Is Perruchet's Dissociation Between Eyeblink ... - Semantic Scholar

Journal of Experimental Psychology: Animal Behavior Processes 2009, Vol. 35, No. 2, 169 –176

© 2009 American Psychological Association 0097-7403/09/$12.00 DOI: 10.1037/a0013294

Is Perruchet’s Dissociation Between Eyeblink Conditioned Responding and Outcome Expectancy Evidence for Two Learning Systems? Gabrielle Weidemann

Jason M. Tangen

University of New South Wales

University of Queensland

Peter F. Lovibond and Christopher J. Mitchell University of New South Wales P. Perruchet (1985b) showed a double dissociation of conditioned responses (CRs) and expectancy for an airpuff unconditioned stimulus (US) in a 50% partial reinforcement schedule in human eyeblink conditioning. In the Perruchet effect, participants show an increase in CRs and a concurrent decrease in expectancy for the airpuff across runs of reinforced trials; conversely, participants show a decrease in CRs and a concurrent increase in expectancy for the airpuff across runs of nonreinforced trials. Three eyeblink conditioning experiments investigated whether the linear trend in eyeblink CRs in the Perruchet effect is a result of changes in associative strength of the conditioned stimulus (CS), US sensitization, or learning the precise timing of the US. Experiments 1 and 2 demonstrated that the linear trend in eyeblink CRs is not the result of US sensitization. Experiment 3 showed that the linear trend in eyeblink CRs is present with both a fixed and a variable CS–US interval and so is not the result of learning the precise timing of the US. The results are difficult to reconcile with a single learning process model of associative learning in which expectancy mediates CRs. Keywords: eyeblink conditioning, expectancy, humans, conditioned stimulus, unconditioned stimulus

systems operate independently, it is postulated that CRs may occur without conscious awareness of the contingency between the CS and the US and that under some circumstances, US expectancy will not influence CR production. Although this dual-system view of associative learning is widely accepted, the empirical evidence on which it is based is often quite weak. For example, it has been argued that differential delay eyeblink conditioning can occur in the absence of conscious awareness of the contingency relations (Clark & Squire, 1998, 1999; C. N. Smith, Clark, Manns, & Squire, 2005). However, such single dissociations between awareness and conditioned responding can provide only very weak evidence for the existence of two separate learning systems. As Lovibond and Shanks (2002) have argued, all that is required to produce such dissociation is the use of an insensitive measure of awareness. The alternative to the dual-system view is that there is only a single learning system that gives rise to both conscious awareness of the contingency between the CS and the US and production of the CR (Brewer, 1974; Lovibond & Shanks, 2002). Bolles (1972) proposed one mechanism by which knowledge of the contingency could give rise to the CR: The CS “retrieves” the CS–US contingency, leading to a representation or expectancy for the US and thereby eliciting an appropriate CR (see also Perruchet, 1985a). More generally, though, the idea is that whatever learning processes give rise to US expectancies that can be verbally reported also produce the CR. According to this hypothesis, CR production should not occur in the absence of any expectation of the US. In a very thorough review of the literature, Lovibond and Shanks (2002) concluded that almost all of the available evidence supports the single-system view of conditioning. However, they did highlight two experiments that appear to provide some

Pavlovian conditioning is an associative process that is the result of the pairing of a neutral conditioned stimulus (CS; e.g., a tone) with a motivationally significant unconditioned stimulus (US; e.g., an airpuff to the eye) in a predictive fashion. A consequence of these pairings that indicates that learning has occurred is the development of the conditioned response (CR; e.g., an eyeblink) to the CS. Human participants can also demonstrate associative learning by simply reporting when they expect the US to occur, by describing the nature of the contingency between the CS and the US, or both. There is much debate as to whether these two types of response, CRs and verbal reports (of the US expectancy or the CS–US contingency), are the consequence of the same psychological processes. Some authors have contended that, at least in some circumstances, there are two independent learning systems in operation in Pavlovian conditioning (Clark & Squire, 1998; Reber & Squire, 1994). One system leads to declarative or consciously available knowledge of the stimulus contingencies. The other system is more primitive and involves the automatic formation of excitatory links between nodes representing the CS and the US. Because these two

Gabrielle Weidemann, Peter F. Lovibond, and Christopher J. Mitchell, School of Psychology, University of New South Wales, Sydney, New South Wales, Australia; Jason M. Tangen, School of Psychology, University of Queensland, Brisbane, Queensland, Australia. This research was supported by Australian Research Council Grant DP0774395. Correspondence concerning this article should be addressed to Gabrielle Weidemann, School of Psychology, University of New South Wales, Sydney NSW 2052, Australia. E-mail: [email protected] 169

