Pergamon

Behav. Res. Thu. Vol. 32, No. 3, 301-305, 1994 pp. Copyright 0 1994 Elsevier Science Ltd

Printed in Great Britain. All rights reserved ooO5-7967/94 $6.00 + 0.00

THE EVALUATIVE RESPONSE: PRIMITIVE BUT NECESSARY

IRENE MARTIN’ and ARCHIE LEVEY~

‘Department of Psychology, Institute of Psychiatry, De Crespigny Park, Denmark Hill, London SE5 SAF, England

*MRC Applied Psychology Unit, Cambridge, England

(Received 27 January 1993)

Summary-We do not believe the claim that “evaluative conditioning is a qualitatively distinct form of classical conditioning” [Davey (1994). Behaviour Research and Therapy, 32,291-2991. We view all classical conditioning as a learning process which leads organisms to assign a positive or negative value to previously neutral stimuli and to respond to them accordingly. The evaluative response is a necessary component of this process and hence it is central to all classical conditioning, not a separate type. Davey concentrates on a signal-based information processing view of learning, i.e. the formation of linear associations between CS and UCS of which human subjects are aware and which they can verbalize. We propose a more primitive and more general model in which stimulus evaluation (like/dislike) occurs with a minimal degree of processing, and enters into a representation of stimulus (CS and UCS) and response (CR and UCR) characteristics which is redintegrative rather than associative.

Davey’s (1994) thesis is that classical conditioning is a complex cognitive activity requiring controlled information processing. His prototypical model supposes that the CR is mediated by an internal representation of the UCS and, in humans, that the CS-UCS association is reflected by conscious awareness of the CS-UCS contingency. The prototypical experiment in human classical conditioning is alleged to be the differential electrodermal paradigm. Conditioning is viewed in terms of the learning of associations between stimuli, hence it is referential, or signal learning. Evaluative conditioning does not fit into this model because of two important differences: it appears to occur without awareness of CS-UCS contingencies, and it appears to be strongly resistant to extinction. It is labelled ‘intrinsic’ learning, as distinct from signal learning.

We do not question the evidence that signal learning occurs but regard this-in the sense that Davey describes it in relation to verbalizable awareness-as rather a special aspect of classical conditioning. A more general and more important issue is what is learned/stored/represented when two events occur contiguously in time, and we wish to emphasize the role of valence, hedonic shifts and the acquisition of likes and dislikes within the representational structure. Evidence for this kind of learning exists in a wide-ranging literature on simple conditioning in primitive single-cell organisms, in invertebrates [bees, Limax, molluscs (cf. Carew & Sahley, 1986)], the newborn and infants of many species (cf. Sullivan & Leon, 1987) and in response systems [interoceptive, immune, the acquisition of food likes and dislikes (e.g. Rozin & Zellner, 1985; Garcia, 1990)], involving procedures in which awareness cannot be present. It is also described in experiments on conditioning of attitudes as in consumer research (Stuart, Shimp & Engel, 1987).

Evaluative learning occurs without awareness (Baeyens, Eelen & Van den Bergh, 1990; Levey & Martin, 1990), in common with many other studies of conditioning. It fails to extinguish because, we suggest, the CS and UCS are represented in a CS-UCS complex which does not imply an association between the CS and UCS but an integrated fusion of the two. Thus, every presentation of the CS must evoke the UCR and hence brings about its own reinforcement.

Cognitive views of conditioning have been promoted extensively in recent years, and it is a welcome development. However, there has been an unwelcome tendency to try to account for all conditioning in terms of cognitive generalizations, and to over-interpret the empirical data in terms which lack both precision and meaning. The terms ‘cognitive’ and ‘controlled information processing’, as used by Davey (1994), do not make explicit some important differences in explanatory mechanisms invoked by animal and human researchers. Rescorla’s theory (Rescorla

301

302 IRENE MARTIN and ARCHIE LEVEY

& Wagner, 1972) is based solely on an associative mechanism, and as far as we are aware does not make use of ‘controlled information processing’. The specific view of classical conditioning as mediated by conscious awareness of the CS-UCS contingency seems to arise from the electrodermal differential conditioning paradigm. The implication that Rescorla’s and the human autonomic conditioners’ views can be combined within an all-embracing ‘cognitive’, ‘controlled information processing’ theory of classical conditioning requires justification. We, like Davey, welcome attempts to bring animal and human research into closer contact, but see an alternative route to the one he proposes. Ours would stress the more primitive responses which humans and animals share. It would be regrettable if the strong claim for cognitive factors in conditioning were to displace the more biological approach to learning with its emphasis on the affective significance of events in the external world.

REPRESENTATIONS

Some of the features of evaluative conditioning may need clarification in view of Davey’s criticisms. Specifically, is it an example of S-R or S-S learning? What is the nature of the representation?

