Contextual Pointers and Sensory Codes in Working Memory: Toward a Hierarchical Framework

Working memory (WM) research has long debated whether its storage operates via discrete slots or distributed resources. However, emerging neuroscientific evidence suggests that neither account alone fully captures the complexity of WM function. Here, I propose a context-content hierarchical framework that unifies these views. A capacity-limited “context” system maintains discrete spatiotemporal pointers that individuate and prioritize items, while a precision-limited “content” system encodes detailed sensory information in modality-specific population codes whose fidelity varies with task demands. These systems interact through predictive top-down feedback and rhythmic coordination to support “holographic” reconstruction, in which a low-dimensional pointer selectively reinstates high-dimensional content on demand. This architecture explains how stable capacity limits can coexist with graded precision, activity-silent states, and rapid cue-driven reinstatement, and clarifies when contextual retrieval protects performance from interference. The framework yields clear experimental predictions and motivates computational models in which pointer indexing is coupled to sensory reconstruction. By integrating discrete control signals with continuous sensory representations, the context–content framework reconceptualizes WM as a dynamic, generative system that balances sparse indexing with on-demand reconstruction to guide adaptive behavior.