170

WEIDEMANN, TANGEN, LOVIBOND, AND MITCHELL

support for a dual-system model, an evaluative conditioning experiment using flavors conducted by Baeyens, Eelen, Van den Bergh, and Crombez (1990) and an eyeblink conditioning study conducted by Perruchet (1985b). This article focuses on the latter study. Perruchet (1985b) provided evidence for a double dissociation between US expectancy and eyeblink CRs in eyeblink conditioning using a 50% random partial reinforcement schedule. In Perruchet’s (1985b) study, participants provided ratings during the intertrial interval specifying the degree to which they expected the airpuff to occur after the presentation of the next CS. In this schedule, the probability of the CS being reinforced is 0.5, regardless of the nature of the preceding trial.1 However, Perruchet found that after a run of nonreinforced trials, participants were more likely to expect the US to occur on the next trial. Conversely, following a run of reinforced trials, participants were less likely to expect the US to occur on the next trial. That is, they showed the common “gambler’s fallacy” (Burns & Corpus, 2004; Keren & Lewis, 1994). What makes Perruchet’s study so important is that participants’ eyeblink CRs revealed the opposite pattern: As the number of consecutive nonreinforced trials increased, participants’ eyeblink CRs progressively decreased, and as the number of consecutive reinforced trials increased, their eyeblink CRs progressively increased (see also Higgins & Prokasy, 1968; Prokasy & Kumpfer, 1969). This double dissociation between US expectancy and the CR, hereinafter referred to as the Perruchet effect, is obviously a significant problem for single-system models of conditioning that predict that CRs and US expectancy should be correlated. Although the Perruchet effect is certainly intriguing and seems to lend support to dual-system models of conditioning, some questions remain about the origins of the observed dissociation. As Lovibond and Shanks (2002) pointed out, it is unclear whether the effect of reinforced and nonreinforced runs on the production of the CR is the result of a change in the associative strength of the CS or is the result of some other nonassociative influence on responding. For example in a partial reinforcement design, CS–US pairings are perfectly confounded with the occurrence of the US, and CS-alone trials are perfectly confounded with the absence of the US. Thus, the effect of reinforced and nonreinforced runs on responding may be the result of the pattern of recent US exposure, not a change in the associative strength of the CS. Perhaps recent presentations of the US sensitize the participant to the US through increased activation of the US representation (Rescorla & Heth, 1975). Perruchet (1985b) provided some evidence against this US sensitization explanation through the inclusion of an unpaired control group, which received CSs and USs in an explicitly unpaired fashion. Consistent with the idea that the Perruchet effect is not the consequence of US sensitization, participants in this group did not show changes in eyeblink CRs as a consequence of US recency. However, US sensitization in this control group may have been obscured by the lack of any associative connection between the CS and the US. If the excitatory strength of the CS was very low, even a US representation that was strongly activated (as a consequence of US sensitization) might not have been triggered by the CS, and so little responding would be observed even after many US presentations. Thus, the unpaired control group is not the most appropriate test of the US sensitization hypothesis. In Experiment 1, we examined the possibility that the pattern of CRs

observed by Perruchet (1985b) was because of US sensitization under the partial reinforcement conditions in which the Perruchet effect was originally observed.

Experiment 1 This experiment used a within-subject design in which participants were exposed to a partial reinforcement schedule, similar to that used by Perruchet (1985b). The focus of the experiment was on the factors influencing CR production in the Perruchet effect. Consequently, we did not include a trial-by-trial measure of expectancy to show what we know from previous experiments is a very robust gambler’s fallacy effect in US expectancy (Clark, Manns, & Squire, 2001; Perruchet, 1985b; Perruchet, Cleeremans, & Destrebecqz, 2006). The sequences of trials were arranged such that each participant received runs of CS–US and CS-alone trials, in the same proportion as Perruchet (1985b), but intermixed with runs of US-alone and blank trials. If US sensitization is responsible for the linear trend in responding after runs of reinforced and nonreinforced trials, then CRs should increase progressively after runs of US-alone trials just as they do after runs of CS–US pairings (and they should decrease progressively after runs of blank trials, just as they do after CS-alone presentations). If, as Perruchet (1985b) argued, the changes in CRs observed in the Perruchet effect are the consequence of changes in the strength of a CS–US association, then the introduction of runs of US-alone presentations should not increase (and blank trials should not decrease) the probability that a CR will be observed on the following trial.

Method Participants. The participants were 46 students (31 women and 15 men) from the University of New South Wales (Sydney, New South Wales, Australia), who received credit toward a course requirement for their participation. They averaged 20.3 years of age (range ⫽ 17– 46). Apparatus. The CS was an 85-dB, 1000-Hz tone of 1,350-ms duration superimposed over a background of a 75-dB white noise. The auditory CS and background white noise were delivered via Sennheiser HD515 headphones. The background white noise was used to mask any sound occurring outside of the conditioning room and thus reduce any external auditory distractions. The US was a 100-ms, 15-psi puff of air (measured at the point of generation) by an eyeblink airpuff unit (San Diego Instruments, San Diego, CA). It was delivered to the left eye via 2 m of flexible plastic tubing terminating in a 1-mm nozzle attached to the left frame of a pair of spectacles worn by the participant. An infrared emitter and an infrared detector for measuring eyeblinks were also attached to the nozzle. Conditioning and recording were carried out by an experimental interface (SG-500, Med Associates Inc., St. Albans, VT) connected to an Intel Pentium computer. Med-PC experimental control software was used to program the conditioning session and to record the eyeblink data. Procedure. Participants gave informed consent and were then seated in a dimly lit room separate from the apparatus room. 1

With the exception that after a run of four consecutive trials of the same type, there was always an alternation in the nature of reinforcement.

