Contingency learning without awareness: Evidence for implicit control

https://doi.org/10.1016/j.concog.2006.06.010Get rights and content

Abstract

The results of four experiments provide evidence for controlled processing in the absence of awareness. Participants identified the colour of a neutral distracter word. Each of four words (e.g., MOVE) was presented in one of the four colours 75% of the time (Experiments 1 and 4) or 50% of the time (Experiments 2 and 3). Colour identification was faster when the words appeared in the colour they were most often presented in relative to when they appeared in another colour, even for participants who were subjectively unaware of any contingencies between the words and the colours. An analysis of sequence effects showed that participants who were unaware of the relation between distracter words and colours nonetheless controlled the impact of the word on performance depending on the nature of the previous trial. A block analysis of contingency-unaware participants revealed that contingencies were learned rapidly in the first block of trials. Experiment 3 showed that the contingency effect does not depend on the level of awareness, thus ruling out explicit strategy accounts. Finally, Experiment 4 showed that the contingency effect results from behavioural control and not from semantic association or stimulus familiarity. These results thus provide evidence for implicit control.

Introduction

Cognitive processes that are controlled are conventionally assumed to operate in a slow, effortful, and voluntary manner (Posner and Cohen, 1984, Posner and Snyder, 1975, Shiffrin and Schneider, 1977). Thus, when researchers discuss the influence of “controlled” processes, it is typically assumed that such processes are explicit (i.e., conscious; cf., Besner & Stolz, 1999). As such, the term “implicit control” would seem to be nonsensical, because “implicit” (i.e., unconscious) seems to preclude the possibility of control. However, etymologically speaking this is not a necessary conclusion. The Oxford English Dictionary (2001) defines control as “the power to influence people’s behaviour or the course of events.” Similarly, Merriam-Webster (2005) defines control as the “power or authority to guide or manage.” Nothing in these definitions necessitates conscious intent. Whether implicitly controlled processes are actually observable is an empirical question. As the experiments reported here demonstrate, control can be dissociated from consciousness.

Evidence for cognitive control, which is assumed to be explicit and strategic in nature, has been drawn from the Stroop literature (Stroop, 1935). In the Stroop task, identification of the print colour of colour words is slower when the word and ink colour are incongruent (e.g., the word GREEN in orange; GREENorange) than when they are congruent (e.g., ORANGEorange; see MacLeod, 1991, for a review). Probably the most important demonstration of putatively controlled processes in the Stroop literature is the proportion-congruent effect. The proportion-congruent effect refers to the finding that the size of the Stroop effect is influenced by the proportion of congruent items in a block of trials (Lindsay and Jacoby, 1994, Logan and Zbrodoff, 1979). Specifically, the Stroop effect is much larger in a high proportion-congruent block of trials than in a low proportion-congruent block of trials. This effect is commonly attributed to participants explicitly learning to predict the colour from the word. Specifically, because the word usually matches the colour in a high proportion-congruent block of trials, participants can capitalize on this relationship and predict the colour based on the word (e.g., if the word is BLUE, then the colour is probably blue), thus leading to a larger Stroop effect because responses would be especially fast on the expected congruent trials and especially slow for the infrequent incongruent trials. In contrast, the Stroop effect is reduced in a low proportion-congruent block of trials, because participants can learn that the word rarely predicts the correct response and therefore there is no payoff that leads to fast responses on congruent trials and large costs on incongruent trials. This has often been explained in terms of a deliberate, task-wide strategy.

The idea that cognitive control over the amount of Stroop interference requires a task-wide strategy has recently been challenged. In the item-specific proportion-congruent (ISPC) manipulation participants are presented with high and low proportion-congruent stimuli within the same block. This is accomplished by presenting some words (e.g., BLUE and GREEN) mostly with their congruent colour (e.g., BLUE is presented 75% of the time in blue and 25% of the time in green) and other words (e.g., YELLOW and ORANGE) mostly with incongruent colours (e.g., YELLOW is presented 25% of the time in yellow and 75% of the time in orange). The Stroop effect is still larger for high proportion-congruent words relative to low proportion-congruent words (Jacoby et al., 2003, Trainham et al., 1997). This ISPC effect indicates that the “strategy” participants use is not (at least always) task-wide (i.e., participants cannot simply predict that the words will match the colours, because this only applies to the high proportion-congruent words). Rather, participants may generate a contingency estimate associated with each word (e.g., BLUE likely indicates the colour blue, whereas YELLOW likely indicates orange).

