The question of 'What is Learned?' in Pavlovian conditioning amounts to asking about the associative structures that are assumed to underlie Pavlovian conditioning. The issue has been approached in three different ways. First, the historical approach has been to ask whether learning was best described in terms of S-S or S-R associations. The second approach has been to emphasize the question of whether the CS forms associations with either the sensory or the affective/motivational components of the US (or both). The third, more recent, approach has been to ask if different types of associative structures that differ in their complexity are acquired. In particular, it has been determined if hierarchical as well as binary associations can be learned. Below we consider some of the research that addresses the question of what is learned from each of these perspectives.
A. S-S vs S-R associations
1. In Sensory Preconditioning (SPC). Early work by Brogden demonstrated that it was possible to demonstrate that learning takes place between two stimuli, neither of which by itself evokes an appreciable unconditioned response. If it was possible to demonstrate learning in this situation, the idea is that such learning could not be described in terms of S-R associations being established for the simple reason that no response is present. His particular experiment was a 3 phase study in which two "neutral" CSs were paired with one another during phase 1 (i.e., CS1 -> CS2). In a second phase, a conditioned response was established to CS2 by pairing it repeatedly with the US. In phase 3, Brogden went on to test if the CR conditioned to CS2 in phase 2 would transfer back to CS1, the CS that had signaled CS2 previously. It did, and this was taken as evidence that the learning established in phase 1 was of an S-S association. In other words, the claim is that the sensory properties of CS1 associated with the sensory properties of CS2. The observation of a CR to CS1 in the test phase is attributed to the possibility that CS1 evokes a representation of CS2 (because of the S-S association between those two stimuli), which in turn evokes the CR conditioned to it during phase 2.
Two additional points are worth mentioning. Can you think of an appropriate control group that should be run in order to demonstrate that this "transfer" of the CR from CS2 to CS1 was NOT due, simply, to stimulus generalization? Also, I should point out that the interpretation of this result in terms of an S-S association did not go on uncontested. Other psychologists went on to show how it was possible for an S-R association model to explain this result. However, that argument is considerably more complex than we have time for here. For now we can take the SPC experiment as providing some provisional support for the view that S-S associations are possible.
2. Pavlovian 1st Order Conditioning. There have been several experiments that have shown that S-S associations can be learned when you pair a CS with a US (i.e., a biologically important event that itself evokes a UR). One experiment we considered in class was conducted by Holland (1990). In this experiment, during phase 1 two auditory CSs were each paired with a sucrose solution that differed in its sensory characteristics (T1 -> Cinn+Sugar, T2 -> Wintergreen+Sugar). In phase 2, one of the USs was devalued by pairing it separately with LiCl. Magazine approach CRs were conditioned to T1 and T2 during phase 1 and they were assessed when these stimuli were presented without the USs during the test session. Holland observed that there were more magazine approach CRs evoked by the CS whose associated US was not devalued compared to the CS whose associated US was devalued. This result can be understood in terms of an S-S association having been acquired in the first phase. The idea is that the stimulus properties of the sugar solution was devalued when it was paired with LiCl, and that the reduced magazine approach CRs reflected the fact that the CS evoked a representation of those now devalued stimulus properties of the US. However, another important aspect of this data is that although there was less responding to the CS whose associated US was devalued, there was still more responding to this CS compared to the Pre-CS interval. In other words, the reinforcer devaluation effect was incomplete. It is possible to argue that this is because there is an aspect of learning that is totally insensitive to the current value of the US. This aspect of learning may very well be the S-R association. Thus, we can take these results as supporting the view that both S-S and S-R associations can be established in Pavlovian 1st order conditioning.
3. Pavlovian 2nd Order Conditioning. There are several experiments that have examined whether S-S or S-R associations are learned during 2nd order conditioning. The literature is a bit complex. However, we have the luxury of being very selective. In the experiment reported by Nairne and Rescorla (1981) it looks as though only S-R associations are learned. Their study used pigeons in an autoshaping task and they used the same logic as that used in the reinforcer devaluation study we just considered. Two CSs were first established as 1st order CSs (CS1 -> grain, CS2 -> grain). Each of these CSs were then used to reinforce 2nd order conditioning to two additional CSs (CS3 -> CS1, CS4 -> CS2). Then, one of the 1st order CSs was extinguished while the other continued to be paired with the US. Finally, they tested responding to CS3 and CS4 to determine if it depended upon the current value of the two 1st order CSs. It did not. This supports the S-R association view. Can you articulate why? Also, note how this procedure (2nd order conditioning) differs from the SPC procedure only in the order of the first two phases. It is interesting that the two procedures lead to different results.
