CONSCIOUSNESS AND COGNITION 6, 519–544 (1997) ARTICLE NO. CC970322 Brain Indices of Nonconscious Associative Learning P. S. Wong,* ,1 E. Bernat,², S. Bunce,² and H. Shevrin² *Department of Psychology, New School for Social Research, New York, New York 10003; and ²Department of Psychiatry, University of Michigan Medical Center, Ann Arbor, Michigan Using a classical conditioning technique, this study investigated whether nonconscious associative learning could be indexed by event-related brain activity (ERP). There were three phases. In a preconditioning baseline phase, pleasant and unpleasant facial schematics were presented in awareness (suprathreshold). A conditioning phase followed, in which stimuli were presented outside awareness (subthreshold, via energy masking), with an unpleasant face (CS1) linked to an aversive shock and a pleasant face (CS2) not linked to a shock. The third, postconditioning phase, involved stimulus presentations in awareness (suprathreshold). Evidence for acquisition of a conditional response was sought by compar- ing suprathreshold pre- and postconditioning phases, as well as in the subthreshold condi- tioning phase itself. For the pre-postconditioning phase analyses, significant ERP compo- nent differences differentiating CS1 and CS2 were observed for N1, P2, and especially P3. For the conditioning phase, significant differences were observed in the 100–400 ms. post-stimulus region reflecting a CS1 processing negativity. Brain activity does indeed index the acquisition of a conditional response to subthreshold stimuli. Associative learning can occur outside awareness. 1997 Academic Press INTRODUCTION In the past 15 years, there has been increasing interest in the experimental investi- gation of conscious and nonconscious processes (Shevrin & Dickman, 1980; Kihl- strom, 1987). 2 In one approach to this issue, investigators have relied on classical conditioning paradigms to explore the extent to which a response established to stim- uli in awareness could be elicited at a later time when the stimuli were presented outside awareness (Lazarus & McCleary, 1951; Corteen & Wood, 1972; Dawson & Schell, 1982). These studies relied on dichotic listening paradigms (often using aver- sive shock as unconditioned stimulus) with electrodermal measures serving as the main index of responsivity. Other studies, particularly by Ohman and colleagues (Oh- man, Dimberg, & Esteves, 1988; Ohman & Soares, 1993; 1994), have relied on visual backward masking techniques to demonstrate the same effect. Recently, we conducted a study that paralleled the Ohman et al. work by using energy masked facial schematics as conditional stimuli (CS) in a classical condition- 1 Address correspondence and reprint requests to: Philip S. Wong, Department of Psychology, New School for Social Research, 65 Fifth Avenue, New York, NY 10003. E-mail: [email protected]. This research was supported in part by a grant from the Ford Motor Company to H.S. We thank Michael Snodgrass, Jennifer Stuart, and William J. Williams for their assistance in various aspects of the study. Portions of this study were presented at the 1995 meetings of the American Psychological Society and American Psychological Association in New York. 2 We use the term, nonconscious, to refer to activity outside awareness that has a mental referent. Much physiological activity, of course, may not be represented mentally in any form. 519 1053-8100/97 $25.00 Copyright 1997 by Academic Press All rights of reproduction in any form reserved.
Transcript
CONSCIOUSNESS AND COGNITION 6, 519–544 (1997)ARTICLE NO. CC970322
Brain Indices of Nonconscious Associative Learning
P. S. Wong,*,1 E. Bernat,†, S. Bunce,† and H. Shevrin†
*Department of Psychology, New School for Social Research, New York, New York 10003; and†Department of Psychiatry, University of Michigan Medical Center, Ann Arbor, Michigan
Using a classical conditioning technique, this study investigated whether nonconsciousassociative learning could be indexed by event-related brain activity (ERP). There werethree phases. In a preconditioning baseline phase, pleasant and unpleasant facial schematicswere presented in awareness (suprathreshold). A conditioning phase followed, in whichstimuli were presented outside awareness (subthreshold, via energy masking), with anunpleasant face (CS1) linked to an aversive shock and a pleasant face (CS2) not linkedto a shock. The third, postconditioning phase, involved stimulus presentations in awareness(suprathreshold). Evidence for acquisition of a conditional response was sought by compar-ing suprathreshold pre- and postconditioning phases, as well as in the subthreshold condi-tioning phase itself. For the pre-postconditioning phase analyses, significant ERP compo-nent differences differentiating CS1 and CS2 were observed for N1, P2, and especiallyP3. For the conditioning phase, significant differences were observed in the 100–400 ms.post-stimulus region reflecting a CS1 processing negativity. Brain activity does indeedindex the acquisition of a conditional response to subthreshold stimuli. Associative learningcan occur outside awareness. 1997 Academic Press
INTRODUCTION
In the past 15 years, there has been increasing interest in the experimental investi-gation of conscious and nonconscious processes (Shevrin & Dickman, 1980; Kihl-strom, 1987).2 In one approach to this issue, investigators have relied on classicalconditioning paradigms to explore the extent to which a response established to stim-uli in awareness could be elicited at a later time when the stimuli were presentedoutside awareness (Lazarus & McCleary, 1951; Corteen & Wood, 1972; Dawson &Schell, 1982). These studies relied on dichotic listening paradigms (often using aver-sive shock as unconditioned stimulus) with electrodermal measures serving as themain index of responsivity. Other studies, particularly by Ohman and colleagues (Oh-man, Dimberg, & Esteves, 1988; Ohman & Soares, 1993; 1994), have relied on visualbackward masking techniques to demonstrate the same effect.
Recently, we conducted a study that paralleled the Ohman et al. work by usingenergy masked facial schematics as conditional stimuli (CS) in a classical condition-
1 Address correspondence and reprint requests to: Philip S. Wong, Department of Psychology, NewSchool for Social Research, 65 Fifth Avenue, New York, NY 10003. E-mail: [email protected].
This research was supported in part by a grant from the Ford Motor Company to H.S. We thankMichael Snodgrass, Jennifer Stuart, and William J. Williams for their assistance in various aspects ofthe study. Portions of this study were presented at the 1995 meetings of the American PsychologicalSociety and American Psychological Association in New York.
2 We use the term, nonconscious, to refer to activity outside awareness that has a mental referent.Much physiological activity, of course, may not be represented mentally in any form.
5191053-8100/97 $25.00Copyright 1997 by Academic PressAll rights of reproduction in any form reserved.
520 WONG ET AL.
ing paradigm (Wong, Shevrin, & Williams, 1994). Several methodological advancesin this study were noteworthy, including careful measurement of individual visualthresholds operationalizing awareness, and the use of event-related brain potentials(ERP) as an index of the learning process. We systematically replicated previouslyreported results (Ohman et al., 1988) demonstrating that facial schematics condi-tioned in awareness could elicit an electrodermal response (SCR) when later pre-sented outside awareness. The results from both our study and those of Ohman et al.were consistent with earlier findings based on dichotic listening, and support theconclusion that a conditional autonomic response can be elicited even when the CSsare inaccessible to awareness.
Extending the SCR results, we also discovered a brain response to subthresholdpresentations of the conditioned facial stimuli. Distinct slow wave activity in responseto the CS1 (unpleasant face), and not to the CS2 (pleasant face), emerged in theregion prior to the point at which the shock had been delivered during acquisition.This ERP activity was similar to what others have described as an expectancy wave(Simons, Ohman, & Lang, 1979), indicating that an anticipatory process could beelicited entirely outside awareness. In other words, nonconscious activity involvesprocesses associated with expectation or anticipation—processes considerably morecomplex than often attributed to activity outside awareness (Greenwald, 1992). Addi-tionally, our findings highlighted the usefulness of measuring brain activity in re-sponse to stimuli presented in and out of awareness, and added to a growing literatureusing ERPs and subthreshold stimuli (e.g., Brandeis & Lehmann, 1986; Libet, Al-berts, Wright, & Feinstein, 1967; Kostandov & Arzumanov, 1986, 1977; Shevrin,1973, 1988; Shevrin & Rennick, 1967; Shevrin & Fritzler, 1968; Shevrin, Williams,Marshall, Hertel, Bond & Brakel, 1992).
One can conclude, from the studies using autonomic measures and from our resultsextending the effect into central nervous system activity, that a previously learnedresponse can be elicited or activated by stimuli outside awareness. This conclusionhas important implications not only for theories of learning, but also for theories ofpsychopathology. For example, these findings can serve as the basis for neural modelsof anxiety disorders, and how learning during aversive circumstances (i.e., with anunpleasant shock) can have lasting effects that may be elicited, outside awareness,in a seemingly automatic way.
