372 Conditioning and Learning
An alternative to models aiming for broad generality over
tasks and species is to develop separate models for each task
(e.g., foraging, mating, defense, shelter from the elements)
and species, consistent with the view that the mind is modu-
lar (e.g., Garcia, Lasiter, Bermudez-Rattoni, & Deems,
1985). This approach has been championed by some re-
searchers (Cosmides & Tooby, 1994), but faces challenges
because the resulting models can become very complex and
are limited in their potential to generate unambiguous
testable predictions.
Spatiotemporal Contiguity (Similarity). Despite the
empirical importance of contiguity as a determinant of ac-
quired behavior, it is surprising that many associative models
give short shrift to this critical variable. One common tactic
has been to incorporate contiguity indirectly through changes
in the predictive status of the context that on subsequent trials
modulates the associative status of the cue (e.g., Mackintosh,
1975; Pearce & Hall, 1980; Rescorla & Wagner, 1972). The
associative models that do squarely address the effects of
temporal contiguity are real-time models (see Temporal Win-
dow of Analysis on p. 374; e.g., McLaren & Mackintosh,
2000; Sutton & Barto, 1981; Wagner; 1981).
Objective Contingency. The attenuation of acquired
behavior through degradation of contingency has rarely been
addressed as a unified problem. Most associative models of
acquired behavior have accounted for extinction through
either (a) weakening of the cue-outcome association (e.g.,
Rescorla & Wagner, 1972), or (b) the development of an
inhibitory relationship between the cue and outcome that op-
poses the expression of the initial excitatory association (e.g.,
Hull, 1952; Pearce & Hall, 1980; Wagner, 1981). Attenuated
responding due to partial reinforcement (i.e., nonreinforced
cues interspersed among the cue-outcome pairings) is ordi-
narily explained through mechanisms similar to those used to
account for extinction. The CS-preexposure effect has been
explained both in terms of (a) a decrease in the associability
(attention) to the cue as a result of nonreinforced pretraining
exposure (e.g., Pearce & Hall, 1980); and (b) the development
of a strong context-cue association that attenuates acquisition
of the cue-outcome association (e.g., Wagner, 1981). The
context specificity of the CS-preexposure effect seemingly
lends support to this latter view, but at least one attentional ap-
proach can also accommodate it (Lubow, 1989). Notably,
some prominent models simply fail to account for the CS-
preexposure effect (e.g., Rescorla & Wagner, 1972).
Attenuated responding achieved by degrading contin-
gency through unsignaled USs interspersed among the CS-US
pairings and the US-preexposure effect are both explained by
most associative models in terms of context-outcome associ-
ations, which then compete with the cue-outcome association.
This is consistent with the context specificity of these effects
(i.e., CS preexposure in one context retards subsequent stim-
ulus control during cue-outcome pairings much less if the
preexposure occurred outside of the training context). How-
ever, habituation to the outcome can also contribute to the ef-
fect in certain cases (Randich & LoLordo, 1979). Only a few
associative models can account for reduced responding as a
result of unsignaled outcome exposures after the termination
of cue training (Dickinson & Burke, 1996; Van Hamme &
Wasserman, 1994). However, confirmation of this prediction
is only a limited success because the effect is difficult to
obtain experimentally (see Denniston et al., 1996).Cue and Outcome Durations. Models that parse time
into trials usually account for the generally weaker stimulus
control observed when cue duration is increased by changing
the cue’s associability-salience parameter (e.g., Rescorla &
Wagner, 1972). This mechanism is largely post hoc. Changes
in outcome duration might be addressed in the same manner,
but they have received little attention because results of stud-
ies that have varied outcome duration are mixed, presumably
because the motivational properties of the outcome changed
with the duration of its presentation. A far better account of
cue and outcome durations is provided by real-time associa-
tive models (McLaren & Mackintosh, 2000; Sutton & Barto,
1981; Wagner, 1981). According to these models, the associa-
tive strength of a cue changes continuously when it is present,
depending on the activity of the outcome representation.Reinforcement Theory. For the first 60 years of the
twentieth century, various forms of reinforcement theory
dominated the study of acquired behavior. The history of re-
inforcement theory can be traced from Thorndike’s strong law
of effect (1911; see section entitled “Instrumental Respond-
ing”) through Hull’s several models (e.g., 1952). The basic
premise of reinforcement theory was that learning did not
occur without a biologically significant reinforcer. Although
this view was long dominant, as early as Tolman (1932) there
were objections, often framed in terms of reinforcement’s
having more impact on the expression of knowledge than on
the encoding of it. Although reinforcement during training
may well accelerate the rate at which a cue-outcome relation-
ship is learned, encoding of stimulus relationships does occur
in the absence of reinforcement (unless one insists on making
esoteric arguments that every stimulus about which organ-
isms can learn has some minimal reinforcing value). This is
readily demonstrated in Pavlovian situations by the sensory
preconditioning effect (X→A training followed by A→US