Where does working memory collide?

In Baddeley and Hitch's account, working memory is not one store but a set of specialised subsystems run by a central executive. Their 1974 chapter proposed a phonological loop that holds speech-based material and a visuospatial sketchpad that holds visual and spatial material, each with its own limited capacity, coordinated by an attentional controller. Baddeley added a fourth part in 2000, the episodic buffer, a limited store that binds information from the subsystems and from long-term memory into unified episodes. What this model names is the architecture: the capacity is divided by the kind of material held and by the job of control. The referent is a structure of components, not a single pool, and it is studied by showing that loading one subsystem leaves the others working. This is a distinct construct from memory in general, which the page for that word treats as the broad family of storage and retrieval.[1][2]

Unit

A person's temporary holding and control of material, split by subsystem

Recovery means

Rehearsal in the phonological loop refreshes decaying traces; without it, verbal material fades within seconds

Measured by

Dual-task and interference studies; digit and word span, plus the word-length and irrelevant-speech effects

Cowan's embedded-processes model treats working memory not as a separate structure but as the currently activated part of long-term memory, with a narrower focus of attention inside it. Information that is activated stays available for a short time; the small subset that attention holds is what is immediately usable, and Cowan put its capacity at about four chunks in adults when rehearsal and grouping are prevented. His 2001 target article argued that this limit of four, not the older seven, is the real capacity of the focus of attention. His 2010 review restated the case and tied the four-chunk limit to a wide range of tasks. What this model names is a state of attention and activation rather than a set of boxes: the same capacity that Baddeley carves into subsystems, Cowan locates in a single focus of attention. The meaning here is a graded degree of activation, not a fixed set of stores.[3][4]

Unit

The number of chunks the focus of attention can hold at once

Recovery means

Activation decays over time unless attention or rehearsal renews it; the focus itself is limited to about four chunks

Measured by

Span tasks with rehearsal and chunking blocked; change detection and running-span estimates of the focus of attention

Engle and his colleagues define working memory capacity as executive or controlled attention: the ability to keep goal-relevant information active and to resist interference, especially when distraction is strong. On this view the differences between people on span tasks reflect not the size of a store but how well attention can be controlled, and that control is what makes span measures predict fluid intelligence and reading comprehension. Engle's 2002 paper set out the case that complex span tasks, which mix storage with processing, tap this attentional control rather than a passive buffer. Kane and Engle extended the argument, tying the capacity to holding goals in mind, to overriding habitual responses, and to the prefrontal cortex. What this model names is a person's power to direct attention, indexed through memory tasks, so the same word points at an individual-differences trait rather than at an architecture or a state. The referent is controlled attention, close to but not the same as general intelligence.[5][6]

Unit

A person's capacity to control attention, indexed by complex span

Recovery means

Not applicable

Measured by

Complex span tasks such as operation and reading span; attention-control tasks under interference, related to fluid intelligence

Ericsson and Kintsch's long-term working memory challenges the idea that the working store is always small and fixed. They argued that skilled performers, such as expert chess players, skilled readers, and mental calculators, keep far more task-relevant information available than a short-term limit allows, by encoding it in long-term memory with retrieval structures that give fast, reliable access. On this account the effective capacity for a familiar domain is not the handful of chunks in temporary storage but a large body of knowledge that acquired skill makes instantly retrievable. Their 1995 paper in Psychological Review laid out the mechanism and its evidence from expertise and text comprehension. What this model names is a skill-dependent extension of capacity through long-term memory, so the same phrase, working memory, points at something that grows with practice rather than a fixed bottleneck. The referent is retrieval skill, indexed by capacity, not a store of a fixed size.[7]

Unit

The task-relevant information an expert can retrieve on demand within a familiar domain

Recovery means

Not applicable

Measured by

Expert-novice comparisons; recall after interruption in skilled domains such as chess, reading, and mental calculation

Miller's 1956 paper gave the field its first famous number: the immediate memory span, the amount that can be repeated back right after presentation, sits at about seven items, plus or minus two. Miller noted that this span holds roughly constant in items rather than in raw information, so recoding material into larger chunks raises how much can be held. The paper is often read as the founding statement of a fixed short-term capacity, though Miller himself was cautious and drew together several separate limits under one memorable figure. Later work, including Cowan's, argued that the real limit, once chunking and rehearsal are controlled, is closer to four than seven. What this label names is the classic span, the plain count of items reproduced in order, the ancestor that the other models on this map revise or replace. The referent is a behavioural span and the root of the modern idea of a memory capacity, not yet a theory of structure or attention.[8]

Unit

The number of items reproduced in order right after presentation

Recovery means

Not applicable

Measured by

Immediate serial recall; digit span and item span, with chunking used to raise the count