A closely related issue, and one that may have considerably more theoretical sig-nificance, is whether learning itself can occur outside awareness. Could psychopatho-logical conditions develop from aversive circumstances during which an individualwas unaware of the object (CSs) or of the link betweeen object and event? In otherwords, can nonconscious associations occur, and do these associations have identifi-able effects over time?
Studies on nonconscious learning have increased in recent years in parallel withan increased interest in conscious and nonconscious processes. Using primarily cog-nitive experimental techniques, investigators have revealed that learning withoutawareness can occur in two basic ways. First, acquisition of knowledge can occurwith an individual unaware of what is being learned (e.g., Reber (1989) on artificialgrammar; Lewicki (1992) on covariation in social judgment tasks; for a recent alterna-tive view, see Shanks & St. John, 1994). Typically, these paradigms involve demon-
NONCONSCIOUS ASSOCIATIVE LEARNING 521
strations of a person acquiring a complex set of rules without awareness of doingso, with stimuli that are themselves accessible to awareness (i.e., suprathreshold).The rule learning is implicit. Second, learning associated with subthreshold stimuli—stimuli that are inaccessible to awareness—has been demonstrated primarily withthe mere exposure effect (Kunst-Wilson & Zajonc, 1980; see also Murphy & Zajonc,1993).3 Stimulus presentations outside awareness can bias a person’s preference forthe stimuli, even when the person cannot recognize the stimuli as having been pre-sented previously. Learning in this sense is entirely outside awareness.
With some exceptions, little work has been done to elucidate the neural systemsthat underlie either kind of nonconscious learning—the implicit, knowledge-basedkind or the unconscious perceptually based kind. For example, a few investigatorshave discovered neural indices of incidental learning using electrophysiological ap-proaches (Paller et al., 1987; Begleiter et al., 1967, 1969). Investigations of associa-tive learning (using classical conditioning techniques and autonomic measures) havebeen far more prevalent, although the results seem not entirely consistent with theaforementioned cognitive studies. A review of this literature leads one to the conclu-sion that in order to acquire a differential autonomic conditional response, the CS,among other things, has to be attended to and easily discriminable (Dawson & Schell,1985). That is, the CS needs to be in or accessible to awareness. Acquisition of aconditional response to subthreshold stimuli has yet to be demonstrated definitively(although see Esteves, Parra, Dimberg, & Ohman, 1994, for recent electrodermalevidence). The conclusion that the CS needs to be in awareness is at odds with resultsfrom purely cognitive experiments such as the subthreshold mere exposure paradigm.
The present study explored the question of whether nonconscious learning (theperceptually based kind) could occur using subthreshold CSs in a conditioning para-digm. We were specifically interested in whether this kind of learning could be de-tected in central nervous system activity. Such information would facilitate a morecomprehensive understanding of the nature of nonconscious learning from both cog-nitive and neural perspectives and have implications for models of psychopathology.
In structure, the present study parallels our previous study (Wong et al., 1994) anddraws upon research reported by Ohman and colleagues (Dimberg, 1986; Esteves,Dimberg, & Ohman, 1994; Ohman & Dimberg, 1978; Ohman et al., 1994). Oneconsistent finding reported by Ohman and colleagues has been the salience of an-gry faces in conditioning. After pairing with an aversive shock (US), angry faceshave exhibited slower extinction (Dimberg, 1986) and survived backward masking(Esteves et al., 1994) when compared to happy faces paired with the US. Based onthese findings, we relied on a partial-factorial design in our previous study by pairingunpleasant faces with aversive shock, to explore subsequent subthreshold conditionalresponses (Wong et al., 1994). The rationale was straightforward: given the subtleeffects often elicited by subthreshold presentations (e.g., Shevrin et al., 1992), itseemed important to provide a strong test of whether any subthreshold response wasdetectable. If unpleasant faces showed no effect, then it could be concluded that even
3 The effects of subthreshold stimulus presentations, including semantic activation, has been the subjectof much investigation (e.g., Holender, 1986; Greenwald, 1992). These studies often posit brief activationdue to a subthreshold presentation, and are only indirectly related to learning.
522 WONG ET AL.
less salient stimuli would be unlikely to work. However, if unpleasant faces gaverise to an effect, then subsequent studies could explore the parameters of that effect.The present study is based on the same logic, and addresses another conceptual ques-tion—does nonconscious associative learning occur?
There were three phases in the experiment. In the first preconditioning phase, facialschematics depicting pleasant and unpleasant expressions were presented in aware-ness (suprathreshold). These presentations established baseline activity to the CSs. Inthe second conditioning-acquisition phase, stimuli were presented outside awareness(subthreshold). Stimuli were linked in a differential conditioning paradigm to an aver-sive shock in a partial-factorial design (interstimulus interval 5 800 ms; CS1/un-pleasant face; CS2/pleasant face; probability 5 0.8). In the third postconditioningphase, the stimuli were presented in awareness (suprathreshold). Evidence for theacquisition of a conditional response was sought in brain activity (ERP) when com-paring the suprathreshold pre- and postconditioning phases (an indirect index of ac-quisition), as well as in the subthreshold conditioning phase itself (a direct index ofacquisition).4,5
We addressed questions in three related areas.In the Wong et al. (1994) study, P3 amplitude differences (unpleasant face .
pleasant face) were observed in the suprathreshold acquisition phase. These ampli-tude differences reflected the acquisition process using suprathreshold stimuli, andwere consistent with existing theories that P3 is an index of the relative salience ofstimuli (e.g., Begleiter et al., 1983; Donchin & Coles, 1988). In the present study,although P3 would still index stimulus salience it would not reflect direct acquisitionas much as subsequent processes associated with extinction. Our first hypothesis(a priori ) was that an ERP P3 component amplitude advantage would exist for CS1compared to CS2 as a function of previous subthreshold conditioning. We expectedthat P3 component amplitude would remain stable for CS1 and would decrease forCS2, from pre- to postconditioning. Such evidence would be consistent with a ‘‘re-sistance to extinction’’ effect associated with CS1; CS2 responses, in contrast,would reflect a decrease in amplitude to a greater extent than found for CS1. Afinding of differential component effects (such as P3) for CS1 and CS2 would beconsistent with extinction and sensitization results reported in the literature by Ohmanand others using SCR with unpleasant faces (e.g., Esteves et al., 1994).
Second, would we observe any other component differences (e.g., in N1) betweenpre- and postconditioning phases? Researchers using suprathreshold stimuli haveidentified early negative component differences (N1, Nd) possibly related to atten-tional processes (e.g., Naatanen, 1990) primarily in the auditory sphere. No earlycomponent differences were discovered in the Wong et al. (1994) study for the supra-threshold acquisition phase, although evidence of early attentional selectivity wouldbe consistent with nonconscious associative learning.
Third, would there be systematic differences in processing CS1 and CS2 in the
4 We collected electrodermal measures; unfortunately technical problems in the data acquisition proce-dure invalidated the data.
5 This report focuses on early ERP activity (800 ms post-stimulus); results from facial EMG responsesand other ERP activity will be reported elsewhere.
NONCONSCIOUS ASSOCIATIVE LEARNING 523
subthreshold conditioning phase itself? Some investigators have identified N1-P2 dif-ferences associated with subthreshold stimuli (Shevrin, 1973), although no such dif-ferences were observed to previously conditioned subthreshold stimuli (Wong et al.,1994). There have been no reports in the literature, however, examining directly theeffects of acquisition of a conditional response using subthreshold stimuli on ERPs.
In each of the three areas identified above, positive results would provide converg-ing evidence to support the conclusion that acquisition of a conditional response canoccur when stimuli are perceptually inaccessible to awareness.
METHOD
Subjects
Subjects were recruited through advertisements in a university community for apsychology study on perception. Remuneration was $25 for approximately 3.5 to4.0 h. Subjects were screened for general health and for handedness. Those with ahistory of neurological disorder or who were currently experiencing significant prob-lems with their physical or emotional health were excluded. Subjects were scheduledfor a laboratory appointment and were asked to come to the appointment well-restedand to refrain from drinking alcoholic beverages the evening prior to the experiment.
In total, 10 subjects participated in the study. All subjects were right-handed men,with vision correctible to 20/20. The mean age was 21.6 years (SD 5 1.4). Onesubject was eliminated due to a high level of physiological artifact (excess muscletension and movement) in the data. Analyses are reported for the remaining sampleof 9 subjects.
Procedure
The overall procedure paralleled that reported by Wong et al. (1994). After a gen-eral orientation (including screening and other questionnaires), electrodes wereattached for physiological recording. Aversive shock and visual threshold routines(described below) were conducted prior to beginning the main experiment.
