Learning: From Association To Cognition: Further

ANRV398-PS61-11 ARI 5 November 2009 13:5
ANNUAL
REVIEWS Further Learning: From Association
Click here for quick links to
Annual Reviews content online,
including:
to Cognition
• Other articles in this volume
• Top cited articles David R. Shanks
• Top downloaded articles
• Our comprehensive search
Annu. Rev. Psychol. 2010.61:273-301. Downloaded from www.annualreviews.org
Division of Psychology and Language Sciences, University College London,

by University of South Carolina - Columbia on 07/31/13. For personal use only.
London WC1H 0AP United Kingdom; email: d.shanks@ucl.ac.uk
Annu. Rev. Psychol. 2010. 61:273–301 Key Words

First published online as a Review in Advance on association, awareness, blocking, cognition, conditioning, inference,
September 15, 2009
reasoning
The Annual Review of Psychology is online at
psych.annualreviews.org Abstract
This article’s doi: Since the very earliest experimental investigations of learning, tension
10.1146/annurev.psych.093008.100519
has existed between association-based and cognitive theories. Associa-
Copyright c 2010 by Annual Reviews. tionism accounts for the phenomena of both conditioning and “higher”
All rights reserved
forms of learning via concepts such as excitation, inhibition, and re-
0066-4308/10/0110-0273$20.00 inforcement, whereas cognitive theories assume that learning depends
on hypothesis testing, cognitive models, and propositional reasoning.
Cognitive theories have received considerable impetus in regard to both
human and animal learning from recent research suggesting that the key
illustration of cue selection in learning, blocking, often arises from in-
ferential reasoning. At the same time, a dichotomous view that separates
noncognitive, unconscious (implicit) learning from cognitive, conscious
(explicit) learning has gained favor. This review selectively describes key
findings from this research, evaluates evidence for and against associa-
tive and cognitive explanatory constructs, and critically examines both
the dichotomous view of learning as well as the claim that learning can
occur unconsciously.
273
dark days. The present review takes Brewer’s

Contents seminal re-evaluation of the relationship
between learning and cognition as its starting
INTRODUCTION . . . . . . . . . . . . . . . . . . 274
point. One conclusion of Brewer’s assault was
ROLE OF COGNITION
the notion that processes of association forma-
IN LEARNING . . . . . . . . . . . . . . . . . . . 275
tion and reinforcement are largely irrelevant
Blocking: Associative Accounts . . . . . 276
for the understanding of human learning.
Blocking: Cognitive Accounts . . . . . . 277
Another conclusion was that learning invari-
Retrospective Revaluation . . . . . . . . . . 281
ably arises from the operation of conscious
Blocking: Memory for the
cognitive processes. This review reassesses
Blocked Cue . . . . . . . . . . . . . . . . . . . . 283
these fundamental hypotheses about learning
AWARENESS AND LEARNING . . . . 285
in the light of contemporary research.
Eyeblink Conditioning . . . . . . . . . . . . . 285
Associationism has dominated the history
Fear Conditioning . . . . . . . . . . . . . . . . . 286
of learning theory. This prominence originates
Evaluative Conditioning. . . . . . . . . . . . 287

from the work in the early part of the past cen-

The Perruchet Effect . . . . . . . . . . . . . . 289
tury by Edward Thorndike, for whom an un-
Sequence Learning . . . . . . . . . . . . . . . . . 290
derstanding of the reinforcement processes that
Visual Search and Contextual
cause strengthening of S-R connections (e.g.,
Cuing . . . . . . . . . . . . . . . . . . . . . . . . . . 292
the Laws of Effect and Exercise) was a cen-
CONCLUSIONS AND FUTURE
tral concern. Related to his associationist phi-
DIRECTIONS . . . . . . . . . . . . . . . . . . . . 293
losophy, Thorndike also believed that learning
could occur unconsciously or implicitly: he was
the pioneer of what we now know as the field
INTRODUCTION
of implicit learning (Thorndike 1931). Because
It seems unlikely that any investigator would he viewed reinforcing events as being effects
carry out an Instructed Conditioning study that automatically stamped in S-R bonds, rather
with a motor response since, in plain English, than as providing information, Thorndike was
this would consist of telling S to move his fin- eager to show that rewards and punishments
ger. However, common sense is not a good can affect learning even when the individual
guide when predicting the behavior of investi- is unaware that such learning is taking place,
gators operating in an S-R framework. Exper- and he conducted numerous studies (reviewed
S-R: stimulus- iment 6 in Hunter (1938) consists of produc- by Postman 1962) attempting to establish this.
response
ing finger withdrawal conditioning by saying, Thus Thorndike combined two positions, as-
Associationism: the “Lift your finger,” or “Don’t lift your finger.” signing a key role to reinforcement processes
assumption that
If Ss didn’t make a finger withdrawal response in learning together with the claim that these
learning can be
understood in terms of to “Lift your finger,” they were shocked. To in- learning processes can proceed independently
the formation and sure objectivity the commands were presented of awareness.
expression of by telephone. The Ss conditioned. (Brewer A very different view of learning can be
excitatory or inhibitory 1974, p. 14) traced back to the Gestaltists (especially Köhler
associations, formed
via reinforcement
and Koffka) and field theorists (particularly
Brewer’s lampooning of studies conducted Tolman) who took an opposing position on
Implicit learning:
synonymous with
within the framework of stimulus-response the effect versus information issue. For them,
unconscious learning. (S-R) behaviorism, on the grounds that they events did not provide the opportunity for
Also termed ignored the ubiquitous involvement of con- gradual trial-and-error learning, but rather set
“procedural” in scious thinking in learning and memory, will the context for the development of insight and
contrast to doubtless make many modern researchers problem solving. Tolman in particular argued
“declarative”
(conscious) learning
reflect with amusement and relief on how far that reinforcement is not a necessary condition
experimental psychology has come since those for learning; instead, cognitive relationships
274 Shanks
(hypotheses or cognitive maps) are formed be- contributions from the areas of associative and
tween stimuli or events as a result of contiguity contingency learning, and implicit learning
and organizational principles (see the review and memory. A guiding question is whether our
Conditioning: in
of learning by Melton 1950, in the very first explanatory scope for explaining the richness classical or Pavlovian
Annual Review of Psychology). Debate between of learning would be seriously curtailed if con- conditioning, a
S-R and field theories, and the importance of cepts such as excitation and reinforcement were conditioned stimulus
phenomena such as latent learning, dominated abandoned. Plainly, it is not sufficient to reject (CS; e.g., a tone)
predicts an
much of experimental psychology up until the traditional S-R theory. Instead, cognitive the-
unconditioned
cognitive revolution (see Holland 2008). ories that dispense with associative constructs stimulus (US; e.g.,
By the time Brewer (1974) came to review must be contrasted with those modern theories food or shock). As a
the literature, it was abundantly plain that the that incorporate cognitive constructs such as result of learning, the
traditional S-R theory of conditioning, includ- attention and awareness while also assigning response normally
evoked by the US
ing core notions such as the automatic and a fundamental role to association formation.
(e.g., salivation or
unconscious stamping in (i.e., reinforcement) of Extraordinarily rich explanations of learning
freezing) or a similar
associative links, was not appropriate for either phenomena have been achieved by models built response comes to be
classical or operant (instrumental) condition- out of automatic link machinery (i.e., connec- evoked by the CS
ing, nor did it provide an adequate framework tionist, neural network, or parallel distributed CS: conditioned
for such important applied problems as processing accounts) in which representations stimulus
understanding the development of phobias are coded subsymbolically (McClelland & US: unconditioned
(Rachman 1977). Evidence that conditioning Rumelhart 1986, Rumelhart & McClelland stimulus
of autonomic, motor, as well as more complex 1986, Thomas & McClelland 2008). Such
responses only occurs in parallel with expectan- models demonstrate massive “emergentism,”
cies and awareness convinced Brewer that all in that processes that seem cognitive and high
conditioning is the result of cognitive process- level emerge from the operations and inter-
ing, in particular of the formation and testing actions of very elementary processing units.
of conscious hypotheses. Although tension has These processes yield knowledge structures
existed between associative and cognitive views and states of activation which, when sufficiently
of learning for a century, the topic has received strong and stable, constitute the contents of
renewed attention in the past few years. The consciousness (see Maia & Cleeremans 2005).
considerable evidence that awareness necessar- It is against this contemporary associationist
ily accompanies learning has been challenged framework that the present review compares
recently as the study of implicit learning has the alternative propositional-cognitive theory.
gained momentum. The present article reviews
some of the current evidence from this area
and asks whether Brewer’s position remains ROLE OF COGNITION
the most viable interpretation: Is learning IN LEARNING
intrinsically a conscious process? On the other Should we think of learning as the automatic
hand, much recent research has retained the formation of a mental link or bond between a
strongly antiassociationist perspective, which cue (or CS) and an outcome (or US), or instead
Brewer championed, regarding concepts like as the acquisition of a propositional belief
reinforcement as superfluous. Recent evidence representing the relationship between them?
bearing on this issue is also reviewed. One counterintuitive but well-documented
In exploring the relationship between empirical observation forcefully captures the
associative and cognitive processing, this paradox of learning and cognition. This is
article examines a fundamental theme that has the finding that verbal instructions seem to
remained unresolved since the early days of be largely interchangeable with experienced
experimental psychology. Recent research is re- events. The mere instruction that a tone will
viewed, mostly published since 2000, covering be followed by shock is sufficient to cause an
www.annualreviews.org • Learning 275

increase in skin conductance that is indistin- history of learning theory and has been central
guishable from that obtained when the tone in the debate between proponents of associative
is actually paired with shock (e.g., Cook & and cognitive theories. Blocking is the funda-
Blocking: the
fundamental Harris 1937, McNally 1981). In more recent mental demonstration that learning about the
demonstration of cue research, Lovibond (2003) has shown that relationship between two events depends on not
selection in learning. this interchangeability extends to more subtle just their frequency of pairing, but also on the
When a pair of cues A designs including blocking (described below). extent to which one provides information about
and B predicts an
The paradox is that if true conditioning, with the other. Consider the pairing of a cue and an
outcome, the degree of
learning about the events that are actually experienced, engages outcome, or of a conditioned stimulus (CS) with
A-outcome (Brewer notwithstanding) an automatic and an unconditioned stimulus (US). When cues A
relationship is reduced unconscious learning mechanism, then how and B are paired together and predict an out-
(blocked) if B has can verbal, cognitive, conscious instructions come, the extent to which learning accrues to
previously on its own
make contact with that mechanism? Contrary A is diminished if B has previously, in the ab-
predicted that
to commonly held belief, we seem to be faced sence of A, been paired with the outcome (see
outcome
with the conclusion that conditioning in fact Table 1). Thus blocking refers to a two-stage
gives rise to conscious, cognitive, propositional design in which B is established as a reliable pre-
representations rather than to automatic, un- dictor of the outcome in stage 1 (denoted B+),
conscious ones. This is precisely the position with A and B occurring together and jointly pre-
adopted by many contemporary researchers dicting the outcome in stage 2 (denoted AB+).
(De Houwer 2009, Mitchell et al. 2009), in As a concrete example, laboratory studies with
which associative processes are jettisoned. human participants might describe a hypothet-
In the past few years, the cognitive approach ical individual suffering from allergic reactions
(now commonly referred to as propositional to some foods but not others. On day 1 the indi-
or inferential) has kindled renewed attention vidual eats tomatoes and suffers an allergic re-
and emphasis. To what extent, then, do cogni- action, and on day 2 she eats tomatoes together
tive processes penetrate learning? Should con- with pasta and again suffers a reaction. The par-
structs such as reinforcement and excitation ticipant’s task is to judge the extent to which
play any role in our explanatory frameworks? pasta is associated with the reaction, and block-
ing refers to the fact that the initial tomatoes-
allergy pairing will weaken the perceived pasta-
Blocking: Associative Accounts allergy connection.
The signature phenomenon of blocking Standard associationist accounts of block-
(Kamin 1969) has played a key role in the ing rely on concepts such as excitation,
Table 1 Experimental designs relating to blocking

