HSW-0225 · How Studying Works
A student is solving a science problem while holding three quantities, a diagram, the name of a process and the position of two labels in mind. Another student is assembling a geometric proof while keeping a verbal condition, a sketch, several symbols and the order of earlier steps available. If working memory were simply one tiny box with a fixed number of slots, both situations should be easy to describe: count the items, compare the count with the limit, and predict success or failure.
Real working memory is not that simple. What we can keep accessible depends not only on how much information arrives, but also on what kinds of information it is, how similar the pieces are, how they are grouped, what has already been learned, what must be recalled first and what can be represented by a compact pointer rather than by every detail at once.
The direct answer is therefore: different kinds of information can sometimes coexist in working memory more successfully than the same number of highly similar items, but this does not mean working memory becomes unlimited. Distinct sets may be grouped, indexed or partly offloaded from the narrow focus of attention, while interference and output costs still remain. The practical lesson is not “mix modalities to beat the memory limit.” It is to design study so the learner can recognise meaningful groups, retrieve their contents when needed and avoid forcing every detail to compete in the same undifferentiated mental space.
A capacity limit is real, but the unit being limited matters
Working-memory research has long shown severe capacity limits. In many laboratory tasks, only a small number of unfamiliar items can be maintained with high accuracy. This has led to memorable summaries such as “about four items,” which are useful as warnings against overload but dangerous when turned into universal laws.
The hidden question is: four of what? Four isolated colours are not psychologically identical to a familiar word, four related steps that have become one procedure, a labelled diagram that compresses several spatial relations, or three separate categories that can be accessed through distinct cues. A learner does not always hold raw detail. Sometimes the mind holds a compact representation that points toward a larger structure.
This is one reason prior knowledge matters so much. An expert can treat a meaningful configuration as one organised unit where a novice sees many separate elements. But even without expertise, the composition of a temporary memory load can matter. Similar items may compete with one another; distinct sets may provide better boundaries and retrieval routes.
What the 2026 multiple-set study actually tested
In 2026, Luísa Superbia-Guimarães and Nelson Cowan reported a series of experiments in Memory & Cognition designed to examine a problem that ordinary working-memory tasks often simplify away: people frequently need to remember several different sets of information at once. Their paper, “Colors, characters, locations, and shapes: The capacity of working memory for multiple, dissimilar sets of items”, used sequentially presented sets drawn from colours, characters, locations and shapes. Across conditions, the researchers varied total load, number of sets and similarity between sets.
The experiments involved 181 participants. The key finding was not that capacity disappeared. Instead, recall of multiple distinct sets could exceed what a naive single-pool interpretation of classic capacity limits might predict. Grouping and separation between sets appeared to matter. The authors proposed a dual-stage account in which a capacity-limited focus of attention can retain pointers to larger chunks that are temporarily represented in an activated region of long-term memory. Retrieving later sets was not free: there were costs, including loss from earlier material as more information was recalled.
That distinction is crucial. The experiments do not show that students can escape working-memory limits by colouring notes, alternating verbal and visual material, or dividing a chapter into four categories. They show that a simple count of surface items can miss how temporary memory is organised. The evidence comes from controlled serial-recall tasks, not from a classroom trial comparing study methods.
The mechanism: focus, grouping, pointers and retrieval
One useful way to think about the mechanism is to separate four jobs.
- Selection: Which information is currently in the focus of attention?
- Grouping: Which elements are treated as belonging together?
- Addressing: What cue or pointer lets the learner return to a group that is not fully active?
- Reconstruction: Can the learner unpack the group accurately when the next step needs its details?
A weak study design asks working memory to perform all four jobs at once with poorly structured material. A strong design reduces avoidable competition. It makes boundaries visible, uses stable labels, builds meaningful chunks through learning, and gives the learner practice in reconstructing rather than merely recognising those chunks.
Suppose a student must remember the stages of cellular respiration, several locations in a cell, and the molecules entering and leaving each stage. A page that presents all labels as one undifferentiated list may produce interference. A better representation can separate stage, location and input/output while preserving the relations between them. The learner still has to understand the biology. The structure simply gives working memory a cleaner address system.
Distinctness can help without making integration optional
There is a tempting but wrong conclusion here: if distinct sets are easier to keep apart, perhaps studying should maximise separation. That would solve one problem by creating another. Education usually requires integration. The science student must eventually connect the process name to the location and the molecular change. The mathematics student must connect the diagram to the algebra. The reader must connect a claim to its evidence.
The design problem is therefore two-stage. First, create enough distinctness to prevent accidental confusion. Then train the connections that make the separate sets function as one usable model. Separation supports access; integration supports reasoning.
This is why simply putting information into different colours, boxes or columns may improve navigation yet fail to improve learning. A visual boundary can tell the eye that two things differ. It cannot tell the learner why they differ, when the distinction matters or how one part constrains another.
A worked example: one algebra problem, several temporary sets
Consider this illustrative problem: a student must solve a quadratic word problem. She needs to keep the story condition, the algebraic representation, the current equation, the meaning of the roots and a restriction on the answer available while moving through the solution.