The experimental phases and data collection sequence in the main experiment arepresented in Fig. 1.
In the data collection sequence, a tone (T1) signaled the beginning of a trial, andthe subject responded by saying ‘‘ready’’ (T2) when he was looking at the fixationpoint. Four to 6 s after T2 the data collection cycle began; the cycle lasted for 4400ms. The prestimulus interval was 400 ms. S1 denotes presentation of a masked facialschematic (conditioning phase) or unmasked facial schematic (pre- and postcondi-tioning phase). S2 denotes the shock/no-shock event, which in the conditioning phaseoccurred 800 ms after S1. Two tones signaled the end of the data collection cycle,with the intertrial interval approximately 10–15 s. Individual presentations were auto-mated and required no interaction with the experimenter.
During all trials, subjects were instructed to remain as still as possible, look at thefixation point, and keep eye blinks to a minimum. Subjects were told that at somepoint soon after saying ‘‘ready’’ there would be a quick flash of something on thescreen, which might or might not be followed by a shock several seconds later. Sub-
jects were reminded periodically to keep looking at the fixation point and to minimizeeye blinks during trials.
There were three experimental phases: preconditioning, conditioning, and postcon-ditioning. The preconditioning and postconditioning phases consisted of 48 randompresentations (24 each of CS1 and CS2); the conditioning phase consisted of 72random presentations (36 each of CS1 and CS2). In the preconditioning phase, theCSs were shown in awareness (suprathreshold) to establish baseline activity. In theconditioning phase, CSs were shown outside awareness (subthreshold) and linked ina differential conditioning paradigm to an aversive shock (ISI 5 800 ms; CS1, un-pleasant face; CS2, pleasant face; probability 5 0.8). In the postconditioning phase,the schematics were again shown in awareness.
At the end of the main experiment, subjects were re-tested on visual threshold,unhooked, debriefed, paid, and dismissed.
Visual Stimuli, Apparatus, and Masking Technique
The pleasant and unpleasant facial schematics were identical to those used in theWong et al. (1994) study, and consisted of line drawings of faces in schematic form.Stimuli were equated for perceptual characteristics, e.g., number and width of lines,and rated on the extremes of an evaluative scale. Two equivalent sets of stimuli wereused; each set consisted of one pleasant and one unpleasant schematic. One set wasused in the visual threshold procedure, and the other in the main conditioning experi-ment, with sets counterbalanced between subjects.
The stimuli were presented on 3 3 5 white cards in two fields of a three-fieldGerbrands Model T3-8 tachistoscope. The CS1 and CS2 fields were counterbal-anced between subjects; the third field was used as a fixation field. Field brightnesswas tested for luminance level and pulse width, and equated for each field. Luminancelevels for the fields were 5 foot-lamberts; ambient room light conditions were approx-imately the same. The stimuli were circles subtending 1.9 degrees visual angle indiameter. Stimulus duration for the energy masked (subthreshold) presentations was2 ms. Stimulus duration for the unmasked (suprathreshold) presentations was 50 ms.
NONCONSCIOUS ASSOCIATIVE LEARNING 525
Visual Threshold Technique
In our previous study (Wong et al., 1994), visual threshold was established andtested for each subject, in part to address previous criticism in the literature regardingthe potential for significant between-subject variability in psychometric responsefunctions. In the Wong et al. study we found that individual thresholds, as identifiedby an adaptive staircase technique, ranged from 2 to 4 ms with a mean of 2.35 ms.
In the present study, we decided to forego establishing individual thresholds andinstead presented masked stimuli at a duration of 2 ms for all subjects. Although thisapproach does not control for individual variability in response functions, we choseit for a number of reasons. First, the variability in thresholds in the Wong et al. (1994)study was relatively narrow (i.e., 2 to 4 ms). Second, since we were using identicalvisual presentation conditions to the earlier study, a 2-ms threshold duration wouldbe at the low end of the expected range (which, if anything, would decrease theprobability of finding positive results since some subjects presumably would havethresholds above 2 ms). And third, presenting the same threshold duration to all sub-jects is considerably more efficient from a procedural standpoint.
Subjects were tested at the 2 ms duration in a forced-choice two-alternative task.Subjects were first shown one suprathreshold presentation of each facial schematicand then told that for each subsequent presentation he should decide whether thefacial expression was pleasant or unpleasant. Subjects were told each facial expres-sion would be presented an equal number of times in random order, that responsesshould be distributed equally, and to guess if uncertain about a response. The testconsisted of 40 trials and was administered once before and once after the mainexperiment to monitor any potential changes in threshold level.
The mean correct for the 40-trial pre-test was 19.77 (SD 5 4.29), and for the post-test was 19.67 (SD 5 2.74). A discordancy test for single outliers (Barnett & Lewis,1984; Snodgrass et al., 1993) was applied to the extreme low and high values in thesample to determine whether or not a subject was performing within an expectablechance distribution. Each outlier value was assessed for discordancy relative to itsimmediate sample (i.e., a value in the pre-test condition was evaluated relative tothe pre-test sample); in addition, each value was assessed relative to the combinedpre- and post-test samples. None of the extreme values qualified for outlier status(p . .05), indicating that all subjects performed at expectable chance levels (neithertoo high nor too low) on the pre- and post-test trials and in the combined trials. Thepre- and post-test trials also were subject to an analysis of variance, which was notsignificant.
Subjects were carefully questioned about their subjective visual experiences duringpre- and post-tests, as well as during the conditioning phase. None of the subjectsreported seeing an internal feature of the circle with any certainty.
ERP Measures
Standard Grass Instrument silver-silver chloride electrodes were used. Prior toelectrode application, sites were cleaned with a mild abrasive solution, and then af-fixed with Grass electrode paste. Recording sites were F3, F4, Cz, and Pz using theInternational (10–20) Electrode Placement System, with linked earlobes as referenceand left mastoid as ground. Electrode impedance was under 10 Kohms. Eye activity
526 WONG ET AL.
was monitored by electrodes placed on the outer canthus and suborbital ridge of theright eye. Recordings were monitored on-line by a Grass Model 8-24, digitized at250 Hz, and stored in computer files for off-line analysis. The high frequency cut-off was 70Hz. The Grass AC amplifiers were set to a low frequency cut-off of 0.01Hz.
Individual trials contaminated by artifacts (eye blinks, muscle tension, or any suspi-cious activity which would render a trial unusable) were rejected by visual inspectionand replaced on-line.
ERP measures for the pre- and postconditioning phases were based on 24 presenta-tions (per stimulus category for each experimental phase); the conditioning phasemeasures were based on 36 presentations (per stimulus category). The analyses inthis report include the 400-ms prestimulus interval and the 800-ms post-stimulusinterval (at which point the shock event occurred in the conditioning phase).
The overall analysis, unless otherwise specified, was divided into two parts:(1) Precon (preconditioning) vs Postcon (postconditioning) and (2) Con (condi-tioning). The Precon vs Postcon analysis of variance involved a Phase(2; pre-post)3 Face(2; pleasant-unpleasant) 3 Electrode(4) analysis, with the main (includinga priori) hypotheses centered around a Phase 3 Face interaction. The Con analysisinvolved a Face(2) 3 Electrode(4) analysis of variance.
Standard component measures (peak amplitude and latency, and area) were ob-tained on each individual subject’s averaged ERP profile for N1 (90–150 ms), P2(160–220 ms), and P3 (248–548 ms) in all analyses. Early (100–400) and Late (400–700) area measures were obtained in the Con analyses.
An initial analysis of the pre-stimulus interval was undertaken in order to determinewhether there was systematic variation in this interval which might bias subsequentanalyses. Indeed, a Phase(3; pre-con-post) 3 Face(2) 3 Electrode(4) analysis of vari-ance yielded a three-way interaction (F(6, 48) 5 2.53; p 5 .05; e 5 .7695) and atrend for the main effect of phase (F(2, 16) 5 3.48; p 5 .06). No other resultsapproached significance. These results indicated that some variability existed in thepre-stimulus intervals, which was not surprising especially across phase. Conse-quently, for each subject, the ERP average for face and electrode within phase wasadjusted so that the pre-stimulus interval averaged to zero. That is, the average pre-stimulus interval for each profile was subtracted from all post-stimulus values for thatprofile, effectively eliminating any pre-stimulus variability and allowing for directcomparison of component and area measures across phase, face and electrode.
Aversive Shock Procedure
Stimulating electrodes were attached to the distal phalanges of the index and ringfingers of the preferred hand (right). The stimuli were single 200-ms constant-currentsquare wave pulses, delivered by a Grass Model S-88 Stimulator and completelyisolated from ground by a Stimulus Isolation Unit (SIU-7).