PERRUCHET EFFECT

171

500-ms interval before the onset of the US on CS–US paired trials). In addition, to be scored as a CR, the maximal amplitude of the blink had to be equal to or greater than 20% of the same participant’s maximum blink amplitude in response to the first five US presentations of the conditioning session. Data analysis. Data from the experiment were analyzed by means of a set of planned contrasts using a multivariate, repeated measures model (O’Brien & Kaiser, 1985). For each contrast, corresponding Bonferroni-corrected (95%) standardized confidence intervals (CIs) were calculated, using the procedure described by Bird (2002).

Participants were told that they would be presented with tonealone trials, tone-followed-by-airpuff trials, and airpuff-alone trials. They were asked to not promote or inhibit their eyeblink response or close their eyes for long periods of time. Altogether, there were 144 conditioning trials, of which 52 were CS–US trials, 52 were CS-alone trials, 20 were US-alone trials, and 20 were blank trials on which no stimulus events were presented. For the CS–US trials, the interstimulus interval (ISI) between the presentation of the CS and the US was 1,250 ms, and the CS and US coterminated. The number of runs of Lengths 1, 2, 3, and 4 and the number of trials are depicted in Table 1. The number of CS-present runs (shown in the upper section of Table 1) preserved the ratio of runs of CS-alone and CS–US trials used by Perruchet (1985b), but reduced the overall number of such runs to limit the overall length of the conditioning session. Thus, as the length of the run increased by 1, the number of runs of this length decreased by half. There were two CS-absent (i.e., blank and US-alone) runs of each length. This, and the corresponding number of trials, is depicted in the lower section of Table 1. Four different counterbalanced trial orders were used. The original randomized sequence was created by randomly selecting CS-alone and CS–US runs alternately, such that a CS-alone run was always followed by a CS–US run and a CS–US run was always followed by a CS-alone run. Then blank and US-alone runs were randomly inserted after CS-alone and CS–US runs of Length 1. The reason for placing each CS-absent trial after a run of Length 1 is as follows. It is not possible to collect CR data on CS-absent (blank and US-alone) trials because they do not include a CS. The CS-present runs of Length 1 are both more numerous and theoretically less interesting than those of Length 2 and above. Thus, if CS-absent trials are placed only after runs of Length 1, then it is possible to collect CR data after all of the, theoretically more interesting, CS-present runs of Lengths 2, 3, and 4, as well as some runs of Length 1. A second trial order was created by replacing the CS-alone trials from the original sequence with CS–US trials and vice versa. A third trial order was created by replacing the USalone trials from the original sequence with blank trials and vice versa. Finally, a fourth trial order was created by applying both of these trial substitutions simultaneously. CR definition. An eyeblink CR was defined as a blink occurring between 750 and 1,250 ms after the CS onset (i.e., in the

Results and Discussion The percentage of trials on which a CR was observed as a function of the length and the nature of the run preceding those trials is plotted in Figure 1. CRs after runs of trials on which the CS was present (solid black squares) showed a strong positive linear trend as a function of recent reinforcement history. Specifically, participants were most likely to produce a CR when they had just experienced a run of three or four CS–US trials and least likely to produce a CR when they had just experienced a run of three or four CS-alone trials. Statistical analysis confirmed that there was a significant linear trend in CRs after runs of CS-alone and CS–US trials, F(1, 45) ⫽ 12.77, p ⬍ .05, 95% CI ⫽ 0.50 – 1.80. This finding is a replication of Perruchet (1985b) and also of an earlier study by Higgins and Prokasy (1968). However, CRs after runs of trials on which the CS was absent (open circles) showed no evidence of a positive linear trend as a function of recent US exposure history (F ⬍ 1). Overall, there was no statistically significant difference in the mean level of CRs after runs in which the CS was present and runs in which the CS was absent (F ⬍ 1). However, there was a statistically significant interaction between the linear trend after runs in which the CS was present compared with runs in which the CS was absent, F(1, 45) ⫽ 7.30, p ⬍ .05, 95% CI ⫽ 0.017– 0.117. The different pattern in responding after runs in which the CS was present and runs in which the CS was absent suggests that the increase in responding after runs of CS–US trials is not because of US sensitization and the decrease in responding after runs of CS-alone trials is not because of a diminution of US sensitization.

Table 1 Organization of Trials in Experiment 1 CS present Parameter Run length No. runs No. trials

US absent (CS alone) 4 2 8

3 4 12

2 8 16

US present (CS–US) 1 1 16 16 16 16 CS absent

US absent (blank) Run length No. runs No. trials Note.

4 2 8

3 2 6

2 2 4

2 8 16

3 4 12

Total 4 2 8

60 104

4 2 8

16 40

US present (US alone) 1 2 2

1 2 2

2 2 4

3 2 6

Run length refers to the number of trials of the same type presented consecutively.


172

Perce ent CRs

50

CS Present CS Absent

40

30

20

4

3

2

US Absent

1

1

2

3

4

US Present

Figure 1. Mean percentage of conditioned responses (CRs) as a function of the run length (one to four trials) and nature of recent unconditioned stimulus (US) exposure history (US absent vs. US present) of the preceding run in Experiment 1. The US-absent stipulation in the conditioned stimulus (CS)-present condition refers to CS-alone trials and in the CS-absent condition refers to blank trials. The US-present stipulation in the CSpresent condition refers to CS–US paired trials and in the CS-absent condition refers to US-alone trials. Solid black squares represent CS present; open circles represent CS absent.