4. Pavlovian Conditioned Inhibition. The S-S vs S-R question has also been addressed in the study of conditioned inhibition. A noteworthy study was conducted by Delamater, LoLordo, & Sosa (2003). The S-S view of inhibitory learning asserts that the CS associates with sensory-specific attributes of US omission. The animal, in other words, learns to anticipate when a specific event is not going to happen. According to the S-R view, the inhibitory CS is assumed to associate with the response that occurs on trials in which the US is omitted, but does not learn anything regarding the specific event that was omitted. In the study by Delamater, et al., an appetitive conditioning procedure was used with rats. A Pavlovian backward conditioned inhibition procedure was used during the Pavlovian phase and a Pavlovian-to-instrumental transfer test procedure was used to assess the US-specificity of backward inhibitory learning. In the first phase, the rats learned to make two different instrumental responses for qualitatively different, but motivationally similar, food USs. For instance, Lever pressing was reinforced with pellets and Chain pulling was reinforced with sucrose water. During the Pavlovian Backward conditioning phase, pellets preceded a Tone on one type of trial and sucrose water preceded Light on a second type of trial. A long time interval (about 7 min on average) separated different trials, and the two types of trials occurred in an irregular sequence. The Lever and Chain were NOT present during this phase, so that the animals had no opportunity to learn anything about these stimuli and responses during this phase. In the final test phase, the effects of the Tone and Light stimuli upon choice between Lever pressing and Chain pulling was assessed in order to determine if sensory-specific learning developed in the backward conditioning procedure. If the CSs associated with a common motivational response to the noncontinuation of pellets and sucrose (i.e., the S-R view), then Tone and Light would not be expected to exert selective control over choice responding during the Pavlovian-to-instrumental transfer test. What happened, however, was that by the end of the stimuli the CSs selectively decreased the response that had earlier been reinforced with the same outcome as that paired backwardly with the CSs. For example, Tone decreased Lever pressing while Light decreased Chain Pulling. This result supports the notion that the CSs had indeed entered into an inhibitory association with the specific USs with which they were backwardly paired. In other words, at least in the backward conditioning procedure, it seems as though sensory-specific S-S inhibitory associations are learned.
B. Sensory and affective/motivational
properties of the US
1. Blundell, Hall, & Killcross (2001). In this study, Blundell, et al. examined whether it was possible to demonstrate that a CS was capable of associating either aspect of the US and also to assess whether the basolateral amygdala was critical in sensory learning. Two groups of rats were trained to associate each of two CSs with different reinforcing outcomes (i.e., CS1 with US1 and CS2 with US2). The two USs differed in their sensory but not affective properties. Prior to this conditioning, rats had earlier been trained to make one lever press response for US1 and a second lever press response for US2. During a test phase, the rats chose between the two levers and each CS was tested for its effects on choice. In the control group, each CS biased instrumental response choice in favor of the lever press response that shared an outcome with the CS. This selective effect of the CS on choice implies that the CS conveyed information about the specific US with which it was associated. This can be realized by assuming that the CSs associated with the specific sensory qualities of the USs. In contrast, the rats whose basolateral amygdala was lesioned (ie., removed) showed a different pattern of responding. The CSs in this group non-selectively elevated lever press responding relative to when neither CS was present during the test session. In other words, both lever press responses were made more frequently in the presence of each CS compared to when no CS was presented in the test. This result implies that the CSs had associated with the affective/motivational properties of the US, but not the sensory properties. It is as though the basolateral amygdala is responsible for the association that ordinarily forms between the CS and the sensory qualities of the US. When you remove this brain structure, this component of the learning is also removed.