Much previous work was based on an S-R view of conditioning, while current conceptions favour an S-S view. Thus, the argument goes, if evaluative conditioning is not a form of signal (S-S) learning it must fit somehow into the S-R category. This is not how we view it. The evaluative response is an internal, not external response. It does not necessarily generate overt action. We use verbalizations and ratings to infer an evaluative response, and to infer that learning has occurred. We believe that both S-S and S-R learning take place and are registered within a representation which contains stimulus elements (of both CS and UCS) and also the evaluation of the UCS, the evaluative response. We refer to it as a holistic representation in which CS, the UCS and, probably, the evaluative response itself are fused or integrated.

The implication is that the once neutral CS contains the liked or disliked elements of the UCS simply because they are fused in the representation and the CS cannot therefore be perceived without the UCS. After conditioning, the CS evokes the same reaction as the UCS. It follows from this that the representation can be modified by such techniques as counterconditioning and re-evaluation. Our model of ‘fusion’ predicts a re-evaluation effect because presentation of the UCS necessarily evokes the CS (just as the CS evokes the UCSjUCR). The subject does not need to ‘know’ that the CS signals something desirable or nasty. This issue has been examined in a paper by Baeyens, Eelen, van den Bergh and Crombez (1992) on US-re-evaluation, which tackles the issue of the kind of representational structure which is built up in human evaluative conditioning. Assuming that such a structure involves associative processes, they make predictions about post-acquisition re-evaluation which depend upon whether the CS-valence is based on an S-S or an S-R type of association:

“To the extent that the CS tracks the changed valence of the US, it has to be concluded that the acquired valence of the CS is based on its referential (i.e. signal, S-S) value to the US. If, on the other hand, the CS does not track the changed value of the US, it can be concluded that the CS has acquired intrinsic valence”.

As described above, our formulation of the representational structure is different from the associative model which they describe and our interpretation of the Baeyens et al. (1992) US-re-evaluation results differs from theirs. They conclude that because the evaluative CR was influenced by US-re-evaluation it follows that the change must be achieved via an S-R association. We would predict the change to occur because of the redintegrative model which we propose.

AWARENESS

We begin with some reminders of the awareness issue in other response systems than the electrodermal. For decades, researchers have examined the role of awareness in human eyelid conditioning, using a variety of techniques. The general conclusion is that information given to or gained by subjects (Ss) has some effect on both simple and differential conditioned eyelid responding, but that the effect is limited. Although an article by Baer and Fuhrer (1982) is often cited as supporting the view that awareness is necessary, the findings are not greatly different from

Evaluative conditioning 303

others. Ss who could verbalize the CS-UCS contingency showed somewhat better discrimination, but those who could not nevertheless conditioned. Baer and Fuhrer (1982) conclude that “unaware subjects did not show conditioning”, but this is misleading. Unaware Ss conditioned perfectly well but did not show differential conditioning. There has been a failure to recognize that even if differential conditioning requires awareness (a controversial view) it need not follow that simple conditioning cannot take place without awareness.

Trial-by-trial analyses with eyelid conditioning, as with electrodermal conditioning, have failed to demonstrate anything like a strong relationship between Ss’ knowledge or expectancy of the UCS and their responding in a differential paradigm. Trial-by-trial measures of expectancy, a technique used by several workers, inevitably draws Ss’ attention to the contingency and leads to enhanced awareness. Furedy has shown, using his trial-by-trial measure, that there is a poor relationship between subjective contingency and the conditioned electrodermal response (Furedy & Schiffman, 1973). Baer and Fuhrer (1982) report that Ss whose awareness developed relatively late in the trial sequence did not exhibit significant differential eyelid responding. Lovibond, Siddle and Bond (1988) in their blocking studies observed a number of discrepancies between the expectancy measure and electrodermal responding. It is worthwhile recalling Razran’s point, made many years ago, that there is no way of knowing whether the developing awareness brings about ‘conditioning’, or vice versa. The conditioning process may itself lead to awareness.

There are well-recognized problems with the assessment of awareness, and these have been tackled in a variety of ways. We accept Davey’s questions about the problems in the evaluative conditioning paradigm, but feel that they are present in different forms in all human conditioning studies. Baeyens and his colleagues have acknowledged the specific problem of pre- and post-assessment of awareness in evaluative conditioning and have carried out studies using both methods. They and we are well aware of the issues and are pursuing them. However, there is a problem with the assessment of awareness which seems unanswerable: until every possible question and method of assessment has been tried, how can we be ‘sure’ that the S did not ‘know’ or was unaware? To demonstrate ‘no awareness’ raises insoluble philosophical issues.

