Memory and instructions

Definition

Working memory is the system that holds and manipulates information for a short time during a task, such as a value read from a display or the next action of a procedure. Long-term memory stores knowledge and experience over long periods. Chunking is the grouping of items into larger meaningful units, which allows more information to be held. Recognition, the identification of something presented again, is generally easier than recall, the retrieval of information without a cue. Prospective memory is remembering to carry out an intended action at a later point, such as returning to an interrupted step.

Historical development

Frederic Bartlett (1932) showed in studies of recalled stories that memory reconstructs rather than copies, guided by schemas — organized knowledge about familiar situations. George A. Miller’s 1956 paper “The magical number seven, plus or minus two” discussed limits of immediate memory and absolute judgment and introduced chunking as the way these limits are extended. Richard Atkinson and Richard Shiffrin (1968) described a model of separate memory stores. Alan Baddeley and Graham Hitch (1974) replaced the single short-term store with a multicomponent working memory, and Baddeley added the episodic buffer in 2000. Endel Tulving (1972) distinguished episodic from semantic memory. Nelson Cowan (2001) reviewed the evidence and estimated the capacity of working memory at about three to five chunks when rehearsal and grouping are prevented.

Main models

1. Multicomponent working memory (Baddeley and Hitch). A central executive controls a phonological loop for verbal material, a visuospatial sketchpad for visual and spatial material and, since 2000, an episodic buffer that integrates information.
2. Capacity limits. Working memory capacity is small and measured in chunks, not in words or items; what counts as a chunk depends on prior knowledge.
3. Schema theory. Knowledge structures guide expectation, interpretation and recall, and can lead to errors when a situation differs from the familiar pattern.
4. Memory systems. Episodic, semantic and procedural knowledge are distinguished; skilled procedures can run with little conscious control.
5. Prospective memory. Intentions for later actions are easily lost after interruptions unless cues are present.

Function and purpose

These models explain why readers lose track in long or interrupted procedures, why familiar terms and layouts are processed more easily than unfamiliar ones, and why expectations based on experience can override what a text actually says.

Applications in technical communication

Documentation practice draws several consequences, set out in the context card Working memory and procedure steps. Each step contains one action, and values the reader must carry from one place to another are stated where they are needed instead of earlier in the text. Long procedures are divided into meaningful subtasks, which act as chunks; the number of steps follows from the task, not from a fixed limit. Navigation and interfaces rely on recognition — visible headings, labels and menus — rather than on recall of hidden terms. Interruptions are anticipated with clear step numbers and stated results, so that a reader can find the point of resumption. Where readers are likely to apply an incorrect schema, for example when a product behaves differently from its predecessor, the difference is stated explicitly. These choices concern presentation; they never justify omitting a step, a warning or other information that the instructions for use must contain.

Limitations and evidence

Miller’s paper is often cited as a rule that lists, menus or procedures should have at most seven items. The paper does not support such a rule: it concerned immediate recall of unrelated items, and the capacity it discussed depends on chunking. Cowan’s lower estimate refers to conditions without rehearsal and is not a design limit either. Capacity varies between people and materials, and readers of instructions can look back at the page, which changes the demand on memory. The popular idea that people learn better when teaching matches a preferred “learning style” has not been supported by controlled studies (Harold Pashler and colleagues, 2008) and is not a basis for documentation design.

Relation to other areas

Memory connects attention, which selects what enters working memory, with comprehension, which builds a representation from it. Cognitive load theory (see Cognitive load) rests on the limits of working memory, and the analysis of slips and lapses (see Judgment and human error) draws on prospective memory and schemas.

Conclusion

Memory research supports instructions that place information where it is used, structure tasks into meaningful units and rely on recognition. It does not support numerical rules such as “seven plus or minus two” for the length of procedures.