Condition Pretraining Stage 1 Stage 2 Test Comment
Blocking B+ AB+ A B+ trials reduce responding to A at test
Blocking control AB+ A Responding to A at test is strong
Backward blocking AB+ B+ A B+ trials reduce responding to A at test
Subadditivity C+/D+/CD+ B+ AB+ A Pretraining reduces blocking (i.e., enhances
responding to A)
Additivity C+/D+/CD++ B+ AB+ A Pretraining enhances blocking (i.e., reduces
responding to A)
Maximality + B+ AB+ A Pretraining reduces blocking (i.e., enhances
responding to A)
Submaximality ++ B+ AB+ A Pretraining enhances blocking (i.e., reduces
responding to A)
Note: A–D: cues or conditioned stimuli. + indicates occurrence of the outcome or unconditioned stimulus. ++ indicates an outcome of larger magnitude.
276 Shanks
reinforcement, associability, and surprise ciative theories have placed great emphasis on
(Dickinson 1980, Mackintosh 1983, Rescorla attentional, controlled processing and related
& Wagner 1972). In the classic Rescorla- concepts such as limited-capacity working
Wagner theory, for instance, an excitatory memory (Mackintosh 1975, Pearce & Hall
association is assumed to form in stage 1 1980, Wagner 1981), and the role of attention in
between B and the outcome, with the latter human as well as animal conditioning has been
playing the role of reinforcer. In stage 2, recognized for many years (Dawson & Schell
the reinforcing power of the outcome is 1982). The important question, of course, is
diminished because its occurrence is no longer whether cognitive concepts such as these, and
surprising—it is predicted by the presence of the evidence for their role, ultimately require
cue B. Thus the outcome does not serve as a abandoning the associationist perspective.
reinforcer of the A-outcome association, and
little learning accumulates to A. This theory,
whose history and influence is reviewed by Blocking: Cognitive Accounts
Miller et al. (1995), has played such a central Waldmann & Holyoak (1990, 1992) were the
role in recent learning theory that it has in instigators of the cognitive challenge to the as-
effect acted as the departure point for almost sociationist account of blocking. Their key ob-
all subsequent theories. Its influence arose in servation was that associations have no seman-
considerable part from its ability to explain the tic properties, whereas human learning seems
blocking effect Kamin discovered. to be highly sensitive to such properties. For
Although subject to numerous theoretical instance, on the associationist theory, it should
subtleties (Mackintosh 1975, Pearce & Hall not matter whether cues and outcomes are de-
1980), this explanatory framework is supported scribed as causes and effects, respectively, or
by a wealth of evidence (Mackintosh 1983), in- as effects and causes. Imagine an experimen-
cluding such diverse findings as that learning is tal task in which the participant is shown in-
retarded for a blocked cue that is subsequently formation about substances in the blood of a
paired with an entirely new outcome (Le hypothetical patient together with information
Pelley et al. 2007), that blocking occurs in or- about whether the patient is suffering from a
ganisms not generally assumed to be endowed target disease. From an associative perspective,
with cognitive capacities (including the marine the task involves learning connections between
mollusc Hermissenda; Rogers & Matzel 1996), the substance (cue or CS) and the disease (out-
and that the dopamine system in the brain ap- come or US). However, the substance could be
pears to provide a reinforcing mechanism with described as a cause of or as an effect of the dis-
just the required formal properties (Fletcher ease, and Waldmann & Holyoak (1990, 1992)
et al. 2001, Waelti et al. 2001). Although the showed that from a rational perspective these
modern associationist perspective differs in nu- interpretations can have radically different im-
merous ways from the classic S-R theories of plications. Indeed, they argued that blocking
Tolman, Guthrie, Hull, and others, in its fun- would only be expected in a cause-effect sce-
damental form it incorporates many of the con- nario and not in an effect-cause scenario, and
cepts Brewer was so determined to reject (for a they presented experimental evidence consis-
review of the intellectual transition from early tent with this prediction.
to contemporary learning theories, see Mowrer Later research has confirmed that the inter-
& Klein 2001). pretation of events, over and above their simple
It is also important to note that associa- pairing, can substantially modulate blocking,
tive models of learning have often inaccurately but this research has also shown that learning
been described as assuming automatic trans- is sometimes quite immune to the way events
fer of activation from the CS representation to are described (e.g., Cobos et al. 2002, López
the US representation. In reality, modern asso- et al. 2005, Waldmann 2000). For example,

López et al. (2005) presented participants with AB+ in stage 2). It has been known for many
a simulated device in which switches and lights years (Dickinson et al. 1984) that associative or
on one side of a box were connected to lights causal judgments for A will be low in compar-
on the other side of the box. Participants ei- ison to a control condition in which the initial
ther saw trials in which the cue played the role B+ trials are omitted (see Table 1). But studies
of a cause and the outcome the role of an ef- have also shown that judgments of the proba-
fect, or vice versa. The main finding was that bility of the outcome given cue A will manifest
this causal interpretation had no effect except a blocking-like effect (De Houwer et al. 2007,
in conditions where López et al. (2005) went to Lagnado & Shanks 2002, López et al. 1998,
extreme lengths to emphasize the causal nature Price & Yates 1993). This is a striking result
of the task to participants. Thus there do seem because the probability of the outcome given
to be conditions in which the associationist the- cue A, P(O/A), is unaffected by whether or not
ory provides an appropriate account of partici- B+ trials are presented in stage 1. Associative
pants’ behavior, although we are some way from theories explain this effect by proposing that
a full understanding of the boundary conditions A’s associative weight, which is reduced in the
for such an account. blocking condition, is part of the evidence used
In related work, Matute and her colleagues to make a probability judgment. Even though
(Matute et al. 2002, Vadillo et al. 2005) have ex- the probabilities are objectively the same in the
tensively studied aspects of the precise format of two conditions, the cue much more strongly
questions probing associative knowledge. Sup- makes us think of the outcome in the control
pose participants observe trials in which vari- than in the blocking condition, and this men-
ous medicines are related to allergic reactions. tal activation unavoidably biases our probability
Probe questions might ask to what extent a judgment. Explaining the bias in terms of infer-
medicine causes the allergic reaction, the extent ential processes appears difficult. Similar effects
to which it predicts or indicates the reaction, emerge in terms of memory for the relation-
or the extent to which medicine and allergy ship between a blocked cue and its associated
co-occur. Contrary to the simple idea that a outcome (discussed below).
single associative connection underlies all such De Houwer et al. (2002) took a different
judgments, these studies reveal qualitative dif- approach to explore cognitive influences in
ferences between different judgment questions. blocking. They proposed that blocking arises
For example, Vadillo et al. (2005) found that from a chain of reasoning from the premise
causal and prediction judgments could be dis- “the outcome is as probable and intense after B
sociated. Yet as Vadillo et al. themselves noted, a as after AB” to the conclusion “therefore A is
slightly more subtle associative account would not a cause of the outcome.” They also noted,
have no difficulty accounting for this finding. however, that such an inference is valid only if
Such an account would assume that causal judg- the effect is not occurring at its maximal pos-
ments are assessed by simply probing the asso- sible level. The inference only follows if there
ciative strength of the target cue on its own, is room for the outcome to occur with greater
while prediction judgments are based on the probability or intensity. To test this analysis,
combined associative strength of the target cue they described the outcome in a blocking
plus the context. design as occurring with an intensity of 10
Recall that the essence of associative ap- on a scale from 0–10 (maximal condition). De
proaches is that they assume that presentation Houwer et al. (2002) speculated that if blocking
of a cue calls to mind (given sufficient atten- arises as a result of reasoning, rather than from
tional resources and so on) the outcome with associative processes, then their participants
which it was associated. Now consider a situ- should be unsure about the status of the blocked
ation in which a cue competition effect such cue A under these conditions. In stage 1 they
as blocking occurs (B+ in stage 1 followed by learned that cue B predicted the outcome, and
278 Shanks
then in stage 2 they observed A and B occur- effects in conventional conditioning procedures
ring together with the same outcome. Such with rats and concluded that similar inferential
information is insufficient to distinguish reasoning processes and “remarkable cognitive
Maximality:
between A having no relationship with the abilities” (p. 100) are employed. In their stud- information (e.g.,
outcome versus it having a strong relationship ies, both subadditivity (C+, D+, CD+) and ad- 20/20) implying that
that is masked by a ceiling effect. If B is already ditivity (C+, D+, CD++, where ++ denotes an outcome is
causing the outcome with maximal intensity, a US with higher magnitude) pretraining in- occurring at its ceiling
level. Contrast with
then A cannot increase or alter that outcome, fluenced blocking, the former reducing it and
submaximality (e.g.,
even if it does have a causal relationship with the the latter enhancing it. Moreover, maximality 10/20)
outcome. De Houwer et al. (2002) contrasted pretraining also had an influence: specifically,
Additivity: trials (e.g.,
the maximal condition with a submaximal one a pretreatment in which the animals received C+, D+, CD++,
in which the same events were presented, but unsignaled large magnitude (++) as well as where ++ denotes a
the outcome was always described as occurring smaller (+) shocks led to an enhancement of high-magnitude
with intensity 10/20. Hence, in this case the blocking, consistent with the inference that the outcome) that imply
that the predictive

failure of A to change the magnitude of the blocked cue could have, but did not, increase
value of cues adds
outcome in stage 2 is significant, as it could the magnitude of the US and therefore must when they are
have done so: The outcome still has room have been an ineffective cue. combined. Contrast
to increase in magnitude, but did not do so. Are these findings inconsistent with associ- with subadditivity
Consistent with this analysis, De Houwer et al. ationist accounts, as has often been claimed? (e.g., C+, D+, CD+)
(2002) observed much stronger blocking in Perhaps the presence of the effects in laboratory Inferential theory:
the submaximal condition and argued that rats should lead us to look particularly carefully assumes that cue
selection effects are
this finding is fundamentally inconsistent with for simpler explanations. And indeed there is a
the result of inferences
associative theories of blocking. strong counterargument that additivity effects from beliefs. In
In a related modification to the standard might have a rather different basis, at least in blocking, for example,
blocking design (see Table 1), a number of au- some circumstances. Wheeler et al. (2008) re- little weight is assigned
thors (Beckers et al. 2005, Lovibond et al. 2003, ported an additivity pretraining effect, namely to the blocked cue
because participants
Mitchell & Lovibond 2002) have given their that subadditivity pretraining attenuated block-
assume cue additivity
participants pre-exposure designed to confirm ing. Specifically, they presented rats in a fear- and reason that the cue
or contradict the idea that cues have additive conditioning preparation with trials in which does not increase the
causal value. If participants believe cues are ad- C, D, and the compound CD all signaled the magnitude of the
ditive, then the failure of cue B to cause an US (C+, D+, CD+), thus providing informa- outcome
increase in the outcome’s magnitude indicates tion that the causal consequences of combin-
that it is ineffective, and blocking should be sub- ing cues were not additive. Then in the main
stantial. However, if they believe that cues are training stage the rats received B+ followed by
subadditive (i.e., that two cues, each of which AB+ trials in a typical blocking arrangement
causes an outcome with magnitude M, do not prior to an assessment of conditioned respond-
produce an outcome of magnitude 2M when ing to the target CS A. Wheeler et al. (2008)
paired together), then the failure of B to in- found that the subadditivity pretreatment sub-
crease the outcome’s magnitude is not evidence stantially weakened the blocking effect—that
that B lacks any causal power. In that event, is, enhanced learning about A—consistent with
blocking should be much weaker in the subad- the inferential theory. The subadditive trials in-
ditive condition, and this is the pattern com- dicate that the outcome of the combination of
monly observed (Beckers et al. 2005, Lovibond two reliable cues is not additive, and hence in
et al. 2003, Mitchell & Lovibond 2002). the main stage of the experiment the animals
These maximality and additivity effects have might have reasoned that A was a perfectly valid
been the subject of extensive further research predictor of the US, despite not increasing the
since their original discovery. For instance, US’s magnitude. Yet Wheeler et al. (2008) went
Beckers et al. (2006) have reported additivity on to show that their subadditivity pretreatment

was sensitive to a number of manipulations Schmajuk & Larrauri (2008) also tested a
typically found to impair generalization. For connectionist model with the events of a typical
example, a change of context from the pretreat- additivity design. In their model, associative
ment to the blocking phase reduced the impact connections are updated by a reinforcement
of the subadditivity trials. This suggests the pos- process, and the connections have no sym-
sibility that the effect emerges not because of bolic reference. Although different from
controlled inference, but rather because the Haselgrove’s (2009) model in some important
pretreatment allows the formation of some as- respects, it also reproduced the additivity
sociative structures that generalize (or fail to effect. Schmajuk & Larrauri (2008) conclude
generalize under a context switch) to the block- that it is not necessary to interpret additivity
ing phase. effects in terms of propositional reasoning.
The nature of this putative generalization Another line of research has considered the
mechanism has been explored by Haselgrove possibility that maximality and additivity effects
(2009). One common way of building gen- have their influence not via changing hypothe-
eralization into associative theories is to as- ses and reasoning about cue combinations but
sume that stimuli are composed of multiple el- rather by inducing shifts between elemental and
ements and that some of these elements may configural processing. There is now a consider-
be shared between nominally distinct stimuli. able body of evidence suggesting that organisms
Consider the additivity design employed by (both human and nonhuman) can represent
Beckers et al. (2006) and Wheeler et al. (2008) stimuli in rather flexible ways (Melchers et al.
in which blocking of A after B+ and AB+ trials 2008). In particular, the same nominal stimulus
is attenuated if subadditivity is demonstrated in can be coded either as an irreducible configu-
a previous learning phase with C+, D+, and ration or as the sum of a set of elements. For
CD+ trials. Haselgrove (2009) noted that most instance, in a human Pavlovian conditioning
of the stimuli used in these experiments came study by Melchers et al. (2004a), participants
from the same modality (auditory) and sug- initially saw either a feature-neutral discrim-
gested that this might enhance generalization ination or a matched control discrimination.
between them. To model this, Haselgrove con- The control discrimination (A+, AB+, C−,
ducted a simple simulation of this experimen- CB−) afforded an elemental solution: Correct
tal design using the Rescorla-Wagner theory, responses are made on all trials if the participant
but assuming that one additional element or learns a strong positive association between cue
cue was present on every trial. Surprisingly, this A and the US, with all other cues having associa-
simple model reproduced the key finding that tive weights of zero, and assuming additivity of
the pretreatment stage reduced blocking, and cue weights. Despite being almost identical, the
the reason for this was that the common ele- feature-neutral discrimination (A−, AB+, C+,
ment accrued asymptotic associative strength in CB−) cannot be solved this way; there is no set
the pretreatment stage, such that when A was of weights for the individual elements that yield
presented in the blocking test (with the com- correct responses across all trial types. Hence
mon element assumed also to be present), a high this problem requires a more complex solution,
level of responding was predicted. Haselgrove such as the formation of configural representa-
(2009) also showed that the model predicted tions in which AB and CD are represented as
the contrasting effect of an enhancement of distinct from the sum of their constituent parts.
blocking after additivity pretraining (i.e., C+, To test the hypothesis that participants in-
D+, CD++), as observed by Beckers et al. deed solved these discriminations in qualita-
(2006), as well as after submaximality pretrain- tively different ways, Melchers et al. (2004a)
ing, and concluded that the inferential basis for next gave both groups a new discrimination
the additivity effect, at least in rats, is far from (EX+, FX−, followed by test trials with E
proven. and F) and found very different behavior as a
280 Shanks
function of the initial problem. Specifically, examples of retrospective revaluation. In a

