Schemas, Chunking and Working Memory
Working memory (WM) is a, if not the, central component of human cognition. WM is responsible for temporarily storing and manipulating information required for reasoning, comprehension, and learning (Baddeley, 2000). However, WM is extremely limited in capacity and duration. Classic research by Miller published in 1956 on how many numbers we could remember (carried out for an American telephone company to determine how many digits a telephone number should have) suggested that individuals can hold approximately seven (plus or minus two) items in WM. Het called this the magical number seven, plus minus two. More recent studies suggest that the true number of items that can be held in our WM is, on average, closer to four discrete units, particularly when rehearsal strategies are restricted (Cowan, 2001). Also, information in our WM typically decays within 15 to 30 seconds if not actively maintained (Brown, 1958; Peterson & Peterson, 1959). That is, it disappears in less than half a minute! These constraints raise an important question, namely: How do we manage to process and retain complex or lengthy information—such as sentences, problem-solving steps, or visual configurations—when our WM capacity is so limited?
The answer to this question is that we have two interrelated cognitive mechanisms that help us address this challenge: chunking and schemas. Chunking refers to the process of grouping individual elements into larger, meaningful units (chunks; Miller, 1956), while schemas are structured representations of knowledge that are stored in our long-term memory (Bartlett, 1932). Together, these two mechanisms reduce the effective load on our WM, allowing us to process more information than would otherwise be possible within its natural limitations.
A foundational framework for understanding WM is the multi-component model proposed by Baddeley and Hitch in 1974, which was later expanded by Baddeley in 2000. This model proposes that WM consists of a central executive (disputed by John Sweller[1]) and two domain-specific storage systems: the phonological loop for verbal and auditory information and the visuo-spatial sketchpad for visual and spatial data. A third component, the episodic buffer, was later introduced to explain how WM integrates information across modalities and links it with long-term memory. The episodic buffer allows us to create unified “episodes” by combining verbal, visual, and semantic information from both WM and long-term memory (LTM). This integrative function is particularly relevant when considering the role of schemas and chunking.
Chunking is one of the most effective strategies for bypassing WM’s limited capacity. When discrete bits of information are grouped into larger units, they are treated as a single item in WM. For example, the number sequence 1-9-4-5-2-0-2-4 can be encoded as two familiar years—1945 and 2024—thus reducing the number of items to be remembered from eight to two. This process relies heavily on prior knowledge since a chunk is only meaningful if the individual can relate it to a familiar concept stored in LTM (Gobet et al., 2001). In this sense, chunking isn’t a standalone process but one closely dependent on activatingrelevant existing schemas in our LTM.
Schemas are cognitive structures that represent organized knowledge about objects, situations, or events (Bartlett, 1932; Rumelhart, 1980). They function as mental frameworks that allow us to efficiently interpret and structure incoming information.
A helpful example of schema use can be found in how we understand and categorise dogs. For many people, the schema of a dog as a house pet includes expectations such as that dogs live with humans, are generally friendly, bark, wag their tails, can have fleas, and enjoy play. They need daily care—food, water, walks, and attention—and form close bonds with their owners (some say caretakers). This schema allows us to instantly interpret what we see when a dog chases a stick or curls up on a couch. We don’t perceive these actions as isolated facts but as coherent parts of the familiar dog-as-pet framework.
However, our mental framework for dogs is not limited to pets. We also understand dogs within the broader category of mammals. This schema includes knowledge such as: dogs have fur, are warm-blooded (endothermic), give birth to live young, have mammary glands to nurse their offspring, have a backbone (are vertebrates), have three middle ear bones, have lungs for breathing… The mammal schema links dogs to other animals like cats, whales, bats, mice, or humans. It also helps us reason about biological similarities with other mammals, such as how dogs grow, breathe, have puppies, or need sleep.
At an even more general level, we recognise dogs as animals. This schema captures shared features of all animals, such as that they are multicellular, move, consume food as they are what is known as heterotrophic (cannot make their own food), respond to stimuli, and reproduce. It helps us distinguish living beings from non-living things and informs our reasoning in contexts such as ecology, ethics, or biology.
What’s important is that these schemas are nested. The dog-as-pet schema fits within the mammal schema, which in turn fits within the animal schema. Depending on the situation—playing at the park, reading a biology textbook, or visiting a zoo—different levels of the schema hierarchy are activated. In daily life, we may use the pet schema. In a scientific context, the mammal or animal schema becomes more relevant. This nested structure allows for both quick interpretation of familiar experiences and flexible reasoning in unfamiliar ones.
In unfamiliar situations, for example, schemas allow us to make educated guesses. If we encounter a breed of dog we’ve never seen, we can usually still identify it as a dog because the animal fits the familiar dog schema. Similarly, when encountering a new animal with fur, a backbone, and live birth, we classify it as a mammal based on the broader schema. In this way, schemas not only support memory and recognition but also enable generalisation and learning.
