In each of them, the extinguished response returns to performance. All of them therefore indicate that extinction is not the same as unlearning, and because all of them can be seen as context effects, they also support the idea that performance after extinction is context-dependent. Extinction involves new learning, and it therefore leaves the CS with two available “meanings” or associations with the US. As is true for an ambiguous word, the context is crucial in selecting between them.
Perhaps the most fundamental of these effects is the renewal effect. In this phenomenon, a change of context after extinction can cause a robust return of conditioned responding. Several versions of the renewal effect have been studied. In the most common one, “ABA renewal,” conditioning is conducted in one context (context A) and extinction is then conducted in a second one (context B). (The contexts are typically separate and counterbalanced apparatuses housed in different rooms of the laboratory that differ in their tactile, olfactory, and visual respects.) When the CS is returned to the original conditioning context (context A), responding to the CS returns. In a second version, “ABC renewal,” conditioning is conducted in context A, extinction is conducted in context B, and then testing is conducted in a third, “neutral” context—context C. Here again, a renewal of responding is observed. In a final version, conditioning and extinction are both conducted in the same context (context A) and then the CS is tested in a second context (context B). Here again, conditioned responding returns, although there is currently less evidence of this “AAB renewal” effect in operant conditioning than in Pavlovian conditioning.
Several facts about the renewal effect are worth noting. First, it has been observed in virtually every conditioning preparation in which it has been investigated. Second, it can occur after very extensive extinction training. In fear conditioning (conditioned suppression) in rats, Bouton and Swartzentruber (1989) observed it when 84 extinction trials followed eight conditioning trials. Other evidence suggests that it can occur after as many as 160 extinction trials, although a recent report suggests that it might not survive an especially “massive” extinction treatment (800 extinction trials after eight conditioning trials. Third, the role of the context is different from the one anticipated by standard models of classical conditioning. Those models accept the view that the context is merely another CS that is presented in compound with the target CS during reinforcement or nonreinforcement. It therefore enters into simple excitatory or inhibitory associations with the US. In the ABA renewal effect (for example), context A might acquire excitatory associations with the US, and context B might acquire inhibitory associations. Either kind of association would summate with the CS to produce the renewal effect (inhibition in B would reduce responding to the CS, whereas excitation in A would enhance it). However, a number of experiments have shown that the renewal effect can occur in the absence of demonstrable excitation in context A or inhibition in context B. These findings, coupled with others showing that strong excitation in a context does not influence performance to a CS unless the CS is under the influence of extinction, suggest that direct associations in a context are neither necessary nor sufficient for a context to influence responding to a CS. The implication is that the contexts modulate or “set the occasion” for the current CS-US or CS-no US association. Put another way, they activate or retrieve the current relation of CS with the US.
A further important characteristic of the renewal effect is that it implies that extinction learning is more context-specific than is original conditioning. Notice that this must be true if one observes ABC and AAB renewal; in either case, conditioning transfers better to the final test context than extinction. But our experiments on renewal have often involved comparisons of groups that received extinction training in the context in which conditioning had occurred or in a discriminably different context. Strikingly, there was no measurable effect of switching the context after conditioning on responding to the CS. In contrast, extinction itself was relatively context-specific, as the renewal effect itself suggests. Recent research suggests that both conditioning and extinction become somewhat context-specific after extinction has occurred. But there is little question that extinction is still more context-dependent than is the original conditioning. We have therefore emphasized the fact that extinction learning is especially context-dependent.
A final fact about the renewal effect is that it appears to be supported by many kinds of contexts. For example, when fear extinction was conducted in the interoceptive context provided by benzodiazepine tranquilizers chlordiazepoxide and diazepam, renewed fear was observed when the rat was tested in the original nondrug state had reported compatible evidence with alcohol, and we have recently collected similar observations with the benzodiazepine midazolam. State-dependent learning or retention can be conceptualized as the drug playing the role of context.
Spontaneous Recovery
The passage of time might also bring about changes in internal and external stimulation that provide a gradually-changing context. Pavlov (1927) first observed another well-known extinction effect. In spontaneous recovery, if time is allowed to pass following extinction, the extinguished response can recover. There are several available explanations of spontaneous recovery, and it seems likely to be multiply determined. However, we have argued that just as extinction is relatively specific to its physical context, so it may be specific to its “temporal context.” Spontaneous recovery can be seen as the renewal effect that occurs when the CS is tested outside its temporal context. Both are due to a failure to retrieve memories of extinction outside the extinction context. Consistent with this perspective, a cue that is presented intermittently during the extinction session can attenuate either spontaneous recovery or renewal if it is presented just before the final test. The parallel results suggest that the two effects might be controlled by a common mechanism: a failure to retrieve a memory of extinction outside the extinction context. Interestingly, changing the physical context and temporal context together can have a bigger effect than changing either context alone, as if their combination creates an even larger context change.