Although it is possible that participants could be predicting congruency with the word (e.g., BLUE likely indicates a matching colour, whereas YELLOW likely indicates a mismatching colour), it is probably more likely that participants use the word to predict a specific colour response. Thus, for instance, if BLUE is presented most often in blue, then BLUE will indicate a blue verbal or key press response. Similarly, if ORANGE is presented most often in yellow, then ORANGE will indicate a yellow verbal or key press response. Thus, participants may be using the contingencies between words and colours to predict a specific response rather than congruency in general. In support of this contention, Musen and Squire (1993) demonstrated contingency learning independent of any congruency relations between the words and colours. In their Experiment 2, they paired each of seven arbitrary words (e.g., SOCKS) with a colour. Words were always presented in their assigned colours, but participants were instructed to ignore the word and respond to the colour. Halfway through the experiment the words were re-paired with new colours without notice. This led to an increase in response latencies. Thus, participants learned the contingencies (i.e., not congruency) between the arbitrary words and the colours and used these to facilitate responding.

One methodological concern with Musen and Squire’s (1993) design is that the words are perfectly correlated with the responses. Thus, it is difficult rule out the possibility that some or all of the participants were making responses by identifying the word rather than the colour, thus leading to significant impairment in responding when the words were re-paired with new colours. In order to overcome this shortcoming, the following two experiments present participants with four arbitrary words that are presented in all four ink colours, but each word is presented most often (75% in Experiment 1 and 50% in Experiment 2) in one colour (e.g., MOST is presented 75% of the time in blue and 25% equally often in the remaining colours).

Interestingly, Musen and Squire (1993) also report that contingency learning appears to be implicit, which runs counter to the assumption that contingency learning effects (e.g., the proportion-congruent effect) result from explicit strategy use. Experiment 1 in their paper was similar to their Experiment 2, except that instead of arbitrary words, incongruent colour words were paired with colours and then re-paired partway through (e.g., BLUE was presented in green for half the experiment, then in orange for the rest). At the end of Experiment 1 (they do not report a similar analysis for Experiment 2), participants were asked to guess which colour each word was presented in for both halves of the experiment (order unimportant). Thus, participants made a seven-alternative forced choice (first guess with all seven colours) followed by a six-alternative forced choice (second guess with the remaining six colours). Importantly, Musen and Squire claim that participants did not guess contingencies at a rate higher than expected by chance. Unfortunately, they miscalculated chance to be 33.3% when chance was actually 28.6%.1 From their graph, normal participants appeared to respond at or above 35%. Reference to the same graph clearly indicates that the 28.6% chance level is not captured by the error bars. Thus, Musen and Squire’s data indicate that participants are guessing contingencies at a rate better than chance when chance is calculated correctly. As such, it remains to be determined whether participants can learn contingencies implicitly.

Because Musen and Squire’s (1993) re-analysed results indicate that participants guessed the contingencies at a rate greater than chance, it may be the case that all contingency learning in their experiments was explicit (i.e., conscious). On the other hand, this is not necessarily the case. First, the observation of some explicit contingency learning does not mean that implicit contingency learning never occurred. Second, participants in Musen and Squire’s study may not have been aware of the contingencies, because unconscious processes can alter a person’s behaviour without them knowing it. Thus, while participants may have guessed contingencies at a rate higher than chance, it does not necessarily follow that these participants were aware of the contingencies (and, indeed, aware that they were guessing at a rate higher than chance).