greater responding to E than to F was evident backward blocking design (Table 1), trials in
in the group pretrained on the control discrim- which A and B are paired with the outcome pre-
Retrospective
ination but not in the one pretrained on the cede B-outcome trials. According to traditional revaluation: indirect
feature-neutral one. Such a pattern is consis- theories (Rescorla & Wagner 1972), blocking change in a cue’s
tent with the hypothesis that the group trained (reduced judgments of the relationship between associative strength
on the control discrimination transferred an el- A and the outcome) should not occur under resulting from later
information. An
emental strategy to the new problem, broke the such circumstances; yet numerous studies have
example is backward
EX and FX compounds into elements, and as- confirmed that it does. The inferential expla- blocking (when B+
signed a positive weight to E and weights of nation is that individuals reason just as they do trials follow AB+ ones,
zero to F and X, whereas the group trained on in forward blocking. B and AB predict the same they cause a reduction
the feature-neutral discrimination transferred outcome, thus they conclude that A is not an in responding to A)
a configural strategy in which compounds and independent cause.
elements are treated as independent. But once again the main findings are not
Building on this idea, Livesey & Boakes incompatible with associationism. In an elegant
(2004; see also Williams et al. 1994) reported analysis, for instance, Ghirlanda (2005) showed
that a range of manipulations thought to en- that a range of retrospective effects including
hance elemental processing increased blocking backward blocking are in fact entirely compati-
(in much the same way as additivity instruc- ble with the Rescorla-Wagner theory, provided
tions), whereas manipulations assumed to en- a suitable scheme for stimulus representation
hance configural processing decreased blocking is employed. Traditional applications of the
(as with subadditivity instructions). Strikingly, theory simply assume that each stimulus is rep-
by presenting the cues in a configural manner, resented by a single unit or node in a network.
Livesey & Boakes (2004) were able to elimi- Under such circumstances, it is true that back-
nate blocking even when an additivity pretreat- ward blocking cannot be predicted. However,
ment stage was provided. Such a result raises the Ghirlanda (2005) showed that this conclusion
strong possibility that maximality and additiv- does not hold when each cue is instead rep-
ity effects have some of their influence via shift- resented via a pattern of activity distributed
ing the balance between elemental and config- over a large number of units. Ghirlanda
ural processing. However, whether this account (2005) reported simulations in which arrays of
can explain all of the relevant results is unclear. 50 units coded each elemental stimulus, yield-
Beckers et al. (2005) found, for example, that ing backward blocking (reduction of judgments
additivity information affected blocking even for A after AB+, B+ training) as well as its con-
when it was presented after the target trials, an verse, unovershadowing (increase in judgments
outcome difficult to reconcile with the idea that for A after AB+, B− training) and backward
the locus of additivity effects resides solely in conditioned inhibition (negative judgments
their influence on the way the blocking stimuli for A after AB−, B+ training). A key element
are coded. Lastly, a subtractivity pretreatment of distributed coding schemes is that they
employed by Mitchell et al. (2005b) affected allow distinct cues to activate overlapping sets
blocking in the way predicted by a reasoning of elements, thus capturing the fundamental
account. principle of stimulus generalization. Kruschke
(2008) has reported similar theoretical de-
velopments for dealing with retrospective
Retrospective Revaluation revaluation, but in the context of models
Cognitive interpretations have also been of- driven by Bayesian considerations. Kruschke
fered for the phenomenon of backward block- assumes that knowledge is not captured by a
ing (Dickinson & Burke 1996, Shanks 1985, single strength of association, but instead by a
Van Hamme & Wasserman 1994) and other distribution of degrees of belief over a range of

hypotheses, with these beliefs being optimally reported that retrospective changes are related
adjusted in response to learning feedback. to the strength of the critical within-compound
Another prominent development is association. In another set of tests, Melchers
Dickinson & Burke’s (1996) proposal that the et al. (2004b, 2006; see also Vandorpe et al.
formation of a within-compound association 2007) found that the magnitude of the ret-
between cues A and B in a backward blocking rospective change was related to participants’
design plays a key role in the retrospection memory for the trial pairings and proposed that
effect. In the first stage, these cues are com- it is only if the AB pairing can be recalled that
bined and jointly predict the outcome (AB+). downward adjustments in A’s strength can be
It is assumed that participants not only learn induced. Importantly, Melchers et al. (2004b,
associations between each of the cues and the 2006) and Vandorpe et al. (2007) also found
outcome, but also between the cues themselves. that there was no corresponding correlation be-
In the second stage, B is presented in isolation tween forward blocking and participants’ mem-
and predicts the outcome (B+). Dickinson ory for the compound trials. This is as expected
& Burke’s (1996) revised associative theory by standard associative theories, because for-
proposes that B activates the representation of ward blocking (B+ followed by AB+) is driven
A in the second stage via the within-compound simply by competition between A and B on the
association formed in the first stage. While AB trials. Any association between A and B is
the physically presented cue, B, undergoes a irrelevant for such competition.
normal increment in associative strength when As noted by Mitchell et al. (2005a) and
the outcome is presented, the associatively ac- Vandorpe et al. (2007), however, these retro-
tivated representation of A undergoes a change spective revaluation effects are also consistent
of the opposite polarity. Hence the B+ trials with cognitive accounts. Having observed AB+
lead to a reduction in A’s associative strength. trials, participants can infer that A has some
Learning is governed by associative processes predictive value, but when they subsequently
and by reinforcement as conceptualized in the observe B+ trials, they should now alter that
traditional theory. conclusion because they have information
Melchers et al. (2004b) proposed an alterna- indicating that A adds nothing beyond what B
tive account based more on memory retrieval signals. Such a chain of inference assigns an im-
processes than on associative activation. On portant role to memory in that the adjustment
this account, participants access trial types from of belief about A requires recollection of the AB
memory and replay them mentally [a similar no- pairing. On the other hand, forward blocking
tion is central to McClelland et al.’s (1995) con- is not expected to be similarly dependent on
nectionist theory of the relationship between memory. In this case (B+ followed by AB+),
the hippocampal and neocortical memory sys- individuals can infer during the AB trials that
tems]. Hence AB+ trials are recalled and re- A is nonpredictive so long as they remember
hearsed during stage two when B+ trials are the B+ trials, and there is no necessity for
observed. If such replayed trials function much memory of the conjunction of A and B. Hence
like experienced trials, then backward block- the asymmetry observed by Melchers et al.
ing would emerge as a simple result of the on- (2004b, 2006) and Vandorpe et al. (2007),
going adjustment of associative strengths and where backward but not forward blocking
would not be the product of true reasoning. correlates with memory for the cue compound,
The concepts of within-compound associations is compatible with a cognitive account.
(Dickinson & Burke 1996) and memory-based Indeed, Mitchell et al. (2005a) and Vandorpe
rehearsal (Melchers et al. 2004b) are obviously et al. (2007) have argued that the patterns of
closely related. behavior in the retrospective case are better ex-
These associative accounts are supported plained by inference than by associative mech-
by a range of evidence. Aitken et al. (2001) anisms. In an ingenious experiment, Mitchell
282 Shanks
et al. (2005a) succeeded in reversing the & Beckers (2002a,b). The occurrence of the
normal backward blocking effect such that B+ allergy in the presence of banana (B+) led to
trials presented subsequent to AB+ ones actu- a concomitant increase in the judged relation-
ally caused an increase rather than a decrease ship between cheese (C) and the allergy, while
in ratings for cue A. They achieved this re- conversely its nonoccurrence led to a decrease.
sult by employing a method akin to the ad- Although it is true that such a result might sig-
ditivity manipulation described previously, in nal the importance of reasoning in blocking-
which they pretrained participants to believe like designs, Melchers et al. (2004b) found that
that compounds predictive of the outcome must the magnitude of the retrospective change in
be composed of elements with equivalent pre- second-order blocking was again related to par-
dictive values. Mitchell et al. (2005a) thus con- ticipants’ memory for the trial pairings and ar-
cluded that their participants solved the task via gued that the data were therefore compatible
deliberate inference, akin to that observed in with an alternative account based on associa-
other rule-learning designs (Shanks & Darby tive reinforcement combined with rehearsal of
1998). Associative approaches might seek to ex- previous trial types.

plain this result in terms of a shift between ele-
mental and configural processing. The reversal
of blocking observed by Mitchell et al. (2005a) Blocking: Memory for the
is consistent with configural-based associative Blocked Cue
theory, assuming there was some generalization The role of memory in cue learning has been
between the A and B cues. central to the evaluation of another prediction
An even more striking empirical example of that distinguishes associative and cognitive
backward blocking is the second-order effect models. As noted above, the classic associative
reported by De Houwer & Beckers (2002a,b; analysis of blocking (B+ trials followed by AB+
Melchers et al. 2004b). In their three-stage pro- trials) proposes that the accumulation of asso-
cedure, participants first saw cues AB paired ciative strength by B in the first stage endows it
with the outcome (AB+) and then were shown with the capacity to compete more effectively
trials in which cues AC were paired with the with cue A in the compound stage. This com-
outcome (AC+). Finally, trials were presented petition can arise because the outcome, being
in which B occurred either with or without unsurprising, is not processed (Rescorla &
the outcome (B+ or B−). To give a concrete Wagner 1972) or because cue A is only
illustration, imagine that the foods avocado weakly attended to and hence not processed
and banana cause an allergic reaction and that, (Mackintosh 1975, Pearce & Hall 1980). Ei-
subsequently, avocado and cheese also cause ther way, the consequence of these processing
the allergy. Finally, banana in isolation either failures is that only a very weak association (if
causes or does not cause the allergy. Consider a any) is formed between cue A and the outcome,
participant for whom B was paired with the out- resulting in low judgments of the A-outcome
come in the final stage. On a cognitive reason- association. A clear prediction from such an
ing account, such a participant might be ex- analysis is that there should be a further man-
pected to infer that since B is a good predic- ifestation of blocking, namely poor memory
tor, then A must be a poor predictor: This is for the outcome paired with A. Thus not only
a standard backward blocking effect. However, should participants judge the association to be
the participant might go on to infer that if A is weak, but they also should find it hard to recall
a poor predictor, then C is a good one. Thus, that the outcome co-occurred with A because
even though B is never paired with C, the final that cue will fail to activate the outcome
stage trials (B+ or B−) might alter the attri- representation associatively. The opposite
bution of predictive significance to C, and this prediction, however, follows from inferential
is exactly the result observed by De Houwer accounts. Indeed, these must assume that the

A-outcome relationship is strongly recalled. accounts. The basic extinction result (on con-
Participants are assumed to reason roughly as tingency judgments) can be explained very eas-
follows: “If A predicts the outcome, then the ef- ily by assuming that participants compute some
fect of AB would have been greater than that of measure of the statistical contingency between
B alone. B and AB predicted the same outcome, the cue and the outcome. For example, an in-
thus A is not predictive.” Intrinsic to this chain ferential account might incorporate rules based
of inference is that participants must remember on the metric P [defined as the difference be-
that AB predicted the outcome, which is to say tween the probability of the outcome given the
they must code the fact that A (in conjunction cue, P(O/C), and the probability of the out-
with B) was paired with the outcome. come in the absence of the cue, P(O/∼C)],
This hypothesis has been tested in recent which would yield a lower measure for an ex-
experiments. Mitchell et al. (2006) devised a tinguished cue compared to a control cue be-
blocking task with many different foods as the cause the extinction (A−) trials reduce P(O/C).
cues and with allergies as the outcomes, such (For discussion of the possible psychological re-
that recall could be tested. In addition to reality of these and similar rules, see Cheng 1997;
vealing blocking on judgments, the results were Cheng & Holyoak 1995; Perales & Shanks
clear in showing that recall of the outcome as- 2007, 2008). However, such an account pro-
sociated with cue A was poor in comparison vides no reason to anticipate poor memory for
with appropriate control cues. This pattern is the initial A+ pairings. Indeed, to the extent
exactly as predicted by the associative account. that the calculation of P(O/C) requires a record
A further important aspect of Mitchell et al.’s of the history of trial outcomes, good recollec-
(2006) experiments was that these results were tion of the pairings would seem to be necessary.
obtained in a task in which subadditivity pre- Of course, inferential accounts could be supple-
training was employed. Recall that such pre- mented with additional processes to accommo-
training ought to eliminate or at least reduce date memory failure, but these processes would
blocking. In providing evidence that the out- have to include precisely the sorts of interfer-
come is no larger following a compound of ence mechanisms that associative theory has de-
predictive cues than with the cues in isolation, voted much effort to developing (Bouton 1993).
subadditivity pretraining leaves the predictive Whereas the Mitchell et al. (2006) study de-
status of B ambiguous. The fact that forward scribed above assessed memory for the extent to
blocking (on both predictive judgments and re- which a blocked cue can prompt recall of its as-
call measures) was observed thus seems to offer sociated outcome, other research has asked the
strong support for the associative analysis and is related question of whether a weakened mem-
difficult to reconcile with inferential accounts, ory representation is formed for the blocked
although other data involving more complex cue itself. It was noted above that some (but not
designs have favored the latter (Mitchell et al. all) associative analyses assume that blocking
2005c, 2007). arises from a failure adequately to process the
Scully & Mitchell (2008) conducted a sim- blocked cue (Mackintosh 1975, Pearce & Hall
pler test of cue memory in the context of ex- 1980). Griffiths & Mitchell (2008) devised an
tinction rather than blocking. After pairing a ingenious modification of a blocking task again
cue with an outcome (A+), the cue was then using foods as the cues and allergies as the out-
extinguished (A−). In a subsequent test, ratings come, but added the feature that categories of
of the A-outcome relationship were reduced as foods (e.g., fruits) played the role of cues A and
a result of the extinction phase. More impor- B in a blocking design. These categories were
tantly, cued recall of the outcome paired with composed of instances (e.g., apples, bananas)
A in the first stage was significantly impaired in such that each actual instance only appeared
comparison to a nonextinguished control cue. once in the training stage. At test, new and
Such a result is again problematic for inferential old instances from the critical categories were
284 Shanks
presented, and participants rated the likelihood a middle way between the extreme cognitivism
of each instance causing the food allergy, as well of inferential accounts and the equally extreme
as making a conventional old/new recognition reductionism of S-R theory.
memory judgment. Blocking of predictive
ratings was, as usual, observed. Griffiths &
Mitchell’s (2008) most striking finding, how- AWARENESS AND LEARNING
ever, was that recognition of the instances of the Brewer’s (1974) view of conditioning was that
blocked category was significantly poorer than in humans it is invariably accompanied by con-
that of appropriate control cues, consistent tingency awareness. There has been a wealth
with associative theories that attribute blocking of research on this topic since the 1970s, so it
to a reduction in attention to and processing is natural to ask whether a contemporary per-
of the blocked cue. In sum, then, these studies spective would lead us to revise this conclusion.
examining memory in the context of blocking The outcome of one recent review (Lovibond
designs provide quite strong support for the in- & Shanks 2002) tended to confirm Brewer’s
volvement of associative processes. As Griffiths position. After analyzing experimental results