Rapid Reacquisition
A third effect further indicates that conditioning is not destroyed in extinction. In rapid reacquisition, when new CS-US pairings are introduced after extinction, the reacquisition of responding can be more rapid than is acquisition with a novel CS, indicating that the original learning has been “saved” through extinction. Unfortunately, the early literature on rapid reacquisition was often difficult to interpret because many early designs were not equipped to rule out less interesting explanations. To add to the complexity, studies of fear conditioning and flavor aversion learning, have shown that reacquisition can be slower than acquisition with a new CS. (It is more rapid than initial acquisition with a CS that has received the same number of nonreinforced trials without conditioning.) In fear conditioning, slow reacquisition requires extensive extinction training; more limited extinction training yields reacquisition that is neither fast nor slow (Bouton 1986). At least part of the reason these preparations support slow reacquisition is that both typically involve very few initial conditioning trials. In contrast, procedures in which rapid reacquisition has been shown (rabbit nictitating membrane response (NMR) conditioning and rat appetitive conditioning) have usually involved a relatively large number of initial conditioning trials. Consistent with a role for number of trials, Ricker and Bouton (1996) demonstrated that slow reacquisition occurred in an appetitive conditioning preparation when the procedure used the number of conditioning and extinction trials that had been used in previous fear conditioning experiments. In rabbit NMR and heart rate conditioning, extensive extinction training has abolished rapid reacquisition, although slow reacquisition has yet to be observed.
Ricker and Bouton (1996) suggested that rapid reacquisition may partly be an ABA renewal effect that occurs when the animal has learned that previous USs or conditioning trials are part of the original “context” of conditioning. That is, the animal might learn that recent CS-US pairings are part of the context of conditioning, whereas recent CS-only presentations are part of the context of extinction. When CS-US pairings are resumed after extinction, they would thus return the animal to the original conditioning context. The hypothesis is compatible with Capaldi’s (1967, 1994) sequential analysis of extinction, which has made excellent use of the idea that responding on a particular trial is determined by how the animal has learned to respond in the presence of similar memories of previous trials (see below). Presumably, conditioning preparations that use a relatively large number of conditioning trials allow ample opportunity for the animal to learn that previous reinforced trials are part of the context of conditioning. Ricker and Bouton (1996) also reported evidence that high responding during the reacquisition phase was more likely after a reinforced than a nonreinforced trial, which presumably signaled conditioning and extinction, respectively.
In more recent experiments, Bouton et al. (2004) reasoned that if rapid reacquisition is caused by recent reinforced trials generating ABA renewal, then an extinction procedure that includes occasional reinforced trials among many nonreinforced trials should slow down rapid reacquisition by making recent reinforced trials part of the context of both conditioning and extinction. Consistent with this hypothesis, a very sparse partial reinforcement procedure in extinction slowed reacquisition in a final phase compared with a group that had received simple extinction. Such a result is consistent with the idea that rapid reacquisition is at least partly an ABA renewal effect. Because the partial reinforcement treatment involved many more CS-US pairings than did simple extinction, it is difficult to reconcile with the view that rapid reacquisition is a simple function of the strength of an association that remains after extinction.
Reinstatement
A fourth context-dependent extinction phenomenon is reinstatement. In this effect, the extinguished response returns after extinction if the animal is merely reexposed to the US alone. If testing of the CS is contemporaneous with US delivery, then the USs may cause a return of responding because they were encoded as part of the conditioning context. On the other hand, in many studies of reinstatement, testing is conducted at an interval of at least 24 h after US re-exposure; here one still observes reinstatement compared with controls that were not re-exposed to the US. In this case, evidence strongly suggests that the effect is due to conditioning of the context. When the US is presented after extinction, the organism associates it with the context; this contextual conditioning then creates reinstatement. For example, if the reinstating USs are presented in an irrelevant context, there is no reinstatement when the CS is tested again. Independent measures of contextual conditioning also correlate with the strength of reinstatement. And if the animal receives extensive extinction exposure to the context after the reinstatement shocks are presented, reinstatement is not observed. These results indicate that mere re-exposure to the US is not sufficient to generate reinstatement. It is necessary to test the CS in the context in which the US has been re-exposed.
This effect of context conditioning is especially potent with an extinguished CS. For example, Bouton (1984) compared the effects of US exposure in the same or a different context on fear of a partially extinguished CS or another CS that had reached the same low level of fear through simple CS-US pairings (and no extinction). Although contextual conditioning enhanced fear of the extinguished CS, it had no impact on the nonextinguished CS. This result is consistent with the effects of context switches mentioned above: An extinguished CS is especially sensitive to manipulations of the context. One reason is that contextual conditioning may be another feature of the conditioning context; its presence during a test may cause a return of responding after extinction because of another ABA renewal effects.
In summary, a variety of research indicates that responding to an extinguished CS is susceptible to any of a number of recovery effects, suggesting that extinction is not unlearning. Indeed, based on the results of a number of tests that allow a specific comparison of the strength of the CS-US association before and after extinction has suggested that extinction involves no unlearning whatsoever; the original CS-US association seems to survive essentially intact. Extinction must thus depend on other mechanisms. The renewal effect, and the fact that extinction leaves the CS so especially sensitive to manipulations of context, is consistent with the idea that extinction involves new learning that is especially context-dependent. We have therefore suggested that extinction leaves the CS under a contextually modulated form of inhibition: The presence of the extinction context retrieves or sets the occasion for a CS-no US association.
Other Phenomena With Theoretical Links to Extinction
Several behavioral phenomena have been linked theoretically with extinction, and it is worth considering them to see what insights they provide.