An additional issue is that awareness is probably not a well-defined or unitary state. Ss may be instructed about contingencies but may not use that information. Dawson, Catania, Schell and Grings (1979) noted that some of their fully informed Ss reported no expectancy of the UCS. Are these Ss aware or not? We have observed that some instructed Ss cannot repeat the instructions post-experimentally. Furedy has shown that while Ss can learn positive contingencies with relative ease, they have difficulty with negative contingencies, and particularly with the explicitly unpaired condition. These complexities caution against simplistic conclusions about ‘awareness’.

Davey seems to imply that the ‘prototypical’ or defining paradigm for conditioning is the differential paradigm:

“In the traditional autonomic paradigm. conditioning is said to have occurred when there is a significantly greater amplitude CR to the CS paired with the UCS than to a random control CS which is only randomly paired with the UC%”

Our reading of the literature suggests that the random control is rarely used in human studies. Historically what seems to have happened is that some early studies in electrodermal conditioning revealed that pairings of CS-UCS did not necessarily lead to greater responding to the CS than unpaired (uncoupled) CSs and UCSs. This finding probably contributed to the use of differential paradigms to demonstrate conditioning. This is not true of eyelid conditioning. Unless pairing of the CS-UCS occurs with very close temporal contiguity (the parameters have been thoroughly investigated in both humans and rabbits) no conditioning occurs. It is traditional to assess eyelid conditioning in terms of increments in responding over trials. Typically responding is around zero in early trials and increments significantly in both simple and in discrimination paradigms. Simple conditioning paradigms are very frequently used. Would Davey want to say that these do not demonstrate that conditioning has occurred?

Davey (1994) correctly points out that Shanks and Dickinson (1990) used a non-paired control condition in an evaluative conditioning study. This was, however, a student project on a small number of Ss, and the results were only considered to be suggestive. It is noteworthy that the stimuli used in evaluative conditioning differ from the discrete sensory stimuli typically used, and it is difficult to say how long the stimulus ‘trace’ or the S’s evaluation persists. It seems rash to

304 IRENE MARTIN and ARCHIE LEVEY

conclude that mere separation in time ensures ‘unpaired’ presentation without further investigation of the precise time parameters involved. We have examined CS durations and CS-UCS intervals but they do not seem to contribute strongly to evaluative conditioning.

ASSOCIATIONS

Our earlier work with human classical eyelid conditioning led us to think in terms of a redintegrative model of the CS, UCS, CR and UCR rather than the formation of associations among them. We observed on unreinforced or early extinction trials that the CS alone elicited a response topography in the form of both CR and UCR. This suggested that stimuli paired in time are registered as a CS-UCS stimulus complex, capable of generating both CR and UCR, and involving the storage of the characteristics of both stimuli. The observation that response topography developed over trials towards specifiable endpoints (e.g. appropriate placement in time) and shapes (e.g. CR-UCR blends) further suggested that response characteristics were also involved in the stored model such that it contained both stimulus and response elements.

The development of such a model could be followed over trials, and we proposed the occurrence of stages, an initial stimulus registration the development of a response model and the subsequent integration and shaping of the CR-UCR complex within the CS-UCS complex. To postulate that a connection is formed between them seems unnecessary and unparsimonious. Similarly, to suggest that a connection must be formed between the stimulus complex and the response which it initiates is to introduce a superfluous factor: the association is itself superfluous and the factors which ‘strengthen’ it are no longer required.

According to this view, the resulting stimulus model is integrated in a holistic fashion, and this has inclined us towards those models which illustrate a logical structure such that a single discrete memory can activate the larger information structure of which it is a component (e.g. Hayes-Roth, 1977); to holographic models as a physical instance of redintegration in which a cue can reinstate the total image, and to distributed memory models of the type discussed by Murdock (1983) in which two items are fused or blended such that there is no link or connection at all.

Our model simply stated that when a salient event sequence occurs, the shared characteristics of the stimuli are stored. The evaluative conditioning experiments were designed to analyse the definition of ‘salience’ and to examine the issue of what is stored. In this context, we have argued that visual materials which are strongly liked or disliked will serve as salient stimuli, and that the stimulus characteristics which lead to a S’s like or dislike of the picture will transfer to similar materials to which the S was previously neutral. This implies that salience can be defined for classical conditioning in terms of like and dislike. The immediate registration of an event pair and the analysis of those features of the CS and UCS which are salient and which are stored involve automatic mechanisms of a more primitive sort than the parsing and decision-making which are usually thought of as cognitive processing.

An effect of this formulation is to dispense with the contingency relation as such, and to rely on contiguity to account for the temporal fusion of the stimulus elements. Quite clearly such a formulation must refer to primitive levels of representation, consistent with the exhibition of the basic phenomena of simple conditioning very early in the phylogenesis and ontogenesis of behaviour. The identification of the CS and UCS as separate stimuli is superfluous to this primitive representation, although, of course, it does occur in higher organisms as an additional event. Psychologists have tended to see Pavlovian behaviour through Aristotelian eyes. The CS causes the UCS; the CS is a signal for the UCS. This is neither necessary nor parsimonious. All of the phenomena of simple conditioning can be accounted for in terms of a holistic representation in which the contingency (the referential component) need play no part.

