364 Conditioning and Learning
introduced, independent of their immediate location in the
experimental apparatus. The temporal relationship between a
cue and outcome that existed in training is evidenced in two
ways. First, with asymptotic training, the conditioned re-
sponse ordinarily is emitted just prior to the time at which the
outcome would occur based on the prior pairings (Pavlov,
1927). Second, the nature of the response often changes with
different cue-outcome intervals. In some instances, when an
outcome (e.g., food) occurs at regular intervals, during the in-
tertrial interval subjects emit a sequence of behaviors with a
stereotypic temporal structure appropriate for that outcome in
the species’ ecological niche (e.g., Staddon & Simmelhag,
1970; Timberlake & Lucas, 1991).
Pavlovian conditioned responding often closely resembles
a diminished form of the response to the unconditioned out-
come (e.g., conditioned salivation with food as the outcome).
Such a response topology is called mimetic. However, condi-
tioned responding is occasionally diametrically opposed to
the unconditioned response (e.g., conditioned freezing with
pain as the outcome, or a conditioned increase in pain sensi-
tivity with delivery of morphine as the outcome; Siegel,
1989). Such a conditioned response topology is called com-
pensatory.We do not yet have a full understanding of when
one or the other type of responding will occur (but see this
chapter’s section entitled “What Is a Response?”).
Stimulus Generalization
No perceptual event is ever exactly repeated because of vari-
ation in both the environment and in the nervous system.
Thus, learning would be useless if organisms did not general-
ize from stimuli in training to stimuli that are perceptually
similar. Therefore, it is not surprising that conditioned re-
sponding is seen to decrease in an orderly fashion as the
physical difference between the training and test stimuli
increases. This reduction in responding is called stimulus
generalization decrement. Response magnitude or frequency
plotted as a function of training-to-test stimulus similarity
yields a symmetric curve that is called a generalization
gradient(e.g., Guttman & Kalish, 1956). Such gradients
resulting from simple cue-outcome pairings can be made
steeper by introducing trials with a second stimulus that is not
paired with the outcome. Such discrimination training not
only steepens the generalization gradient between the rein-
forced stimulus and nonreinforced stimulus, but often shifts
the stimulus value at which maximum responding is ob-
served from the reinforced cue in the direction away from the
value of the nonreinforced stimulus (the so-called peak shift;
e.g., Weiss & Schindler, 1981). With increasing retention in-
tervals between the end of training and a test trial, stimulus
generalization gradients tend to grow broader (e.g., Riccio,
Richardson, & Ebner, 1984)Phenomena Involving More Than Two Stimuli:
Competition, Interference, Facilitation,
and SummationWhen more than two stimuli are presented in close proximity
during training, one might expect that the representation of
each stimulus-outcome dyad would be treated independently
according to the laws described above. Surely these laws do
apply, but the situation becomes more complex because in-
teractions between stimuli also occur. That is, when stimuli
X, Y, and Z are trained together, behavioral control by X
based on X’s relationship to Y is often influenced by the pres-
ence of Z during training. Although these interactions (de-
scribed in the following sections) are often appreciable, they
are neither ubiquitous (i.e., they are more narrowly parameter
dependent) nor generally as robust as any of the phenomena
described under “Phenomena Involving Two Stimuli.”Multiple Cues With a Common OutcomeCues Trained Together and Tested Apart: Competition
and Facilitation. For the last 30 years, much attention has
been focused oncue competitionbetween cues trained in com-
pound, particularlyovershadowingandblocking.Overshad-
owing refers to the observed attenuation in conditioned re-
sponding to an initially novel cue (X) paired with an outcome
in the presence of an initially novel second cue (Y), relative to
responding to X given the same treatment in the absence of Y
(Pavlov, 1927). The degree that Y will overshadow X depends
on their relative saliences; the more salient Y is compared to
X, the greater the degree of overshadowing of X (Mackintosh,
1976). When two cues are equally salient, overshadowing is
sometimes observed, but is rarely a large effect. Blocking
refers to attenuated responding to a cue (X) that is paired with
an outcome in the presence of a second cue (Y) when Y was
previously paired with the same outcome in the absence of X,
relative to responding to X when Y had not been pretrained
(Kamin, 1968). That is, learning as a result of the initial Y-
outcome association blocks (so to speak) responding to X that
the XY-outcome pairings would otherwise support. (Thus,
observation of blocking requires good responding to X by the
control group, which necessitates the use of parameters that
minimize overshadowing of X by Y in the control group.)
Both overshadowing and blocking can be observed with a
single compound training trial (e.g., Balaz, Kasprow, &
Miller, 1982; Mackintosh & Reese, 1979), are usually greatest
with a few compound trials, and tend to wane with many