Memory

Memory is the set of processes by which experience leaves a trace that later shapes thought and behaviour.

10 frameworks, from 1750 to today · Cognitive Science · Social Sciences

still a working view ended · each leads to its entry below

Overview

Memory is the set of processes by which experience leaves a trace that later shapes thought and behaviour1. In cognitive science the subject is studied as a system: what gets encoded, how it is stored, how it is retrieved, and why it so often fails1. The stakes are large. Memory is what makes a person continuous across time, what schooling and testimony and clinical care depend on, and what any artificial system that learns from experience must have. The history of the field is a history of models of what that system is made of—and of arguments about whether it is one thing or many.

Associationism and the First Experimental Study of Memory

The oldest framework is associationism: the idea that memories are links between ideas, and that the mind works by connecting them. In the British empiricist tradition, David Hume argued in the eighteenth century that thought follows principles of resemblance, contiguity in time and place, and cause and effect. The tradition's central claim—that what is learned together comes to be recalled together—set the agenda for a century of experiment.

The decisive turn to measurement came with Hermann Ebbinghaus, working alone in the 1880s23. To study memory uncontaminated by meaning, he invented nonsense syllables (consonant–vowel–consonant strings such as "DAX") and memorised lists of them, then tested himself after delays ranging from minutes to a month. The result was the forgetting curve: retention drops steeply at first and then levels off45. He also found that relearning a list took fewer trials than learning it the first time, a measure he called savings, and that distributed practice beat massed practice467. Ebbinghaus founded the experimental study of verbal learning, and his methods—controlled lists, measured retention, savings as a sensitive index—became the template for the field34.

Behaviourism, Interference, and the Problem of the Trace

In the early twentieth century, behaviourism pushed memory study toward observable learning. The dominant account of forgetting became interference theory: memories are not so much erased as crowded out by competing material8. Two forms were distinguished—proactive interference, in which earlier learning disrupts later recall, and retroactive interference, in which later learning disrupts earlier18. The paired-associate experiments of the mid-twentieth century, in which subjects learned pairs like "table–chair" and were tested on them, produced a large body of evidence for interference and for the shape of learning curves.

The behaviourist programme, however, left a gap. It could describe the conditions under which responses changed, but it had no account of the internal representation—the memory trace, or engram—that must carry information from one occasion to the next. That gap, together with the wartime and post-war demand for practical answers about attention, vigilance, and training, helped open the way to the cognitive revolution.

The Multistore Model and the Return of Internal Structure

The pivotal synthesis was the multistore model of Atkinson and Shiffrin (1968)19. Memory was divided into structural components: a sensory register holding raw input very briefly, a short-term store of limited capacity, and a long-term store of essentially unlimited capacity. Control processes—rehearsal, coding, retrieval strategies—moved information between them. The model made memory a flow diagram, and it made testable predictions: that short-term capacity is small (around seven items, following George Miller's 1956 paper on "the magical number seven"), that rehearsal protects items from decay, and that the serial position of an item in a list predicts recall, with primacy reflecting long-term storage and recency reflecting short-term storage.

The model's weakness was that short-term storage behaved less like a passive box than the diagram implied. Baddeley and Hitch (1974) proposed working memory instead: a limited-capacity system for holding and manipulating information during ongoing tasks, composed of a phonological loop for verbal material, a visuospatial sketchpad for images and locations, and a central executive that allocates attention110. Later work added an episodic buffer integrating information across modalities110. The change mattered because it made memory a part of thinking rather than a stage before it—the same resource used to hold a phone number and to reason about it.

Levels of Processing and the Constructive Turn

A different challenge came from Craik and Lockhart (1972), whose levels-of-processing framework argued that what determines retention is not which store an item enters but how deeply it is processed11. Shallow processing—noticing a word's font or its sound—yields poor memory; deep processing—judging whether a word fits a sentence or relates to oneself—yields good memory11. The framework shifted attention from structure to encoding and produced robust findings, such as the self-reference effect, in which material judged in relation to oneself is remembered better than material judged in other ways.

In parallel, Bartlett's earlier work (1932) on the constructive and reconstructive nature of memory gained renewed influence. Bartlett had subjects reproduce stories such as "The War of the Ghosts," a Native American tale unfamiliar to his English readers, and found that recall drifted toward familiar cultural schemas: details were dropped, rationalised, or transformed. Memory, on this view, is not a recording but a reconstruction, shaped by prior knowledge and expectation. The idea was later extended by research on schemas, on false memories, and on the misinformation effect, in which misleading post-event information alters what people report having seen.