Perruchet (1985b) provided some initial evidence against the sensitization account, and our data support his findings. However, Experiment 1 goes further toward demonstrating that the trend in responding after runs of reinforced and nonreinforced trials is not because of US sensitization. In this experiment, participants showed a clear linear trend in CRs after runs of reinforced and nonreinforced trials. If US sensitization was responsible for this trend, then it is clear that the conditions are right to demonstrate an effect of US-alone presentations (and blank trials) on CRs, but no such effect was observed. This argues against a US sensitization explanation of the pattern of eyeblink CRs in the Perruchet effect and therefore supports the idea that this pattern is the result of changes in the strength of the CS–US association. However, there is evidence from other studies of human eyelid conditioning that unpaired presentations of the US can motivate CR performance to a CS that has previously been paired with the US (Dufort & Kimble, 1958; Kimble, Mann, & Dufort, 1955). After 20 initial CS–US paired trials, Kimble et al. (1955) compared responding in a group of participants who received a further 20 CS–US paired trials with a group who received 20 US–alone trials and a group that sat in the apparatus for an equivalent amount of time and looked at a fixation point. They found that when participants were subsequently presented with further CS–US trials, the group that received the US-alone trials showed an equivalent level of CRs to those who received the CS–US trials and more CRs than the fixation group. It is not immediately obvious why a similar effect was not seen in this experiment. One possibility lies in the nature of the dependent variable. The assessment of CRs after US-alone presentations in the Kimble et al. (1955) experiment was an average of CRs emitted during the CS on the 10 ensuing CS–US paired trials. The assessment of CRs after USalone trials in this experiment was restricted to the first CS pre-

sentation after the run of US-alone trials, which was sometimes paired with the US and sometimes presented alone. It may be that the CR to this first CS after a run of CS-absent trials is not a good measure of US sensitization. Thus, one possible explanation for the failure to detect US sensitization in Experiment 1 is that the surprising representation of the CS after a run of CS-absent trials interfered with the expression of US sensitization in some way. It is not immediately obvious how the surprise induced by this unexpected CS might have masked a US sensitization effect. However, in light of Kimble et al.’s (1955) results, it would seem premature to rule out the possibility of a US sensitization account of the CR trend seen in CS-present trials on the basis of a single experiment. Experiment 2 was designed to provide a further test of the sensitization hypothesis. In this experiment, the data were collected not on the first CS presentation after a run of CS-absent trials, but on the second CS presentation.

Experiment 2 The design of Experiment 2 was similar to that of Experiment 1, in that participants were presented with runs of CS–US trials and CS-alone trials intermixed with runs of US-alone and blank trials. Experiment 2 differed from Experiment 1 in that the data collected after CS-absent runs came not from the first but from the second CS that occurred after the US-alone or blank trials. This second CS was expected to be less surprising than the first CS, and therefore may allow any differences in US sensitization produced by the preceding CS-absent trials to be revealed. The way in which this was implemented was simply by taking each CS-absent run of Length 2, 3, and 4 (of which there were two of each, as in Experiment 1) and presenting a CS on the final trial of the run. For example, on a run of four US-alone trials, a CS was presented on the last of those trials (US, US, US, and CS–US). The CR was then measured on the following trial, as in Experiment 1. For the sake of simplicity, the CS-absent runs (on all but the last trial of the run) are referred to as runs of blank trials and runs of US-alone trials, even though a CS was presented on the final trial of the run. This design had two unavoidable consequences. First, of course, one trial of each of the CS-absent runs was in fact a CS-present trial. Thus, any trend observed after CS-absent runs may, to some extent, be because of that CS-present trial. Second, it was not possible to implement any CS-absent runs of Length 1 (because a CS would be presented on that trial). Just as in Experiment 1, if the pattern of eyeblink CRs in the Perruchet effect is the result of US sensitization, then there should be an increase in CRs after runs of US trials, regardless of whether the US is presented alone or whether it is signaled by a CS. Similarly, there should be a decrease in CRs after runs of no-US trials, regardless of whether these trials include a CS or not. If the trend in CRs seen in the Perruchet effect is because of changes in associative strength, then no CR differences as a consequence of CS-absent trials should be observed.

Method Participants. The participants were 40 students (30 women and 10 men) from the University of New South Wales who received credit toward a course requirement for their participation. They averaged 19.6 years of age (range ⫽ 17–30).