One of the goals of the present experiments is to demonstrate that participants can detect and use contingencies even when they are subjectively unaware of (i.e., unable to report noticing) the contingencies between the arbitrary distracter words and colours. To this end, participants were told nothing about the word–colour contingencies until after the completion of the experiment, at which time participants were asked whether they noticed any contingencies. If a contingency effect is observed for those participants who were not subjectively aware of such contingencies, then this would constitute strong evidence that contingency learning can be implicit.

There are two primary goals of the current research. First, if participants use contingency information associated with distracter words to predict specific responses rather than congruency in general, then the distracter words need not be colour words. Any words that are correlated with specific colours (e.g., MOST presented most often in blue) should be able to elicit a contingency effect. Specifically, participants should respond faster on high-contingency trials (e.g., MOSTblue) than on low-contingency trials (e.g., MOSTorange). On the other hand, if distracter words are used to predict congruency, then arbitrary words cannot be used (i.e., because an arbitrary word like MOST is neither congruent nor incongruent with a colour). Thus, no difference between high- and low-contingency trials is expected.

Second, the current research expands on the idea that controlled processes need not be explicit. To foreshadow our results, we reproduce two classical patterns of data that are taken as evidence for cognitive control. Critically, we demonstrate these patterns of data with participants who are subjectively unaware of the contingencies driving these effects, suggesting that controlled processing can be dynamic and implicit, rather than always being slow, effortful, and explicit. In Experiments 1 and 2, we demonstrate proportion-congruent-like contingency learning for arbitrary words (i.e., faster responding for high-contingency trials relative to low-contingency trials). These experiments demonstrate the hallmark pattern of controlled processing that is observed in the proportion-congruent design. This contingency learning was observed even when participants were subjectively unaware of the word–colour contingencies. That is, a contingency effect was found even for participants who reported no awareness of the fact that some words were presented more often in one colour than in the other colours. We then demonstrate that another commonly used example of cognitive control, sequential control of contingency effects, is also seen when participants are unaware of the contingencies present. Finally, we report a block analysis that indicates that contingency learning occurs rapidly in the first block of trials and is maintained throughout the experiment, ruling out an explicit strategy interpretation. Experiments 3 and 4 extend these findings by showing, respectively, that level of awareness is unrelated to the size or presence of the contingency effect and that contingency learning results from behavioural control over responding and not simple semantic associations between word–colour pairs or stimulus familiarity.

Section snippets

Experiments 1 and 2

If the ISPC effect is due solely to contingencies between the distracter words and correct responses, then it should be the case that contingencies can be learned for any set of words. In that vein, Experiment 1 uses a key press Stroop task in which neutral distracter words predict the correct colour response on 75% of the trials (e.g., the word MOVE presented in blue 75% of the time and the remaining 25% of the time in the other three display colours). In Experiment 2, the contingencies are

Sequential analysis

To strengthen the claim that control can be exerted implicitly, we report an analysis of sequential effects that are most easily understood in terms of control. Specifically, it is often observed that participants will alter the way they perform a task based on what happened on the preceding trial (e.g., Besner & Risko, 2005). For instance, it has been observed that Stroop-like effects are larger for trials following a congruent trial than for trials following an incongruent trial (Gratton,

Block analysis

A contingency effect was observed in both Experiments 1 and 2. Because this effect was observed even for participants who were subjectively unaware of contingencies, we have argued that contingency learning is implicit. If this assertion is correct, then it would be expected that contingency learning occurs rapidly at the very beginning of the experiment. If, on the other hand, contingency learning resulted from conscious strategy, then it would be expected that contingency learning would not

Experiment 3

So far, we have argued that our proportion-congruent-like contingency effects and sequential trial effect result from implicit control. The claim that these controlled processes can be implemented implicitly is supported by the fact that these effects were observed even for contingency-unaware participants. The claim that the effect is implicit is also supported by the block analysis, which showed that learning occurs very rapidly. Experiment 3 was conducted to investigate whether level of

Experiment 4

So far we have presented a behavioural control account of our data. That is to say, we argue that participants learn contingency information implicitly and then directly use this information to control responses. There are, however, at least two alternative accounts of the data presented thus far. The first alternative is the semantic association account, which holds that rather than using contingencies to predict responses, participants are simply forming meaningful associations between

