Memories of control: One-shot episodic learning of item-specific stimulus-control associations
Introduction
Cognitive control denotes the ability to use internal goals and knowledge to coordinate behavior in a context-sensitive fashion (Miller & Cohen, 2001). This requires maintaining – and shielding – goal-relevant stimuli and rules linking them to appropriate actions (task sets), as well as updating rules in response to changing circumstances (i.e., task switching; Frank, Loughry, & O'Reilly, 2001; Monsell, 2003). Much research on cognitive control has focused on the key question of how we apply a given control operation in a context-appropriate manner (e.g., Blais, Robidoux, Risko, & Besner, 2007; Botvinick, Braver, Barch, Carter, & Cohen, 2001; Verguts & Notebaert, 2008). One answer to this question appears to be that control processes, such as task switching, can become mnemonically associated with particular contexts or stimuli (reviewed in Egner, 2014; Abrahamse, Braem, Notebaert, & Verguts, 2016; Braem & Egner, 2018). For instance, the repeated pairing of a particular stimulus with the need to switch (as opposed to repeat) a task leads to an item-specific reduction in task switching costs, the relatively longer and more erroneous responses typically observed on task switch compared to task repeat trials (Chiu & Egner, 2017; Leboe, Wong, Crump, & Stobbe, 2008). Similarly, rendering specific target stimuli predictive of conflicting distracters in Stroop-type tasks results in item-specific reductions of interference effects (Bugg & Dey, 2018; Bugg & Hutchison, 2013; Bugg, Jacoby, & Chanani, 2011).
The exact learning processes that mediate the formation of such context- or item-specific stimulus-control associations, however, are not yet well understood (for a recent review, see Chiu & Egner, 2019). One popular conception is that control learning may reflect the formation, and cued retrieval of episodic event files (Egner, 2014; Spapé & Hommel, 2008). That is, a given event (e.g., a trial) is encoded as an episodic memory file that binds together various event features, including different stimulus characteristics (Treisman & Gelade, 1980), as well as actions performed in response to the stimulus (Hommel, 1998). When a feature of that event (e.g., the stimulus) re-occurs in the future, it automatically retrieves the event file with all its component features, which can serve as a shortcut to the appropriate response (reviewed in Hommel, 2004). Applied to context-sensitive cognitive control, this idea has been extended via the proposal that episodic event files – in addition to stimulus and response features - also encode internal states, notably ongoing control operations (Egner, 2014). Thus, when an event feature re-occurs, the retrieval of the event file not only invokes previously co-occurring stimuli and motor responses, but also reinstates the relevant control process, leading to the observed performance benefits. We will here refer to this proposal as the episodic control-binding hypothesis.
A number of studies on contextual adjustments of control have been interpreted within this episodic control-binding framework (e.g., Dignath, Johannsen, Hommel, & Kiesel, 2019; Jiang, Brashier, & Egner, 2015; Spapé & Hommel, 2008), but unambiguous evidence supporting this account is still lacking. This is largely due to previous studies' use of a small number of stimuli repeated many times over the course of an experiment. This procedure was chosen to create probabilistic associations between items and control demands, for instance, in item-specific proportion switch (e.g., Chiu & Egner, 2017) or proportion congruent (e.g., Bugg et al., 2011) manipulations. However, this prevents one from distinguishing whether enhanced performance for items that have been associated repeatedly with a particular control demand is mediated by the retrieval of specific previous episodes involving that item or whether enhanced performance is mediated by non-episodic, incremental learning mechanisms. For instance, a number of previous studies have modeled contextual adjustments in control via reinforcement learning algorithms. There, control is nudged up or down from trial to trial based on cumulative experience rather than the retrieval of specific past episodes (e.g., Botvinick et al., 2001; Jiang et al., 2015; Chiu et al., 2017).
To provide a proper test of the episodic control-binding hypothesis, one would therefore have to document experience-based control learning in the absence of any potential influence of incremental rule learning. One clean way of doing that is to probe whether performance improvements due to putative stimulus-control associations can be observed under one-shot learning: Can a single exposure to a “prime” event result in the reinstatement of the associated control operations in a later, matching “probe” event? We are aware of only one previously-attempted test of one-shot stimulus-control associations (Brosowsky & Crump, 2018). However, Brosowsky and Crump's (2018) experiments were unable to provide unambiguous evidence for this type of learning, as they confounded differences in the overlap of cognitive control requirements between primes and probes with differences in stimulus overlap (cf. Hommel, Proctor, & Vu, 2004).