This focus on what we suggest to be the core of classical conditioning-the evaluative response-is only a first step towards understanding what is learned. We have discussed elsewhere the necessity of supplementing this core learning with learning about recurring event sequences, learning the rules which predict events in the environment, and the related development of plans and actions (Levey & Martin, 1983; Martin & Levey, 1987).

Theoretical accounts of conditioning should take into account the wider range of empirical data referred to above; such data and their implications cannot simply be ignored. Our thesis that it

Evaluative conditioning 305

is the evaluated components of the UCS which are important for conditioning to take place fits well within this wider framework. It is in this sense that we defend the view that evaluative conditioning is of theoretical significance, and that the results are not appropriately discussed in terms of mere procedural detail. Evaluative conditioning is not a qualitatively different form of classical conditioning but lies at the core of the conditioning process itself.

REFERENCES

Baer P. E. & Fuhrer M. J. (1982). Cognitive factors in the concurrent differential conditioning of eyelid and skin conductance responses. Memory and Cognifion, IO, 135-140.

Baeyens F., Eelen P. & Van den Bergh 0. (1990). Contingency awareness in evaluative conditioning: a case for unaware affective-evaluative learning. Cognition and Emofion, 4, 3-l 8.

Baeyens F., Eelen P., Van den Bergh 0. & Crombez G. (1992). The content of learning in human evaluative conditioning: acquired valence is sensitive to US-revaluation. Learning and Motivarion, 23, 200-224.

Carew T. J. & Sahley C. L. (1986). Invertebrate learning and memory: from behaviour to molecules. Annual Revjet< of Neuroscience, 9, 435481.

Davey G. C. L. (1994). Is evaluative conditioning a qualitatively distinct form of classical conditioning? Behaviour Research and Therapy, 32, 291-299.

Dawson M. E., Catania J. J., Schell A. M. & Grings W. W. (1979). Autonomic classical conditioning as a function of awareness of stimulus contingencies. Biological Psychology, 9, 23-40.

Furedy J. J. and Schiffmann K. (1973). Concurrent measurement of autonomic and cognitive processes in a test of the traditional discriminative control procedure for Pavlovian electrodermal conditioning. Journal of Experimental Psychology, 100, 210-217.

Garcia J. (1990). Learning without memory. Journal of Cognifive Neuroscience, 2. 287.-305. Hayes-Roth B. (1977). Evolution of cognitive structures and processes. Psychological Review%,, 84, 26OG278. Levey A. B. & Martin I. (1983). Cognitions, evaluations and conditioning. Rules of Sequence and Rules of Consequence.

Advances in Behaviour Research and Therapy, 4, 181~195. Levey A. B. & Martin I. (1990). Evaluative conditioning: overview and further options. Cognition and Emotion, 4, 31-37. Lovibond P. F., Siddle D. A. T. & Bond N. (1988). Insensitivity to stimulus validity in human Pavlovian conditioning.

Quarterly Journal of Experimental Psychology, 4OB, 377410. Martin I. & Levey A. B. (1987). Learning what will happen next: conditioning, evaluation, and cognitive processes. In Davey

G. (Ed.), Cognitive processes and Pavlovian conditioning in humans (pp. 57-82). New York: Wiley. Murdock B. M. Jr (1983). A distributed memory mode1 for serial-order information. Psychological Review, 90, 316-338. Rescorla R. A. & Wagner A. R. (1972). A theory of Pavlovian conditioning: variations in the effectiveness of reinforcement

and non-reinforcement. In Black A. & Prokasy W. F. (Eds), Classical conditioning II. New York: Appleton-Century- Crofts.

Rozin P. & Zellner D. (1985). The role of Pavlovian conditioning in the acquisition of food likes and dislikes, In Braveman N. S. & Bronstein P. (Eds), Experimental assessments and clinical applications qfconditioned.food aversions (pp. 1899202). New York: New York Academy of Sciences.

Shanks D. R. & Dickinson A. (1990). Contingency awareness in evaluative conditioning: a comment on Baeyens, Eelen, and Van den Bergh. Cognition and Emotion, 4, 19-30.

Stuart E. W., Shimp T. A. & Engle R. W. (1987). Classical conditioning of consumer attitudes: four experiments in an advertising context. Journal qf Consumer Research, 14. 334.-349.

Sullivan R. M. & Leon M. (1987). One-trial olfactory learning enhances olfactory bulb responses to an appetitive conditioned odor in 7 day-old rats. Developmental Brain Research, 35. 307-311.