Figure 1
Illustration of the Binding and Retrieval in Action Control (BRAC) framework.
Note. Sketch of the BRAC framework. Adapted from Frings et al., (2020).
Figure 2
Illustration of the procedural WM model and its main components, the set-selection and response selection module.
Note. Procedural WM is a WM subsystem next to declarative WM. The model developed by Oberauer et al. (2013) consists of two modules, the set-selection module and the response-set module. In the set-selection module the current task cue (C) activates the task set (T) currently relevant. The response-selection module holds the task-relevant bindings between stimuli (S) and responses (R). The activated stimuli are represented in the input layer which forward activations to the responses in the output layer. The model contains an additional third layer, the candidate layer, which is a replica of the output layer.
Table 1
Summary of the comparative analysis of BRAC and WM perspectives on the representation of stimulus-response bindings.
| BRAC FRAMEWORK | WM | ||
|---|---|---|---|
| General System Structure | Separation of Memory Systems | Single memory system with short-lived bindings; recent clarifications suggest transition from WM to LTM based on binding strength, repetition, and spacing. | Separation of WM (limited capacity, fast learning rate) and LTM (large capacity, slow learning rate, stores information permanently). LTM representations built in parallel to WM. |
| Capacity Limits | Implicit capacity limits assumed, based on decay and interference within WM. | WM has limited capacity; LTM has no apparent capacity limit. Interference in WM as a limiting factor. | |
| Representation of Stimuli, Responses | Event files with neuronally distributed, multimodal feature representations; no clear distinction between perception and action. | Stimuli and responses represented as activation patterns in distinct layers. | |
| Representation of Task, and Control Sets | Task and control states included but not specifically detailed. | Task sets are represented as bindings in the response-selection module, with learned sets in the set-selection module. | |
| Structure and Properties of Bindings | Formation of Bindings | Created based on co-occurrence, influenced by attention and saliency; bindings of all features in independent event files. | Rapid Hebbian learning binds active representations, represented in a weight matrix across events; strength influenced by activation levels. Allows for independent pairwise bindings between stimuli and responses. |
| Episode Boundaries for Bindings | Bindings exist for an episode, with boundaries under investigation (decay and unbinding debated). | Bindings remain until actively removed; no causal role for episode boundaries. | |
| Code Occupation | Feature representations can only be in one event file at a time; unbinding needed for new bindings. | No constraint on binding features to multiple stimuli or responses. | |
| Sequential Effects on Bindings | Short-lived binding effects; full repetition benefits vs. partial repetition costs. | Previous actions affect subsequent actions due to temporary strengthening of stimulus-response bindings, incomplete updating of bindings, and residual activation. | |
| Proposed Processes | Selective Attention | Selective attention influences which features are bound; activated features have higher binding probability. | Features activated through attention are bound via Hebbian learning; selective attention modulates activation strength. |
| Learning | Bindings as building blocks for learning; transition to LTM with binding strength. | Separate rapid binding in WM and slow learning in LTM; parallel updates in response-selection and set-selection modules. | |
| Retrieval | Feature repetition serves as retrieval cues; reactivates the entire event file. | Cue-based retrieval; stimuli act as retrieval cues for responses; task cues act as retrieval cues for task sets. | |
| Updating | Not explicitly specified; dissolution through decay or unbinding necessary for new bindings. | Explicit updating through delta rule; iterative process until match criterion is reached. | |
| Practical Application in Experiments | Dependent Variables | Primarily reaction times, with some accuracy measurements; focus on sequential effects in prime-probe designs. | Equal weighting of reaction times and accuracy; general model for action control in simple tasks. |
| Application in Paradigms | Broad application across various paradigms (e.g., stimulus categorization, action planning, task switching, Stroop-like tasks). | Applied to task-set switching, memory-set switching, and object switching within memory sets. |
Figure 3
Predictions form the procedural WM model for partial and full repetition effects.
Note. Data simulated with the WM model applied to an experiment demonstrating partial-repetition costs for relevant and irrelevant stimulus features as well as responses. Reaction times (bottom row) are simulated in an artificial unit (a.u.) and absolute values should not be interpreted. Rep = repetition; Alt = alternation.