If she tries to preserve every sentence verbatim, every intermediate line and every possible method, working memory fills quickly. A better approach is to compress the problem into separate but connected sets:
- Model set: what the variables mean and the relationship given by the story.
- Procedure set: the current algebraic transformation.
- Constraint set: which mathematical roots are physically or contextually admissible.
The labels are not magic. Their value comes from reducing unnecessary search. When the equation is solved, “constraint” becomes a pointer: the learner knows to return to the condition that decides which root survives. With practice, more of the procedure can become a learned chunk, freeing attention for modelling and verification.
Why similarity can be expensive
Working memory is vulnerable not only to quantity but to confusability. Similar formulae, similar vocabulary, similar grammatical structures, similar historical names and similar problem types can interfere with one another. The learner may preserve the general region of knowledge while losing the identity of the item that belongs in the current slot.
This is an important reason to teach discrimination explicitly. If two concepts are often confused, studying them farther apart is not always the solution. Sometimes they need to be compared directly so the learner can build separate retrieval conditions: this cue selects A; that cue selects B. The purpose is not to make everything visually different. It is to make the decision boundary cognitively meaningful.
The existing eduKateSG guide How Working Memory Affects Examination Performance owns the broader working-memory problem under examination conditions. The related guide How Cognitive Load Works explains overload more generally. This article has a narrower job: understanding why the composition and organisation of a temporary load can change what a simple item count predicts.
Three misconceptions worth removing
Misconception 1: “Working memory holds exactly four things.” Capacity estimates depend on task, stimulus, knowledge, grouping, attention and scoring. “A few items” is a useful caution; an exact universal number is not.
Misconception 2: “Different modalities have separate unlimited channels.” Distinct representations can reduce some forms of interference, but attention, control and integration remain limited. A learner can still be overloaded by text, diagrams and speech arriving together.
Misconception 3: “Chunking means renaming a large list.” A true useful chunk has learned internal structure. Calling ten unrelated facts “Unit A” does not make them one cognitive unit. The learner must be able to use the label to recover an organised set.
A diagnostic routine for students
When a problem feels mentally crowded, do not immediately conclude that the topic is too difficult. Ask where the temporary-memory burden comes from.
- What information must remain active for the next decision?
- Which pieces are genuinely separate, and which belong in one learned group?
- Which similar items are being confused?
- Can a stable label, diagram or notation act as a pointer without hiding the meaning?
- What can be written down safely, and what must be retrievable from memory because the final task will not provide the support?
- Can you reconstruct each group after a short delay without looking?
The last question prevents a common failure. External structure can make the task feel easier while the learner remains dependent on it. If the examination or real-world task requires independent reconstruction, the study routine must eventually remove the labels and test whether the internal address system survives.
What parents and tutors can change
A tutor often sees working-memory overload as hesitation, lost steps, repeated rereading or a student asking “Where was I?” The first response should not always be to explain faster. Reduce the coordination burden and observe what remains.
One useful teaching sequence is separate → label → connect → retrieve → remove support. First make the relevant sets visible. Then give each a meaningful name. Next teach the relations between them. Ask the learner to retrieve the groups and their relations from a partially blank representation. Finally remove the scaffold and require independent performance.
This sequence is especially useful when a learner knows individual facts but loses the thread while combining them. It is less useful when the real problem is missing prerequisite knowledge. Structure cannot compensate for knowledge that was never built.
The delayed test: did organisation become knowledge?
Immediate success with a well-organised page is not enough. After a delay, give the learner a fresh problem with the same underlying coordination demand but different surface details. Remove the original colour scheme and layout. Ask the learner to identify the relevant sets, reconstruct what each contains and connect them in the correct order.
If performance collapses only when the page design disappears, the layout was carrying too much of the task. If the learner can rebuild the organisation independently and use it in a changed problem, the structure has begun to become capability.
What the evidence does not yet justify
The 2026 multiple-set experiments sharpen theory about working-memory organisation, but they do not establish a ready-made classroom intervention. They used controlled sequential-recall tasks and specific stimulus sets. Classroom materials are richer, prior knowledge varies widely, and educational tasks require comprehension, reasoning and transfer as well as recall.
The pointer-and-activated-long-term-memory account is an explanatory model, not a photograph of a literal mental filing cabinet. Competing working-memory theories describe capacity and attention differently. The durable conclusion is narrower: temporary memory performance depends on structure, similarity, grouping and retrieval demands, so a single surface item count can be misleading.
A practical rule that survives the caveats
When study material feels too large to hold, do not ask only, “How many things are there?” Ask, “What are the groups, what distinguishes them, what connects them, and how will I retrieve each one when the support disappears?”
That question respects the capacity limit without turning it into a myth. Good studying does not make working memory infinite. It changes the representation so scarce attention is spent on the decision that matters rather than on repeatedly finding the pieces.
Research and further reading
- Superbia-Guimarães, L., & Cowan, N. (2026). Colors, characters, locations, and shapes: The capacity of working memory for multiple, dissimilar sets of items. Memory & Cognition, 54, 1263–1291.
- How Studying Works | Numbered Series Reading Index
- How X Works | Master Hub