& Mitchell (2008) also note, their procedure on Pavlovian autonomic conditioning, condi-
provides a novel bridge between studies of as- tioning with subliminal stimuli, eyeblink con-
sociative learning and research on recognition ditioning, conditioning in amnesia, evaluative
and other forms of memory, and hence opens conditioning, and conditioning under anesthe-
up several promising avenues for further work. sia, Lovibond & Shanks (2002) concluded that
This draws to a close the review of recent there were no strong grounds for revising the
studies guided by cognitive or inferential views view that awareness is a necessary condition for
of learning. There is no doubt that this research learning in these preparations. In the present
has considerably enriched our understanding section, some recent studies of eyeblink, fear,
of basic learning processes in both humans and and evaluative conditioning are described prior
animals. An extensive body of evidence can be to a selective review of implicit learning stud-
marshaled in support of the radical claim that ies using other experimental procedures, such
associative processes play no role in learning. as motor sequence learning.
However, although manipulations designed to
alter individuals’ beliefs have certainly been
shown to influence behavior, it remains possible Eyeblink Conditioning
that the locus of some or all such manipulations The voluminous literature on human skele-
is in their effects on associative rather than tal conditioning has been supplemented by a
inferential processes. Associative theory has, small number of studies published since the re-
over the decades, often succeeded in explaining view period covered by Lovibond & Shanks
phenomena initially thought to be beyond (2002). An important distinction in such stud-
its bounds, and there are solid reasons to ies is between trace conditioning procedures,
believe that the same may apply to some of the in which there is a temporal gap between the
findings described here (e.g., additivity effects). termination of the CS and the onset of the
Results such as the effects of cue competition US, and delay conditioning, in which the on-
on memory and judgment for a blocked cue set of the US occurs before the termination of
seem, in contrast, to provide positive evidence the CS. It is generally agreed that awareness,
for basic associative processes, and it remains typically measured by a postconditioning as-
a challenge, therefore, to inferential accounts sessment of contingency knowledge, is neces-
to explain such findings. A view of learning in sary for the acquisition of responding in typical
which knowledge is formed out of associative laboratory trace eyeblink conditioning prepa-
connections between distributed representa- rations (Clark & Squire 1998, Lovibond &
tions of events remains viable, perhaps offering Shanks 2002). However, the possibility of delay

conditioning in the absence of awareness is continuously registered their expectancy of the

much more controversial. US on a rating scale. Knight et al. (2003) re-
Two recent studies (Bellebaum & Daum ported that unperceived CSs evoked differential
CR: conditioned
response 2004, Knuttinen et al. 2001) failed to obtain ev- CRs but not differential US expectancy ratings.
idence of conditioning in the absence of contin- That is to say, when the tone CS+ could not be
gency awareness. A study by Smith et al. (2005), detected, it failed to evoke a conscious report
however, did obtain such evidence. Specifi- of expectancy of the loud noise but did evoke a
cally, participants were presented with two au- conditioned change in skin conductance.
ditory stimuli (a tone and white noise), one of This impressive demonstration of unaware
which predicted an airpuff to the eye. Smith learning was subsequently replicated in two
et al. (2005) observed reliable delay condi- further studies by Knight et al. (2006, 2009).
tioning in participants classified subsequently Knight et al. (2006) reported at the same time
as being unaware of the correct CS-US rela- that the effect was not observed in trace con-
tionship. The status of this work is difficult ditioning, where differential conditioned re-
to gauge, however, because a subsequent at- sponding only occurred to perceived CSs. In
tempt to replicate the key findings yielded no the delay condition that replicated the earlier
evidence whatsoever of unaware conditioning. Knight et al. (2003) study, statistical analysis
Lovibond and colleagues (P.F. Lovibond, J.J.C. found that differential CRs to unperceived CSs
Loo, G. Weidemann, and C.J. Mitchell, yielded a t value of 1.86, significant at 0.05 (one-
manuscript submitted) reported two experi- tailed, df = 12). A comparable analysis found
ments that used the same methods and pro- that differential expectancy ratings yielded a t
cedure and in addition introduced some fur- value of 1.68 which is nonsignificant at 0.05. In
ther variables (e.g., contrasting short versus a neuroimaging study employing the same pro-
long awareness questionnaires). Even for par- cedure, Knight et al. (2009) obtained similar re-
ticipants treated identically to those of Smith sults, with CR differentiation yielding a reliable
et al. (2005), awareness was a necessary con- t value of 2.46 compared to expectancy ratings
dition for the observation of differential con- reported to be nonsignificant at t = 2.05 (df =
ditioning. Plainly, whatever the basis of these 14 in both cases, though this latter effect is sig-
discrepant findings, it would be premature to nificant one-tailed). Hence, the conclusion of
abandon the conclusion from earlier studies unaware conditioning in these two replications
that awareness and conditioning tend to be as- rests on one effect falling on just one side of the
sociated rather than dissociated. critical t value and the other effect falling on
just the other side. Firmer replication evidence
is needed to allow the Knight et al. (2003) result
Fear Conditioning to be properly evaluated.
In studies of fear conditioning, the US is typ- Moreover, in a careful study employing sig-
ically a shock or loud noise, and the CR is nal detection methods, unperceived CSs in fear
a change in skin conductance. The literature conditioning did not elicit differential skin con-
on awareness and fear conditioning has been ductance responses (Cornwell et al. 2007), and
supplemented by a handful of recent studies. other research has found that fear conditioning
Knight et al. (2003) devised a novel delay con- in delay as well as trace procedures is dependent
ditioning procedure for studying unconscious on awareness. For instance, Weike et al. (2007)
fear conditioning. One tone CS (CS+) pre- used a conventional procedure in which faces
dicted the US (loud noise), whereas another served as CS+ and CS− and shock as the US
tone (CS−) did not, but during acquisition in trace and delay conditioning preparations.
the intensity of the CSs was varied to render Differential skin conductance conditioned re-
some of them imperceptible. Participants in- sponses only occurred in participants classi-
dicated whenever they heard a tone and also fied by a postconditioning test as contingency
286 Shanks
aware. As Weike et al. (2007, p. 178) noted, More recent studies serve to confirm this
“electrodermal conditioning seems to primarily conclusion. For example, Pleyers et al. (2007)
index cognitive learning of the rules or circum- and Stahl et al. (2009) paired images (e.g., con-
stances in which a specific stimulus is signal- sumer products; the CSs) with positively or neg-
ing an aversive event, which is a declarative and atively valenced US pictures. Evaluative condi-
explicit memory.” Weike et al. (2007) did re- tioning only occurred for those CSs for which
port that a different measure of fear, potentiated participants were contingency aware. Dawson
startle responses, developed in participants un- et al. (2007) used faces as their CSs and USs
aware of the CS-US contingency in delay con- and embedded CS-US pairings in a short-term
ditioning, but this result is hard to interpret as visual memory test. Instead of measuring evalu-
it occurred only on trials administered prior to ative conditioning via explicit ratings, they mea-
the awareness questionnaire and not on ones sured skin conductance. Again, evaluative con-
presented after it. Dawson et al. (2007), in conditioning only occurred when participants were
trast, found a good association between startle aware of the CS-US pairings.
modulation and awareness. A study by Walther & Nagengast (2006)

again showed evaluative conditioning accompa-
Evaluative Conditioning nied by above-chance awareness—as measured
Evaluative conditioning refers to the transfer of by a recognition test—at the group level. These
affect or liking from one stimulus to another as authors claimed, however, that the condition-
a result of the pairing of those stimuli. This may ing effect they observed was significant only in
be an important mechanism involved in the de- participants classified as performing at or below
velopment of people’s likes and dislikes for such chance in the recognition test and not in those
everyday things as faces, consumer goods, and performing above chance. Apart from the fact
music. In a typical experiment, pictures judged that the interaction between awareness group
pre-experimentally as being affectively neutral and conditioning was not in fact statistically
are paired with ones rated as highly liked or dis- significant in their data, the method of exam-
liked. The neutral pictures therefore serve as ining learning effects post hoc in a subgroup
CSs and the affectively valenced ones as USs. of participants classified as unaware can lead
The common finding is that affective reactions to highly questionable conclusions. This widely
to the CS pictures are pulled in the direction used technique is almost guaranteed to generate
of the US pictures with which they are asso- apparent but potentially misleading evidence of
ciated, such that a neutral picture paired with unaware learning.
a highly liked picture itself comes to be liked; To illustrate this point, consider the fol-
the converse is true for a neutral picture paired lowing extremely simple simulation. Suppose
with a disliked picture. The important question data for a number of hypothetical participants
is whether participants have to be aware of the are generated, each producing a conditioning
nature of the stimulus pairings in order to show score. Assume also that this score is based on a
these learning effects. Is it necessary to be aware normally distributed variable x with a mean of
that a certain CS picture was paired with a liked 30 and standard deviation of 20. Hence x rep-
or disliked US picture in order for its valence resents the knowledge underlying the partici-
to change? Lovibond & Shanks (2002) (see also pant’s conditioned responding. The mean con-
Field & Davey 1999, Shanks & Dickinson 1990) ditioning score for the group is of course about
highlighted a number of methodological con- 30, but a proportion of participants will score
cerns in some of the studies included in their re- at or below zero, taken here to reflect chance
view but concluded that the more careful ones performance. Next, assume that the awareness
demonstrated (with the possible exception of score is also based on x, but with added noise.
evaluative conditioning of tastes—see below) Specifically, the awareness score is derived for
that contingency awareness is indeed necessary. each value of x by simply adding a uniformly

