406 Animal Memory and Cognition
Achunkconsists of lower order functional elements (e.g.,
letters of the alphabet) combined to form higher order ele-
ments (e.g., words). If items of a series are grouped—say
some on the left, others on the right—rats may tend to chunk
items similarly grouped (see Capaldi, 1992). Chunking is
clearly a form of categorization. Grouping cues that lead to
chunking in people have been used with rats and shown to be
similarly effective. These include the presentation of items
under different brightness conditions, in different spatial lo-
cations, and at different temporal intervals (see, e.g., Capaldi,
1992, 1994).
Another method employed in investigating serial learning
is the displaying of all items simultaneously rather than suc-
cessively. The participants’ task is to respond to the items in
a particular order. One of the more interesting findings ob-
tained employing the simultaneous presentation of items is
that monkeys appear to have a better grasp of an overall se-
quence of events than do pigeons. For example, pigeons
make more errors than monkeys to interior items of, say, a
five-item series (see D’Amato & Colombo, 1988; Terrace,
1986).
Areward schedule investigation consists of presenting re-
wards according to some rule. For example, food reward
might occur on a random half of all trials, nonreward on the
other half. A reward schedule of this sort, called a 50% irreg-
ular partial reward schedule,is clearly of major concern to
various orthodox theories of animal learning. We might ask
what, theoretically speaking, the difference is between a 50%
irregular schedule of partial reward and a serial learning task
in which, say, reward magnitudes become progressively
smaller over successive trials (a decreasing monotonic sched-
ule). It is the case that the two sorts of situation have been
treated differently in that many theories that attempt to deal
with 50% irregular schedules do not attempt to deal with mo-
notonic schedules, and vice versa. Recent evidence suggests
that treating the two sorts of schedules differently appears to
be unjustified (Capaldi & Miller, in 2001b). That is, memory,
which is clearly a major factor controlling performance in
orthodox cases of serial learning (e.g., the monotonic sched-
ule) is also a major factor in controlling performance in
orthodox reward schedule cases (e.g., the 50% irregular
schedule). Capaldi and Miller (2001b) demonstrated, essen-
tially, that similar variables, such as the number of nonre-
warded trials that occur in succession, have identical effects
in the two situations. The general implication of such find-
ings is as follows. If a clearly cognitive process such as
memory is intimately involved in regulating performance
under 50% irregular schedules, it is probably a factor in reg-
ulating instrumental learning generally. Put somewhat differ-
ently, the usual distinction between orthodox learning tasks
(e.g., varieties of instrumental conditioning) and orthodox
cognitive tasks (e.g., complex serial learning) may be artifi-
cial, cognition being involved in both.
The investigation of chunking in serial learning provides a
window into the ability of an animal such as the rat to orga-
nize separately presented items into wholes. Evidence has
been presented indicating that rats can form three different
sorts of chunks of varying degrees of complexity (see
Capaldi, 1992; Haggbloom, Birmingham, & Scranton, 1992).
Consider a rat trained in a runway—an apparatus in which
the animal must run from one end of a confined path to the
other end to obtain food. The first (and lowest order) chunk
formed is what is called the trial chunk. A trial chunk consists
of the animal’s combining into a single whole the separate
events of the trial—for example the opening of a door to
allow the animal access to the runway, to the animal’s run-
ning in the middle section of the runway, to its entering the
goal box at the end of the runway. The next highest chunk is
called a series chunk, which consists of the animal’s combin-
ing trial chunks into a higher level chunk. For example, a rat
trained under four nonrewarded trials followed by a rewarded
trial responds as follows: It begins by running slowly to the
initial nonrewarded trial, the progressively increases its run-
ning speed over the successive nonreward trials, until by the
terminal nonrewarded trial, the animal runs about as fast as it
is able. Such responding indicates that the rat is treating the
five trials, four nonrewarded followed by a reward, as a sin-
gle organized whole or a chunk. The third chunk, alist chunk,
consists of the animal’s using a series chunk as a discrimina-
tive stimulus or signal for a subsequent series chunk. For ex-
ample, rats have learned that a particular series of perhaps
three trials (say, two rewards followed by a nonreward)
will signal, some 1 to 20 min later, another distinctive subse-
quent series of, say, three trials. This only reliable signal of the
subsequent series is theinitial series. Under these conditions,
rats have correctly anticipated the trials of the subsequent se-
ries (e.g., running fast, fast, slow, respectively, over the three
trials consisting of two rewards followed by a nonreward).
List chunks indicate that rats possess a fairly high capacity
to organize discrete events into wholes (see Haggbloom
et al., 1992).
An explanation of how chunks are formed in serial tasks
stressesovershadowing(Capaldi, Birmingham, & Miller,
1999). Overshadowing may occur when two stimuli, A and B,
signal some event, X, when one of the stimuli is a more valid
or reliable signal than the other. In a case of this sort the more
valid or reliable signal becomes the stronger signal for X. In a
serial task, item validity may be reduced when similar or iden-
tical items signal different items. For example, in the series
A-B-C-B-D-E, the validity of B is reduced because it signals