There is general agreement that exposure to a contingency between conditioned stimulus (CS) and US will create an associative process called conditioning (Dickinson, 1980
; Rescorla, 1988
). The difference between human and non-human animal models is that the former can easily produce a verbal or motor report of the relationship between the CS and US. If humans can form an internal representation of the contingency and verbalize it or make a voluntary response, then this behavior is taken as evidence of conscious awareness (Lovibond and Shanks, 2002
What if the CS is masked, or more generally, not accessible to verbal reports? Can a verbal report of the contingency still be present? This question is central to current theoretical discussions of the role of conscious processing in trace conditioning. Is conscious awareness of the stimuli needed in order for trace conditioning to occur? There are primarily two models that account for awareness in trace conditioning. The single-process model, asserts that a sole propositional learning process mediates expression CR and the expectancy of US (full network mapping; Lovibond and Shanks, 2002
). The dual-process model, however, claims that these behavioral responses, both CR expression and US expectancy, are expressions of two independent learning processes (partial network mapping; Perruchet, 1985
; Morris et al., 1999
; Perruchet et al., 2006
There is empirical evidence championing each model. Some studies find evidence in support of the single-process model (Daum et al., 1991
; Manns et al., 2000a
; Weike et al., 2007
). While the series of experiments performed by Perruchet, Destrebecqz, Cleeremans, and colleagues (Destrebecqz and Cleeremans, 2001
; Destrebecqz et al., 2005
; Perruchet et al., 2006
) and experiments performed by others (Ohman and Soares, 1993
; Ohman et al., 1995
; Weidemann et al., 2009
) strongly support a two-way learning process.
The discussion is centered on the verbal reports of awareness of the contingencies (in Questionnaires, motor evaluation or subjective ratings) and the measures of CR. There is either complete agreement or disagreement between the two. This discrepancy raises two relevant methodological aspects about how to measure the awareness. Firstly, post-training questionnaires may elicit metacognitive process. Therefore the participant may verbally report the contingencies, not because they noticed them during the learning phase, but because they were forced to think about the contingencies after conditioning had finished. In this case it is difficult to know when conscious learning of the relationship between the stimuli occurred. It could have been during learning or it could have been when that participant was prompted about the relationship between stimuli. Secondly, if awareness is measured online during the experiment, as to avoid a post hoc metacognitive process, then participants may then take note of the contingencies of which they were previously unaware. This may lead to previously unaware participants becoming explicitly aware of the contingencies. The measurement of both verbal reports of contingencies and the CR is critical.
In a series of studies, Ohman et al. (1995
) and Ohman and Soares (1993
) argue that conditioning of electro-dermal responses to electric shocks can occur with masked, unconsciously perceived, stimuli. It has been found that unconsciously perceived stimuli only elicited a CR when fear-relevant stimuli, such as spiders and snakes (Ohman and Soares, 1993
) or angry faces (Esteves et al., 1994
), were presented as CS+. In support of Ohman’s claims, one recent study showed that sensitivity to masking conditions was related to the CR of a masked CS but not an unmasked CS (Cornwell et al., 2007
). In this study, sensitivity to the masked condition was a marker of unconscious processing, i.e., if participants were not aware of the masked item then this was taken to indicate that items were processed at an unconscious level. The depth of unconscious processing of the CS was linked to the intensity of the CR. Weak perception of the CS through masking may not elicit conscious recognition of it, but the CS may still be above an identification threshold. However, there are criticisms of these findings.
One criticism is that the measure used to assess perceptual awareness of the CS may not be sufficiently sensitive to identify participants with residual awareness of stimulus features (Pessoa, 2005
; Graziano and Sigman, 2009
). Additionally, some fear relevance effects in backward masking conditioning, as observed by Ohman, could be due to selective sensitization rather than unconscious associative processes. Some methodological concerns have also been raised concerning the extent to which participants were truly unaware of the stimuli (Lovibond and Shanks, 2002
). Ohman replied to these criticisms thoroughly, using two main arguments. Firstly, not all verbal discriminative responses indicate awareness; discrimination of stimuli above chance levels does not necessarily imply conscious awareness (Merikle and Daneman, 2000
; Wiens and Ohman, 2002
). Secondly, criticisms have assumed that awareness is a conscious experience. Therefore a measure of awareness must involve a measure of a subjective state.
Another highly debated experiment was performed by Núñez and de Vicente (2004
), they have also showed that CR can be elicited when masked words are paired with a mild shock, and that this response is, as it was in the studies by Ohman, related to the participants’ detection threshold. However, the results of this study are somewhat difficult to interpret as a higher proportion of participants in the unconscious masked condition produced a CR than in the conscious group. This study evaluated masked words paired with electric shocks; they used either a detection threshold, i.e., Was the stimulus a word or a blank?
, or an identification threshold, i.e., Was it word1, word2 or not a word?
When participants failed to detect a stimulus in the tachitoscope, half exhibited a CR above learning criterion (four out of eight). Yet, when participants were above detection threshold (“conscious”) only 11% (2 of 18) showed learning. There was a higher instance of conditioning, using a detection threshold, when stimuli were are presented unconsciously and when CR was measured using autonomic nervous system signals like skin conductance. On the contrary, when an identification threshold was applied, i.e., subjects had to differentiate between two words or two non-words, only 10% (1 of 10 participants) of participants in the unconscious condition exhibited CR, but 58% (7 out of 12) of participants in the conscious group exhibited CR. These contradictory results point to two different learning systems, the unconscious system, which bypasses central processes and consciousness-related workspaces. It also possibly directly links the early visual system with the autonomic nervous system and the conscious associative system, where the activation of the frontoparietal cortices may influence the autonomous nervous system giving rise to a different signal and a different type of learning. It is indeed the case that the variability, speed, and regularity of the CR was higher in the conscious identification group as compared to the unconscious detection learners (Núñez and de Vicente, 2004
There are, however, caveats to this study’s design that must be considered when discussing the conclusions. Firstly, in the identification threshold condition, words were repeated and could have therefore been deprived of their meaning. When a word is repeated it becomes easy to rely on low level features, such as the letter array, to determine the word’s identity without the need to access its meaning. The participants’ decisions in this condition may not have been based on anything more than a surface features. Another methodological problem of this study is that the variability of the perceptual threshold was high and several subjects were therefore excluded leading to a small sample size. The final subgroups of learners were 2/18 and 4/8 for conscious and unconscious detection of a word/blank, respectively, and 7/12 and 1/10 subjects for conscious and unconscious identification of the word, respectively. The low number of participants complicates the statistical analysis and make conclusions difficult to extrapolate to other cases. Thirdly, there is another possible explanation of “unconscious conditioning”. When an electric shock is used as the aversive stimulus, as opposed to a puff of air to the eye or a loud tone, it may induce a general increase in arousal. This may, change detection thresholds and act as a confounding factor, leading to difficulty in interpreting the results.
In short, there seems to be consistent evidence showing that trace conditioning can also be elicited by unconscious stimuli with a strong emotional content, which is not accessible for verbal report. This is reminiscent of the sea slug. There may be some specific residual forms of trace learning that can be mediated by the most likely candidates: unconscious processing and emotional stimuli. However, evidence of subliminal abstract, non-emotional stimuli eliciting trace conditioning is not conclusive. Careful and well-designed experiments are needed to robustly deliminate the boundaries of consciousness thresholds of stimuli in a given task. Their ability to elicit trace learning above and below this threshold should certainly be included in the agenda of highly relevant experiments in the next years.