distributed random number, for example be- her liking ratings for red drinks as a result of
tween −40 and 40. The critical point is that red-banana-polysorbate pairings. At the end of
the conditioning and awareness scores are both the experiment, awareness for the various CS-
based on the same underlying knowledge, x. US contingencies can be assessed by present-
The only difference is that noise is added to ing colorless flavors or flavorless colors without
generate the awareness score. If one now selects any US. Although Dickinson & Brown (2007)
all the participants who score at or below zero introduced a number of procedural improve-
on the awareness measure, what is their aver- ments on the original Baeyens et al. (1990)
age conditioning score? The answer is that it is procedure, they nonetheless replicated the key
about 10 (SD ≈ 15), reliably greater than zero. observation, namely evidence of flavor eval-
This in statistical terms is just a regression-to- uative conditioning in the absence of aware-
the-mean effect (the expected value of a normal ness. More specifically, liking ratings of the fla-
variable x conditional upon x + e < 0, where e vors changed as a consequence of their pairings
is a random variable with mean 0, is not con- with sugar/polysorbate. Strikingly, participants
strained to be less than or equal to zero), but were able to recall the color-US pairings but
it illustrates that the practice of looking at con- not the flavor-US ones. Thus participants did
ditioning (or any other measure) in a sample show some degree of contingency awareness,
of participants selected post hoc as scoring at but it was not related to evaluative conditioning:
or below chance on an awareness measure is They knew the color contingencies and not the
a very dangerous practice. Above-chance con- flavor ones, but showed conditioning of flavors.
ditioning may be evident in such a subsample Wardle et al. (2007) introduced some further
even though conditioning and awareness derive procedural improvements such as counterbal-
solely (except for random noise) from the same ancing the order of presentation of the evalu-
underlying source. ative conditioning and contingency awareness
As mentioned above, there were a small tests and using a better format for the aware-
number of studies (e.g., Baeyens et al. 1990) ness questions. Whereas Dickinson & Brown
on evaluative conditioning with tastes that (2007) asked participants to identify which fla-
Lovibond & Shanks (2002) highlighted as pro- vor went with a particular US, in Wardle et al.’s
viding tantalizing evidence of unaware learn- procedure they tasted a flavor and rated their
ing and thus meriting further research. Two confidence that it went with sugar or polysor-
studies have indeed followed up this earlier bate. Under such circumstances, Wardle et al.
work. Dickinson & Brown (2007) replicated the (2007) found high levels of flavor-US and color-
Baeyens et al. (1990) procedure, which involves US awareness, as well as evaluative condi-
presenting participants with compound drinks tioning of the flavors (but not of the colors).
that are both flavored (e.g., vanilla) and col- Thus the key evidence of unaware condition-
ored (e.g., blue) as well as being pleasant or un- ing was no longer observed. When Wardle
pleasant. Pleasant tastes were created by adding et al. (2007) broke down their data according to
sugar; unpleasant ones had the bitter substance contingency awareness, they found that flavor
polysorbate 20 added to them. From a condi- evaluative conditioning was confined to those
tioning perspective, pleasantness can be consid- participants who demonstrated awareness.
ered the US and the flavors and colors as CSs, Notwithstanding the earlier cautions expressed
and evidence of evaluative conditioning would about such analyses, the results revealed no ev-
comprise pairing-dependent changes in likabil- idence of evaluative conditioning in unaware
ity ratings for the flavors or colors when pre- participants. Moreover, when Wardle et al.
sented in isolation. Thus a participant might (2007) reanalyzed Dickinson & Brown’s (2007)
come to increase her liking of vanilla as a re- data by separating aware from unaware partic-
sult of blue-vanilla-sugar pairings, or decrease ipants, they obtained the same pattern. Thus,
288 Shanks
whatever the final picture to emerge from this showed the gambler’s fallacy pattern, RTs be-
intriguing line of research, it seems safe to came slower with runs of CS-alone trials and
conclude that it has not as yet yielded strong faster with runs of CS-US ones.
Perruchet effect:
evidence of unconscious learning. Recent follow-up studies on this intriguing while CRs increase
phenomenon have raised as many questions as across a run of CS-US
they have answered, however. Both the gam- trials, US expectancy
The Perruchet Effect bler’s fallacy aspect of the results (the changes ratings decrease, and
vice versa for a series
In the Lovibond & Shanks (2002) review, one in conscious expectancy with run length) and
of CS-alone trials
other empirical result was highlighted as pro- the behavioral part of the results are matters
RT: reaction time
viding potentially convincing evidence of un- of dispute. The circumstances for assessing ex-
aware learning. This is the Perruchet effect, pectancy are hardly ideal. In Perruchet et al.’s
which arises in conditions of partial reinforce- (2006) studies, participants moved a pointer on
ment. Perruchet (1985) used an eyeblink con- a dial to indicate their moment-by-moment ex-
ditioning preparation in which the CS was pectancy that the target would occur on the next
reinforced on 50% of trials. In the trial se- trial. These expectancies, measured at a point
quence, runs of CS-alone or CS-US trials oc- just before presentation of the target, were the
curred. When a run comprised CS-alone trials, primary data. But suppose the participant only
Perruchet found that the likelihood of the con- changed his/her expectancy rating every few
ditioned response declined across trials. For seconds. With a very short cue (<1 sec) and
example, if there happened to be a sequence a stimulus-onset asynchrony of only 500 msec,
of four CS-alone trials, the likelihood of a the participant has virtually no time to register
CR would tend to decline across these four any CS-dependent change in expectancy. Sup-
trials. Conversely, in a run of CS-US trials, pose that, for whatever reason, expectancy of
the likelihood of the CR would tend to in- the target is not the same in the intertrial in-
crease across trials. Such changes are consis- terval as it is when the CS is actually present.
tent with the increments and decrements in The method would provide almost no chance of
associative strength that would be predicted detecting such changes. Perruchet et al. (2006)
by any reinforcement-based learning theory. acknowledged this issue and took some steps
Perruchet (1985) reported, however, that par- to ameliorate it, but without a much longer
ticipants’ expectation of the US showed exactly stimulus-onset asynchrony it is hard to be con-
the opposite pattern. In a run of CS-alone tri- fident in their results. In short, although the
als, expectation of the US increased the longer gambler’s fallacy is unquestionably observed in
the run, consistent with the well-known gam- many situations, it is difficult to accept uncrit-
bler’s fallacy. In a run of CS-US trials, expec- ically the claim that it occurs here and that ex-
tation of the US decreased the longer the run. pectancy ratings are largest after runs of cue-
Thus, the Perruchet effect comprises a striking alone trials.
dissociation between conditioned responding Turning to the other part of the dissociation,
and awareness, and hence evidence for unaware recall that CRs and speeded button-presses
learning. Participants seem to have thought that show a pattern whereby responses are less likely
the next trial must contain an outcome different or rapid after a run of CS-alone trials than a
from the preceding ones in order to maintain run of CS-US trials. A key question is whether
the 50% reinforcement probability. Perruchet these changes in responding are genuinely as-
et al. (2006) obtained a similar pattern in a re- sociative in nature. If they are not, then the
action time (RT) experiment. Here, the CS was result may have less to do with learning and
a warning tone and the US an imperative stim- more to do with performance aspects of the
ulus to which participants had to respond with task, such as US-recency effects. To investigate
a rapid button-press. While expectancy ratings whether the effect is truly associative, Mitchell

et al. repeated the Perruchet et al. (2006) RT dependencies in speeded RT settings, and the
study but arranged backward pairings or ran- learning of contextual cues in visual search.
dom occurrences of the warning stimulus (CS) We now turn to a brief review of some of this
and imperative stimulus (US). Under such cir- evidence.
cumstances, participants have no opportunity
to use the warning stimulus to prepare for the
US. Despite this, Mitchell et al. (2009) saw Sequence Learning
the same behavioral effect as in the standard Nissen & Bullemer (1987) devised a simple pro-
forward-pairing condition. They also observed cedure in which a visual target appeared on
the typical pattern when the CSs were omitted each trial at one of four locations in a display,
entirely. The obvious conclusion is that changes and the participant’s task was to press the ap-
in RT to the imperative stimulus have nothing propriate key for that location as fast as pos-
to do with learning predictive relationships, but sible. Targets moved from location to location
rather are a consequence of runs of motor re- according to a fixed, but nonsalient, sequence
sponses. When there is a run of CS-US trials, (later studies have employed noisy or proba-
for example, then the frequency of the US (and bilistic sequences). Although participants were
hence of responses to the imperative stimulus) not instructed about the presence of the se-
is high, whereas when there is a run of CS-alone quence, they nonetheless showed evidence of
trials, the frequency of the US declines. It seems sequence learning in that their RTs were faster
that in the RT experiment it is variation in the than in a control condition with random target
frequency of responses that determines reac- locations, an effect that has been replicated in
tion times, unconnected to learning anything numerous subsequent studies. This task, with
about the relationship between the CS and its heavy emphasis on perceptual-motor speed,
US. provides an ideal method for studying implicit
Weidemann et al. (2009) reported a similar learning: If participants readily learn the target
analysis in the context of eyeblink condition- sequence as evidenced by speeded RTs, are they
ing rather than speeded button pressing. In this also aware of the sequence? Or can procedural
case, the associative basis of the effect seems learning occur even in the absence of sequence
more secure. They found that the behavioral awareness?
effect (fewer CRs after a run of CS-alone trials To avoid the likelihood that verbal responses
than a run of CS-US trials) was not the conse- to postexperimental questions underestimate
quence of US (airpuff) recency. Thus, there is awareness, researchers have developed alter-
some reason to believe that the Perruchet effect native tests to assess awareness. In one such
in eyeblink conditioning represents genuine ev- test, participants are asked to report their
idence of unaware learning, in that conditioned conscious sequence knowledge by reproducing
eyeblink responses follow a pattern that is com- or generating the training sequence or parts
pletely opposite to the pattern displayed in ex- of it. In another test, they observe sequence
pectancy ratings. However, the effect in other chunks and signal whether they are old (that
behavioral preparations such as speeded button is, from the learned sequence) or new. With
pressing seems much less secure. Plainly this such tests, Gaillard et al. (2009), Jiménez et al.
is an area where more research is needed and (2006), Norman et al. (2006, 2007), Perruchet
where firm conclusions must await the outcome & Amorim (1992), Perruchet et al. (1997),
of that research. Shanks & Johnstone (1999), Shanks et al.
What of other experimental procedures (2003, 2005, 2006), Stefaniak et al. (2008),
within the broad domain of implicit learning? and Wilkinson & Shanks (2004) have all re-
Numerous tasks have been developed and ex- ported clear associations between learning and
plored over the past few years, such as artificial awareness, and indeed Perruchet et al. (1997)
grammar learning, the learning of sequential and Shanks & Johnstone (1999) have shown
290 Shanks
that above-chance performance on a test of amnesics trained with a deterministic sequence

awareness can precede any reliable degree of were not. Vandenberghe et al. (2006) there-
chronometrically measured sequence learning. fore concluded that in the latter cases, learn-
Bremner et al. (2007) discovered, remarkably, ing was implicit. However, they did not re-
that 2-year-olds were able to report their port statistical tests to confirm the reliability
sequence knowledge in a generation test. of deterministic sequence learning in the am-
Some studies have reported sequence learn- nesic group, and with only six participants in
ing in the absence of awareness as measured this group, it is by no means clear that learning
by generation or recognition tests (e.g., Dennis occurred. The results for the controls trained
et al. 2006; Fu et al. 2009, Vandenberghe et al. on probabilistic sequences must also be treated
2006), but these results are hard to interpret. with caution. Participants in comparable con-
For example, Dennis et al. (2006) reported that ditions of experiments by Shanks et al. (2003,
sequence learning in young participants was 2005) and Fu et al. (2008) showed substan-
accompanied by awareness, but not in older tial levels of sequence awareness, despite be-
participants. Yet, in one experiment, the older ing trained for only about half as many trials.
group did show awareness. Fu et al. (2008) re- Until more is known about the basis of these
ported that after short amounts of sequence discrepant results, null results such as those of
training, participants lacked awareness in that Vandenberghe et al. (2006) should be treated
they generated the learned sequence even when with caution.
instructed to avoid doing so. Taking control Also relevant are data from brain imaging
as an index of awareness ( Jacoby 1991), Fu studies of implicit and explicit learning. Plainly,
et al. (2008) concluded that under these cir- data showing that different neural networks are
cumstances, sequence knowledge was not con- activated under implicit and explicit learning
sciously accessible. However, in a subsequent conditions would provide powerful support
report, Fu et al. (2009) failed to observe this ef- for the idea that learning can be dissociated
fect when they used a slightly different genera- from awareness. In fact, this does not seem to
tion test for assessing awareness, although they be the case. In a careful functional magnetic
did obtain other apparent evidence of implicit resonance neuroimaging study, Willingham
learning. et al. (2002) defined an implicit condition as
Vandenberghe et al. (2006) trained both am- one in which recognition was at chance and
nesic and control participants with a determin- an explicit condition as one above chance in
istic sequence comprising a fixed 12-location recognition. These investigators found that
sequence or with a probabilistic sequence in the same neural systems (e.g., left prefrontal
which the majority of targets appeared at pre- cortex, left inferior parietal cortex, and right
ordained locations while the minority appeared putamen) were activated in both conditions but
at unexpected locations. Vandenberghe et al. that additional regions were activated in the
(2006) concluded that although control par- explicit condition (e.g., premotor cortex). This
ticipants were able to learn both types of seems to challenge the idea of distinct implicit
sequences, amnesics could learn only the de- and explicit learning processes. Even under
terministic ones (although learning of prob- implicit conditions, Willingham et al. (2002)
abilistic sequences in amnesia is possible: see did find a difference (albeit nonsignificant)
Shanks et al. 2006). On subsequent tests (gen- between recognition ratings for old and new
eration and recognition) of sequence awareness, test sequences of a comparable magnitude to
there was evidence that participants in the con- those typically found. Thus, it seems reason-
trol group trained on a deterministic sequence able to conjecture that Willingham et al.’s
were conscious of that sequence (replicating the implicit and explicit groups merely represented
findings of Wilkinson & Shanks 2004), whereas participants with weaker and stronger sequence
controls trained on a probabilistic sequence and knowledge.

Visual Search and Contextual Cuing displays and asked to report which ones they
had seen previously. This test once again re-
Another task that has gained some popularity
vealed above-chance levels of awareness. Such
for the study of implicit learning involves the
above-chance recognition was also observed by
contextual cuing of target locations in a visual
Preston & Gabrieli (2008) and Vaidya et al.
search setting. Suppose that a target such as a
(2007, Exp. 2). In the influential study by Chun
horizontal letter T appears among a set of hor-
& Phelps (1999), contextual cuing was observed
izontal L distracters, and the participant’s task
in control but not amnesic individuals. The
is to indicate the orientation of the T. Suppose
authors argued that the task assesses implicit
also that on some trials the configuration of dis-
learning because recognition was statistically at
tracters is repeated from earlier trials, and that
chance. Yet, the mean recognition hit and false-
in these repeated displays the location of the tar-
alarm rates (0.37 versus 0.32 for the controls,
get is constant. If the participant is able to learn
0.64 versus 0.42 for the amnesics) for old and
about the repeated displays, and that the loca-
new displays suggest the alternative possibility
tion of the target is reliably cued by a familiar

that a real awareness effect was masked by un-

distracter context, then faster localization re-
reliability in their measure and low statistical
sponses might be made to repeated compared
power.
to unique displays. Chun & Jiang (1998), who
Nevertheless, a common finding in this re-
devised this task, showed that participants could
search is that awareness (as measured by recog-
indeed learn about repeated configurations and
nition, for example) and implicit performance
make especially fast responses to them while
tend to be uncorrelated. For instance, in the
apparently being unaware of the repetition of
Preston & Gabrieli (2008) and Smyth & Shanks
displays (see also Chun & Jiang 2003).
(2008) studies, the correlations were close to
Later studies have replicated these results
zero, and implicit performance was no greater
but have also suggested that participants are
for recognized than for unrecognized configu-
generally aware of the display repetitions, pro-
rations. In the Smyth & Shanks (2008) study, it
vided that their awareness is assessed appropri-
was also the case that analyses of contextual cu-
ately. Smyth & Shanks (2008) measured aware-
ing in “unaware” participants showed reliable
ness with two methods. In one, participants
learning effects (see also Howard et al. 2004).
were shown repeated displays, but with the
But for the reason described above, these find-
T replaced by another distracter L, and were
ings still do not constitute clear evidence of
asked to indicate where they thought the miss-
implicit learning. In addition to the possibility
ing T would be located. When this test incor-
that the awareness test was unreliable (Howard
porated only 12 trials, no evidence of aware-
et al. 2004 tested awareness with only 12
ness was obtained (that is, performance was at
test displays), such dissociations are predicted
chance despite the fact that in the preceding
by single-system models that do not incorpo-
target-search part of the experiment, RTs were
rate the implicit–explicit distinction (Shanks &
faster to repeated displays). However, when the
Perruchet 2002, Shanks et al. 2003) when items
number of trials was increased to 48, thus in-
or participants are selected post hoc. Sup-
creasing the reliability and power of the test,
pose that there is a single knowledge base
above-chance performance was observed. For
that controls performance both in an implicit
instance, reliability increased from r = 0.09
test, such as contextual cuing, and in an ex-
to r = 0.46. Thus, evidence that learning in
plicit test, such as recognition. Suppose also,
the contextual cuing task is unconscious comes
however, that independent sources of noise
from studies in which the assessment of aware-
or error contribute to each performance mea-
ness is unreliable. Smyth & Shanks (2008) also
sure. As noted above, under such circum-
evaluated awareness in a test in which partici-
stances, it will inevitably be the case that sim-
pants were simply shown repeated and unique
ulated participants selected after the fact as
292 Shanks
scoring at or below chance on the explicit nevertheless implied by some of the key find-
measure will score above chance on the im- ings considered here.
plicit one, and likewise for configurations se- An important question is the extent to
lected post hoc on the same basis. Indeed, these which so-called dual-process theories have util-
models can even predict correlations of zero ity in helping us to understand learning. These
between implicit and explicit measures, despite theories (Broadbent et al. 1986, Evans 2008,
them arising from the same underlying repre- Kahneman 2003, Sloman 1996) propose that
sentation (Berry et al. 2006, Kinder & Shanks the mind is composed of two systems, one au-
2003). tomatic, implicit, nonrational, and unconscious
Thus the evidence is rather compelling that and the other slow, effortful, rational, explicit,
awareness is a necessary condition for learning and conscious. Thus the first system is consis-
across a wide range of experimental prepara- tent with unconscious learning and the second
tions. On this substantial issue, Brewer (1974) with inferential reasoning. The present review
seems to have been correct in his conclusion. should make it clear that the need for two sys-
tems has yet to be established, at least in rela-

