memory could be increased. In that sense, learning could be used to increase
the eVective capacity of short-term memory. Similarly, while working
memory can only process a limited number of schematically based elements,
what constitutes an element is entirely dependent on what has been learned.
If much has been learned, an element can incorporate a massive amount of
information. Indeed, there may be no limit to the amount of information
incorporated in a schema that acts as a single element in working memory.
In that sense, there is no limit to the amount of information that can be
processed by working memory. The capacity limitations of working memory
appear only when new, unorganized information that has not yet been
organized into schemas must be processed.
Empirically, de Groot (1965) and Chase and Simon (1973) provided the
strongest early evidence for this phenomenon. Chess experts with their
appropriate schemas can hold an entire board of chess pieces taken from a
real game in working memory because they have a schema for that confi-
guration. Novices have to remember each piece individually, which is beyond
working memory capacity, as are random configurations for experts. This
result has been obtained in a wide variety of areas (e.g., Egan & Schwartz,
1979; JeVries, Turner, Polson, & Atwood, 1981; Sweller & Cooper, 1985).
The ability of working memory to hold and process large amounts of
learned information for long periods of time was recognized by Ericsson and
Kintsch (1995). Their concept of ‘‘long-term working memory’’ applies to
very well-learned material. For such material, the capacity limitations of
‘‘short-term working memory’’ disappear. Large amounts of domain-
specific, well-learned material in complex areas such as text comprehension,
chess, and music can be held and processed in working memory for long
periods. The usual capacity and duration limits associated with working
memory are not in evidence for such well-learned material.
In eVect, we are dealing with two continua: A learning continuum and a
working memory limitations continuum. At one extreme, when dealing with
yet-to-be-learned or unlearned material, well-known working memory
limitations are relevant to processing. At the other extreme, when dealing
with well-learned material, the usual working memory limitations are
irrelevant and working memory can best be described in terms of Ericsson
and Kintsch’s (1995) long-term working memory. Thus, in this chapter,
long-term working memory is incorporated at one end of a working memory
continuum rather than as a discrete structure. Rows 1 and 5 of a cognitive
Evolution of human cognitive architecture 225matrix of continua (see Fig. 1) depict the learning and working memory
continua, respectively.
This chapter is concerned primarily with the intervening constructs
relating unlearned material and a limited working memory at one extreme of
Fig. 1. A cognitive matrix of continua.
226 John Swellerthe matrix of continua and relating well-learned material and an unlimited
working memory at the other extreme. While learning through schema
acquisition eliminates the problem of a limited working memory having to
handle large numbers of interacting elements (the right side of the matrix of
continua), the question remains why human cognitive architecture evolved
in this manner rather than following the apparently more straightforward
route of a larger working memory, either as an adjunct to or even as a
substitute for some learning mechanisms. That route of a larger working
memory would have permitted larger amounts of new material to be
processed. There should be evolutionary reasons why that route was not
followed