The intensity level of the stimulus was determined by each subject, and identifiedas the level at which the sensation felt ‘‘annoying or unpleasant.’’ Subjects were toldthat the sensation should not be painful in any way; in no case did the levels gobeyond 5 mA. Subjects rated the degree to which the stimulus was ‘‘unpleasant’’ or‘‘annoying’’ on a 9-point scale (9 5 high; 1 5 low) during threshold determination,
NONCONSCIOUS ASSOCIATIVE LEARNING 527
FIG. 2. ERP grand average at each electrode: Analysis by phase (a, preconditioning; b, postcondition-ing). y axis (at stimulus onset) 5 4 µv increments; x axis 5 400-ms intervals; solid line, CS1; dottedline, CS2.
as well as after the conditioning experiment. These threshold methods parallel thoseused by Wong et al. (1994).
The mean pre-test rating was 5.78 and the mean post-test rating was 3.8. An analy-sis of variance was significant (F(1, 8) 5 10.29; p 5 .01), indicating that a decreasein shock unpleasantness occurred over time. This result is likely associated with anhabituation effect.
RESULTS
Preconditioning vs Postconditioning Phase
The grand average ERP profiles are presented in two ways for purposes of compari-son. Figures 2a and 2b are an analysis by phase; activity in response to the CS2/pleasant is contrasted with CS1/unpleasant in the (a) Precon phase and (b) Postconphase. On inspection of this figure, the most obvious differences emerge at electrodesCz and Pz in the Postcon phase (Fig. 2b). Here, activity in response to CS2 is lowerin amplitude than CS1 from stimulus onset until just prior to where the shock/no-shock event occurred in the subthreshold conditioning phase. No such differences
528 WONG ET AL.
FIG. 3. ERP grand average at each electrode: Analysis by stimulus (a, CS1/unpleasant; b, CS2/pleasant). y axis (at stimulus onset) 5 4 µv increments; x axis 5 400-ms intervals; solid line, precondi-tioning; dotted line, postconditioning.
emerge in the Precon phase. Figures 3a and 3b are an analysis by stimulus; activityin response to the CSs across Precon and Postcon phases is examined for (a) CS1/unpleasant and (b) CS2/pleasant. On inspection of this figure, clear differencesemerge at electrode Cz and Pz for the CS2/pleasant stimulus (Fig. 3b). Here, activityin response to CS2/pleasant is lower in amplitude in the Postcon phase comparedto the Precon phase. Activity in response to the CS1/unpleasant does not changeacross phase.
Further elaboration of the methods applied to assess the statistical significance ofthe ERP grand average differences in the context of our experimental hypotheses aredescribed below.6 Table 1 includes values for all component measures.
Main Component Findings
P3 Amplitude. Peak amplitude of the P3 component was subjected to a Phase(2)3 Face(2) 3 Electrode(4) analysis of variance. There was a significant main effectof Electrode (F(3, 24) 5 5.94; p 5 .029; e 5 .4213; Pz . Cz . F3-F4) and a
6 All statistical tests reported regard a significance level of p , .05 (two-tailed) as consistent withrejection of the null hypothesis. For repeated measures analyses, the Huynh-Feldt correction procedureis applied with epsilon reported.
NONCONSCIOUS ASSOCIATIVE LEARNING 529
TABLE 1Component Mean (SD) Values
Amplitude (µv) Latency (mseconds) Area (µv/region)
significant Phase 3 Face 3 Electrode three-way interaction (F(3, 24) 5 4.56; p ,.011; e 5 1.000). Post-hoc test7 of relevant contrasts revealed that the most notablecontribution was at Pz, where CS2 amplitude was significantly smaller in the Postconphase than in the Precon phase (F(1, 8) 5 7.82; p 5 .02), whereas CS1 amplitudedid not change (F(1, 8) 5 .16; n.s.). No other effects were observed within electrodealthough the means are all in directions consistent with Pz.
N1 Amplitude. Peak amplitude of the N1 component was subjected to a Phase(2)3 Face(2) 3 Electrode(4) analysis of variance. A highly significant Phase 3 Face3 Electrode three-way interaction emerged (F(3, 24) 5 13.34; p , .001; e 5 .6396).Post-hoc tests of the relevant contrasts revealed that the most notable contributionto the interaction occurred at electrode Pz, where CS2 amplitude was more negativethan CS1 in the Postcon (F(1, 8) 5 4.34; p 5 .07) but not in the Precon (F(1, 8)5 .66; n.s.). At Cz, there was an overall phase effect, where activity in the Postconwas more negative than Precon for both CS1 and CS2 (F(1, 8) 5 6.62; p 5 .03). Noother effects within electrode were observed although the means were in directionsconsistent with Pz.
P2 Amplitude. Peak amplitude of the P2 component was subjected to a Phase(2)3 Face(2) 3 Electrode(4) analysis of variance. A marginally significant Phase 3Face 3 Electrode three-way interaction emerged (F(3, 24) 5 2.72; p , .076;e 5.8802). Inspection of the relevant cell means revealed that the most notable effectswere at Pz, with the amplitude of the CS2 increasing from Precon to Postcon, whilethe amplitude of the CS1 remained the same.
Summary of main component findings. The P3 amplitude findings indicated thatactivity associated with the CS2 was smaller (less positive) in the Postcon phasecompared to the Precon phase, whereas CS1 did not change (and in fact becameslightly larger in the Postcon). The effects are strongest at Pz and Cz. These P3findings are highly consistent with our first hypothesis: that we would find a P3 com-ponent amplitude advantage of CS1 over CS2. The findings support the conclusionthat P3 changes systematically as a function of previous subthreshold conditioning.
The N1 amplitude findings indicated that activity associated with the CS2 wasmore negative than activity associated with the CS1 in the Postcon phase but notin the Precon phase. These effects were most prominent at Pz. The P2 amplitudefindings indicated that P2 activity was smaller (less positive) for the CS2 than forthe CS1 in the Postcon phase, with no difference in the Precon phase (althoughspecific P2 contrasts could not be tested statistically because the relevant interactionsreached only marginal significance levels). Taken together, the N1 and P2 amplitudefindings shed light on our second area of inquiry: previous subthreshold conditioningappears to affect not only the P3 component, but also N1 and P2 component activity.
Corollary Component Findings
Analyses of component latencies for N1, P2, and P3 also were performed. Al-though some statistical trends were observed, there were no consistent latency differ-ences either within or across components.
7 All post-hoc contrasts were adjusted using the Scheffe-type method (O’Brien & Kaiser, 1985).
NONCONSCIOUS ASSOCIATIVE LEARNING 531
Analyses of component areas for N1, P2, and P3 were performed in parallel withthe component amplitude analyses. Area results were consistent with the componentamplitude findings, highlighting the differences in component responses across pre-and postconditioning phases.
Summary of Preconditioning vs Postconditioning Phase Findings
The results from the Preconditioning vs Postconditioning phase analyses bear onseveral questions set forth in the introduction to this paper. The first question con-cerned whether we would observe component differences associated with the condi-tioning process; it was predicted that P3 amplitude would reflect the conditioningprocess across pre- and postconditioning phases. This prediction was confirmed: P3amplitude associated with the CS2 came significantly smaller (less positive) in thePostcon phase compared to the Precon phase, whereas CS1 did not change fromPrecon to Postcon, and was larger than CS2 in the Postcon. This finding supportsthe general experimental hypothesis that acquisition of a conditional response canoccur with subthreshold stimuli.
Upon closer inspection of the grand average profiles (Figs. 4 and 5 for Pz), weobserved in greater detail how the experimental effect was expressed. In Fig. 4 (theanalysis by phase), the CS2 profile is more negative than the CS1 in the Postconphase (Fig. 4b) but not in the Precon phase (4a). This reflects an overall decrease inthe ERP amplitude for CS2, from stimulus onset to approximately 800-ms post-stimulus. Similarly, in Fig. 5 (the analysis by stimulus), the CS2 profile decreasesnotably from Precon to Postcon phases (Fig. 5b), whereas the CS1 does not (Fig.5a). Taking these observations into account, the effects observed for the N1, P2, andP3 components combined are highly consistent with each other. That is, for CS1,we observe a smaller (less negative) N1, a larger (more positive) P2, and a largerP3, with CS2 having opposite results.
These results suggest that the component effects observed may be contributing toa generalized decrease in amplitude for CS2. This result is consistent with our mainhypothesis. One effect of the acquired conditional response is that the CS1 maintainsits activation (as reflected, for example, in P3 component differences) while CS2decreases substantially. This result is consistent with the ‘‘resistance to extinction’’effect to particularly salient stimuli (e.g., Ohman & Soares, 1993). In contrast to theCS1 activity, CS2 activity decreases through repetition in the Postcon phase com-pared to the Precon phase.
