19
THE LEARNING-WITHOUT-AWARENESS QUESTION IN HUMAN PAVLOVIAN CONDITIONING
Back to Realism Applied to / Home PageIntegrative Physiological and Behavioral Science, January-March 2000, Vol. 35, No. 1, pp. 17-34.
Rev_Learning.doc
Revisiting the Learning-Without-Awareness
Question in Human Pavlovian Autonomic
Conditioning: Focus on Extinction in a
Dichotic Listening Paradigm
John J. Furedy1
Boris Damke2
Wolfram Boucsein2
1 Department of Psychology, University of Toronto, Canada 2physiological Psychology, University of Wuppertal, Wuppertal, Germany
Abstract—-Numerous studies have indicated that, consistent with current "cognitive" accounts of information processing, human Pavlovian autonomic discrimination acquisition cannot occur without awareness of the CS-US relationship. However, extinction studies have suggested that awareness is not necessary, findings that, in information-processing terms, have been explained by assuming that the processing by the extinction stage is parallel (automatic) rather than serial (controlled). This explanation was tested in an 80-subject study. The first, acquisition phase was a standard semantic differential conditioning arrangement with a 96-db white noise as US, and a "long" CS-US interval of 8 s, with ten trials each of CS+ (paired with US) and CS- (unpaired) trials. In extinction (USs omitted), in order to obtain non-autonomic indices of processing and thereby test the information-processing account of "unaware" autonomic conditioning during extinction, a dichotic listening task was implemented, with the CSs presented in the unattended channel (ear), while the subject had to perform a semantic differential reaction task in an attended-to channel (other ear). In early extinction, the electrodermal response occurring at an interval of 9-15 s after CS onset (i.e., following placement of the US during acquisition) and the finger-pulse-volume response occurring at an interval of 4--I1 s after CS onset both showed reliable conditioning, but reaction-time and subjective-report data for the recognized critical words indicated serial rather than parallel processing of the CSs during extinction.
Key Words—-learning without awareness, human Pavlovian long-interval discrimination conditioning, acquisition vs. early extinction, information-processing accounts, serial versus parallel processing, electrodermal responses.
Introduction
The learning-without-awareness phenomenon has long been of central interest to students of conditioning. In the heyday of S-R behaviorism that was the dominant Zeitgeist until the late fifties, awareness was considered a mere epiphenomenon of conditioning, and reports of learning-without-awareness (of the CS-US or operant-reinforcer relationship) phenomena abounded in both classical (Pavlovian) and operant conditioning. In many of these studies awareness was measured crudely, so that its apparent absence could well
Address correspondence to: John J. Furedy, Ph.D., Department of Psychology, University of Toronto, 100 St. George Street, Toronto, Ontario M5S 3G3, Canada. E-mail:
Integrative Physiological and Behavioral Science, January-March 2000, Vol. 35, No. 1, 17-34.
19
THE LEARNING-WITHOUT-AWARENESS QUESTION IN HUMAN PAVLOVIAN CONDITIONING
have been due simply to measurement insensitivity. In contrast, the currently dominant Zeitgeist assigns conditioning the merely epiphenomenal role, and awareness the exclusive causal one. Perhaps the apogee of this form of cognitivism applied to Pavlovian conditioning is represented by the authoritative assertion that Pavlovian conditioning is "now described as the learning of relations among events" (Rescorla, 1988: 151, our emphasis). Clearly, in this view, learning without awareness is almost a logical impossibility. Essentially, this cognitive approach views the organism in terms of the information-processing, computer metaphor, so that conditioning is "described as" the analysis of (CS-US) contingencies, or of "the relations between events," and the organism is seen, essentially, as an computer-like analyzer of contingencies.
However, to those for whom the issue of the role of awareness in conditioning raises questions of an empirical sort rather than more or less convenient metaphors, human Pavlovian autonomic conditioning (HPAC) is of particular interest. This is so because the dependent variable (usually the electrodermal response [EDR]) is unavailable to consciousness, and because CS-US contingency awareness can be readily measured independently of EDR conditioning either following the conditioning experiment (e.g., Baer and Fuhrer, 1968), or, more accurately, during the experiment (e.g., Furedy and Schiffmann, 1973). An additional reason for special interest is that when the US is aversive (e.g., electric shock or loud noise), there is the possibility that the conditional EDR reflects the learning of an emotional non-computer-like process, i.e., fear.