2. Betts, Brandon, & Wagner (1996) study. In this study, Betts et al. examined the question of whether under normal circumstances learning about the sensory and affective properties of the US is independent of one another. They examined this question by determining if it was possible to learn about the sensory properties of a US under conditions in which learning about the affective properties was blocked. One group of rabbits were taught initially to expect a left eye shock after one auditory CS (A1 -> shock (left)), a right eye shock after a second auditory CS (A2 -> shock (right)), and no shock after a third auditory CS (A3-). During phase 2, three separate groups were trained to expect a left eye shock US after an auditory-visual compound CS. Conditioning to the added visual CS was assessed in a subsequent phase. One subgroup was trained with A1+L -> Shock (left) trials. The second subgroup was trained with A2+L -> Shock (left) trials. The third subgroup was trained with A3+L -> Shock (left) trials. Conditioning to the L CS was assessed in two ways : (1) by monitoring left eyeblink CRs directly conditioned to L and (2) by determining to what extent L would potentiate the startle response to an airpuff stimulus. Relative to Group 3, Group 1 displayed little evidence of conditioning with either measure. However, Group 2 displayed comparable amounts of eyeblink CRs to the L as did the Group 3 (the control group), but they also showed no evidence of learning in the potentiated startle test. One can take the eyeblink measure as an indication of the association between the CS and the specific sensory qualities of the shock US. One may also take the potentiated startle response measure as an indication of the association between the CS and the affective/motivational qualities of the US. These data show that there was blocking of learning about both of these attributes in Group 1, but blocking only of the association with the motivational component but not the sensory component of the US in Group 2. Thus, it seems possible for the normal animal to learn about the sensory component of the US even when there is no evidence of learning about the motivational component. This would seem to suggest that learning about these two associations can proceed independently.
C. Binary or Hierarchical Associative
Structures.
1. Feature Positive Occasion Setting. In this task one CS indicates when another CS is paired with the US. For example, in a pigeon autoshaping study, Rescorla (1985) taught birds that a Red keylight was followed by grain only when it was preceded by a Noise stimulus. There were N -> R -> grain trials interspersed with R- trials. The birds learned to peck the Red keylight more when it was presented after N than when it was presented alone. This could mean that two independent associations (N - grain and R - grain) summate their effects to cause more responding on the N -> R trials. Alternatively, it could be that the N enters into a hierarchical association with the R - grain association (an N - [R-grain] hierarchical association). This possibility was examined by determining if separate N- extinction trials conducted after the training phase would undermine N's ability to elevate responding to the Red keylight CS. In fact, this is what Rescorla observed in a final test. He compared control by the N "occasion setting" stimulus to another occasion setting stimulus that was trained identically to the N stimulus in phase 1 but that did not also receive separate extinction trials. Comparing the control by these two occasion setting stimuli revealed that separate extinction given to N had no impact on N's control over R responding. This result cannot be reconciled with the view that the original discrimination was learned on the basis of two binary associations summating their effects during the N -> R trials. Instead, the data can be understood in terms of the N stimulus entering into a hierarchical association with the R-grain link.
2. Feature Negative Occasion Setting. This task is the conceptual opposite to feature positive occasion setting. Initially, the Red keylight is paired with the grain US except when it follows the Noise stimulus. Over training, more keypeck CRs develop on trials with the Red keylight by itself compared to trials in which the Noise precedes it. Is this due to simple conditioned inhibition to Noise summating with conditioned excitation to the Red keylight? Or, is this due to the Noise acquiring hierarchical control over keylight responding? This was answered by turning Noise into an excitatory CS by giving the pigeons separate Noise -> grain pairings for one group but not for a control group in the next phase. Finally, the inhibitory effects of Noise upon responding to the Red keylight as well as to a transfer excitator were assessed in a test phase. Both groups displayed equal inhibitory control by the Noise, in spite of the fact that Noise was paired repeatedly with the grain US during the second phase in one of the groups. Those separate pairings, in other words, had absolutely no impact on N's ability to inhibit responding to the Red keylight. Both of these studies suggest that it is possible for a stimulus to enter into an association with other CS-US pairs, and that this association displays different properties than would be expected of a binary association. In other words, they provide support for the view that hierarchical associative structures can be learned.
3. Renewal. The renewal phenomenon has also been understood in terms of occasion setting. In this case (e.g., Bouton & King, 1983), the context is said to develop occasion setting control over extinction to a CS. This may be revealed by the fact that when the extinguished CS is tested outside of the experimental context in which extinction took place, the CR returns to the CS. This fact could reflect the possibility that the new learning that occurs during the extinction phase is what becomes the target of the occasion setting by the context. This would mean that in the absence of that contextual stimulus, it should be difficult to retrieve the information about extinction. This would result in more responding when the test occurs outside of the extinction context providing that the acquisition memory is more readily retrieved (i.e., is less context specific than is the extinction memory). Can you think of another interpretation of the results from the Bouton & King (1983) study in terms of the learning of binary associations? How would you go about ruling out that alternative interpretation?