General discussion

The present series of experiments provide a clear demonstration of implicit control. The results of all four experiments indicate that participants are able to learn and use contingencies between colour-unrelated words and colours without awareness. The specific mechanism by which participants use contingency information is further clarified by Experiment 4, where simple semantic association and stimulus familiarity accounts were ruled out and support for the behavioural control account was

Conclusions

The received view of controlled processes is that, in contrast to automatic processes, they are effortful, require more processing resources and attention, and are conscious in nature. The present results are not consistent with this simplified view of controlled processes. In particular, we are able to learn contingency information, exert control over responding with this information, and even control the degree to which such contingency information is used on a trial-by-trial basis. All of

References (39)

  • G.E. Bodner et al.

    Prime validity affects masked repetition priming: Evidence for an episodic resource account of priming

    Journal of Memory and Language

    (2001)
  • M.J. Nissen et al.

    Attentional requirements of learning: Evidence from performance measures

    Cognitive Psychology

    (1987)
  • D. Besner et al.

    Stimulus-response compatible orienting and the effect of an action not taken: Perception delayed is automaticity denied

    Psychonomic Bulletin & Review

    (2005)
  • D. Besner et al.

    What kind of attention modulates the Stroop effect?

    Psychonomic Bulletin & Review

    (1999)
  • G.E. Bodner et al.

    Beyond spreading activation: An influence of relatedness proportion on masked semantic priming

    Psychonomic Bulletin & Review

    (2003)
  • M.J. Crump et al.

    The context-specific proportion congruent Stroop effect: Location as a contextual cue

    Psychonomic Bulletin & Review

    (2006)
  • J. De Houwer

    On the role of stimulus–response and stimulus–stimulus compatibility in the Stroop effect

    Memory & Cognition

    (2003)
  • A. Destrebecqz et al.

    Can sequence learning be implicit? New evidence with the process dissociation procedure

    Psychonomic Bulletin & Review

    (2001)
  • P.A. Frensch et al.

    Implicit learning

    Current Directions in Psychological Science

    (2003)
  • G. Gratton et al.

    Optimizing the use of information: Strategic control of activation of responses

    Journal of Experimental Psychology: General

    (1992)
  • A.G. Greenwald et al.

    Long-term semantic memory versus contextual memory in unconscious number processing

    Journal of Experimental Psychology: Learning, Memory, and Cognition

    (2003)
  • H. Hendrickx et al.

    Hidden covariation detection might be very hidden indeed

    Journal of Experimental Psychology: Learning, Memory, and Cognition

    (1997)
  • H. Hendrickx et al.

    The hide-and-seek of hidden covariation detection: Reply to Lewicki, Hill, and Czyzewska (1997)

    Journal of Experimental Psychology: Learning, Memory, and Cognition

    (1997)
  • J. Hoffmann et al.

    Location-specific target expectancies in visual search

    Journal of Experimental Psychology: Human Perception and Performance

    (1999)
  • L.L. Jacoby et al.

    Item-specific control of automatic processes: Stroop process dissociations

    Psychonomic Bulletin & Review

    (2003)
  • P. Jaskowski et al.

    How the self controls its “automatic pilot” when processing subliminal information

    Journal of Cognitive Neuroscience

    (2003)
  • L. Jiménez et al.

    Which attention is needed for implicit sequence learning?

    Journal of Experimental Psychology: Learning, Memory, & Cognition

    (1999)
  • W. Kunde

    Sequential modulations of stimulus–response correspondence effects depend on awareness of response conflict

    Psychonomic Bulletin & Review

    (2003)
  • P. Lewicki

    Processing information about covariations that cannot be articulated

    Journal of Experimental Psychology: Learning, Memory, and Cognition

    (1986)
  • Cited by (0)

    This research was supported by Grant A0998 from the Natural Sciences and Engineering Research Council of Canada (NSERC) to D.B. and by an NSERC Undergraduate Student Research Award (Saskatchewan samples) and Canada Graduate Scholarship (Waterloo samples) to J.R.S.

    View full text