376 Conditioning and Learning
signals a reduction in the rate or probability of reinforcement
relative to the baseline occurrence of the reinforcer during
trainingin the absence of the cue. This position avoids the
theoretical quandary faced by the associative views of con-
ditioned inhibition concerning the representation of (a) no-
outcome, or (b) a below-zero expectation of the outcome.
Acquisition Rules. As previously stated (Acquisition-
Focused (Associative) Models), models of acquired behav-
ior must include both acquisition rules and response rules.
In contrast to acquisition-focused models, which generally
have simple response rules and leave accounts of behavioral
differences largely to differences in what is encoded during
training, expression-focused models have simple rules for
acquisition and rely on response rules for an account of
most behavioral differences. Thus, the attenuated respond-
ing to a target cue observed, for example, in a blocking or
contingency-degrading treatment is assumed to arise not
from a failure to encode the target cue-outcome pairings, but
rather from a failure to express this information in behavior.
Accounts of Retrospective Revaluation
In the section entitled “Retrospective Revaluation,” we
described retroactive revaluation of response potential, in
which, after training with a target cue in the presence of other
stimuli (discrete or contextual), treatment of the companion
stimuli (i.e., presentation of a companion stimulus with or
without the outcome) can alter responding to the target cue.
Examples include such mediational phenomena as sensory
preconditioning—in which the mediating stimulus is paired
with the outcome; see section entitled “Second-Order Condi-
tioning and Sensory Preconditioning”—and recovery from
overshadowing as a result of extinguishing the overshadow-
ing cue (e.g., Dickinson & Charnock, 1985; Kaufman &
Bolles, 1981; Matzel et al., 1985).
Expression-focused models that accommodate multiple
cues (e.g., the comparator hypothesis and RET) generally
have no difficulty accounting for retrospective revaluation
because new experience with a companion stimulus changes
its predictive value, and responding to the cue is usually as-
sumed to be inversely related to the response potential of
companion stimuli. Thus, a retrospective change in a cue’s
response potential does not represent new learning about the
absent cue, but rather new learning concerning the compan-
ion stimuli.
In contrast, empirical retrospective revaluation is prob-
lematic to most traditional acquisition-focused models. This
is because these models assume that responding reflects
the associative status of the target cue, which is generally
assumed not to change during retrospective revaluation trials
(on which the cue is absent). But given growing evidence of
empirical retrospective revaluation, several researchers have
proposed models that allow changes in the associative status
of a cue when it is absent. One of the first of these was a re-
vision of the Rescorla-Wagner (1972) model by Van Hamme
and Wasserman (1994), which allows changes in the associa-
tive strength of an absent target cue, provided that some
associate of the target cue was present. This simple modifica-
tion successfully accounts for most instances of retrospective
revaluation, but otherwise has the same failings and suc-
cesses as the Rescorla-Wagner model (see R. R. Miller et al.,
1995). An alternative associative approach to retrospective
revaluation is provided by Dickinson and Burke (1996), who
modified Wagner’s (1981) SOP model to allow new learning
about absent stimuli. As might be expected, the Dickinson
and Burke model has many of the same successes and prob-
lems as Wagner’s model (see section entitled “Where Have
the Models Taken Us?”). A notable problem for these asso-
ciative accounts of retrospective revaluation is that other
researchers have attempted to explain mediated learning
(e.g., sensory-preconditioning and mediated extinction) with
similar models, except that absent cues have an associability
of opposite sign than that assumed by Van Hamme and
Wasserman and by Dickinson and Burke (Hall, 1996;
Holland, 1981, 1983b). Without a principled rule for deciding
when mediation will be positive (e.g., second-order condi-
tioning) as opposed to negative (e.g., recovery from
overshadowing achieved through extinction of the overshad-
owing cue), there seems to be an arbitrariness to this
approach. In contrast, the expression-focused models unam-
biguously predict negative mediation (and fail to account for
positive mediation when it is observed). That is, a change in
the response potential of a companion stimulus is always
expected to be inversely related to the resulting change in the
response potential of the target cue.Where Have the Models Taken Us?As previously noted (in our discussion of acquisition-focused
models), theorists have been able to develop models of
acquired behavior that can potentially account for many
observations after the fact. Any specificmodel can, in princi-
ple, be refuted, but classes of models, such as the families of
acquisition-focused or expression-focused models, allow
nearly unlimited possibilities for future models within that
family (R. R. Miller & Escobar, 2001). If the goal is to deter-
mine precisely how the mind processes information at the psy-
chological level, contemporary theories of learning have not
been successful because viable post hoc alternatives are often