By the early 1970s there was extensive evidence to indicate that, at least in acquisition, awareness of the CS-US relation was necessary for the discrimination form of this conditioning—mainly because autonomic dependent variables like the EDR are grossly affected by individual differences, most current forms of autonomic conditioning are of the discrimination sort, i.e., where both CS+ (associated with the US) and CS- (not associated with the US) are presented to all subjects, and where conditioning is defined as occurring when responding to CS+ significantly exceeds responding to CS-(CS+>CS-). For example, a review of the acquisition literature by Dawson (1973) indicated that all prior reports of apparent unaware conditioning were easily accounted for in terms of an insensitivity of the measures of awareness employed, and that, in terms of the question raised in the title of his paper, classical conditioning could not occur without contingency learning.
It bears emphasis that demonstrations of necessity, no matter how convincing, do not permit an inference of causality, for which a demonstration of sufficiency is minimally required. Dawson's (1973) demonstration of necessity was accompanied in the same journal by Furedy's (1973) refutation of sufficiency, in a review that presented evidence of such dissociations between awareness and conditioning as the lack of any correlation between awareness of the CS-US relations and the extent of discriminative autonomic conditioning. A resolution of the apparent conflict between the two reviews was offered in the "necessary-gate" hypothesis (Dawson and Furedy, 1976), but, as detailed most recently by Furedy and Kristjansson (1996), this resolution itself is ambiguous concerning whether awareness is not sufficient but nevertheless important for conditioning, or whether it is a relatively unimportant factor in the acquisition of autonomically controlled and apparently useless (i.e., without a clear function) responses such as the EDR, as well as of more potentially useful responses like heart-rate deceleration (Furedy, 1992).
For those who would argue that awareness is important for acquisition, one problem is that the experiments demonstrating necessity with normal human subjects have employed
19
THE LEARNING-WITHOUT-AWARENESS QUESTION IN HUMAN PAVLOVIAN CONDITIONING
preparations where the CS-US contingency is masked (e.g., Baer and Fuhrer, 1968) because, otherwise, almost all subjects quickly become aware of the CS-US contingency. Masking is an appropriate methodological device for assessing necessity, but it may beg the question of importance. A thought experiment proposed by Furedy and Kristjansson (1996) illustrates this point. In that hypothetical experiment, the CS+/CS- difference is one of wavelength, where the stimuli are so desaturated that, for half the subjects, the CS-US contingency is effectively masked. The usual result that only those subjects who were aware of the CS-US contingency (for which, obviously, CS+/CS- discrimination was necessary) showed discrimination acquisition EDR conditioning would demonstrate only the necessity of wavelength discrimination under these masked conditions, but not the importance of CS-US awareness in conditioning. Nevertheless, the fact that the evidence is so clear on the necessity of awareness for acquisition conditioning has led most current students of conditioning to follow Brewer's (1974) lead, and ascribe to awareness an influence of considerable importance, if not one of sole causality or as being the terms in which conditioning is to be "described" (Rescorla, 1988).
In contrast, the evidence on the relation between awareness and (early) extinction has not provided clear support for even the relatively weak necessity hypothesis. It is true that early reports of conditioning in early extinction by unaware subjects by Diven (1937) and Haggard (1943) were shown to have methodological flaws by Dawson (1973). This is not surprising, because these early reports would also have constituted examples of unaware acquisition conditioning. However, another requirement for the necessity of awareness is that, once subjects become aware that the CS and US are no longer associated, autonomic discrimination conditioning should also disappear. This has not been the case, especially with "prepared" CSs of the sort studied by Oehman and his associates (e.g., Oehman et al., 1976) and have led to the suggestion that at least certain sorts of extinction do not operate according to cognitive CS-US contingency awareness, but rather according to non-cognitive S-R processes (e.g., Oehman, 1979).
A version of this view has been formulated in cognitive models of human information processing (Posner and Boies, 1971; Shiffrin and Schneider, 1977). These models draw a distinction between parallel and serial processing, and apply, respectively, the terms "automatic" and "aware" to the two sorts of processing. Applied to conditioning, parallel processing is clearly of the S-R learning sort, for which awareness of the CS-US contingency is not necessary, whereas serial processing involves "the learning of relations between events" (Rescorla, 1988) for which contingency awareness is not only necessary, but also very important.
The information-processing position also asserts that, to be established, automatic information processing must be preceded by training. Applied to conditioning, repetition of CS-US trials constitutes training, and, in the typical conditioning preparation these training (CS-US acquisition trials) should produce a transition from serial to parallel processing. Accordingly, during early extinction, awareness may not be necessary for the CS+>CS-autonomic conditioning phenomenon, if sufficient training has produced a shift from serial to parallel processing during acquisition.