PERRUCHET EFFECT

Apparatus and procedure. The apparatus and procedure were the same as those used in Experiment 1, except as indicated below. There were 140 conditioning trials, of which 58 were CS–US trials, 58 were CS-alone trials, 12 were US-alone trials, and 12 were blank trials. The numbers of CS-present runs were the same as in Experiment 1 (see upper section of Table 1). In addition, there were two CS-absent runs of Lengths 2, 3, and 4, but no CS-absent runs of Length 1. Four different original sequences and two different counterbalanced orders of those sequences were used. The original sequences were created by randomly alternating between the set of US-absent (CS-alone and blank) runs, and the set of US-present (CS–US and US-alone) runs, with the restriction that blank and US-alone runs could not occur consecutively and could not occur after CS-alone and CS–US runs of Length 4. This restriction was included so that it was possible to collect data on all of the, very rare, CS-present runs of Length 4 (as pointed out earlier, it is not possible to collect CR data on CS-absent trials). Thus, the less frequently occurring runs of blank and US-alone trials, and the CS-alone and CS–US runs of Length 4, were always followed by a CS-present trial, ensuring the collection of CR data. The counterbalancing of the original sequences was created by replacing the CS-alone trials with CS–US trials and by replacing the US-alone trials with blank trials and vice versa.

Results and Discussion Percentage of CRs as a function of the length and the nature of the run preceding each trial are plotted in Figure 2. As in Experiment 1, CRs after runs of trials on which the CS was present (solid black squares) showed a strong positive linear trend as a function of recent reinforcement history. That is, participants showed a

50

Percentt CRs

CS Present CS Absent 40

30

20

4 3 2 1 1 2 3 4 US Absent

US Present

Figure 2. Mean percentage of conditioned responses (CRs) as a function of the run length (one to four trials) and nature of recent unconditioned stimulus (US) exposure history (US absent vs. US present) of the preceding run in Experiment 2. The US-absent stipulation in the CS-present condition refers to CS-alone trials and in the CS absent condition refers to one to three blank trials followed by a CS-alone trial. The US-present stipulation in the CS-present condition refers to CS–US paired trials and in the CS-absent condition refers to one to three US-alone trials followed by a CS–US paired trial.

173

progressive increase in responding after runs of CS–US trials and a progressive decrease in responding after runs of CS-alone trials. Statistical analysis confirmed that there was a significant linear trend in CRs as a function of recent runs of reinforced and nonreinforced trials, F(1, 39) ⫽ 10.29, p ⬍ .05, 95% CI ⫽ 0.030 – 0.133. However, CRs after runs of trials on which the CS was absent (open circles) showed no consistent linear trend in CRs as a function of recent US exposure history. Statistical analysis of the linear trend in CRs as a function of recent US exposure history in the CS-absent condition was not significant (F ⬍ 1).2 Overall, there was no statistically significant difference in the mean level of CRs after runs in which the CS was present and those in which the CS was absent (F ⬍ 1). However, there was a statistically significant interaction between the linear trend across runs and whether the CS was present or absent during those runs, F(1, 39) ⫽ 7.01, p ⬍ .05, 95% CI ⫽ 0.001– 0.110. Overall, the findings of Experiment 2 replicated those of Experiment 1: Eyeblink CRs increased with repetitions of CS–US pairings and decreased with repetitions of CS-alone trials. This linear trend in responding after runs of reinforced and nonreinforced trials was not reproduced after runs of US-alone and blank trials. Hence, recent reinforcement history and not-recent US exposure history seems to be the critical variable influencing subsequent CR likelihood in the Perruchet effect. This result suggests that the trend in eyeblink CRs in the Perruchet effect is not because of the effects of US sensitization and is consistent with an associative strength explanation of eyeblink CRs.

Experiment 3 It seems clear that when participants emit a strong CR in the Perruchet effect, they do so as a consequence of some process that is unrelated to whether they expect a US to be presented on that trial. The question then is what this second factor is. Clark and Squire (1998) have argued that this second factor is a separate procedural learning system in which associative links are automatically formed between the CS and US. However, there are other possibilities. When a CS is repeatedly paired with a US, there are many things that can be learned, about both each stimulus individually and the relationship between them. Experiments 1 and 2 ruled out US sensitization as the second factor that produces the trend in CRs in the Perruchet effect. Another possibility is that participants learn the precise timing of the CS–US interval. Thus, as well as learning whether the US is likely to occur on a given trial, participants can also learn when the US will occur (see Bouton, 1997, for a discussion of this issue). Perhaps US expectancy on the following trial (which is unlikely to be affected by timing issues) is determined by what the participant believes about whether the US is likely to occur, but the CR is largely determined by what the participant believes about the precise interval between the CS and US. In other words, the two responses are not generated by two separate learning systems, but by knowledge acquired by the same system about two different features of the CS–US relationship. It is widely recognized that the time course of the CR in eyeblink conditioning in rabbits is highly attuned to the ISI 2

This linear trend was calculated across the runs of four, three, and two blank trials and the runs of two, three and four US-alone trials.