Multiple Memory Systems

The most consequential development of the late twentieth century was the recognition that "memory" names several systems with different properties and different brain bases. The clinical evidence was decisive. The patient known as H.M., who underwent bilateral removal of the medial temporal lobes in the 1950s, could not form new long-term memories for facts and events, yet he could learn new motor skills and showed priming—improved performance on previously seen material without recollection of it. The dissociation implied that declarative memory (conscious recollection of facts and events) and nondeclarative memory (skills, habits, priming, conditioning) are separable112.

The multiple memory systems framework, articulated influentially by Squire and colleagues, organised memory into declarative and nondeclarative branches, with working memory as a short-lived store and procedural memory as the home of skills. Endel Tulving's distinction between episodic memory (memory for specific events, tied to a time and place) and semantic memory (general knowledge) cut across the declarative branch and gave the field a vocabulary that is still standard213.

Consolidation and the Hippocampus

The question of how memories become durable was transformed by hippocampal consolidation theory. The standard account holds that new declarative memories are initially dependent on the hippocampus and related medial temporal structures, and that over time—often during sleep—they are reorganised into neocortical networks, becoming independent of the hippocampus1415. The theory explained why H.M. could recall old memories but not form new ones, and it motivated decades of work on replay, in which hippocampal activity patterns observed during learning recur during rest and sleep. The theory is not settled: some evidence supports a more permanent role for the hippocampus in detailed episodic recollection, and the debate over the scope of consolidation remains active.

Connectionism and Distributed Representation

The final major strand is connectionism: the modelling of memory as patterns of activation across networks of simple units rather than as discrete stored items16. In distributed models, a memory is a pattern of weights across many units, and retrieval is the reconstruction of a pattern from a partial cue17. This gave a natural account of generalisation, of graceful degradation after damage, and of content-addressable retrieval—the ability to recover a whole from a part. It also reframed interference and forgetting as properties of overlapping representations rather than of separate slots. Connectionist models of memory, from the 1980s onward, connected the psychology of memory to computational neuroscience and to modern machine learning, and they remain a major framework for thinking about how memory could be implemented in a brain.

The Present

Today memory research is plural. The multistore and working-memory models remain the backbone of experimental work on short-term retention and executive control. Multiple memory systems and consolidation theory organise the neuropsychology of memory and its clinical assessment. Levels-of-processing and constructive accounts shape research on encoding, schemas, and false memory. Connectionist and distributed models, increasingly joined to large-scale neural recording and to machine learning, supply the computational vocabulary. The open questions are substantial: how consolidation and reconsolidation actually work at the synaptic and systems levels, how episodic and semantic memory interact, how memory is reconstructed in the moment of retrieval, and how far the lessons of human memory transfer to artificial systems that learn from experience.

References

  1. Memory (introduction, Interference, Multi-store model, Working memory), Wikipedia
  2. Explicit memory (History), Wikipedia
  3. Hermann Ebbinghaus, Wikipedia
  4. Forgetting curve (introduction, History), Wikipedia
  5. Forgetting (History), Wikipedia
  6. Recall (memory) (History), Wikipedia
  7. Spacing effect, Wikipedia
  8. Interference theory, Wikipedia
  9. Atkinson–Shiffrin memory model, Wikipedia
  10. Baddeley's model of working memory, Wikipedia
  11. Levels of processing model, Wikipedia
  12. Procedural memory (History), Wikipedia
  13. Episodic memory, Wikipedia
  14. Memory consolidation, Wikipedia
  15. Hippocampus, Wikipedia
  16. Connectionism, Wikipedia
  17. Hopfield network, Wikipedia

The frameworks of Memory

Each framework is a way of working the field has used: when it held, and the question it answers.

  1. 1750-1950

    Associationism

    How do ideas become linked in the mind?

    Fly to its tower in Nousfarer

  2. 1885-1960
  3. 1900-1970

    Interference theory

    Why does old learning disrupt new recall?

    Fly to its tower in Nousfarer

  4. 1932-Present

    Reconstructive memory

    Is remembering a recording or a reconstruction?

    Fly to its tower in Nousfarer

  5. 1957-Present
  6. 1957-Present
  7. 1968-1990

    Atkinson–Shiffrin memory model

    What stages does information pass through?

    Fly to its tower in Nousfarer

  8. 1972-Present

    Levels of processing model

    What makes some experiences stick?