tion to learning. It is quite debatable whether
CONCLUSIONS AND FUTURE any significant forms of learning can proceed
DIRECTIONS automatically or unconsciously, nor is it the
This review of recent research on the roles of case that apparently rational forms of learning
associative and cognitive processes in human can be explained only by reasoning-like pro-
learning points to two main conclusions: first, cesses: We have seen several examples of phe-
that learning across a range of preparations and nomena, such as retrospective revaluation, that
conditions is almost invariably accompanied by can be explained associatively as well as they can
awareness of the experimental contingency; and inferentially.
second, that although manipulations of indi- Of course, advocates of dual-process theo-
viduals’ beliefs can profoundly affect the way ries draw support from a range of domains in ad-
they acquire information, associationist condition to learning, including decision-making,
cepts continue to play a key role in explaining reasoning, and many aspects of social cognition.
learning. There are numerous examples of what appear
These conclusions run counter to the cur- to be powerful unconscious influences on be-
rent zeitgeist in the psychology of learning. havior (Nisbett & Wilson 1977, Wilson 2002).
The study of implicit (i.e., unconscious) learn- Yet it is equally important to note that sophis-
ing has become a very substantial and distinct ticated theoretical and methodological analysis
research topic, influencing other areas such as of consciousness has been applied to few do-
developmental psychology and cognitive neu- mains more thoroughly than it has to the field
roscience, and perpetuating a belief going back of learning. When examples of apparent uncon-
at least as far as Thorndike (1931) that learning scious influences in other domains have been
can proceed independently of awareness. The subjected to the same level of scrutiny, it has not
present review suggests that the core concept been uncommon to find that the evidence for
at the heart of this research area commands re- such influences is considerably weakened; good
markably little empirical support. The present illustrations are Maia & McClelland’s (2004)
review also suggests that the current fashion demonstration of the role of awareness in the
for viewing learning as an entirely cognitive, Iowa Gambling Task, and the extensive evi-
inferential process faces a number of chal- dence that attitudes measured implicitly (e.g.,
lenges. Certainly, cognitive processes penetrate by the Implicit Association Test) are often not
very deeply into many of the subprocesses that truly unconscious (Gawronski et al. 2006). It is
contribute to learning, yet associationist con- plain that future research with more demand-
cepts such as reinforcement and activation are ing tests of awareness will allow much firmer

conclusions to be drawn about the limits of un- meaningful (e.g., linguistic: Leow & Bowles
conscious processing. 2005) materials and numerous perceptual and
The phenomenon of blocking has been cen- motor as well as cognitive tasks, without yield-
tral to the research reviewed here. It is of course ing stronger evidence.
possible that what appear to be instances of The major alternative to the inferential and
the same phenomenon in fact arise for differ- dual-process perspectives derives both from
ent reasons. For instance, blocking might be traditional associative learning theory but also
a single name for a set of related outcomes from the success of the connectionist project
that arise from overlapping or even indepen- in human cognition (McClelland & Rumelhart
dent processes. It was mentioned above that 1986, Rumelhart & McClelland 1986, Thomas
blocking can be observed in marine molluscs & McClelland 2008). It is not possible to re-
as well as in humans and numerous species in view this project in detail here, but its essential
between; plainly that does not imply that it oc- components are the critical role played by asso-
curs for the same reason. Perhaps associative ciative processes together with the appreciation
principles are more important in some species that when such processes operate across dis-
or circumstances, and cognitive ones in others. tributed representations, high-level reasoning-
Although such a possibility must be borne in like behavior may emerge. The simulation by
mind, it must also be recognized that the sim- Ghirlanda (2005) described above is a striking
ilarities between blocking in different species illustration of this, in that it showed how ret-
run very deep. In both humans and rats, for rospective revaluation (the apparently inferen-
example, blocking is sensitive to manipulations tial readjustment of the weight assigned to a
of additivity information. Indeed, it has proven cue) may result from activation and incremental
quite difficult to find any associative learning learning within a simple network of neuron-like
phenomena that appear to be qualitatively dif- processing units. Although connectionist mod-
ferent across species or across contexts within els are in principle consistent with unconscious
a species (such as across implicit and explicit learning, it is often assumed that stable states of
learning conditions in humans). Thus, support- activation within the brain, subject to selective
ers of dual-process theories face the important attention, are precisely those states of which we
challenge in future work of demonstrating such are conscious (Maia & Cleeremans 2005).
qualitative differences more convincingly than Although the present review has focused on
has been achieved so far. studies explicitly designed to test and compare
It is also entirely reasonable to take the view associative and inferential accounts, other find-
that whereas unconscious learning has proven ings not reviewed here continue to defy expla-
hard to demonstrate in the sorts of preparations nation in anything other than associative terms.
reviewed here (e.g., eyeblink conditioning), it Principle among these is the nonrational aspect
does nonetheless occur in other circumstances. of many conditioned behaviors, a point made
Typical experimental procedures for studying forcefully by Dickinson (1980, 2009; see also
implicit learning tend to employ single labo- Shanks 1990). In a procedure such as fear con-
ratory sessions with meaningless stimuli (tones ditioning, for example, presentation of a CS that
and lights), so perhaps forms of more gradual predicts an aversive US such as shock evokes a
learning with meaningful stimuli are being ig- range of responses including sweating and el-
nored. It is undoubtedly important that future evated heart rate, yet there is no rational basis
work attempts to extend beyond simple labo- for these responses. Or consider the even more
ratory tasks. Yet a good deal of research has striking example of pigeon autoshaping, a ma-
been undertaken in which learning is studied jor source of evidence in the study of condi-
over quite prolonged periods during which ex- tioned behavior. If a keylight signals the pre-
pertise is established (e.g., 10,000 repetitions sentation of food, the pigeon will approach and
in table tennis: Koedijker et al. 2008) and with peck it. Although the approach behavior might
294 Shanks
be the product of rational inference—if the pi- item for the research agenda is to gain a fuller
geon “believes” that approach will cause food to understanding of the limits of associative prin-
be obtained—it is hard to see any chain of infer- ciples in explaining additivity and related ef-
ences that would cause the pigeon to peck the fects on blocking. As described in this review, it
keylight. Such pecking does not affect the de- has been clearly demonstrated that the outcome
livery of food, and indeed pigeons will continue of a blocking experiment, in both humans and
to peck even when pecking causes omission of animals, can be radically altered by pretreat-
the food. The form of the response (the pigeon ments designed to demonstrate the additivity
appears to try to eat a keylight that predicts food or nonadditivity of cues. Yet the precise locus of
and drink one that predicts water) strongly sug- these effects is still unknown. It is possible that
gests Pavlov’s original mechanism of stimulus they operate by altering the beliefs that form
substitution, whereby the CS becomes associ- the basis for inferences about predictive value,
ated with and takes the place of the US. It is but it is also possible that configural process-
very hard to see any involvement of cognitive ing or generalization are the key mechanisms
or inferential operations in this type of learning that are affected by such manipulations. These
and behavior. and related questions will surely feature promi-
Many gaps remain in our knowledge of the nently in future research seeking to under-
contributions of cognitive and associative pro- stand the inferential and associative aspects of
cesses to learning. Perhaps the most important learning.
SUMMARY POINTS
1. A review of the literature illustrates that although they are based on radically different
principles, cognitive and associative accounts of learning can both encompass a broad
range of empirical phenomena. Decisive tests are difficult to devise.
2. Forty years after its discovery, the phenomenon of blocking continues to lie at the heart
of theoretical debate. Explanations in terms of the automatic formation of associations
are certainly inadequate.
3. Learning is strongly influenced by pretreatments that provide information about the
additivity or nonadditivity of cue weights and by other manipulations of beliefs relevant
to inferential reasoning.
4. There is a substantial body of results, including the effects of cue competition on memory
and judgment, that provides evidence for associative processes such as activation and
reinforcement.
5. Robust and replicable instances of unconscious learning have failed to emerge in the
experimental literature, consistent with the view that awareness is a necessary condition
for all forms of learning, including conditioning.
6. Although its basis is not yet fully understood, the Perruchet effect represents perhaps the
most clear-cut current evidence for independent implicit and explicit learning systems.
DISCLOSURE STATEMENT
The author is not aware of any biases that might be perceived as affecting the objectivity of this
review.

LITERATURE CITED
Aitken MRF, Larkin MJW, Dickinson A. 2001. Re-examination of the role of within-compound associations
in the retrospective revaluation of causal judgements. Q. J. Exp. Psychol. 54B:27–51
Baeyens F, Eelen P, Van den Bergh O, Crombez G. 1990. Flavor-flavor and color-flavor conditioning in
humans. Learn. Motiv. 21:434–55
Beckers T, De Houwer J, Pineño O, Miller RR. 2005. Outcome additivity and outcome maximality influence
cue competition in human causal learning. J. Exp. Psychol.: Learn. Mem. Cogn. 31:238–49
Argues that inferential Beckers T, Miller RR, De Houwer J, Urushihara K. 2006. Reasoning rats: forward blocking in Pavlo-
causal reasoning vian animal conditioning is sensitive to constraints of causal inference. J. Exp. Psychol.: Gen.
underlies blocking in 135:92–102
animal conditioning. Bellebaum C, Daum I. 2004. Effects of age and awareness on eyeblink conditional discrimination learning.
Behav. Neurosci. 118:1157–65
Berry CJ, Henson RNA, Shanks DR. 2006. On the relationship between repetition priming and recognition
memory: insights from a computational model. J. Mem. Lang. 55:515–33
Bouton ME. 1993. Context, time, and memory retrieval in the interference paradigms of Pavlovian learning.
Psychol. Bull. 114:80–99