174


between the onset of the CS and the US (e.g., Gormezano, Kehoe, & Marshall, 1983; M. C. Smith, 1968; White, Kehoe, Choi, & Moore, 2000). During conditioning, the rabbit learns the precise timing of the US and that the US will follow the CS. This timing is seen in the gradual acquisition of an eyeblink CR whose maximal closure coincides with time of US delivery across a wide variety of ISIs (e.g., Gormezano et al., 1983; M. C. Smith, 1968; White et al., 2000). There is also some evidence in human eyeblink conditioning to suggest that the eyeblink CR is sensitive to the ISI. During conditioning, there is an increase in CRs that coincide with the US (McAllister, 1953). If the ISI is subsequently shifted to a new ISI, then CRs that coincide with the old ISI gradually decrease and CRs that coincide with the new ISI gradually increase (Boneau, 1958). It may be that the linear trend in eyeblink CRs in the Perruchet effect reflects a refinement of the timing of CRs after CS–US trials and, conversely, a greater variance in the timing of CRs after the CS-alone trials, rather than a change in the strength of the CS–US association. Experiment 3 used a two-group design to examine this possibility. In one group of participants, the fixed group, the length of the CS was fixed at 900 ms. In a second group of participants, the variable group, the length of the CS was varied between 500 ms and 1,300 ms with an average CS duration of 900 ms. The consequence of these CS durations was that in the fixed group, the ISI between the onset of the CS and the onset of the US was always 800 ms, whereas in the variable group, the ISI varied between 400 ms and 1,200 ms, with an average ISI of 800 ms. If the trend in CRs seen in the Perruchet effect is the consequence of learning about the precise CS–US interval (i.e., when the US will be presented), then the fixed and variable groups should show very different patterns of CR production. The fixed group would be expected to replicate the trend in CRs seen on the CS-present trials of Experiments 1 and 2 (no CS-absent trials were presented to either group). The variable group would, by contrast, show no trend because the lack of temporal constancy between the onset of the CS and the US in this group precludes the formation of a precise temporal representation that would allow appropriate timing of the CR. If, however, the CRs are largely generated by acquisition of knowledge as to whether the US is likely to occur, then the trend seen in Experiments 1 and 2 should be observed in both the fixed and the variable groups.

Method Participants. The participants were from the University of New South Wales and received credit toward a course requirement for their participation. Participants were assigned to one of two groups. There were 12 individuals in the fixed group (6 women and 6 men averaging 19.2 years of age; range ⫽ 18 –23) and 12 individuals in the variable group (6 women and 6 men averaging 19.5 years of age; range ⫽ 17–23). Apparatus and procedure. The apparatus was the same as that used in Experiment 1. The procedure used in Experiment 3 differed from that used in Experiment 1 in that there were no US-alone and blank trials (those indicated in the lower part of Table 1). The number of trials of each type was the same as the number of CS-present trials in Experiment 1 (see the upper part of Table 1).

Experiment 3 contained two groups, the fixed group and the variable group. The timing of stimulus events on CS–US trials in both groups is represented in Figure 3. The top section of Figure 3 illustrates the timing of stimulus events on CS–US trials in the fixed group, where the length of the CS was fixed at 900 ms and the ISI between the presentation of the CS and the US was fixed at 800 ms. The timing of stimulus events in the fixed group on CS-alone trials was exactly the same as on CS–US trials except the US was not presented. The lower section of Figure 3 illustrates the timing of stimulus events on CS–US trials in the variable group. The gray section of this figure represents the 800-ms window within which CS onset occurred. The 800-ms variable window was followed by a fixed 500-ms period during which the CS remained on. The US was presented in the last 100 ms of this latter period. The timing of stimulus events in the variable group on CS-alone trials was exactly the same as on CS–US trials except the US was not presented. The average duration of the CS in the variable group was 900 ms and varied between 500 ms and 1,300 ms in a continuous fashion. The average ISI in the variable group was 800 ms and varied between 400 ms and 1,200 ms. Four different original sequences and two different counterbalanced versions of those sequences were created in the same way as in Experiment 2. CR definition. An eyeblink CR was defined as a blink occurring between 500 ms and 100 ms before the CS offset (i.e., in the 400-ms interval before the onset of the US on CS–US paired trials).

Results and Discussion Percentage of CRs as a function of the length and the nature of the run preceding each trial is plotted in Figure 4. For both the fixed group (open circles) and the variable group (solid black squares), there was a progressive increase in responding after runs of CS–US trials and a progressive decrease in responding after runs of CS-alone trials. Statistical analysis confirmed that there was a significant linear trend across runs of CS-alone and CS–US trials in both groups, F(1, 20) ⫽ 9.04, p ⬍ .05, 95% CI ⫽ 0.057– 0.361. Examination of Figure 4 reveals that overall there were fewer CRs in the variable group than in the fixed group. However, this difference was not statistically significant, F(1, 20) ⫽ 2.19, p ⬎ .05. More important, the difference in the linear trend across runs of CS-alone and CS–US trials between the fixed group and the variable group was not statistically significant (F ⬍ 1). The results of Experiment 3 indicate that varying the duration of the CS and the duration of the ISI between the presentation of the

Figure 3. Relative stimulus onsets and offsets in Experiment 3 on conditioned stimulus– unconditioned stimulus (CS–US) trials in the fixed group and the variable group. The gray section of the CS-US trials in the variable group indicates the window in which the CS could onset.

PERRUCHET EFFECT

100

Percentt CRs

80

Group p Fixed Group Variable

60 40 20 0

4 3 2 1 1 2 3 4

non reinforcements reinforcements Figure 4. Mean percentage of conditioned responses (CRs) in both the fixed group and the variable group as a function of run length (one to four trials) and nature (nonreinforcement vs. reinforcement) of the preceding run in Experiment 3.