Conditioning Phase
The grand average ERP profiles for the Conditioning phase of the experiment arepresented in Fig. 6. On inspection, the most salient differences emerge in electrodesCz and Pz. Soon after stimulus presentation, the CS1/unpleasant face activity be-comes more negative than the CS2/pleasant face activity. The differences betweenCS1 and CS2 disappear shortly before the shock-noshock event occurs at 800-mspost-stimulus (at the arrow). Further elaboration of the methods applied to assess thesignificance of the differences observed in the ERP grand averages are describedbelow. Component and area values are presented in Table 2.
532 WONG ET AL.
FIG. 4. Detailed analysis by phase for Pz (a, preconditioning; b, postconditioning). y axis (at stimulusonset) 5 4 µv increments; x axis 5 400-ms intervals; solid line, CS1; dotted line, CS2.
Component Findings
Component amplitude, latency, and area measures were subjected to a Face(2) 3Electrode(4) analysis of variance for N1, P2, and P3.8
The N1 peak amplitude analysis revealed a significant main effect of Face (F(1,8) 5 5.91; p , .04; CS1 more negative than CS2) and a marginally significantinteraction of Electrode 3 Face (F(3, 24) 5 2.96; p , .06; e 5 .8977). Analysis ofthe latency of N1 peak amplitude revealed no significant differences. The N1 areaanalysis revealed a significant effect of Face (F(1, 8) 5 6.22; p , .04; CS1 morenegative than CS2).
8 As in many ERP averages associated with subthreshold presentations, distinct component processes(such as N1-P2-P3) are not readily visible in the grand averages. This is due primarily to the nature ofbrain responses to subthreshold stimuli, and is an area of important future investigation (see Shevrin,1973, for an early example of this analysis). We conducted this analysis for comparative purposes inorder to be consistent with the approach taken in the Precon-Postcon phase analysis.
NONCONSCIOUS ASSOCIATIVE LEARNING 533
FIG. 5. Detailed analysis by stimulus for Pz (a, CS1/unpleasant; b, CS2/pleasant). y axis (at stimulusonset) 5 4 µv increments; x axis 5 400-ms intervals; solid line, preconditioning; dotted line, postcondi-tioning.
The P2 peak amplitude analysis revealed significant main effects of Electrode (F(3,24) 5 3.20; p , .04; e 5 1.0; F3-F4 . Pz-Cz) and Face (F(1, 8) 5 13.73; p ,.006; CS1 less positive than CS2). Analysis of P2 latency revealed a significanteffect of Electrode (F(3, 24) 5 7.41; p , .005; F3 . others). The P2 area analysisrevealed a significant effect of Face (F(1, 8) 5 13.91; p , .006; CS1 less positivethan CS2).
Neither P3 amplitude nor area measures were significant, although the means werein a direction consistent with P2. Analysis of P3 latency of peak amplitude revealeda significant Electrode 3 Face interaction (F(3, 24) 5 4.97; p , .009) and a signifi-cant effect of Face (F(1, 8) 5 5.99; p , .05; CS1 . CS2), with the interactionthe result of a reversal in effect for electrode F3 (CS2 . CS1).
On closer inspection of the averages, we observed that the activity in responseto presentations of CS1 is generally more negative over time than in response toCS2, from stimulus onset to the shock-noshock event. We decided to assess thesignificance of the CS1 negativity using area measures that spanned a wider timeinterval than any specific component area measure.
534 WONG ET AL.
FIG. 6. ERP grand average at each electrode: Conditioning phase. y axis (at stimulus onset) 5 2 µvincrements; x axis 5 400-ms intervals; arrow, shock-nonshock event at 800 ms; solid line, CS1; dottedline, CS2.
Area Findings
We obtained area measures from three regions: (1) Total area (100–700 ms post-stimulus), (2) Early area (100–400 ms), and (3) Late area (400–700 ms). Each ofthe regions was subject to a Face(2) 3 Electrode(4) analysis of variance. For theTotal area, there was a marginally significant effect of Face (F(1, 8) 5 4.71; p ,.06) in the expected direction (CS1 more negative). No other effects were significant.In the Early area, there was a significant effect of Face (F(1, 8) 5 7.43; p , .03),again, in the expected direction. No other effects were significant. Finally, in the Latearea, there were no significant effects though the means for Face were in the expecteddirection (Face F(1, 8) 5 1.69; n.s.).
NONCONSCIOUS ASSOCIATIVE LEARNING 535
TABLE 2Conditioning Phase Mean (SD) Values
a. Component measure
Amplitude (µv) Latency (mseconds) b. Area measure
CS1 CS2 CS1 CS2 CS1 CS2
Pz PzN1 22.7 (1.0) 2.9 (1.6) 130 (20) 128 (18) Early area 226 (72) 72 (115)P2 1.8 (1.4) 2.7 (1.9) 198 (25) 191 (19) Late area 24 (126) 74 (215)P3 3.4 (1.8) 4.7 (2.8) 473 (74) 368 (65) Total area 225 (141) 116 (205)
Cz CzN1 22.8 (1.1) 21.2 (2.1) 121 (21) 136 (20) Early area 233 (64) 61 (128)P2 1.2 (1.3) 2.6 (1.0) 198 (25) 184 (21) Late area 8 (99) 61 (180)P3 3.4 (1.1) 4.1 (2.3) 439 (69) 374 (66) Total area 231 (125) 99 (219)
F4 F4N1 21.4 (1.1) 2.9 (1.5) 118 (21) 128 (19) Early area 35 (97) 90 (83)P2 2.0 (1.2) 3.3 (1.2) 188 (22) 178 (19) Late area 83 (146) 122 (122)P3 4.0 (1.6) 4.3 (1.6) 440 (52) 430 (85) Total area 81 (169) 163 (147)
F3 F3N1 21.6 (2.0) 21.1 (2.4) 118 (17) 120 (19) Early area 41 (86) 75 (99)P2 2.4 (.5) 3.5 (1.2) 186 (20) 168 (10) Late area 95 (134) 100 (159)P3 4.2 (2.0) 4.3 (1.6) 422 (113) 436 (99) Total area 86 (146) 130 (180)
The results from the area analyses indicate that there were significant ERP differ-ences during the Conditioning phase. Specifically, the CS1 became more negativethan CS2 shortly after stimulus presentation, and returned to baseline prior to theshock-noshock event. Statistically, this effect was strongest in the Early post-stimulusregion (100–400 ms) and decreased in the Late region (400–700 ms).
Given that the results of the Conditioning phase are a direct index of the acquisitionprocess, and that the Early area measure is a good indicator of the differences betweenCS1 and CS2, we conducted an analysis exploring the trial-by-trial developmentof the Early area difference in this phase. Using trial as a continuous independentvariable, we conducted a Face(2) 3 Trial repeated-measures analysis of variance forthe Early area. Frontal electrodes yielded no significant effects. For Cz, a Face 3Trial interaction emerged as a nonsignificant trend in the expected direction (F(1,8) 5 2.39; p 5 .16); for Pz, a significant trend emerged (F(1, 8) 5 4.70; p 5 .06).As the trials progressed, CS1 became more negative than CS2 in the Early area,especially at Pz. These findings should be understood as a conservative index thatunderestimates existing differences, given the relatively low sensitivity of single-trialERPs. Thus, the finding of trends in the expected direction likely indicate actualdifferences that are much larger and more significant than those reported here.9
9 One could also justify conducting this analysis with a directional hypothesis, specifically that CS1will become increasingly negative across trials. Such a directional hypothesis would change the criterionfor significance to p 5 .1 (Howell, 1987). The findings for Pz then would technically reach a significantlevel.
536 WONG ET AL.
Subjective Ratings
Ratings obtained after the experiment revealed that most subjects did not haveconsistent ideas regarding when the shock would occur. Of the three subjects whoventured hypotheses, one believed that the unpleasant face was linked to the shockand two believed that the pleasant face was linked to the shock.
Summary of Conditioning Phase Findings
The results of the Conditioning phase analyses provide positive evidence in ourthird area of inquiry: acquisition of a conditional response to subthreshold stimuli isreflected immediately and directly in scalp recorded brain activity. The componentanalyses revealed that CS1 is more negative than CS2 for N1 and less positive thanCS2 for P2 and P3. The area analyses were quite consistent with the componentresults. The CS1 became more negative than the CS2 shortly after stimulus presen-tation, and returned to baseline prior to the shock-noshock event. Furthermore, theCS1 negativity increased across trials, indicating that this negativity indexed a learn-ing process. The CS1 negativity, best reflected in the area measure, is a new findingthat supports the experimental hypothesis that acquisition of a conditional responsecan occur with subthreshold stimuli.