Bremner AJ, Mareschal D, Destrebecqz A, Cleeremans A. 2007. Cognitive control of sequential knowledge in
2-year-olds: evidence from an incidental sequence-learning and-generation task. Psychol. Sci. 18:261–66
Brewer WF. 1974. There is no convincing evidence for operant or classical conditioning in adult humans. In
Cognition and the Symbolic Processes, ed. WB Weimer, DS Palermo, pp. 1–42. Hillsdale, NJ: Erlbaum
Broadbent DE, Fitzgerald P, Broadbent MHP. 1986. Implicit and explicit knowledge in the control of complex
systems. Br. J. Psychol. 77:33–50
Cheng PW. 1997. From covariation to causation: a causal power theory. Psychol. Rev. 104:367–405
Cheng PW, Holyoak KJ. 1995. Complex adaptive systems as intuitive statisticians: causality, contingency,
and prediction. In Comparative Approaches to Cognitive Science, ed. HL Roitblat, J-A Meyer, pp. 271–302.
Cambridge, MA: MIT Press
Chun MM, Jiang Y. 1998. Contextual cueing: implicit learning and memory of visual context guides spatial
attention. Cogn. Psychol. 36:28–71
Chun MM, Jiang Y. 2003. Implicit, long-term spatial contextual memory. J. Exp. Psychol.: Learn. Mem. Cogn.
29:224–34
Chun MM, Phelps EA. 1999. Memory deficits for implicit contextual information in amnesic subjects with
hippocampal damage. Nat. Neurosci. 2:844–47
Clark RE, Squire LR. 1998. Classical conditioning and brain systems: the role of awareness. Science 280:77–81
Cobos PL, López FJ, Caño A, Almaraz J, Shanks DR. 2002. Mechanisms of predictive and diagnostic causal
induction. J. Exp. Psychol.: Anim. Behav. Process. 28:331–46
Cook SW, Harris RE. 1937. The verbal conditioning of the galvanic skin reflex. J. Exp. Psychol. 21:202–10
Cornwell BR, Echiverri AM, Grillon C. 2007. Sensitivity to masked conditioned stimuli predicts conditioned
response magnitude under masked conditions. Psychophysiology 44:403–6
Dawson ME, Rissling AJ, Schell AM, Wilcox R. 2007. Under what conditions can human affective conditioning
occur without contingency awareness? Test of the evaluative conditioning paradigm. Emotion 7:755–66
Dawson ME, Schell AM. 1982. Electrodermal responses to attended and nonattended significant stimuli
during dichotic listening. J. Exp. Psychol.: Hum. Percept. Perform. 8:315–24
The first report of
De Houwer J. 2009. The propositional approach to associative learning as an alternative for association
higher-order formation models. Learn. Behav. 37:1–20
retrospective De Houwer J, Beckers T. 2002a. Higher-order retrospective revaluation in human causal learning. Q.
revaluation. J. Exp. Psychol. 55B:137–51
De Houwer J, Beckers T. 2002b. Second-order backward blocking and unovershadowing in human causal
learning. Exp. Psychol. 49:27–33
Demonstrates effects on
De Houwer J, Beckers T, Glautier S. 2002. Outcome and cue properties modulate blocking. Q. J.
blocking of maximality/
Exp. Psychol. 55A:965–85
submaximality
information.
De Houwer J, Vandorpe S, Beckers T. 2007. Statistical contingency has a different impact on preparation
judgements than on causal judgements. Q. J. Exp. Psychol. 60:418–32
296 Shanks
Dennis NA, Howard JH, Howard DV. 2006. Implicit sequence learning without motor sequencing in young
and old adults. Exp. Brain Res. 175:153–64
Dickinson A. 1980. Contemporary Animal Learning Theory. London: Cambridge Univ. Press
Dickinson A. 2009. What are association formation models? Learn. Behav. 37:21–24
Dickinson A, Brown KJ. 2007. Flavor-evaluative conditioning is unaffected by contingency knowledge during
training with color-flavor compounds. Learn. Behav. 35:36–42
Dickinson A, Burke J. 1996. Within-compound associations mediate the retrospective revaluation of Illustrates the role of
causality judgements. Q. J. Exp. Psychol. 49B:60–80 within-compound
Dickinson A, Shanks DR, Evenden JL. 1984. Judgement of act-outcome contingency: the role of selective associations in
retrospective
attribution. Q. J. Exp. Psychol. 36A:29–50
revaluation.
Evans JSBT. 2008. Dual-processing accounts of reasoning, judgment, and social cognition. Annu. Rev. Psychol.
59:255–78
Field AP, Davey GCL. 1999. Reevaluating evaluative conditioning: a nonassociative explanation of condition-
ing effects in the visual evaluative conditioning paradigm. J. Exp. Psychol. Anim. Behav. Process. 25:211–24
Fletcher PC, Anderson JM, Shanks DR, Honey R, Carpenter TA, et al. 2001. Responses of human frontal
cortex to surprising events are predicted by formal associative learning theory. Nat. Neurosci. 4:1043–48
Fu QF, Dienes Z, Fu XL. 2009. Can unconscious knowledge allow control in sequence learning? Conscious.
Cogn. In press
Fu QF, Fu XL, Dienes Z. 2008. Implicit sequence learning and conscious awareness. Conscious. Cogn. 17:185–
202
Gaillard V, Destrebecqz A, Michiels S, Cleeremans A. 2009. Effects of age and practice in sequence learning:
a graded account of ageing, learning and control. Eur. J. Cogn. Psychol. 21:255–82
Gawronski B, Hofmann W, Wilbur CJ. 2006. Are “implicit” attitudes unconscious? Conscious. Cogn. 15:485–99
Ghirlanda S. 2005. Retrospective revaluation as simple associative learning. J. Exp. Psychol.: Anim. Demonstrates that
Behav. Process. 31:107–11 retrospective
Griffiths O, Mitchell CJ. 2008. Selective attention in human associative learning and recognition memory. revaluation effects can
be simulated by a simple
J. Exp. Psychol.: Gen. 137:626–48
associative network,
Haselgrove M. 2009. Reasoning rats or associative animals? A common-element analysis of the effects of
with a suitable stimulus
additive and subadditive pretraining on blocking. J. Exp. Psychol.: Anim. Behav. Process. In press
coding scheme.
Holland PC. 2008. Cognitive versus stimulus-response theories of learning. Learn. Behav. 36:227–41
Howard JH, Howard DV, Dennis NA, Yankovich H, Vaidya CJ. 2004. Implicit spatial contextual learning in
healthy aging. Neuropsychology 18:124–34
Jacoby LL. 1991. A process dissociation framework: separating automatic from intentional uses of memory.
J. Mem. Lang. 30:513–41
Jiménez L, Vaquero JMM, Lupiáñez J. 2006. Qualitative differences between implicit and explicit sequence
learning. J. Exp. Psychol.: Learn. Mem. Cogn. 32:475–90
Kahneman D. 2003. A perspective on judgment and choice: mapping bounded rationality. Am. Psychol. 58:697–
720
Kamin LJ. 1969. Selective association and conditioning. In Fundamental Issues in Associative Learning, ed.
NJ Mackintosh, WK Honig, pp. 42–64. Halifax: Dalhousie Univ. Press
Kinder A, Shanks DR. 2003. Neuropsychological dissociations between priming and recognition: a single-
system connectionist account. Psychol. Rev. 110:728–44
Knight DC, Nguyen HT, Bandettini PA. 2003. Expression of conditional fear with and without awareness.
Proc. Natl. Acad. Sci. 100:15280–83
Knight DC, Nguyen HT, Bandettini PA. 2006. The role of awareness in delay and trace fear conditioning in
humans. Cogn. Affect. Behav. Neurosci. 6:157–62
Knight DC, Waters NS, Bandettini PA. 2009. Neural substrates of explicit and implicit fear memory.
NeuroImage 45:208–14
Knuttinen M-G, Power JM, Preston AR, Disterhoft JF. 2001. Awareness in classical differential eyeblink
conditioning in young and aging humans. Behav. Neurosci. 115:747–57
Koedijker JM, Oudejans RRD, Beek PJ. 2008. Table tennis performance following explicit and analogy learning
over 10,000 repetitions. Int. J. Sport Psychol. 39:237–56

Kruschke JK. 2008. Bayesian approaches to associative learning: from passive to active learning. Learn. Behav.
36:210–26
Lagnado DA, Shanks DR. 2002. Probability judgment in hierarchical learning: a conflict between predictive-
ness and coherence. Cognition 83:81–112
Le Pelley ME, Beesley T, Suret MB. 2007. Blocking of human causal learning involves learned changes in
stimulus processing. Q. J. Exp. Psychol. 60:1468–76
Leow RP, Bowles MA. 2005. Attention and awareness in SLA. In External and Internal Factors in Adult SLA,
ed. C Sanz, pp. 179–203. Washington, DC: Georgetown Univ. Press
Livesey EJ, Boakes RA. 2004. Outcome additivity, elemental processing and blocking in human causality
judgements. Q. J. Exp. Psychol. 57B:361–79
López FJ, Cobos PL, Caño A. 2005. Associative and causal reasoning accounts of causal induction: symmetries
and asymmetries in predictive and diagnostic inferences. Mem. Cogn. 33:1388–98
López FJ, Cobos PL, Caño A, Shanks DR. 1998. The rational analysis of human causal and probability
judgment. In Rational Models of Cognition, ed. M Oaksford, N Chater, pp. 314–52. London: Oxford Univ.
Press
Lovibond PF. 2003. Causal beliefs and conditioned responses: retrospective revaluation induced by experience
and by instruction. J. Exp. Psychol.: Learn. Mem. Cogn. 29:97–106
Lovibond PF, Been S-L, Mitchell CJ, Bouton ME, Frohardt R. 2003. Forward and backward blocking of
causal judgment is enhanced by additivity of effect magnitude. Mem. Cogn. 31:133–42
Lovibond PF, Shanks DR. 2002. The role of awareness in Pavlovian conditioning: empirical evidence and
theoretical implications. J. Exp. Psychol.: Anim. Behav. Process. 28:3–26
Mackintosh NJ. 1975. A theory of attention: variations in the associability of stimuli with reinforcement.
Psychol. Rev. 82:276–98
Mackintosh NJ. 1983. Conditioning and Associative Learning. Oxford, UK: Clarendon
Maia TV, Cleeremans A. 2005. Consciousness: converging insights from connectionist modeling and neuro-
science. Trends Cogn. Sci. 9:397–404
Maia TV, McClelland JL. 2004. A reexamination of the evidence for the somatic marker hypothesis: what
participants really know in the Iowa gambling task. Proc. Natl. Acad. Sci. USA 102:16075–80
Matute H, Vegas S, De Marez PJ. 2002. Flexible use of recent information in causal and predictive judgments.
J. Exp. Psychol.: Learn. Mem. Cogn. 28:714–25
McClelland JL, McNaughton BL, O’Reilly RC. 1995. Why there are complementary learning systems in the
hippocampus and neocortex: Insights from the successes and failures of connectionist models of learning
and memory. Psychol. Rev. 102:419–57
McClelland JL, Rumelhart DE. 1986. Parallel Distributed Processing: Explorations in the Microstructure of Cogni-
tion. Vol. 2: Psychological and Biological Models. Cambridge, MA: MIT Press
McNally RJ. 1981. Phobias and preparedness: instructional reversal of electrodermal conditioning to fear-
relevant stimuli. Psychol. Rep. 48:175–80
Melchers KG, Lachnit H, Shanks DR. 2004a. Past experience influences the processing of stimulus compounds
in human Pavlovian conditioning. Learn. Motiv. 35:167–88
Melchers KG, Lachnit H, Shanks DR. 2004b. Within-compound associations in retrospective revaluation and
in direct learning: a challenge for comparator theory. Q. J. Exp. Psychol. 57B:25–53
Melchers KG, Lachnit H, Shanks DR. 2006. The comparator theory fails to account for the selective role of
within-compound associations in cue selection effects. Exp. Psychol. 53:316–20
Melchers KG, Shanks DR, Lachnit H. 2008. Stimulus coding in human associative learning: flexible repre-
sentations of parts and wholes. Behav. Process. 77:413–27
Melton AW. 1950. Learning. Annu. Rev. Psychol. 1:9–30
Miller RR, Barnet RC, Grahame NJ. 1995. Assessment of the Rescorla-Wagner model. Psychol. Bull. 117:363–
86
Mitchell CJ, De Houwer J, Lovibond PF. 2009. The propositional nature of human associative learning.
Behav. Brain Sci. 32:183–246
Mitchell CJ, Killedar A, Lovibond PF. 2005a. Inference-based retrospective revaluation in human causal
judgments requires knowledge of within-compound relationships. J. Exp. Psychol.: Anim. Behav. Process.
31:418–24
298 Shanks
Mitchell CJ, Livesey E, Lovibond PF. 2007. A dissociation between causal judgement and the ease with which
a cause is categorized with its effect. Q. J. Exp. Psychol. 60:400–17
Mitchell CJ, Lovibond PF. 2002. Backward and forward blocking in human electrodermal condition- Early demonstration of
ing: blocking requires an assumption of outcome additivity. Q. J. Exp. Psychol. 55B:311–29 additivity effects in
Mitchell CJ, Lovibond PF, Condoleon M. 2005b. Evidence for deductive reasoning in blocking of causal human blocking.
judgments. Learn. Motiv. 36:77–87
Mitchell CJ, Lovibond PF, Gan CY. 2005c. A dissociation between causal judgment and outcome recall.
Psychon. Bull. Rev. 12:950–54
Mitchell CJ, Lovibond PF, Minard E, Lavis Y. 2006. Forward blocking in human learning sometimes reflects
the failure to encode a cue-outcome relationship. Q. J. Exp. Psychol. 59:830–44
Mitchell CJ, Wardle SG, Lovibond PF, Weidemann G, Chang BPI. 2009. Is Perruchet’s dissociation between
cued reaction time and outcome expectancy evidence for two separate associative learning processes?
J. Exp. Psychol.: Learn. Mem. Cogn. In press
Mowrer RR, Klein SB. 2001. The transitive nature of contemporary learning theory. In Handbook of Contem-
porary Learning Theories, ed. RR Mowrer, SB Klein, pp. 1–21. Mahwah, NJ: Erlbaum
Nisbett RE, Wilson TD. 1977. Telling more than we can know: verbal reports on mental processes. Psychol.
Rev. 84:231–59
Nissen MJ, Bullemer P. 1987. Attentional requirements of learning: evidence from performance measures.
Cogn. Psychol. 19:1–32
Norman E, Price MC, Duff SC. 2006. Fringe consciousness in sequence learning: the influence of individual
differences. Conscious. Cogn. 15:723–60
Norman E, Price MC, Duff SC, Mentzoni RA. 2007. Gradations of awareness in a modified sequence learning
task. Conscious. Cogn. 16:809–37
Pearce JM, Hall G. 1980. A model for Pavlovian learning: variations in the effectiveness of conditioned but
not of unconditioned stimuli. Psychol. Rev. 87:532–52
Perales JC, Shanks DR. 2007. Models of covariation-based causal judgment: a review and synthesis. Psychon.
Bull. Rev. 14:577–96
Perales JC, Shanks DR. 2008. Driven by power? Probe question and presentation format effects on causal
judgment. J. Exp. Psychol.: Learn. Mem. Cogn. 34:1482–94
Perruchet P. 1985. A pitfall for the expectancy theory of human eyelid conditioning. Pavlov. J. Biol. Sci.
20:163–70
Perruchet P, Amorim M-A. 1992. Conscious knowledge and changes in performance in sequence learning:
evidence against dissociation. J. Exp. Psychol.: Learn. Mem. Cogn. 18:785–800
Perruchet P, Bigand E, Benoit-Gonin F. 1997. The emergence of explicit knowledge during the early phase
of learning in sequential reaction time tasks. Psychol. Res. 60:4–13
Perruchet P, Cleeremans A, Destrebecqz A. 2006. Dissociating the effects of automatic activation and Demonstration of the
explicit expectancy on reaction times in a simple associative learning task. J. Exp. Psychol.: Learn. Perruchet effect in a
Mem. Cogn. 32:955–65 human reaction time
task.
Pleyers G, Corneille O, Luminet O, Yzerbyt V. 2007. Aware and (dis)liking: Item-based analyses reveal that
valence acquisition via evaluative conditioning emerges only when there is contingency awareness. J. Exp.
Psychol.: Learn. Mem. Cogn. 33:130–44
Postman L. 1962. Rewards and punishments in human learning. In Psychology in the Making: Histories of Selected
Research Problems, ed. L Postman, pp. 331–401. New York: Knopf
Preston AR, Gabrieli JDE. 2008. Dissociation between explicit memory and configural memory in the human
medial temporal lobe. Cereb. Cortex 18:2192–207
Price PC, Yates JF. 1993. Judgmental overshadowing: further evidence of cue interaction in contingency
judgment. Mem. Cogn. 21:561–72
Rachman S. 1977. The conditioning theory of fear-acquisition: a critical examination. Behav. Res. Ther. 15:375–
87
Rescorla RA, Wagner AR. 1972. A theory of Pavlovian conditioning: variations in the effectiveness of rein-
forcement and nonreinforcement. In Classical Conditioning II: Current Theory and Research, ed. AH Black,
WF Prokasy, pp. 64–99. New York: Appleton-Century-Crofts