Some preparations like that of eyelid conditioning allow the use of a large number of acquisition trials, with asymptotic responding to CS+ being reached and maintained for as many as 100 acquisition trials. By this stage, the shift from serial to parallel processing can safely be assumed to have occurred, thus permitting a clear test of the awareness-necessity assumption for extinction eyelid conditioning. However, autonomic responses like the EDR allow only about eight CS-US acquisition trials, before responding begins to habitu-
19
THE LEARNING-WITHOUT-AWARENESS QUESTION IN HUMAN PAVLOVIAN CONDITIONING
ate (see e.g., Kimmel, 1966), so the method of using a large number of acquisition trials to guarantee the serial-to-parallel processing shift is not available.
A methodological alternative is to test for the presence of the processing shift rather than guarantee its presence. A preparation which appears to be suited for this purpose is that of a dichotic listening task (e.g., Corteen and Wood, 1972; Corteen and Dunn, 1974) which can be superimposed on the conditioning "task" (i.e., discriminating between CS+ and C-) during acquisition or, as in the present case, during extinction. Briefly, after subjects had engaged in the acquisition phase of a semantic differential conditioning task with CSs (words in two categories) presented binaurally through earphones, the extinction period (a "new task" for the subject) was introduced, in which the CS words without the noise US would be presented in one ear. Simultaneously, in the other ear (to which alone subjects were instructed to attend), there would be presented new words (from two different categories) which subjects would have to discriminate by minimal reaction time (RT) to press a button whenever the word fell into one category (fruit) rather than the other (vegetable).
The shift from serial to parallel processing is assumed to occur if there is no interference between the two tasks, namely CS and RT discrimination during the binaural, dichotic phase. It will be noted that because this issue can be assessed independently of whether early extinction has yielded the learning-without-awareness phenomenon (i.e., autonomic discrimination in the absence of a belief in differential CS+/CS- CS-US contingency), it becomes possible to provide an empirical, and non-circular test of the serial-to-parallel processing account of the early-extinction autonomic-learning-without-awareness phenomenon.
Specifically, the account was tested by determining whether two hypotheses deduced from the account were confirmed. The first hypothesis was that there would be no performance (reaction-time) deficit in the attended channel when, in the unattended ("parallel") channel a CS+ rather than a CS- was presented. This hypothesis follows from the assumption that the early-extinction CS+>CS- autonomic effect is automatic and "parallel" to the reaction-time task in the attended channel.
The second hypothesis was that, in a post-experimental questionnaire, subjects would not report having recognized the critical (CS+) words in the unattended channel. This hypothesis also follows from the assumption that the processing is parallel rather than serial during early extinction.
The main dependent autonomic conditioning variable here, as in most studies, was the EDR. The relatively long (8-s.) CS-US interval employed allowed separation of CS-elicited EDRs into the first-interval response (FIR-—-occurring 1-5 s. following CS onset), second-interval response (SIR—-occurring 5-8 s. following CS onset), and third-interval response (TIR-occurring, on CS-alone extinction trials, 9-15 s. following CS onset). Different mechanisms have been ascribed to these latency-specified EDRs (for an overview, see Boucsein, 1992). The most common view (e.g., Stewart et al., 1961) holds only the SIR to be a "true" CR. reflecting anticipation of the US, with FIR being merely an orienting reaction to the CS (analogous to the short-latency, "alpha" response in eyelid conditioning, which habituates rather than increases as a function of CS-US conditioning trials—-see also Prokasy and Ebel, 1967; Furedy and Poulos, 1977), and the TIR being an orienting reaction to the omission of the US (which represents a novel stimulus leading to the reinstatement of the orienting reaction-—see also Sokolov, 1960).
These hypotheses about mechanisms, however, have not always been confirmed by the actual EDR results. For example, it has been frequently observed (e.g., Furedy and
19
THE LEARNING-WITHOUT-AWARENESS QUESTION IN HUMAN PAVLOVIAN CONDITIONING
Schiffmann, 1973; Oehman et al., 1976) that the SIR fails to show any reliable evidence for the basic discrimination conditioning CS+>CS- outcome, which makes further speculations about the conditioning mechanisms that underlie the conditional SIR rather moot. It appears necessary to observe, empirically, in each study which of the EDR components show evidence for the basic acquisition CS+>CS- conditioning effect, and then of the same effect during early extinction. We took this approach here, and also recorded the finger pulse volume response (PVR).
The PVR is another autonomically controlled response which behaves similarly to the EDR in discrimination conditioning studies (e.g., Furedy and Schiffmann, 1973), except for having a somewhat longer latency (so that, over an 8-sec CS-US interval, only one response can be observed). Finally, heart-rate (HR) was measured on a beat-by-beat basis. As detailed elsewhere (Furedy and Poulos, 1976), with loud-noise and shock USs (though not with negative tilt), the topography of the conditional HR response is multiphasic and complex, and the discrimination conditioning phenomenon itself is less than robust. Further analyses of the HR data would proceed only if there was a significant difference between mean post-CS+ and post-CS- HR trends during acquisition.