    About Levels of processing model
    What determines whether we remember something: how long we hold it in mind, or how deeply we think about it? In the early 1970s, Fergus Craik and Robert Lockhart proposed a radical answer that shifted the focus from memory stores to the quality of mental processing. Their Levels of Processing framework argued that memory is not a set of separate boxes but a continuum of analysis, from shallow sensory features to deep semantic meaning. The deeper the processing, the more durable the memory trace. This simple idea challenged the dominant Multi-Store Model and reshaped how cognitive scientists think about encoding, rehearsal, and forgetting.
    At the heart of the framework is a single organizing distinction: shallow versus deep processing. Shallow processing involves attending to surface features—the shape of a word, its sound, or its typography. Deep processing engages meaning, associations, and personal relevance. The framework's core curriculum begins with this continuum and its consequences. Key concepts include: Depth of Processing, the central claim that deeper encoding leads to stronger, longer-lasting memories; Elaboration, the idea that richer, more connected processing at a given depth further strengthens the trace; and the Distinction between Maintenance Rehearsal (rote repetition that keeps information active at a shallow level) and Elaborative Rehearsal (thinking about meaning and connections, which deepens encoding). The framework also emphasizes the Transfer-Appropriate Processing principle: memory performance depends on the match between how information was encoded and how it is later retrieved. Other core concepts include the Self-Reference Effect (processing information in relation to oneself produces especially deep encoding), the Generation Effect (actively producing information leads to better memory than passively reading it), and the distinction between Incidental Learning (learning without intention to remember) and Intentional Learning (learning with the goal of later recall). The framework also addresses Encoding Specificity (the idea that retrieval cues are most effective when they match the original encoding context) and the notion of Distinctiveness (unique or unusual items are better remembered because they stand out). Finally, the framework's emphasis on the Role of Meaning in memory challenged the field to move beyond simple rehearsal counts and structural models.
    The Levels of Processing framework emerged in the early 1970s as a direct challenge to the Multi-Store Model, which dominated memory research after 1968. The Multi-Store Model, proposed by Richard Atkinson and Richard Shiffrin, divided memory into sensory, short-term, and long-term stores, with rehearsal as the mechanism for transferring information between them. Craik and Lockhart, both at the University of Toronto, argued that this structural approach was misleading: what matters is not how many stores information passes through, but the depth at which it is processed. Their seminal 1972 paper, "Levels of Processing: A Framework for Memory Research," published in the Journal of Verbal Learning and Verbal Behavior, laid out the core ideas. In a classic experiment, participants were asked to make different judgments about words: some judged the case (shallow), some judged the rhyme (intermediate), and some judged whether the word fit into a sentence (deep). Later memory tests showed that deeper judgments produced far better recall, even though participants did not expect a memory test. This finding was a powerful demonstration that the nature of encoding, not rehearsal time, determines memory strength. The framework was further developed by Craik and Endel Tulving in 1975, who showed that elaborative processing at a deep level—thinking about multiple aspects of meaning—enhances memory even more. The framework's influence peaked in the 1970s and 1980s, but it coexisted with and was eventually supplemented by Memory Systems Theory, which distinguished multiple memory systems (e.g., episodic, semantic, procedural) that operate at different levels of processing. The Levels of Processing framework remains a core part of the curriculum because it foregrounds the active, constructive role of the learner and the importance of meaning in memory. It also inspired practical applications in education, such as encouraging elaborative rehearsal and self-referencing to improve learning.
    The 12 concepts its tower teaches
    • Depth of Processing
    • Distinctiveness
    • Elaboration
    • Elaborative Rehearsal
    • Encoding Specificity
    • Generation Effect
    • Incidental Learning
    • Intentional Learning
    • Maintenance Rehearsal
    • Role of Meaning
    • Self-Reference Effect
    • Transfer-Appropriate Processing

    Fly to its tower in Nousfarer

  9. 1974-Present

    Baddeley's model of working memory

    How do we hold information while thinking?

    Fly to its tower in Nousfarer

  10. 1980-Present

    Connectionism

    How can memory emerge from simple units?

    Fly to its tower in Nousfarer

How the ideas connect

Each arrow runs from an earlier idea to the later one it shaped. The list below says what passed between them.

AssociationismExperimental verbal learningInterference theoryReconstructive memoryHippocampal consolidation theoryMultiple memory systemsAtkinson–Shiffrin memory modelLevels of processing modelBaddeley's model of working memoryConnectionism

Beyond the field

Where Memory's ideas came from in other fields, and what they led to there: the strongest 12 of 20 links.

A city in Nousfarer: a red-carpeted stair climbing terraces to a crowned tower on an acropolis, more towers round it, forest and snowy mountains beyond.
A city in Nousfarer, the game played in the browser.

Memory is a planet in Nousfarer

Its city has a tower for each of these 10 frameworks. Land, walk in, and climb one concept at a time.

Land on Memory

Related fields

Fields its ideas are linked to

More in Cognitive Science

See where Memory sits on the Map of Human Knowledge