CS and the US did not interfere with the linear trend in eyeblink CRs observed in the Perruchet effect. This result suggests that the progressive increase in eyeblink CRs after runs of CS–US trials and the progressive decrease in eyeblink CRs after runs of CSalone trials is not the result of participants learning the precise timing of US onset. This result is consistent with the suggestion that the linear trend in eyeblink CRs in the Perruchet effect is the result of participants learning about whether the US is going to occur.

General Discussion In Experiments 1 and 2, we found that eyeblink CRs are progressively strengthened through repetitions of CS–US pairings and weakened through repetitions of CS-alone trials, and the same pattern of responding is not observed after repetitions of US-alone and blank trials. These experiments provide evidence that the pattern of eyeblink CRs in the Perruchet effect is not the result of US sensitization. Furthermore, in Experiment 3, we found that the increase in responding after CS–US trials and the decrease in responding after CS-alone trials was apparent whether the ISI between the presentation of the CS and the US was constant or whether it varied from trial to trial. These data suggest that the data pattern is not the consequence of participants refining their timing of the CR after runs of CS–US trials. Hence, recent reinforcement history, regardless of the precise temporal relationship between the CS and the US, seems to be the critical variable influencing subsequent CR likelihood in the Perruchet effect. Because US expectancy is known to exhibit the opposite pattern of responding (US expectancy is lower after many CS–US pairings), the present findings are consistent with the idea that CRs are generated by a system that is quite separate from that which produces US expectancies. The present findings do not necessarily imply that the learning system that generates CRs operates unconsciously through the automatic formation of an associative link between the CS representation and the US representation. In these experiments, as in the original demonstration by Perruchet (1985b), participants were not

175

prevented from paying attention to the contingency between the CS and the US or to the sequence of trials that they were presented with. In fact, participants were informed about the critical relationship between the CS and the US before the experiment began, and to produce expectancy ratings that are sensitive to the history of CS–US and CS-alone runs, participants must also be aware of those runs. Consequently, on the basis of the present findings, it is unwarranted to conclude that the learning system that generates eyeblink CRs operates unconsciously. Other eyeblink conditioning data have been offered as evidence that the learning system that generates eyeblink CRs can operate in the absence of conscious awareness of the contingency (Clark & Squire, 1998). However, as argued earlier, this evidence is based on a single dissociation—the presence of conditioning and the absence of awareness—and so may simply reflect the use of an insensitive measure of awareness. Similar results have been observed in evaluative conditioning procedures in which neutral stimuli are paired with positive or negative (visual or flavor) stimuli with the consequence that the neutral stimulus inherits the valence of its partner (see De Houwer, Baeyens, & Field, 2005, for review). In this case, however, it would seem that the dissociation between awareness and conditioning at the group level (when CRs and awareness are collapsed across all participants) disguises a strong relationship between awareness and conditioning when the data are analyzed at the level of individual CSs (Pleyers, Corneille, Luminet, & Yzerbyt, 2007; Wardle, Mitchell & Lovibond, 2007). Lovibond and Shanks (2002) found no other convincing evidence for conditioning outside of awareness. It is odd, given that the Perruchet effect is so robust, that there is so little evidence for separate learning systems to be found using other procedures. One possible reason for this may lie in the way in which US expectancy is manipulated in the Perruchet effect. To provide really strong evidence for dissociation between two systems, those systems must be placed in opposition to each other; this is the logic underlying the Perruchet effect. However, it is usually assumed that eyeblink CRs can be produced by both the expectancy-based system and the procedural learning system (Clark & Squire, 1998). Thus, if the two systems are set in opposition to each other, to reveal a strong influence of the procedural system the impact of the expectancy system must be kept to a minimum. The Perruchet effect seems ideal for this purpose. At the start of the experiment, participants are informed of the 50% partial reinforcement contingency. Thus, US expectancy might be expected to remain somewhere near 50% throughout the experiment, with perhaps a very slight trial-by-trial bias produced by the gambler’s fallacy effect. These small fluctuations in US expectancy produced by the gambler’s fallacy may be insufficient to counteract the effect of the changes in procedural learning. Therefore, it may be that minimizing the influence of US expectancy in this way is critical to dissociating the expectancy system from the procedural learning system. Perhaps further experiments in which the potential role of US expectancy is kept to a minimum will provide further evidence for a procedural associative learning system. In conclusion, the experiments presented here found no evidence for any kind of performance-based artifact that might undermine the conclusion that the Perruchet effect is evidence for two learning systems. However, because it is almost the only such evidence, this effect should be subject to further close scrutiny.


176

Future studies may uncover some other factor, not yet considered, that allows the Perruchet effect to be assimilated within the (empirically more strongly supported) single-system approach to Pavlovian learning (Lovibond & Shanks, 2002). We are divided as to the likelihood of this outcome. However, we are all in agreement that these experiments, and Perruchet’s original finding, lend support to a dual-system model of learning.

