Cognitive Processes 405occur between category slides than between noncategory
slides. An associative approach can explain much of the
available category data (Wasserman & Astley, 1994). Feature
theory, the view that a set of conjoined features separates
category members from nonmembers, has been applied
to category data (Watanabe et al., 1993). Feature theory, too,
is compatible with an associative analysis.
In categorization experiments animals may come to form
aprototype,an exemplar representing the central tendency of
all of the individual exemplars. For example, a robin is a bet-
ter prototype of bird than is, say, a penguin. There is no com-
pelling evidence that animals form prototypes (see, e.g.,
Mackintosh, 1995). Rather, available data in animals can be
interpreted in terms of exemplars. An extensive discussion of
the use of concepts and categories by humans is to be found
in the Goldstone and Kersten chapter in this volume.
Serial Learning
The specific experiments cited in the previous section, as
well as many other types, are explicitly recognized instances
of discrimination learning. However, there are many other
types of investigations that are clear instances of discrimina-
tion learning but are not generally considered under that
heading. A case in point isserial learning,a procedure
popular in animal and human learning alike. In one type of
serial learning task, items are presented successively in a
particular order (e.g., A-B-C-D) and the animal’s task is to
learn both the items and the specific order in which they
occur. People master many sorts of successive serial tasks:
days of the week, months of the year, the alphabet, and so on.
In a serial learning task in which items are presented succes-
sively, the animal must learn to respond differentially to
different stimuli; thus serial learning is a variety of discrimi-
nation learning.
Not surprisingly, the issues raised in animal serial learning
are quite similar to those raised in connection with explicitly
recognized instances of discrimination learning. In some re-
spects, however, contrary to popular opinion, serial learning
data are more germane to the issue of animal cognition than
are currently available, explicitly recognized instances of dis-
crimination learning, including category learning. For one
thing, there can be little doubt that serial learning, as investi-
gated using animals, involves cognition of some sort, partic-
ularly memory, as we shall see. For another, it is clear that
categorization (called chunking) is involved in serial learn-
ing, and it apparently cannot be explained in terms of stimu-
lus generalization.
Consider an animal that learned to respond appropriately
to a progressively decreasing series of reward magnitudes
terminating in nonreward (e.g., large reward, medium reward,
small reward, nonreward). Appropriate responding might
consist of progressively weaker responding over the series.
What has an animal learned in such a case? Consider three
different interpretations. The animal may, as the Gestalt psy-
chologists have suggested, have learned a relationship among
the items; that is, it may be that reward magnitude decreases
monotonically over trials (e.g., Hulse & Dorsky, 1977). The
animal may learn an association between the item and its po-
sition in the series—that is, that Position 1 signals large re-
ward, Position 2 signals medium reward, and so on (Burns,
Dunkman, & Detloff, 1999). The animal may also learn an as-
sociation between the memory of one or more prior items and
the current item; that is, the memory of Item A (large reward)
signals B (medium reward), the memories of Items A and B
signal C (small reward), and so on (see Capaldi, 1994). Re-
cent evidence suggests that rats are able to employ either item
cues or position cues in learning a successively presented
series of food items (Burns et al., 1999; Capaldi & Miller,
2001a). The conditions under which rats may tend to employ
either position cues or item cues or both have yet to be iso-
lated and identified clearly.
In the sort of serial learning task examined in this section,
items are separated by a retention interval. For example, a
given item, say, A, may be presented and removed minutes or
hours before the participant is asked to respond to the next
item, B. Appropriate serial responding under retention inter-
val conditions necessarily involves employing the memory or
representation of some prior event (item memory, position
memory, or both) in order to anticipate the current event cor-
rectly. Series may be employed such that the memory or
representation involved is necessarily that of one or more
prior items. As an example, consider rats that have received
two slightly different series in irregular order: XNY and ZNN
(Capaldi & Miller, 1988a). X, Y, and Z are discriminably dif-
ferent food items; N is nonreward. Items of each series were
separated by shorter intervals than that separating the series
themselves. Rats trained XNY and ZNN learn to respond cor-
rectly to the third item of the series—that is, to respond more
vigorously to Y than to N. Trial 3 running cannot be mediated
by the Trial 2 event because it is the same in both series, N.
Whatever else may be the case, therefore, discriminative re-
sponding on Trial 3 requires that the rat remember on Trial 3
the item presented on Trial 1; that is, the rat must respond
more vigorously when X occurred on Trial 1 than when Z oc-
curred on Trial 1. Further implicating memory, rats have re-
sponded more vigorously on rewarded (R) than nonrewarded
(N) trials when the R and N trials were alternated (R, N, R, N,
etc.) and the retention interval was 24 hr (e.g., Capaldi &
Lynch, 1966; Capaldi & Minkoff, 1966).