Stimulus/response learning in masked congruency priming of faces: Evidence for covert mental classifications?

Abstract

Reaction times for categorization of a probe face according to its sex or fame were contrasted as a function of whether the category of a preceding, sandwich-masked prime face was congruent or incongruent. Prime awareness was measured by the ability to later categorize the primes, and this was close to chance and typically uncorrelated with priming. When prime faces were never presented as visible probes within a test, priming was not reliable; when prime faces were also seen as probes, priming was only reliable if visible and masked presentation of faces were interleaved (not simply if primes had been visible in a previous session). In the latter case, priming was independent of experimentally induced face–response or face–category contingencies, ruling out any simple form of stimulus–response learning. We conclude that the reliable masked congruency priming reflects bindings between stimuli and multiple, abstract classifications that can be generated both overtly and covertly.

Keywords:

• Priming
• Learning
• Subliminal perception
• Response priming
• Unconscious
• Semantic priming

Acknowledgments

This manuscript was funded by Swiss National Foundation Fellowship No. PA001–113106/1 granted to the first author and by the UK Medical Research Council (MC_US_A060_0046). We thank Elias Mouchlianitis and Jason Taylor for collecting some of the data, Dennis Norris and Aidan Horner for helpful discussions, and the Editor and two reviewers for their thoughtful comments.

Notes

¹ While these might also be called “novel” primes (e.g., Kunde et al., 2003), we prefer to reserve that label for stimuli that are only ever presented once as a prime throughout an experiment, unlike here, where prime-only stimuli were repeated as primes on multiple trials within the experiment.

² To test the hypothesis that S–R learning was incremental, we performed a post hoc analysis where priming in the prime + probe condition was split as a function of the number of prior presentations of the face as a probe, binned every 4 presentations (i.e., 0–3 prior presentations as a probe, 4–7, etc., leading to 6 levels of exposure). The interaction between congruency and bin did not reach significance, F(5, 18) = 1.04, MSE = 423, p > .20. However, the congruency effect was not significantly different from zero after the first 0–3 presentations, M = 5.7 ms, SE = 7.0 ms, t(23) = 0.81, p > .20, whereas it reached significance in all later bins (M > 11 ms, SE < 6.3 ms), t(23) > 1.75, p < .05, except Bin 5 (M = 8.7, SE = 7.0 ms), t(23) = 1.24, p = .11. Thus, while noisy (owing to small numbers of trials), these binned data suggest fairly rapid learning of an S–R association, in which a face requires only a few trials of being categorized as a (visible) probe before masked priming by that face (as an invisible prime) occurs.

³ In order to test for spurious associations between the fame decision and the wide-apart decision, a wideness score was computed for each face—that is, the average proportion of “wide-apart” decisions for a probe face across participants. Scores ranged from .09 to .83, whereby 4 famous and 4 nonfamous faces had scores lower than or equal to .5, and 4 famous and 4 nonfamous faces had scores higher than .5. This suggests that there was no correlation across items between the two decisions.

⁴ In fact, such prime + probe faces were seen less often on average within a session than prime-only faces were across session, though it should be noted that prime + probe faces were nonetheless seen more recently on average than prime-only faces. This issue is considered in the General Discussion.

⁵ We use “fixed” versus “variable”, rather than “consistent” versus “inconsistent”, in that “yes” and “no” responses are mutually exclusive (hence consistent vs. inconsistent responses), whereas “male” and “famous” are not (hence fixed vs. variable classifications).

⁶ In other experiments, Horner and Henson (2009) also found evidence for binding of stimuli with both decisions (yes/no) and actions (e.g., left vs. right finger press)—that is, evidence for simultaneous S–R bindings at multiple levels of response code. It is not clear why there was no evidence for such decision/action bindings in addition to classification bindings in the present paradigm, but we suspect that the frequent task switching (and hence decision/action switching) in Experiments 4–5, which was not the case in the single repetition priming of Horner and Henson, caused participants to devalue, or reduce attention to, any retrieval of responses at the level of decisions or actions.

⁷ It is of course also possible that participants learned something about responses (“R learning”)—for example, forming stronger associations between a decision (e.g., “yes”) and an action (e.g., left index finger press). However, such learning would facilitate RTs to both congruent and incongruent trials, so not produce priming (given that congruency depends on the relationship between prime and probe stimulus, not the response per se).