Habits, Goals, and Behavioral Signs of Cognitive Perseveration on Wisconsin Card-Sorting Tasks

Abstract

Wisconsin card-sorting tasks provide unique opportunities to study cognitive flexibility and its limitations, which express themselves behaviorally as perseverative errors (PE). PE refer to those behavioral errors on Wisconsin card-sorting tasks that are committed when cognitive rules are maintained even though recently received outcomes demand to switch to other rules (i.e., cognitive perseveration). We explored error-suppression effects (ESE) across three Wisconsin card-sorting studies. ESE refer to the phenomenon that PE are reduced on repetitive trials compared to non-repetitive trials. We replicated ESE in all three Wisconsin card-sorting studies. Study 1 revealed that non-associative accounts of ESE, in particular the idea that cognitive inhibition may account for them, are not tenable. Study 2 suggested that models of instrumental learning are among the most promising associative accounts of ESE. Instrumental learning comprises goal-directed control and the formation of corresponding associative memories over and above the formation of habitual memories according to dual-process models of instrumental learning. Study 3 showed that cognitive, rather than motor, representations of responses should be conceptualized as elements entering goal-directed instrumental memories. Collectively, the results imply that ESE on Wisconsin card-sorting tasks are not only a highly replicable phenomenon, but they also indicate that ESE provide an opportunity to study cognitive mechanisms of goal-directed instrumental control. Based on the reported data, we present a novel theory of cognitive perseveration (i.e., the 'goal-directed instrumental control' GIC model), which is outlined in the Concluding Discussion.

Keywords: GIC; Wisconsin card sorting; cognitive flexibility; cognitive perseveration; error suppression; goal-directed control; instrumental learning.

Figure 1

An outline of error-suppression effects (ESE). Figure 1A shows the typical layout of the Wisconsin stimulus material. All Wisconsin stimulus cards can be described in terms of a three-dimensional rule space (with the rules COLOR/SHAPE/NUMBER of the depicted items). Each rule is instantiated by four distinct features (with COLOR features: red, green, yellow, blue/SHAPE features: triangle, star, cross, circle/NUMBER features: #1, #2, #3, #4). On each trial, a response is requested that assigns the current target card to one of four simultaneously presented key cards (i.e., outside-left key card: ‘1 red triangle’, inside-left key card: ‘2 green stars’, inside-right key card: ‘3 yellow crosses’, outside-right key card: ‘4 blue circles’). The task is selecting—on each trial—the key card that shares the feature with the to-be-prioritized rule-contingent target feature (be that the COLOR, SHAPE or NUMBER feature). This is exemplified on trial t − 1 of Figure 1A by selecting the inside-left key card ‘2 green stars’ that shares the COLOR feature ‘green’ with the target card by pressing the response button that is labeled ‘2’. (A) In a previous study [9], we stratified perseverative errors (PE) by trial-by-trial transitions of selected key cards/corresponding responses. Key card/response transitions are highlighted in light blue (see the framed key cards/response buttons on trial t). In the depicted example, re-application of the COLOR rule would result in a PE; potential PE are shown as red response buttons on trial t. We compared repetitive PE that implied repeated key cards/responses (right panels) with non-repetitive PE that implied altered key cards/responses (left panels). Repetitive PE comprise the repetition of key cards (incl. prioritized features, i.e., ‘green’ on both trials in the example) and of responses (‘inside-left’ on both trials in the example). Non-repetitive PE comprise the alteration of key cards (incl. prioritized features, i.e., from ‘green’ on trial t − 1 to ‘blue’ on trial t in the example) and of responses (from ‘inside-left’ on trial t − 1 to ‘outside-right’ on trial t in the example). The y-axis shows time in discrete trial-based units (t − 1, t). (B) ESE refer to the finding that conditional probabilities of repetitive PE (involving key card and response repetitions) are lower than conditional probabilities of non-repetitive PE (involving key card and response alterations). Our initial ESE study [9] was based on a sample of neurology inpatients who were assessed by means of a paper-and-pencil version of the Wisconsin card-sorting task. (C) In a follow-up study [10], we replicated ESE in a relatively large sample of young volunteers who were assessed by means of a computerized version of the Wisconsin card-sorting task (cWCST). (B,C) The y-axes show conditional PE probabilities (sample means, inter-individual variabilities).

