An Episodic Model of Task Switching Effects: Erasing the Homunculus from Memory

Schmidt, James R.; Liefooghe, Baptist; De Houwer, Jan · 2020 · OpenAlex-citations

DOI: 10.5334/joc.97

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This review article presents an episodic account of task switching, challenging dominant theories that attribute switch costs to executive control processes like task-set reconfiguration or inertia. The authors argue that much of the observed performance difference between task repetitions and switches is actually driven by feature integration biases—systematic effects arising from the repetition or alternation of specific stimuli, responses, and cues. To demonstrate this, the paper extends the Parallel Episodic Processing (PEP) model, a neural network based on exemplar memory, to simulate human performance in cued task switching paradigms. The methodology involves adapting the PEP model to handle multi-goal situations by encoding task instructions (cue-to-goal and decision-to-response mappings) as episodic memories rather than hardwired parameters. This allows the model to "follow instructions" through retrieval processes similar to those used for learning from experience. The extended model was used to simulate key findings from the task switching literature, including the switch cost, task-rule congruency effects, response repetition asymmetries, and cue repetition benefits. Specifically, the authors simulated data from a feature integration decomposition study (Schmidt & Liefooghe, 2016), where trials were categorized by the repetition or alternation of cues, stimuli, decisions, and responses. The simulations tested whether the model could reproduce the full pattern of means from this complex experimental design without arbitrary parameter adjustments. The results show that the extended PEP model fits participant data well and naturally reproduces the observed effects. The model generates switch costs primarily through feature integration biases: recently encoded episodes exert a strong influence on current retrieval, facilitating performance when features repeat (complete repetitions) and hindering it when they conflict (partial repetitions). For instance, repeated cues speed up goal retrieval, while repeated stimuli bias responses based on the previous trial, creating costs on task switches where the required response changes. The paper demonstrates that competing models, which rely on task-set control mechanisms, fail to account for these specific binding biases. Furthermore, the PEP model possesses fixed qualitative predictions, meaning it cannot be flexibly tuned to fit any arbitrary pattern of results, thereby validating its explanatory power. The significance of this work lies in its parsimonious explanation of task switching phenomena. By erasing the "homunculus" (vague executive control agents) from the model, the authors show that episodic memory processes alone can explain complex switching effects. The findings suggest that feature integration biases have a far greater impact on task-switching performance than previously assumed, reducing the need for dedicated control mechanisms to explain the switch cost. This supports a unified view of cognition where instruction following, learning, and task execution are all supported by the same underlying episodic encoding and retrieval processes.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• task switching