Schemas also develop with experience. A child may start with a basic schema—“dogs are furry and bark”—but over time this becomes more detailed. They might learn that some dogs are trained to assist people, that breeds differ greatly, or that dogs share anatomical and behavioural traits with other mammals. The schema grows more elaborate (it gets both broader and deeper) and more useful as new experiences and knowledge are incorporated. Jearn Piaget spoke of assimilation and accommodation. New information is assimilated into already existing schemas, and already existing schemas are changed (accommodated) as new and sometimes incongruous information (knowledge) is acquired.
In addition to supporting capacity, schemas also contribute to the duration of memory retention. Structured, coherent information is easier to rehearse and maintain in WM, and if some parts are forgotten, schemas can help us to reconstruct missing elements by filling in likely details based on past experience (Alba & Hasher, 1983). In this way, schemas enhance the stability and resilience of working memory contents.
The interaction between WM and LTM is central to these processes. WM constantly draws upon LTM to make sense of new inputs, organise them into meaningful structures, and guide their processing. The episodic buffer plays a key role in this interaction by binding current WM contents with information retrieved from LTM (Baddeley, 2000). For example, in narrative comprehension, individuals rely on story schemas to chunk and integrate events into a coherent structure, linking current input with previously acquired knowledge.
There are many real-world applications of these mechanisms. In language processing, people rely on stored grammatical and semantic schemas to chunk phrases and sentences efficiently. In number recall, such as remembering a phone number or credit card number, people tend to group digits into segments based on familiar patterns, such as area codes or key dates. These examples illustrate how prior knowledge structures enable more effective chunking, enhancing WM performance.
In summary, the limitations of working memory—its restricted capacity and duration—can be substantially mitigated by the use of chunking and schemas. Chunking reduces the number of items we need to maintain by grouping them into larger, meaningful units, while schemas provide the structure and interpretive framework necessary to form those units. These mechanisms depend on continuous interaction between WM and LTM, illustrating how prior experience and knowledge scaffold ongoing cognitive processes. Ultimately, the synergy between chunking and schemas allows individuals to function effectively despite the narrow constraints of WM and highlights the role of expertise, practice, and education in expanding cognitive efficiency.
References
Alba, J. W., & Hasher, L. (1983). Is memory schematic? Psychological Bulletin, 93(2), 203–231. https://doi.org/10.1037/0033-2909.93.2.203
Baddeley, A. D. (2000). The episodic buffer: A new component of working memory? Trends in Cognitive Sciences, 4(11), 417–423. https://doi.org/10.1016/S1364-6613(00)01538-2
Baddeley, A. D., & Hitch, G. (1974). Working memory. In G. H. Bower (Ed.), The psychology of learning and motivation (Vol. 8, pp. 47–89). Academic Press.
Bartlett, F. C. (1932). Remembering: A study in experimental and social psychology. Cambridge University Press.
Brown, J. (1958). Some tests of the decay theory of immediate memory. Quarterly Journal of Experimental Psychology, 10(1), 12–21. https://doi.org/10.1080/17470215808416249
Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behavioral and Brain Sciences, 24(1), 87–114. https://doi.org/10.1017/S0140525X01003922
Gobet, F., Lane, P. C. R., Croker, S., Cheng, P. C.-H., Jones, G., Oliver, I., & Pine, J. M. (2001). Chunking mechanisms in human learning. Trends in Cognitive Sciences, 5(6), 236–243. https://doi.org/10.1016/S1364-6613(00)01662-4
Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63(2), 81–97. https://doi.org/10.1037/h0043158
Peterson, L. R., & Peterson, M. J. (1959). Short-term retention of individual verbal items. Journal of Experimental Psychology, 58(3), 193–198. https://doi.org/10.1037/h0049234
Rumelhart, D. E. (1980). Schemata: The building blocks of cognition. In R. J. Spiro, B. C. Bruce, & W. F. Brewer (Eds.), Theoretical issues in reading comprehension (pp. 33–58). Lawrence Erlbaum.
Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12(2), 257–285. https://doi.org/10.1207/s15516709cog1202_4
[1] John Sweller challenged the concept of a central executive in working memory by arguing that its role is effectively fulfilled by schemas stored in LTM, making the central executive redundant.
One of Sweller’s key objections to the idea of a central executive is what is known as the infinite regression problem. The central executive is often conceived as a “mini-mind” responsible for controlling and directing cognitive processes. However, this raises the next question: what controls the mini-mind? Such reasoning leads to a paradoxical infinite regress, where each controlling entity (each mini-mind) requires yet another controller (another mini-mind), ultimately undermining the coherence of the model. Sweller saw this as a fundamental flaw in the concept of a central executive. For Sweller, schemas in LTM serve as the true directors of cognition. In his view, schemas provide the cognitive framework that governs WM processes without requiring a separate executive module.