DISCUSSION
The nature of nonconscious learning has been a topic of increasing interest inpsychology. The cognitive literature indicates that such learning can occur in twodistinct ways: in an implicit, knowledge-based way and in a subthreshold, percep-tually based way. Implicit learning involves stimuli that are accessible to awareness,but with the rules or algorithms acquired by the individual inaccessible. Subthresholdor subliminal learning, in contrast, involves perceptually inaccessible stimuli. Al-though the cognitive literature has progressed toward elaborating the nature of non-conscious learning, little work has been accomplished in elucidating the neural sys-tems involved in such processes. The classical conditioning literature that addressessimple associative learning using physiological indices is one area where this topichas generated interest, but the results have been equivocal and at odds with the ex-isting cognitive literature. A major conclusion in the classical conditioning literatureis that acquisition of a conditional response cannot occur when stimuli are degradedor rendered inaccessible to awareness (Dawson & Schell, 1985). The present studysought to examine whether nonconscious associative learning (of the perceptual kind)could occur in the context of a classical conditioning paradigm, and whether neuralindices of such learning could be measured by scalp recorded electrical activity. Af-firmative answers to these questions would provide an important link between thecognitive and neural basis of nonconscious associative learning.
The results of this study provide two lines of evidence supporting the conclusionthat associative learning can occur outside awareness. First, in the pre- vs postcondi-tioning phase analyses (an indirect measure of acquisition), we discovered differentialactivity in CS1 and CS2, especially in the predicted P3 component amplitude. Andsecond, in the conditioning phase analyses (a direct measure of acquisition), we dis-
NONCONSCIOUS ASSOCIATIVE LEARNING 537
covered a processing negativity associated with CS1, which appears to index acquisi-tion itself. We now turn to a more extended discussion of these results.
Preconditioning vs Postconditioning Phase Findings
The major finding in the suprathreshold pre- vs postconditioning phase analyseswas of differential effects for CS1 and CS2. First, as predicted, P3 amplitude forCS1 did not change from preconditioning to postconditioning phase, and was largerthan CS2 in the postconditioning phase. This effect is consistent with what has beendescribed as a ‘‘resistance to extinction’’ effect (e.g., Ohman & Soares, 1993).
The P3 findings also are consistent with other results using conditioning paradigms.In our previous study, for example, P3 amplitude for CS1 was larger than CS2 ina suprathreshold acquisition series (Wong et al., 1994). Increased component ampli-tudes (for CS1) as a result of conditioning also were reported by Begleiter and Platz(1969). Though P3 differences in the present study are a consequence of prior sub-threshold acquisition of a conditional response (and not a direct index of acquisition),the effects on P3 amplitude are similar. The overall results are consistent with theview that P3 is an index of stimulus salience or emotional value (Begleiter et al.,1983; Johnston et al., 1986; Donchin & Coles, 1988).
The second main finding in this analysis is of an overall decrease in amplitudeassociated with the CS2 in the postconditioning phase. In contrast, the CS1 resultsdemonstrate a ‘‘resistance’’ to the process observed in CS2. These results are consis-tent with other research findings. Schell et al. (1991), for example, reported thatpotentially phobic stimuli were more prone to resist extinction than neutral stimuli.It seems likely that certain stimuli, such as faces with expressions of negative affect,may be especially salient (Dimberg, 1986; Esteves et al., 1994a). Supporting thisconclusion, Esteves et al. (1994b) recently demonstrated a differential extinction ef-fect in SCR with facial stimuli that had been conditioned outside awareness. Thiseffect was found with angry faces only.
One alternative interpretation of the results in the postconditioning phase shouldbe addressed. One could argue that the differences observed are in fact due to intrinsicqualities of the stimuli and not due to conditioning. In other words, differences thatemerged later in the postconditioning phase were a function solely of the additionalstimulus repetitions, with no effect from the subthreshold acquisition series. Whilethis alternative interpretation (of differential sensitization) is viable in principle, itshould be noted that the present study relied on established standards to evaluatestimulus differences in the preconditioning phase. Thus, this interpretation wouldneed to hinge on the premise that the measurement of intrinsic stimulus differencesneeds more powerful signal-to-noise ratios (i.e., more repetitions) than is ordinarilyrequired.
Evidence directly contradicting a differential sensitization hypothesis can be foundin the recent work by Esteves et al. (1994b). In two separate experiments, usingconceptually identical experimental procedures to the present study, Esteves et al.found no evidence that intrinsic effects of angry and happy faces interacted withconditioning. In Experiment 1, one group included neutral faces as subthreshold CSs,with one face linked to a shock and the other not, during the acquisition or condition-ing phase. In the postconditioning extinction phase, a happy and angry face were
538 WONG ET AL.
substituted for the (previously neutral) conditional stimuli; no evidence was obtainedfor a differential SCR response to the happy and angry faces. The only evidence fordifferential responses to happy and angry faces in a postconditioning extinction phasewas when these faces had been used as conditional stimuli in a prior acquisitionphase. In Experiment 2, subthreshold angry and happy faces were paired randomlywith a US (aversive shock). Again, no evidence was obtained for differential SCRresponses to the happy and angry faces when they were later presented in awareness.Both experiments by Esteves et al. (1994b) consistently demonstrate that sensitizationeffects do not interact with subthreshold conditioning of faces with emotional expres-sions.
In sum, both results—the P3 amplitude advantage for CS1 and the decrease inresponsivity for CS2—provide converging evidence that a conditional response wasacquired to subthreshold presentations of the facial schematics. While these findingsare an indirect index of the acquisition process, they nonetheless constitute strongconverging evidence that a conditional response had been acquired previously tosubthreshold presentations of the stimuli.
Conditioning Phase Findings
Direct evidence for the acquisition of a conditional response was found in thesubthreshold conditioning phase of the experiment. Here, the CS1 elicited greaternegativity in the brain potential than did the CS2. This negativity emerged shortlyafter stimulus presentation and diminished as the shock-noshock event neared. Fur-thermore, the CS1 negativity increased with each additional trial on a trial-by trialanalysis, indicating that the acquisition process was indexed by the CS1 negativityand that subthreshold learning was unfolding over time. These findings are notewor-thy, and are highly consistent with the main experimental hypothesis in that differen-tial responses to the CS1 and CS2 were obtained during the acquisition phase itself.
The ERP negativity for CS1 involves an initial negative shift, which then gradu-ally diminishes as the shock-noshock event approaches (cf. Early and Late area re-sults). What does this CS1 negativity reflect? We might logically conclude that itreflects some aspect of the acquisition process; however, one could consider twoalternative hypotheses. Suppose, for example, the negativity reflects a preexistingdifference in the stimuli (a differential sensitization hypothesis), or a consequenceof the subthreshold presentation of the stimuli? A number of observations, however,directly contradict these alternative hypotheses. First, a differential sensitization hy-pothesis would involve the assertion that sensitization occurs in different directionsfor sub- and suprathreshold stimuli, e.g., CS1 is more positive in the suprathresholdpostconditioning phase and more negative in the subthreshold conditioning phase.There is no data in the literature indicating that this pattern exists, and there is no apriori reason to reject the more parsimonious assertion that sensitization effectsshould be directionally consistent. Second, one way to address the alternative hypoth-eses would be to present subthreshold affectively valent facial stimuli, without condi-tioning, to test for intrinsic stimulus differences. This approach, in fact, was takenin our earlier study (Wong et al., 1994). In that study, the same facial stimuli, eachpresented under virtually identical subthreshold conditions in a baseline, precondi-tioning phase, elicited no ERP differences between the facial stimuli. Third, a trial-
NONCONSCIOUS ASSOCIATIVE LEARNING 539
by-trial analysis of the Wong et al. (1994) data (which was identical to the analysisconducted in the present study in the Conditioning phase for the Early area) yieldedno significant differential effects across time in the baseline phase. Based on theseobservations, it is unlikely that intrinsic differences in the stimuli emerged in thecurrent subthreshold conditioning phase, nor can it be argued that it is the subthresh-old stimulus presentations as such that produce the effect insofar as no such differ-ences were found in a preconditioning baseline subthreshold phase for either compo-nent or trial-by-trial analyses. It thus seems likely that the effect observed in theconditioning phase of the present study is a reflection of the acquisition process itself.