Rogers RF, Matzel LD. 1996. Higher-order associative processing in Hermissenda suggests multiple sites of
neuronal modulation. Learn. Mem. 2:279–98
Rumelhart DE, McClelland JL. 1986. Parallel Distributed Processing: Explorations in the Microstructure of Cogni-
tion. Vol. 1: Foundations. Cambridge, MA: MIT Press
Schmajuk N, Larrauri J. 2008. Associative models can describe both causal learning and conditioning. Behav.
Process. 77:443–45
Scully AL, Mitchell CJ. 2008. Extinction in human learning and memory. Q. J. Exp. Psychol. 61:1472–78
Shanks DR. 1985. Forward and backward blocking in human contingency judgement. Q. J. Exp. Psychol.
37B:1–21
Shanks DR. 1990. On the cognitive theory of conditioning. Biol. Psychol. 30:171–79
Shanks DR, Channon S, Wilkinson L, Curran HV. 2006. Disruption of sequential priming in organic and
pharmacological amnesia: a role for the medial temporal lobes in implicit contextual learning. Neuropsy-
chopharmacology 31:1768–76
Shanks DR, Darby RJ. 1998. Feature- and rule-based generalization in human associative learning. J. Exp.
Psychol.: Anim. Behav. Process. 24:405–15
Shanks DR, Dickinson A. 1990. Contingency awareness in evaluative conditioning: a comment on Baeyens,
Eelen, and Van den Bergh. Cogn. Emot. 4:19–30
Shanks DR, Johnstone T. 1999. Evaluating the relationship between explicit and implicit knowledge in a
sequential reaction time task. J. Exp. Psychol.: Learn. Mem. Cogn. 25:1435–51
Shanks DR, Perruchet P. 2002. Dissociation between priming and recognition in the expression of sequential
knowledge. Psychon. Bull. Rev. 9:362–67
Shanks DR, Rowland LA, Ranger MS. 2005. Attentional load and implicit sequence learning. Psychol. Res.
69:369–82
Shanks DR, Wilkinson L, Channon S. 2003. Relationship between priming and recognition in deterministic
and probabilistic sequence learning. J. Exp. Psychol.: Learn. Mem. Cogn. 29:248–61
Sloman SA. 1996. The empirical case for two systems of reasoning. Psychol. Bull. 119:3–22
Smith CN, Clark RE, Manns JR, Squire LR. 2005. Acquisition of differential delay eyeblink classical condi-
tioning is independent of awareness. Behav. Neurosci. 119:78–86
Smyth AC, Shanks DR. 2008. Awareness in contextual cuing with extended and concurrent tests. Mem. Cogn.
36:403–15
Stahl C, Unkelbach C, Corneille O. 2009. On the respective contributions of awareness of unconditioned stim-
ulus valence and unconditioned stimulus identity in attitude formation through evaluative conditioning.
J. Personal. Soc. Psychol. 97:404–20
Stefaniak N, Willems S, Adam S, Meulemans T. 2008. What is the impact of the explicit knowledge of sequence
regularities on both deterministic and probabilistic serial reaction time task performance? Mem. Cogn.
36:1283–98
Thomas MSC, McClelland JL. 2008. Connectionist models of cognition. In The Cambridge Handbook of
Computational Psychology, ed. R Sun, pp. 23–58. London: Cambridge Univ. Press
Thorndike EL. 1931. Human Learning. New York: Century
Vadillo MA, Miller RR, Matute H. 2005. Causal and predictive-value judgments, but not predictions, are
based on cue-outcome contingency. Learn. Behav. 33:172–83
Vaidya CJ, Huger M, Howard DV, Howard JH. 2007. Developmental differences in implicit learning of spatial
context. Neuropsychology 21:497–506
Van Hamme LJ, Wasserman EA. 1994. Cue competition in causality judgments: the role of nonpresentation
of compound stimulus elements. Learn. Motiv. 25:127–51
Vandenberghe M, Schmidt N, Fery P, Cleeremans A. 2006. Can amnesic patients learn without awareness?
New evidence comparing deterministic and probabilistic sequence learning. Neuropsychologia 44:1629–41
Vandorpe S, de Houwer J, Beckers T. 2007. The role of memory for compounds in cue competition. Learn.
Motiv. 38:195–207
Waelti P, Dickinson A, Schultz W. 2001. Dopamine responses comply with basic assumptions of formal
learning theory. Nature 412:43–48
Wagner AR. 1981. SOP: a model of automatic memory processing in animal behavior. In Information Processing
in Animals: Memory Mechanisms, ed. NE Spear, RR Miller, pp. 5–47. Hillsdale, NJ: Erlbaum
300 Shanks
Waldmann MR. 2000. Competition among causes but not effects in predictive and diagnostic learning. J. Exp.
Psychol.: Learn. Mem. Cogn. 26:53–76
Waldmann MR, Holyoak KJ. 1990. Can causal induction be reduced to associative learning? Proceedings 12th
Ann. Conf. Cogn. Sci. Soc., pp. 190–97. Hillsdale, NJ: Erlbaum
Waldmann MR, Holyoak KJ. 1992. Predictive and diagnostic learning within causal models: asymme- Contrasts learning
tries in cue competition. J. Exp. Psychol.: Gen. 121:222–36 about cause-effect
Walther E, Nagengast B. 2006. Evaluative conditioning and the awareness issue: assessing contingency aware- (predictive) and
ness with the four-picture recognition test. J. Exp. Psychol.: Anim. Behav. Process. 32:454–59 effect-cause (diagnostic)
relationships.
Wardle SG, Mitchell CJ, Lovibond PF. 2007. Flavor evaluative conditioning and contingency awareness.
Learn. Behav. 35:233–41
Weidemann G, Tangen JM, Lovibond PF, Mitchell CJ. 2009. Is Perruchet’s dissociation between eyeblink
conditioned responding and outcome expectancy evidence for two learning systems? J. Exp. Psychol.:
Anim. Behav. Process. 35:169–76
Weike AI, Schupp HT, Hamm AO. 2007. Fear acquisition requires awareness in trace but not delay condi-
tioning. Psychophysiology 44:170–80
Wheeler DS, Beckers T, Miller RR. 2008. The effect of subadditive pretraining on blocking: limits on gen-
eralization. Learn. Behav. 36:341–51
Wilkinson L, Shanks DR. 2004. Intentional control and implicit sequence learning. J. Exp. Psychol.: Learn.
Mem. Cogn. 30:354–69
Williams DA, Sagness KE, McPhee JE. 1994. Configural and elemental strategies in predictive learning.
J. Exp. Psychol.: Learn. Mem. Cogn. 20:694–709
Willingham DB, Salidis J, Gabrieli JDE. 2002. Direct comparison of neural systems mediating conscious and
unconscious skill learning. J. Neurophysiol. 88:1451–60
Wilson TD. 2002. Strangers to Ourselves: Discovering the Adaptive Unconscious. Cambridge, MA: Belknap

AR398-FM ARI 7 November 2009 7:47
Annual Review of
Psychology
Volume 61, 2010 Contents
Prefatory
Love in the Fourth Dimension

Ellen Berscheid p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 1
Brain Mechanisms and Behavior
The Role of the Hippocampus in Prediction and Imagination
Randy L. Buckner p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p27
Learning and Memory Plasticity; Neuroscience of Learning
Hippocampal-Neocortical Interactions in Memory Formation,
Consolidation, and Reconsolidation
Szu-Han Wang and Richard G.M. Morris p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p49
Stress and Neuroendocrinology
Stress Hormone Regulation: Biological Role
and Translation Into Therapy
Florian Holsboer and Marcus Ising p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p81
Developmental Psychobiology
Structural Plasticity and Hippocampal Function
Benedetta Leuner and Elizabeth Gould p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 111
Cognitive Neuroscience
A Bridge Over Troubled Water: Reconsolidation as a Link Between
Cognitive and Neuroscientific Memory Research Traditions
Oliver Hardt, Einar Örn Einarsson, and Karim Nader p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 141
Cognitive Neural Prosthetics
Richard A. Andersen, Eun Jung Hwang, and Grant H. Mulliken p p p p p p p p p p p p p p p p p p p p p p 169
Speech Perception
Speech Perception and Language Acquisition in the First Year of Life
Judit Gervain and Jacques Mehler p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 191
vi
Chemical Senses (Taste and Smell)

An Odor Is Not Worth a Thousand Words: From Multidimensional
Odors to Unidimensional Odor Objects
Yaara Yeshurun and Noam Sobel p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 219
Somesthetic and Vestibular Senses
Somesthetic Senses
Mark Hollins p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 243
Basic Learning and Conditioning
Learning: From Association to Cognition
David R. Shanks p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 273

Comparative Psychology
Evolving the Capacity to Understand Actions, Intentions, and Goals
Marc Hauser and Justin Wood p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 303
Human Development: Processes
Child Maltreatment and Memory
Gail S. Goodman, Jodi A. Quas, and Christin M. Ogle p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 325
Emotional, Social, and Personality Development
Patterns of Gender Development
Carol Lynn Martin and Diane N. Ruble p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 353
Adulthood and Aging
Social and Emotional Aging
Susan T. Charles and Laura L. Carstensen p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 383
Development in Societal Context
Human Development in Societal Context
Aletha C. Huston and Alison C. Bentley p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 411
Genetics and Psychopathology
Epigenetics and the Environmental Regulation
of the Genome and Its Function
Tie-Yuan Zhang and Michael J. Meaney p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 439
Social Psychology of Attention, Control, and Automaticity
Goals, Attention, and (Un)Consciousness
Ap Dijksterhuis and Henk Aarts p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 467
Contents vii
Bargaining, Negotiation, Conflict, Social Justice

Negotiation
Leigh L. Thompson, Jiunwen Wang, and Brian C. Gunia p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 491
Personality Development: Stability and Change
Personality Development: Continuity and Change Over the
Life Course
Dan P. McAdams and Bradley D. Olson p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 517
Work Motivation
Self-Regulation at Work
Robert G. Lord, James M. Diefendorff, Aaron C. Schmidt, and Rosalie J. Hall p p p p p p p p 543
Cognition in Organizations
Creativity
Beth A. Hennessey and Teresa M. Amabile p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 569
Work Attitudes ( Job Satisfaction, Commitment, Identification)
The Intersection of Work and Family Life: The Role of Affect
Lillian T. Eby, Charleen P. Maher, and Marcus M. Butts p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 599
Human Factors (Machine Information, Person Machine Information,
Workplace Conditions)
Cumulative Knowledge and Progress in Human Factors
Robert W. Proctor and Kim-Phuong L. Vu p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 623
Learning and Performance in Educational Settings
The Psychology of Academic Achievement
Philip H. Winne and John C. Nesbit p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 653
Personality and Coping Styles
Personality and Coping
Charles S. Carver and Jennifer Connor-Smith p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 679
Indexes
Cumulative Index of Contributing Authors, Volumes 51–61 p p p p p p p p p p p p p p p p p p p p p p p p p p p 705

Cumulative Index of Chapter Titles, Volumes 51–61 p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 710
Errata
An online log of corrections to Annual Review of Psychology articles may be found at

http://psych.annualreviews.org/errata.shtml
viii Contents

Learning: From Association To Cognition: Further

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Learning: From Association To Cognition: Further

Uploaded by

Copyright:

Available Formats

ANRV398-PS61-11 ARI 5 November 2009 13:5

Division of Psychology and Language Sciences, University College London,

London WC1H 0AP United Kingdom; email: d.shanks@ucl.ac.uk

Annu. Rev. Psychol. 2010. 61:273–301 Key Words

dark days. The present review takes Brewer’s

Evaluative Conditioning. . . . . . . . . . . . 287

from the work in the early part of the past cen-

www.annualreviews.org • Learning 275

Table 1 Experimental designs relating to blocking

www.annualreviews.org • Learning 277

that the predictive

www.annualreviews.org • Learning 279

function of the initial problem. Speciﬁcally, examples of retrospective revaluation. In a

www.annualreviews.org • Learning 281

1998). Associative approaches might seek to ex- previous trial types.

www.annualreviews.org • Learning 283

volvement of associative processes. As Grifﬁths position. After analyzing experimental results

www.annualreviews.org • Learning 285

conditioning in the absence of awareness is continuously registered their expectancy of the

modulation and awareness. A study by Walther & Nagengast (2006)

www.annualreviews.org • Learning 287

www.annualreviews.org • Learning 289

that above-chance performance on a test of amnesics trained with a deterministic sequence

www.annualreviews.org • Learning 291

tion of the target is reliably cued by a familiar

that a real awareness effect was masked by un-

tems has yet to be established, at least in rela-

www.annualreviews.org • Learning 293

www.annualreviews.org • Learning 295

Psychol. Bull. 114:80–99

www.annualreviews.org • Learning 297

www.annualreviews.org • Learning 299

www.annualreviews.org • Learning 301

Volume 61, 2010 Contents

Love in the Fourth Dimension

Chemical Senses (Taste and Smell)

David R. Shanks p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 273

Bargaining, Negotiation, Conflict, Social Justice

Cumulative Index of Contributing Authors, Volumes 51–61 p p p p p p p p p p p p p p p p p p p p p p p p p p p 705

An online log of corrections to Annual Review of Psychology articles may be found at

You might also like