References Baeyens, F., Eelen, P., Van den Bergh, O., & Crombez, G. (1990). Flavor-flavor and color-flavor conditioning in humans. Learning and Motivation, 21, 434 – 455. Bird, K. D. (2002). Confidence intervals for effect sizes in analysis of variance. Educational and Psychological Measurement, 62, 197–226. Bolles, R. C. (1972). Reinforcement, expectancy, and learning. Psychological Review, 79, 394 – 409. Boneau, A. C. (1958). The interstimulus interval and the latency of the conditioned eyelid response. Journal of Experimental Psychology, 56, 464 – 471. Bouton, M. E. (1997). Signals for whether versus when an event will occur. In M. E. Bouton (Ed.), Learning, motivation, and cognition: The functional behaviorism of Robert C. Bolles (pp. 385– 409). Washington, DC: American Psychological Association. Brewer, W. F. (1974). There is no convincing evidence for operant or classical conditioning in adult humans. In W. B. Weimer & D. S. Palermo (Eds.), Cognition and the symbolic processes (pp. 1– 42). Oxford, England: Erlbaum. Burns, B. D., & Corpus, B. (2004). Randomness and inductions from streaks: “Gambler’s fallacy” versus “hot hand.” Psychonomic Bulletin & Review, 11, 179 –184. Clark, R. E., Manns, J. R., & Squire, L. R. (2001). Trace and delay eyeblink conditioning: Contrasting phenomena of declarative and nondeclarative memory. Psychological Science, 12, 304 –308. Clark, R. E., & Squire, L. R. (1998, April 3). Classical conditioning and brain systems: The role of awareness. Science, 280, 77– 81. Clark, R. E., & Squire, L. R. (1999). Human eyeblink classical conditioning: Effects of manipulating awareness of the stimulus contingencies. Psychological Science, 10, 14 –18. De Houwer, J., Baeyens, F., & Field, A. P. (2005). Associative learning of likes and dislikes: Some current controversies and possible ways forward. Cognition & Emotion, 19, 161–174. Dufort, R. H., & Kimble, G. A. (1958). Ready signals and the effect of interpolated UCS presentations in eyelid conditioning. Journal of Experimental Psychology, 56, 1–7. Gormezano, I., Kehoe, E. J., & Marshall, B. S. (1983). Twenty years of classical conditioning research with the rabbit. In J. M. Sprague & A. N. Epstein (Eds.), Progress in psychobiology and physiological psychology (Vol. 10, pp. 197–275). New York: Academic Press. Higgins, J. D., & Prokasy, W. F. (1968). The effects of transition probability of reinforcement in intermittent reinforcement schedules in human eyelid conditioning. Psychonomic Science, 12, 69 –70.

Keren, G., & Lewis, C. (1994). The two fallacies of gamblers: Type I and Type II. Organizational Behavior and Human Decision Processes, 60, 75– 89. Kimble, G. A., Mann, L. I., & Dufort, R. H. (1955). Classical and instrumental eyelid conditioning. Journal of Experimental Psychology, 49, 407– 417. Lovibond, P. F., & Shanks, D. R. (2002). The role of awareness in Pavlovian conditioning: Empirical evidence and theoretical implications. Journal of Experimental Psychology: Animal Behavior Processes, 28, 3–26. McAllister, W. R. (1953). The effect on eyelid conditioning of shifting the CS–US interval. Journal of Experimental Psychology, 45, 423– 428. O’Brien, R. G., & Kaiser, M. K. (1985). MANOVA method for analyzing repeated measures designs: An extensive primer. Psychological Bulletin, 97, 316 –333. Perruchet, P. (1985a). Expectancy for airpuff and conditioned eyeblinks in humans. Acta Psychologica, 58, 31– 44. Perruchet, P. (1985b). A pitfall for the expectancy theory of human eyelid conditioning. Pavlovian Journal of Biological Science, 20, 163–170. Perruchet, P., Cleeremans, A., & Destrebecqz, A. (2006). Dissociating the effects of autonomic activation and explicit expectancy on reaction times in a simple associative learning task. Journal of Experimental Psychology: Learning, Memory, and Cognition, 32, 955–965. 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. Journal of Experimental Psychology: Learning, Memory, and Cognition, 33, 130 –144. Prokasy, W. F., & Kumpfer, K. L. (1969). Conditional probability of reinforcement and sequential behavior in human conditioning with intermittent reinforcement schedules. Psychonomic Science, 14, 49 –50. Reber, P. J., & Squire, L. R. (1994). Parallel brain systems for learning with and without awareness. Learning & Memory, 1, 217–229. Rescorla, R. A., & Heth, C. D. (1975). Reinstatement of fear to an extinguished conditioned stimulus. Journal of Experimental Psychology: Animal Behavior Processes, 1, 88 –96. Smith, C. N., Clark, R. E., Manns, J. R., & Squire, L. R. (2005). Acquisition of differential delay eyeblink classical conditioning is independent of awareness. Behavioral Neuroscience, 119, 78 – 86. Smith, M. C. (1968). CS-US interval and US intensity in classical conditioning of the rabbit nictitating membrane response. Journal of Comparative and Physiological Psychology, 66, 679 – 687. Wardle, S., Mitchell, C. J., & Lovibond, P. F. (2007). Flavor evaluative conditioning and contingency awareness. Learning and Behavior, 35, 233–241. White, N. E., Kehoe, E. J., Choi, J.-S., & Moore, J. W. (2000). Coefficients of variation in timing of the classically conditioned eyeblink in rabbits. Psychobiology, 28, 520 –524.

Received February 20, 2008 Revision received June 17, 2008 Accepted June 19, 2008 䡲