Additional evidence indicating that the conditioning phase ERP negativity is in-dexing the acquisition process comes from concurrent measures of facial EMG re-sponsivity using the same paradigm (Bunce, 1996; Bunce et al., 1997). For facialEMG, Bunce et al. demonstrated a trial-by-trial increase in level of activity acrossthe conditioning phase for CS1, while CS2 showed no such increase. This trial-by-trial analysis, which is an even more robust analysis with facial EMG activity dueto higher sensitivity and signal-to-noise ratios than ERPs, is yet another indicationthat subthreshold learning is unfolding during the conditioning phase. Although theexact correspondence between the facial EMG activity and the ERP negativity is stillunknown (and in need of further investigation), it is clear that the facial EMG activityis differentiating CS1 and CS2 in ways that are consistent with what was discoveredin the present study using ERPs. Each measure appears to be indexing, via differentpsychophysiological ‘‘windows,’’ acquisition of a conditional response to the sub-threshold facial stimuli.
In structure, the conditioning phase ERP negativity is similar to the expectancywave (e-wave) observed in the Wong et al. (1994) study. This e-wave was obtainedin response to subthreshold presentations of faces previously conditioned in aware-ness. The results of the present study, however, derive from a slightly different para-digm; with an 800-ms ISI, we were not expecting an e-wave because previous re-search highlighted the importance of longer ISIs in the development of an expectancyprocess (e.g., Backs & Grings, 1985). Thus the negativity discovered in this study,although perhaps related to an expectancy process, also may reflect something differ-ent. What exactly is different, however, is an open question. For example, might thisearly negativity be related to what others have identified in the auditory sphere as a‘‘processing negativity’’ (Naatanen, 1990), or with an anticipatory motor responsepartially reflected in facial EMG (Bunce et al., 1997)? The negativity also may reflectface-specific processing. Hallgren (1992), for example, reports widespread negativityat 225 ms in response to faces; this negativity attains maximal amplitude in the amyg-dala in depth recordings. If the conditioning phase negativity reflects face-specificneuronal responses that involve activity in the amygdala, then a neural basis for sub-threshold stimulus responsivity may be at hand.
In view of the potential role of the amygdala in the conditioning phase negativity,and its importance in processing affectively valent stimuli, one aspect of the studyshould be explored further. We assume (based on well-established findings, e.g.,Esteves et al., 1994) that the affective valence of the CS, especially the unpleasantface, is particularly salient regarding conditioning. In making such an assumption,we were not exploring affective valence per se in the acquisition process (and it is
540 WONG ET AL.
not logically necessary to do so in order to establish that a conditional response canbe acquired). However, given the potential relevance of affective valence of the CSson the conditioning process, further exploration of this issue is warranted. For exam-ple, one might undertake a replication of the present study using a full factorial designduring acquisition (i.e., where both pleasant and unpleasant faces serve as CS1),which would address directly whether subthreshold acquisition requires a negativelyvalent face or whether faces of either valence would work as well. Based on thework of Esteves et al. (1994) where a full-factorial design was used with SCR, onemight anticipate that subthreshold conditioning effects will be observed only withunpleasant faces. Other questions pertaining to the nature of the CSs also can beaddressed, such as whether unconscious associations can be established to affectivelyvalent stimuli other than faces, or whether affectively valent stimuli are needed atall.
Overall, the ERP negativity in the conditioning phase seems to index brain pro-cesses associated with the nonconscious acquisition of a conditional response tofaces. The negativity may reflect an expectancy or anticipatory process that emergesonce acquisition has occurred, some other yet to be identified process related to stimu-lus selection outside awareness, or a combination of the two. The negativity alsomay provide us with neural evidence for subthreshold face-specific processing. Im-portantly, all of these processes occur entirely outside awareness.
Implications for Conscious and Nonconscious Processes
In this study, we provide two lines of evidence supporting the conclusion thatassociative learning can occur outside awareness as reflected in brain activity. First,in the postconditioning phase we find indirect evidence of acquisition reflected inthe differential activity of the suprathreshold CS1 and CS2. And second, we finddirect evidence of acquisition in the conditioning phase, reflected in a processingnegativity associated with the subthreshold CS1 and not CS2. Of note is that thisnegativity increases with additional trials, indicating that a learning process is un-folding during the conditioning phase. Thus, we have converging evidence, both di-rect and indirect, indicating that acquisition of a conditional response can occur withstimuli that are inaccessible to awareness.
Other studies (Ohman & Soares, 1993; Wong et al., 1994) have demonstrated thata previously acquired conditional response can be elicited at a later time by subthresh-old presentations of the stimuli. The present study extends these results into the areaof nonconscious associative learning, and provides evidence that learning can occurwith stimuli rendered perceptually inaccessible to awareness. The combined evidencefrom these recent conditioning studies indicates that mental processes associated withacquisition, expectancy and extinction can be elicited by perceptually inaccessiblestimuli. Nonconscious processes, as indexed by these paradigms, appear to be some-what more complex than others have described (Greenwald, 1992). The present re-sults also are consistent with other experiments demonstrating nonconscious learning(e.g., Kunst-Wilson & Zajonc, 1980; Esteves et al., 1994), and extend these processesinto the physiological domain as reflected in brain activity.
The present study raises several interesting questions regarding the nature of non-conscious learning and its effect on conscious processes. Based on the CS1/CS2
NONCONSCIOUS ASSOCIATIVE LEARNING 541
results in the suprathreshold postconditioning phase, it seems that what is learnedoutside awareness can have an effect on subsequent stimulus presentations that arefully accessible to awareness (i.e., when subjects can see the faces). It’s not evident,however, whether any subjective experience, other than a perceptual experience, waselicited in subjects in the postconditioning phase. For example, in post-experimentquestionnaires and interview data, there was no indication of alteration in subjectiveexperience. Subjects reported no reactions to the stimuli in the postconditioning phasethat were different from what was experienced in the preconditioning phase. Thissuggests that differential responses can occur with perceptually accessible stimuliwithout the person being aware, in a knowledge-based sense, of the processes them-selves. Similarly, post-experiment ratings of the subthreshold conditioning phase in-dicated that even though subjects acquired a differential response to perceptuallyinaccessible stimuli, there were no consistent consciously articulated knowledge-based ideas regarding why or when the shock would occur.
The distinction we are making between perceptual awareness and knowledge-based awareness is important in several regards.10 The research we have describedin this and in our previous report (Wong et al., 1994) suggests that mental processes inthe context of a conditioning paradigm can be elicited when stimuli are perceptuallyinaccessible. The effects of these nonconscious processes on conscious, knowledge-based processes are not yet clear. It seems likely, however, that knowledge-basedprocesses are only partially correlated with an individual’s responses to both percep-tually accessible and inaccessible stimuli. Additional exploration is needed of theinterface between conscious processes, in both the perceptual and knowledge-basedsense and the nonconscious processes. Furthermore, conditioning paradigms such asthe one used in the present experiment, in which stimuli in various stages of theconditioning process are perceptually inaccessible, seem well suited to the investiga-tion of nonconscious learning.
REFERENCES
Backs, R. W., & Grings, W. W. (1985). Effects of UCS probability on the contingent negative variationand electrodermal response during long ISI conditioning. Psychophysiology, 22, 268–275.
Barnett, V., & Lewis, T. (1984). Outliers in statistical data. New York: Wiley.
Begleiter, H., & Platz, A. (1969). Evoked potentials: Modifications by classical conditioning. Science,166, 769–771.
Begleiter, H., Porjesz, B., Chou, C. L., & Aunon, J. I. (1983). P3 and stimulus incentive value. Psycho-physiology, 20, 95–101.
Bower, G. H. (1990). Awareness, the unconscious, and repression: An experimental psychologist’s per-spective. In J. L. Singer (Ed.), Repression and dissociation. Chicago: University of Chicago Press.
Brandeis, D., & Lehmann, D. (1986). Event-related potentials of the brain and cognitive processes:Approaches and applications. Neuropsychologia, 24, 151–166.
Bunce, S. C. (1996). Facial EMG and ERP indices of emotion-relevant unconscious learning. Paperpresented at the annual meeting of the American Psychological Association, Toronto.
Bunce, S. C., Bernat, E., Wong, P. S., & Shevrin, H. (1997). Evidence for unconscious learning: FacialEMG indicators of unconscious associative learning. Under review.
10 Note that other researchers have made similar distinctions, e.g., Bower (1990) and Hirst (1995).
542 WONG ET AL.
Corteen, R. S. & Dunn, D. (1974). Shock associated words in a nonattended message. A test for momen-tary awareness. Journal of Experimental Psychology, 102, 1143–1144.
Corteen, R. S., & Wood, B. (1972). Autonomic responses to shock associated words in an unattendedchannel. Journal of Experimental Psychology, 94, 308–313.