Figure 2

The design (top panels) and results (bottom panels) of Study 1, which examined non-associative and associative accounts of ESE. (Top panels): The allocation of the four key cards to spatial positions was randomly rescheduled on each trial. The main comparison was between those trials t on which key cards that were selected on trial t − 1 retained their spatial positions on trial t (left column) and those trials t on which key cards that were selected on trial t − 1 altered their spatial positions on trial t (central and right columns). The y-axis shows time in discrete trial-based units (t − 1, t). (Bottom panels): The y-axes show conditional PE probabilities (sample means, inter-individual variabilities). (Left column): ESE were discernible (i.e., repetitive PE were reduced compared to non-repetitive PE) when previously selected key cards re-occurred at their previous spatial positions, thereby replicating the ESE (see also Figure 1; [9,10]). (Central column): When previously selected key cards altered their spatial positions, no ESE were discernible when solely the previously prioritized key-card features were repeated (versus altered). (Right column): When altered key cards occurred at previously selected spatial positions, no ESE were discernible when the previous responses were repeated (versus altered). Thus, ESE are dependent on the repeated presentation of identical key cards (incl. repetition of prioritized features) at one and the same spatial position. See text for more details.

Figure 3

The design (top panels) and results (bottom panels) of Study 2, which investigated associative accounts of the origin of ESE. As in standard Wisconsin card-sorting tasks, the key cards occupied constant spatial positions. We manipulated the key-card/response mapping on a trial-by-trial basis, with two different mappings. One mapping is the ‘up-down’ mapping (i.e., ‘1 red triangle’ key card maps to response button 1; ‘2 green stars’ key card maps to response button 2; ‘3 yellow crosses’ key card maps to response button 3; ‘4 blue circles’ key card maps to response button 4). The other mapping is the ‘down-up’ mapping (i.e., ‘1 red triangle’ key card maps to response button 4; ‘2 green stars’ key card maps to response button 3; ‘3 yellow crosses’ key card maps to response button 3; ‘4 blue circles’ key card maps to response button 1). (Left column): ESE were discernible (i.e., repetitive PE were reduced compared to non-repetitive PE) as long as the S-R mapping remained unchanged across trials, thereby replicating the ESE (see also Figure 1; [9,10]). (Right column): No ESE were discernible when the S-R mapping changed across trials. The y-axes show conditional PE probabilities (sample means, inter-individual variabilities). See text for more details.

Figure 4

The design (top panels) and results (bottom panels) of Study 3, which investigated the nature of associative accounts of the origin of ESE. As in standard Wisconsin card-sorting tasks, the key cards occupied constant spatial positions. We manipulated the response/hand mapping on a trial-by-trial basis, with two different mappings. One mapping represented the ‘respond with the right hand’-mapping. The other mapping represented the ‘respond with the left hand’-mapping. (Left column): ESE were discernible (i.e., repetitive PE were reduced compared to non-repetitive PE) when the response/hand mapping remained unchanged across trials, thereby replicating the ESE (see also Figure 1; [9,10]). (Right column): ESE were discernible (i.e., repetitive PE were reduced compared to non-repetitive PE) even though the response/hand mapping changed across trials, indicating that ESE are not effector specific. Presumably, response codes are specified at a conceptual level such as ‘button 1’ = ‘outside-left’, ‘button 2’ = ‘inside-left’, ‘button 3’ = ‘inside-right’, ‘button 4’ = ‘outside-right’. The y-axes show conditional PE probabilities (sample means, inter-individual variabilities). See text for more details.

Figure 5

A graphical outline of the goal-directed instrumental control (GIC) model of cognitive perseveration. The left column shows presumed cognitive events (mental representations are shown black; processes are shown in warm colors (orange, pink, red); their observable behavioral expressions are shown in blue; blue arrows show strength of PE decrease) on exemplary single trials (t − 1 (affording a rule switch, t). The GIC model distinguishes between retrieval-based and feedback-based routes to GIC. Top panel: On repetitive trials, repeating identical S-R conjunctions (S-R) on t − 1 and on t retrieves goal-directed instrumental memories (pink rectangle). This retrieval brings back to mind that the application of the corresponding rule had just now been disconfirmed via O(s). As an effect of this retrieval-based route to GIC, previously prioritized rules (r(t − 1)) are subject to additive (retrieval-based plus feedback-based) inhibition from GIC (shown in red); hence, repetitive PE are strongly suppressed. Bottom panel: On non-repetitive trials, altered S-R conjunctions on t − 1 (S-R) and on t (S’- R’) do not retrieve goal-directed instrumental memories (pink rectangle). As an effect, previously prioritized rules (r(t − 1)) are solely subject to feedback-based inhibition from GIC (shown in red); hence, non-repetitive PE are less strongly suppressed. The right column shows emerging summary statistics of observable PE across multiple trials: Repetitive PE ($p_r$) are reduced compared to non-repetitive PE ($p_n$), paving the way for what we refer to as conjunctive error-suppression effects (ESE). S, S’ = stimulus (rule-contingent feature); R, R’ = response (spatial code); O(s) = outcome (‘switch’); r(t − 1) = prioritized rule on t − 1; PE = perseverative error. Arrows as endpoints: activation; circles as endpoints: inhibition. See text for more details.