Models of Pavlovian Responding: Theory 369negative retrospective revaluation, in which the change in
control by the target is in direct opposition to the change
produced in the companion during retrospective revaluation.
Examples of negative retrospective revaluation include
recovery from overshadowing as a result of extinction of the
overshadowing stimulus (e.g., Matzel et al., 1985), decreases
in conditioned inhibition as a result of extinction of the in-
hibitor’s training excitor (e.g., DeVito & Fowler, 1987), and
backward blocking (AX→outcome, followed by A→out-
come, e.g., Denniston et al., 1996).
The occurrence of both positive and negative mediation
in retrospective revaluation parallels the two opposing ef-
fects that are observed when the companion cue is treated
before or during the compound stimulus trials. In the section
entitled “Multiple Cues With a Common Outcome,” we
described not only overshadowing but also potentiation,
which, although operationally similar to overshadowing, has
a converse behavioral result. Notably, the positive mediation
apparent in potentiation can usually be reversed by post-
training extinction of the mediating (potentiating) cue (e.g.,
Durlach & Rescorla, 1980). Similarly, the negative media-
tion apparent in overshadowing can sometimes be reversed
by massive posttraining extinction of the mediating (over-
shadowing) cue (e.g., Kaufman & Bolles, 1981; Matzel
et al., 1985). However, currently there are insufficient data
to specify a rule for the changes in control by a cue that will
be observed when its companion cue is reinforced or extin-
guished. That is to say, we do not know the critical vari-
ables that determine whether mediation will be positive or
negative. As previously mentioned (see section titled “Con-
ditioned Inhibition”), the two prime candidates for deter-
mining the direction of mediation are the number of pairings
of the target with the mediating cue and whether those
pairings are simultaneous or serial. Whatever the outcome
of future studies, research on retrospective revaluation has
clearly demonstrated that the previously accepted view—
that the response potential of a cue cannot change if it is not
presented—was incorrect.
MODELS OF PAVLOVIAN RESPONDING: THEORY
Here we turn from our summary of variables that influence
acquired behavior based on cue-outcome (Pavlovian) rela-
tionships to a review of accounts of this acquired behavior. In
this section, we contrast the major variables that differentiate
among models, and we refer back to our list of empirical vari-
ables (see section titled “Factors Influencing Acquired Stim-
ulus Control of Behavior”) to ask how the different families
of models account for the roles of these variables. Citations
are provided for the interested reader wishing to pursue the
specifics of one or another model.Units of AnalysisWhat Is a Stimulus?Before we review specific theories, we must briefly consider
how an organism perceives a stimulus and processes its rep-
resentation. Different models of acquired behavior use dif-
ferent definitions of stimuli. In some models, the immediate
perceptual field is composed of a vast number of microele-
ments (e.g., we learn not about a tree, but each branch, twig,
and leaf; Estes & Burke, 1953; McLaren & Mackintosh,
2000). In other models, the perceptual field at any given mo-
ment consists of a few integrated sources of receptor stimula-
tion (e.g., the oak tree, the maple tree; Rescorla & Wagner,
1972; Gallistel & Gibbon, 2000). For yet other models, the
perceptual field at any given moment is fully integrated and
contains only one so-called configured stimulus, which con-
sists of all that immediately impinges on the sensorium (the
forest; Pearce, 1987). Although each approach offers its own
distinct merits and demerits, they have all proven viable.
Generally speaking, the larger the number of elements as-
sumed, the more readily can behavior be explained post hoc,
but the more difficult it is to make testable a priori predic-
tions. By increasing the number of stimuli, each of which can
have its own associative status, one is necessarily increasing
the number of variables and often the number of parameters.
Thus, it may be difficult to distinguish between models that
are correct in the sense that they faithfully represent some
fundamental relationship between acquired behavior and
events in the environment, and models that succeed because
there is enough flexibility in the model’s parameters to ac-
count for virtually any result (i.e., curve fitting). Most models
assume that subjects process representations of a small num-
ber of integrated stimuli at any one time. That is, the percep-
tual field might consist of a tone and a light and a tree, each
represented as an integrated and inseparable whole.
Worthy of special note here is the McLaren and Mackintosh
(2000) model with its elemental approach. This model not
only addresses the fundamental phenomena of acquired be-
havior, but also accounts for perceptual learning, thereby pro-
viding an account of how and by what mechanism organisms
weave the stimulation provided by many microelements into
the perceptual fabric of lay usage. In other words, the model
offers an explanation of how experience causes us to merge
representations of branches, twigs, and leaves into a com-
pound construct like a tree.