Dawson, M. E., & Schell, A. M. (1982). Electrodermal responses to attended and nonattended significantstimuli during dichotic listening. Journal of Experimental Psychology: Human Perception and Per-formance, 8, 315–324.
Dawson, M. E., & Schell, A. M. (1985). Information processing and human autonomic classical condi-tioning. Advances in Psychophysiology, 1, 89–165.
Dawson, M. E., & Schell, A. M. (1987). Human autonomic and skeletal classical conditioning: The roleof conscious cognitive factors. In G. Davey (Ed.), Cognitive processes and pavlovian conditioningin humans. Chichester: Wiley.
Dawson, M. E., Schell, A. M., & Banis, H. T. (1986). Greater resistance to extinction of electrodermalresponses conditioned to potentially phobic CSs: A noncognitive process? Psychophysiology, 23,552–561.
Dimberg, U. (1986). Facial expressions as excitatory and inhibitory stimuli for conditioned autonomicresponses. Biological Psychology, 22, 37–57.
Donchin, E., & Coles, M. G. H. (1988). Is the P300 component a manifestation of context updating?Behavioral and Brain Sciences, 11, 357–374.
Esteves, F., Parra, C., Dimberg, U., & Ohman, A. (1994b). Nonconscious associative learning: Pavlovianconditioning of skin conductance responses to masked fear-relevant facial stimuli. Psychophysiol-ogy, 31, 375–385.
Gaillard, A. W. K. (1977). The late CNV wave: Preparation versus expectancy. Psychophysiology, 14,563–568.
Greenwald, A. G. (1992). New look 3. Unconscious cognition reclaimed. American Psychologist, 47,766–779.
Hirst, W. (1995). Cognitive aspects of consciousness. In M. S. Gazzaniga (Ed.), The cognitive neurosci-ences (pp. 1307–1320). Cambridge, MA: MIT Press.
Holender, D. (1986). Semantic activation without conscious identification in dichotic listening, parafo-veal vision, and visual masking: A survey and appraisal. Behavioral and Brain Sciences, 9, 1–66.
Howell, D. C. (1987). Statistical methods for psychology (2nd ed.). Boston: Duxbury Press.
Huynh, H., & Mandevill, G. K. (1979). Validity conditions in repeated measures designs. PsychologicalBulletin, 86, 964–973.
Kihlstrom, J. F. (1987). The cognitive unconscious. Science, 237, 1445–1452.
Klorman, R., & Ryan, R. M. (1980). Heart rate, contingent negative variation, and evoked potentialsduring anticipation of affective stimulation. Psychophysiology, 17, 513–524.
Kostandov, E., & Arzumanov, Y. (1977). Averaged cortical evoked potentials to recognized and non-recognized verbal stimuli. Acta Neurobiologica Experimentalis, 37, 311–324.
Kostandov, E., & Arzumanov, Y. (1986). The influence of subliminal emotional words on functionalhemispheric asymmetry. International Journal of Psychophysiology, 4, 143–147.
Kunst-Wilson & Zajonc (1980). Affective discrimination of stimuli that cannot be recognized. Science,207, 557–558.
Lazarus, R. S., & McCleary, R. A. (1951). Autonomic discrimination without awareness: A study ofsubception. Psychological Review, 58, 113–122.
Lewicki, P., Hill, T., & Czyzewska, M. (1992). Nonconscious acquisition of information. AmericanPsychologist, 47, 796–801.
Libet, B., Alberts, W. W., Wright, E. W., & Feinstein, B. (1967). Responses of human somatosensorycortex to stimuli below threshold for conscious sensation. Science, 158, 1597–1600.
NONCONSCIOUS ASSOCIATIVE LEARNING 543
McGinnies, E. (1949). Emotionality and perceptual defence. Psychological Review, 56, 244–251.
Murphy, S. T., & Zajonc, R. B. (1993). Affect, cognition, and awareness: Affective priming with optimaland suboptimal stimulus exposures. Journal of Personality & Social Psychology, 64, 723–739.
Naatanen, R. (1990). The role of attention in auditory information processing as revealed by event-related potentials and other brain measures of cognitive function. Behavioral and Brain Sciences,13, 201–288.
O’Brien, R. G., & Kaiser, M. K. (1985). MANOVA method for analyzing repeated measures designs:An extensive primer. Psychological Bulletin, 97, 316–333.
Ohman, A. (1986). Face the beast and fear the face: Animal and social fears as prototypes for evolutionaryanalyses of emotion. Psychophysiology, 23, 123–145.
Ohman, A. (1988). Nonconscious control of autonomic responses: A role for Pavlovian conditioning?Biological Psychology, 27, 113–135.
Ohman, A., & Dimberg, U. (1978). Facial expressions as conditioned stimuli for electrodermal responses:A case of preparedness? Journal of Personality and Social Psychology, 36, 1251–1258.
Ohman, A., Dimberg, U., & Esteves, F. (1988). Preattentive activation of aversive emotions. In T. Ar-cher & L. Nilsson (Eds.), Aversion, avoidance, and anxiety. Perspectives on aversively motivatedbehavior. Hillsdale, NJ: Erlbaum.
Ohman, A., & Soares, J. J. F. (1993). On the automatic nature of phobic fear: Conditioned electrodermalresponses to masked fear-relevant stimuli. Journal of Abnormal Psychology, 102, 121–132.
Ohman, A., & Soares, J. J. F. (1994). ‘‘Unconscious anxiety’’: Phobic responses to masked stimuli.Journal of Abnormal Psychology, 103, 231–240.
Paller, K. A., Kutas, M., & Mayes, A. R. (1987). Neural correlates of encoding in an incidental learningparadigm. EEG and Clinical Neurophysiology, 67, 360–371.
Reber, A. S. (1989). Implicit learning and tacit knowledge. J.Experimental Psychology (General), 118,219–235.
Rohrbaugh, J. W., McCallum, W. C., Gaillard, A. W. K., Simons, R. F., Birbaumer, N., & Papakosto-poulos, D. (1986). ERPs associated with preparatory and movement-related processes. A review.Cerebral Psychophysiology: Studies in Event-Related Potentials, EEG Suppl. 38, 189–229.
Schell, A. M., Dawson, M. E., & Marinkovic, K. (1991). Effects of potentially phobic conditionedstimuli on retention, reconditioning, and extinction of the conditioned skin conductance response.Psychophysiology, 28, 140–153.
Shanks, D. R., & St. John, M. F. (1994). Characteristics of dissociable human learning systems. Behav-ioral and Brain Sciences, 17, 367–395.
Shevrin, H. (1973). Brain wave correlates of subliminal stimulation, unconscious attention, primary-and secondary-process thinking, and repressiveness. Psychological Issues, 8 (Monograph 30), 56–87.
Shevrin, H. (1988). Unconscious conflict: A convergent psychodynamic and electrophysiological ap-proach. In Horowitz, M.J. (Ed.), Psychodynamics and cognition. Chicago: University of ChicagoPress.
Shevrin, H., & Dickman, S. (1980). The psychological unconscious. A necessary assumption for allpsychological theory? American Psychologist, 35, 421–434.
Shevrin, H., & Fritzler, D. E. (1968). Visual evoked response correlates of unconscious mental processes.Science, 161, 295.
Shevrin, H., & Rennick, P. (1967). Cortical response to a tactile stimulus during attention, mental arith-metic, and free associations. Psychophysiology, 3, 381–388.
Shevrin, H., Williams, W. J., Marshall, R. E., Hertel, R. K., Bond, J. A., & Brakel, L. A. (1992). Event-related potential indicators of the dynamic unconscious. Consciousness & Cognition, 1, 340–366.
Simons, R. F. (1986). Can the autonomic nervous system contribute to our understanding of event-related slow brain potentials? Cerebral psychophysiology: Studies in event-related potentials, EEGSuppl. 38, 189–229.
544 WONG ET AL.
Simons, R. F., Ohman, A., & Lang, P. J. (1979). Anticipation and response set: Cortical, cardiac, andelectrodermal correlates. Psychophysiology, 16, 222–233.
Snodgrass, M., Shevrin, H., & Kopka, M. (1993). The mediation of intentional judgments by unconsciousperceptions: The influences of task strategy, task preference, word meaning, and motivation. Con-sciousness & Cognition, 2, 169–193.
Wong, P. S., Shevrin, H., & Williams, W. J. (1994). Conscious and nonconscious processes: An ERPindex of an anticipatory response in a conditioning paradigm using visually masked stimuli. Psycho-physiology, 31, 87–101.
Zajonc, R. B. (1980). Feeling and thinking: Preferences need no inferences. American Psychologist, 35,151–175