In the present study, we therefore developed a novel, direct test of the episodic control-binding hypothesis that allowed for a confound-free test of reinstatement of control based on episodic stimulus-control bindings. We expanded a design that recently documented one-shot acquisition of stimulus-action and stimulus-classification associations (Pfeuffer, Moutsopoulou, Pfister, Waszak, & Kiesel, 2017; Pfeuffer, Moutsopoulou, Waszak, & Kiesel, 2018; see also Moutsopoulou, Yang, Desantis, & Waszak, 2015) to incorporate a test of one-shot learning of item-specific associations between stimuli and the control process of task switching. Across three experiments, we observed robust support for the episodic control-binding hypothesis.
Section snippets
Experiments 1a and 1b
To test whether stimulus-control associations can be acquired through rapid, episodic learning, we employed a modified version of the prime-probe task used by Pfeuffer et al. (2017). In Experiment 1a, we employed two prime exposures: Individual stimuli were paired twice with a specific control demand (i.e., presented as task repeat vs. switch trials) before being probed for an item-specific stimulus-control association. In Experiment 1b, we tested whether item-specific stimulus-control
Experiment 2
The observation of one-shot, item-specific stimulus-control associations in Experiment 1 supports the idea that external stimuli and internal control processes (here: task switching) become bound together in episodic memory. Next, we aimed to determine whether these item-specific stimulus-control associations are context- (or task-) specific by varying stimulus-classification (S-C) mappings in addition to the task sequence factor. We modified the one-shot design of Experiment 1b such that the
Experiment 3
Experiments 1 and 2 provided support for the episodic control-binding hypotheses. However, results were obtained in the context of a highly regularized task structure, where only 4 (Experiments 1a/1b) or 8 (Experiment 2) items were relevant per mini-block and participants could potentially form expectations that a specific subset of prime stimuli would soon be repeated as probe stimuli. In Experiment 3, we sought to test whether one-shot, item-specific stimulus-control associations would also
Discussion
Across three experiments, we consistently demonstrated the formation of one-shot, item-specific stimulus-control associations, as reflected in reduced task-switching costs for probes whose primes were task switch as opposed to repeat trials. This strongly supports the episodic control-binding hypothesis. While in Experiment 1a we documented the creation of item-specific stimulus-control associations given two probe exposures, in Experiment 1b we found that a single, one-shot presentation of a
Author contributions
T. Egner and C.U. Pfeuffer developed the study concept. All authors contributed to the study design. Data collection and analysis were performed by P.S. Whitehead under the supervision of T. Egner. All authors wrote and approved the final version of the manuscript for submission.
Credit authorship contribution statement
Peter S. Whitehead:Methodology, Formal analysis, Investigation, Writing - original draft, Writing - review & editing, Visualization.Christina U. Pfeuffer:Conceptualization, Methodology, Writing - original draft, Writing - review & editing.Tobias Egner:Conceptualization, Methodology, Formal analysis, Investigation, Writing - original draft, Writing - review & editing, Supervision, Funding acquisition.
Acknowledgements
We thank Sam Verschooren for helpful comments on an earlier version of this draft. This work was supported by NIMH R01 MH116967 (T.E.).
References (43)
- et al.
Cortical and subcortical contributions to context-control learning
Neuroscience & Biobehavioral Reviews
(2019) Chapter 20 what are the differences between long-term, short-term, and working memory?
- et al.
Complementary memory systems: Competition, cooperation and compensation
Trends in Neurosciences
(2005) Event files: Feature binding in and across perception and action
Trends in Cognitive Sciences
(2004)Task switching
Trends in Cognitive Sciences
(2003)- et al.
Multiple priming instances increase the impact of practice-based but not verbal code-based stimulus-response associations
Acta Psychologica
(2018) - et al.
A feature-integration theory of attention
Cognitive Psychology
(1980) - et al.
Task-switching and long-term priming: Role of episodic stimulus–task bindings in task-shift costs
Cognitive psychology
(2003) - et al.
Grounding cognitive control in associative learning
Psychological Bulletin
(2016) - et al.
Creating implicit task-sets from categorical stimuli: When structure learning degrades incidental memory
(2018)